Saimm 201410 oct

VOLUME 114

NO. 10

OCTOBER 2014

The MSA

explor and e solutions t The MSA Group i and grow its Mini Interested parties

MSA Johannesb +27 (0)11 880 42 info@msagroup www.msagroups

The Southern African Institute of Mining and Metallurgy OFFICE BEARERS AND COUNCIL FOR THE 2014/2015 SESSION Honorary President Mike Teke President, Chamber of Mines of South Africa Honorary Vice-Presidents Ngoako Ramatlhodi Minister of Mineral Resources, South Africa Rob Davies Minister of Trade and Industry, South Africa Naledi Pandor Minister of Science and Technology, South Africa President J.L. Porter President Elect R.T. Jones Vice-Presidents C. Musingwini S. Ndlovu Immediate Past President M. Dworzanowski Honorary Treasurer C. Musingwini Ordinary Members on Council V.G. Duke M.F. Handley A.S. Macfarlane M. Motuku M. Mthenjane D.D. Munro G. Njowa

T. Pegram S. Rupprecht N. Searle A.G. Smith M.H. Solomon D. Tudor D.J. van Niekerk

Past Presidents Serving on Council N.A. Barcza R.D. Beck J.A. Cruise J.R. Dixon F.M.G. Egerton G.V.R. Landman R.P. Mohring

J.C. Ngoma S.J. Ramokgopa M.H. Rogers G.L. Smith J.N. van der Merwe W.H. van Niekerk

Branch Chairmen DRC

S. Maleba

Johannesburg

I. Ashmole

Namibia

N. Namate

Pretoria

N. Naude

Western Cape

C. Dorfling

Zambia

H. Zimba

Zimbabwe

E. Matinde

Zululand

C. Mienie

Corresponding Members of Council Australia: I.J. Corrans, R.J. Dippenaar, A. Croll, C. Workman-Davies Austria: H. Wagner Botswana: S.D. Williams Brazil: F.M.C. da Cruz Vieira China: R. Oppermann United Kingdom: J.J.L. Cilliers, N.A. Barcza, H. Potgieter USA: J-M.M. Rendu, P.C. Pistorius Zambia: J.A. van Huyssteen

▲

ii

OCTOBER 2014

PAST PRESIDENTS *Deceased * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

W. Bettel (1894–1895) A.F. Crosse (1895–1896) W.R. Feldtmann (1896–1897) C. Butters (1897–1898) J. Loevy (1898–1899) J.R. Williams (1899–1903) S.H. Pearce (1903–1904) W.A. Caldecott (1904–1905) W. Cullen (1905–1906) E.H. Johnson (1906–1907) J. Yates (1907–1908) R.G. Bevington (1908–1909) A. McA. Johnston (1909–1910) J. Moir (1910–1911) C.B. Saner (1911–1912) W.R. Dowling (1912–1913) A. Richardson (1913–1914) G.H. Stanley (1914–1915) J.E. Thomas (1915–1916) J.A. Wilkinson (1916–1917) G. Hildick-Smith (1917–1918) H.S. Meyer (1918–1919) J. Gray (1919–1920) J. Chilton (1920–1921) F. Wartenweiler (1921–1922) G.A. Watermeyer (1922–1923) F.W. Watson (1923–1924) C.J. Gray (1924–1925) H.A. White (1925–1926) H.R. Adam (1926–1927) Sir Robert Kotze (1927–1928) J.A. Woodburn (1928–1929) H. Pirow (1929–1930) J. Henderson (1930–1931) A. King (1931–1932) V. Nimmo-Dewar (1932–1933) P.N. Lategan (1933–1934) E.C. Ranson (1934–1935) R.A. Flugge-De-Smidt (1935–1936) T.K. Prentice (1936–1937) R.S.G. Stokes (1937–1938) P.E. Hall (1938–1939) E.H.A. Joseph (1939–1940) J.H. Dobson (1940–1941) Theo Meyer (1941–1942) John V. Muller (1942–1943) C. Biccard Jeppe (1943–1944) P.J. Louis Bok (1944–1945) J.T. McIntyre (1945–1946) M. Falcon (1946–1947) A. Clemens (1947–1948) F.G. Hill (1948–1949) O.A.E. Jackson (1949–1950) W.E. Gooday (1950–1951) C.J. Irving (1951–1952) D.D. Stitt (1952–1953) M.C.G. Meyer (1953–1954)

* * * * * * * * * * * * * * * * * * * * * * * *

*

*

*

*

*

L.A. Bushell (1954–1955) H. Britten (1955–1956) Wm. Bleloch (1956–1957) H. Simon (1957–1958) M. Barcza (1958–1959) R.J. Adamson (1959–1960) W.S. Findlay (1960–1961) D.G. Maxwell (1961–1962) J. de V. Lambrechts (1962–1963) J.F. Reid (1963–1964) D.M. Jamieson (1964–1965) H.E. Cross (1965–1966) D. Gordon Jones (1966–1967) P. Lambooy (1967–1968) R.C.J. Goode (1968–1969) J.K.E. Douglas (1969–1970) V.C. Robinson (1970–1971) D.D. Howat (1971–1972) J.P. Hugo (1972–1973) P.W.J. van Rensburg (1973–1974) R.P. Plewman (1974–1975) R.E. Robinson (1975–1976) M.D.G. Salamon (1976–1977) P.A. Von Wielligh (1977–1978) M.G. Atmore (1978–1979) D.A. Viljoen (1979–1980) P.R. Jochens (1980–1981) G.Y. Nisbet (1981–1982) A.N. Brown (1982–1983) R.P. King (1983–1984) J.D. Austin (1984–1985) H.E. James (1985–1986) H. Wagner (1986–1987) B.C. Alberts (1987–1988) C.E. Fivaz (1988–1989) O.K.H. Steffen (1989–1990) H.G. Mosenthal (1990–1991) R.D. Beck (1991–1992) J.P. Hoffman (1992–1993) H. Scott-Russell (1993–1994) J.A. Cruise (1994–1995) D.A.J. Ross-Watt (1995–1996) N.A. Barcza (1996–1997) R.P. Mohring (1997–1998) J.R. Dixon (1998–1999) M.H. Rogers (1999–2000) L.A. Cramer (2000–2001) A.A.B. Douglas (2001–2002) S.J. Ramokgopa (2002-2003) T.R. Stacey (2003–2004) F.M.G. Egerton (2004–2005) W.H. van Niekerk (2005–2006) R.P.H. Willis (2006–2007) R.G.B. Pickering (2007–2008) A.M. Garbers-Craig (2008–2009) J.C. Ngoma (2009–2010) G.V.R. Landman (2010–2011) J.N. van der Merwe (2011–2012) G.L. Smith (2012–2013)

Honorary Legal Advisers Van Hulsteyns Attorneys Auditors Messrs R.H. Kitching Secretaries The Southern African Institute of Mining and Metallurgy Fifth Floor, Chamber of Mines Building 5 Hollard Street, Johannesburg 2001 P.O. Box 61127, Marshalltown 2107 Telephone (011) 834-1273/7 Fax (011) 838-5923 or (011) 833-8156 E-mail: journal@saimm.co.za

The Journal of The Southern African Institute of Mining and Metallurgy

Editorial Board R.D. Beck J. Beukes P. den Hoed M. Dworzanowski M.F. Handley R.T. Jones W.C. Joughin J.A. Luckmann C. Musingwini R.E. Robinson T.R. Stacey R.J. Stewart

VOLUME 114

NO. 10

OCTOBER 2014

Contents

by W. Joughin

D. Tudor

Typeset and Published by The Southern African Institute of Mining and Metallurgy P.O. Box 61127 Marshalltown 2107 Telephone (011) 834-1273/7 Fax (011) 838-5923 E-mail: journal@saimm.co.za

Printed by Camera Press, Johannesburg

Advertising Representative Barbara Spence Avenue Advertising Telephone (011) 463-7940 E-mail: barbara@avenue.co.za The Secretariat The Southern African Institute of Mining and Metallurgy ISSN 2225-6253

S E S S

Institute of Mining and Metallurgy. All rights reserved. Multiple copying of the contents of this publication or parts thereof without permission is in breach of copyright, but permission is hereby given for the copying of titles and abstracts of papers and names of authors. Permission to copy illustrations and short extracts from the text of individual contributions is usually given upon written application to the Institute, provided that the source (and where appropriate, the copyright) is acknowledged. Apart from any fair dealing for the purposes of review or criticism under The Copyright Act no. 98, 1978, Section 12, of the Republic of South Africa, a single copy of an article may be supplied by a library for the purposes of research or private study. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means without the prior permission of the publishers. Multiple copying of the contents of the publication without permission is always illegal.

U.S. Copyright Law applicable to users In the U.S.A. The appearance of the statement of copyright at the bottom of the first page of an article appearing in this journal indicates that the copyright holder consents to the making of copies of the article for personal or internal use. This consent is given on condition that the copier pays the stated fee for each copy of a paper beyond that permitted by Section 107 or 108 of the U.S. Copyright Law. The fee is to be paid through the Copyright Clearance Center, Inc., Operations Center, P.O. Box 765, Schenectady, New York 12301, U.S.A. This consent does not extend to other kinds of copying, such as copying for general distribution, for advertising or promotional purposes, for creating new collective works, or for resale.

............................................................

iv

President’s Corner

by J.L. Porter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

v

6th Southern African Rock Engineering Symposium Extending empirical evidence through numerical modelling in rock engineering design

by G.S. Esterhuizen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Time-dependent tensile strengths of Bushveld Complex rocks and implications for rock failure around mining excavations

by D. Nyungu and T.R. Stacey. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts

by B.G. Tarasov. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Unique fall-of-ground prevention strategy implemented at Two Rivers Platinum Mine

by A. Esterhuizen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

In situ monitoring of primary roofbolts at underground coal mines in the USA

by A.J.S. Spearing and A. Hyett. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Pillar behaviour and seismicity in platinum mines

by S.M. Spottiswoode and M. Drummond . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Management of the Nkomati Mine crusher slope failure

by R. Armstrong and K. Moletsane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Grid-based analysis of seismic data

by J. Wesseloo, K. Woodward, and J. Pereira . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Evaluation of the spatial variation of b-value

by J. Wesseloo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Testing tendon support units under a combination loading scenario

by N.L. Ayres and L.J. Gardner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Estimation of future ground vibration levels in Malmberget town due to mining-induced seismic activity

by T. Wettainen and J. Martinsson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

755

765 773 785 791 801 811 815 823 829

835

General Papers Outsourcing in the mining industry: decision-making framework and critical success factors

by C.J.H. Steenkamp and E. van der Lingen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Focal depths of South African earthquakes and mine events

by M.B.C. Brandt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Design and positive financial impact of crush pillars on mechanized deep-level mining at South Deep Gold Mine

by B.P. Watson, W. Pretorius, P. Mpunzi, M. du Plooy, K. Matthysen, and J.S. Kuijpers . . . . . .

The application of geophysics in South African coal mining and exploration

by M. van Schoor and C.J.S. Fourie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

International Advisory Board

845 855

863 875

VOLUME 114

NO. 10

OCTOBER 2014

R. Dimitrakopoulos, McGill University, Canada D. Dreisinger, University of British Columbia, Canada E. Esterhuizen, NIOSH Research Organization, USA H. Mitri, McGill University, Canada M.J. Nicol, Murdoch University, Australia H. Potgieter, Manchester Metropolitan University, United Kingdom E. Topal, Curtin University, Australia

The Journal of The Southern African Institute of Mining and Metallurgy

OCTOBER 2014

iii

â–˛

Editorial Consultant

Journal Comment

â–˛

iv

OCTOBER 2014

The Journal of The Southern African Institute of Mining and Metallurgy

The Journal of The Southern African Institute of Mining and Metallurgy

OCTOBER 2014

v

▲

tʼs iden s e r P er Corn

PAPERS IN THIS EDITION These papers have been refereed and edited according to internationally accepted standards and are accredited for rating purposes by the South African Department of Higher Education and Training

6th Southern African Rock Engineering Symposium 2014

by G.S. Esterhuizen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 755 The use of numerical models to answer questions about specific aspects of excavation stability that existing empirical models were unable to provide is illustrated in two case studies. It is shown that the combined application of numerical and empirical models can help to improve understanding of causes of stability and instability in excavations, resulting in efficient design and increased safety.

by D. Nyungu and T.R. Stacey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 765 Numerical modelling has shown that large zones of extension strain can occur around excavations in the Bushveld Complex, and that the magnitudes of the extension strain can substantially exceed the critical values determined from laboratory testing. These zones may thus be prone to time-dependent spalling, and perhaps more significant failure.

by B.G. Tarasov . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773 Physical and mathematical models have shown that shear ruptures can propagate through a highly confined intact rock mass at shear stresses significantly less than the frictional strength. The failure process is inevitably spontaneous and violent. The fan mechanism allows a novel point of view for understanding the nature of spontaneous failure processes, including shear rupture rockbursts.

by A. Esterhuizen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 785 Since 2005, Two Rivers Platinum Mine has actively monitored and controlled ground conditions on a daily basis by making use of borehole cameras and day-to-day observations of hangingwall conditions. The strategy has greatly reduced the mine’s fall- of-ground frequency and size, while effectively controlling support costs.

by A.J.S. Spearing and A. Hyett . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 791 A study was conducted to assess the performance of primary roofbolts in three underground coal mines in the USA. The results showed that there was no evidence to indicate a difference in performance of active primary roof bolts compared with passive primary roofbolts.

by S.M. Spottiswoode and M. Drummond. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 801 Violent failure of crush pillars is said to be the main cause of seismicity associated with mining of the Merensky Reef.

This paper reports the first results from a new suite of programs to model pillars in platinum mines. The ultimate aim is to provide a tool for on-mine rock engineers to interpret current and planned mining geometry by extrapolating comparisons of historical modelled and observed seismicity into the future for better and safer mining.

by R. Armstrong and K. Moletsane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 811 This paper describes the events leading up to a minor slope failure above the crusher at Nkomati nickel mine.

Real-time monitoring was deployed, and the monitoring data determined a management plan for the failure that resulted in minimal shutdowns of the primary crusher.

by J. Wesseloo, K. Woodward, and J. Pereira . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 815 A grid- based interpretation of seismic data and some examples of results obtained with the method are presented.

Grid- based interpretation allows the spatial variation of seismic source parameters to be evaluated without predetermined analysis volumes. As such, it reduces interpretation bias.

These papers will be available on the SAIMM website

http://www.saimm.co.za

PAPERS IN THIS EDITION These papers have been refereed and edited according to internationally accepted standards and are accredited for rating purposes by the South African Department of Higher Education and Training

6th Southern African Rock Engineering Symposium 2014 Evaluation of the spatial variation of b-value

by J. Wesseloo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The interpretation of b-value of the Gutenberg-Richter relationship is important for both the general interpretation of the mechanism of rock mass response and seismic hazard assessment. This paper discusses the algorithm for spatially sub-sampling the data as well as the algorithm for obtaining the magnitude of completeness mmin and b-value for every spatial sub-sample.

Testing tendon support units under a combination loading scenario

by N.L. Ayres and L.J. Gardner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Tendon support units were tested at different installation angles to establish the performance, mechanical behaviour, and load capacity under different combinations of tensile and shear forces. The results are aimed at improving understanding of how tendons perform under these conditions.

Estimation of future ground vibration levels in Malmberget town due to mining-induced seismic activity

by T. Wettainen and J. Martinsson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

This paper describes investigations into how seismicity will change as production increases at the Malmberget underground iron-ore mine in northern Sweden, and what possible measures could be taken to reduce inconvenience to the residents of the town, which partly overlies the orebodies. Relations between historical seismic events and measured ground vibrations in the town were established, and future ground vibrations caused by expected seismic events were estimated using a probabilistic approach.

823

829

835

General Papers Outsourcing in the mining industry: decision-making framework and critical success factors

by C.J.H. Steenkamp and E. van der Lingen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

A study was conducted to determine whether mining is truly a core competency for a mid-tier commodity specialist mining company. A decision-making framework for mining operations outsourcing was developed and the critical success factors were determined. It is shown that the most important tools at the disposal of a mine owner’s team to manage a contract miner are the social and output control mechanisms.

Focal depths of South African earthquakes and mine events

by M.B.C. Brandt. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The focal depths of 15 tectonic earthquakes and 9 mine-related events were determined using data recorded by the South African National Seismograph Network. Focal depths were estimated for nine stations by visually comparing synthetic waveform phases with recorded waveforms. The results confirm the assumption that focal depths of South African earthquakes and mine-related events are shallow — within the upper third (0 km to 10 km) of the crust.

Design and positive financial impact of crush pillars on mechanized deep-level mining at South Deep Gold Mine

by B.P. Watson, W. Pretorius, P. Mpunzi, M. du Plooy, K. Matthysen, and J.S. Kuijpers. . . . . . . . . . . . . . . . . . . . . . . . . .

Crush pillars have been incorporated into a mechanized, low-profile trackless system at South Deep Gold Mine.

The introduction of these pillars has improved the rockmass conditions because of the active nature of the support, compared to the previous passive backfill method. Importantly, the pillars have increased mining efficiencies and improved face availability.

The application of geophysics in South African coal mining and exploration

by M. van Schoor and C.J.S. Fourie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Geophysics can play a significant role in addressing a wide range of problems in coal mining and exploration.

This paper provides a brief overview of a textbook compiled by Coaltech to guide the application of geophysics to coal mining problems in South Africa, using key sections and selected examples to highlight the value of geophysical techniques.

These papers will be available on the SAIMM website

http://www.saimm.co.za

845

855

863

875

Extending empirical evidence through numerical modelling in rock engineering design

by G.S. Esterhuizen*

Models are used in engineering to reproduce reality as faithfully as possible so that the expected response of a system for given actions or inputs can be determined. In the field of rock engineering, both empirically based and numerical models are widely used to determine the likely response of the rock surrounding excavations. Many of the empirical models are developed from statistical analysis of case histories or from direct observation; however, empirical models are limited because they should be used only within the range of conditions of the observational database. Synergy exists between empirical and numerical models, because empirical models can be used to calibrate and validate numerical models. The empirical approach can benefit from the capability of numerical models to investigate specific mechanisms, which would not be possible using observations alone. Two cases are presented in which the synergy between empirical and numerical models is demonstrated. The first case examines the analysis of discontinuity effects on the strength of slender pillars in limestone mines, and the second case evaluates the effects of stress orientation on coal mine entry stability. It is concluded that numerical model calibration and verification comprises an important first stage in the successful application of models in rock engineering design. Application of numerical models allows mechanisms and interactions of various parameters to be analysed, greatly improving the understanding of the system. The improved understanding can be used to extend the application of empirical design methods, resulting in improved safety and efficiency of rock engineering designs.

Keywords rock engineering, empirical design, numerical modelling.

Introduction The design of excavations in rock is extremely challenging because of the variability of the rock materials, uncertainty about the loading conditions, and the need for cost-effective solutions. Empirical models such as pillar strength equations, excavation stability charts, and classification-based support design charts are widely used in design. These models are popular because they reduce the complexity of the systems into simple charts or equations that can readily be applied. Numerical models, on the other hand, are built upon the mechanics of materials and can be used to conduct experiments in which unknown situations are studied. However, numerical models need to be grounded in reality, which can be achieved only through calibration and validation against empirical evidence. The The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

Models as tools in engineering design Models of various types are used by engineers and scientists to represent reality in a logical and objective way (Frigg and Hartmann, 2006). Models allow us to investigate the behaviour and attributes of a system and can be useful for predicting the response of the system under a different set of conditions. A model can be a physical representation of reality, usually on a reduced scale; or an abstraction such as a set of equations that replicate the system behaviour. Hammah and Curran (2009) state that although models are simplified reflections of reality, they are useful for: ➤ Developing an understanding ➤ Proper formulation of questions ➤ Providing an approximation of behaviour ➤ Providing meaningful predictions ➤ Aiding in design and decision-making.

* National Institute for Occupational Safety and Health, Pennsylvania. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12–14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

755

▲

Synopsis

synergy between empirical and numerical models has helped to improve the effectiveness of both types of models.

This paper discusses modelling in engineering design and examines the various types of models that are used in rock engineering. The synergy between empirical and numerical models is discussed, and two case studies are presented in which empirical models were used to validate numerical models and the numerical model outcomes were used to supplement empirically based design methods.

Extending empirical evidence through numerical modelling in rock engineering design The insight that comes ffrom evaluating model responses greatly helps us in understanding the problem and the development of engineering solutions. Often, the act of simply creating a model can lead to greater insight into the problem and point towards a solution.

Accuracy of models When creating a model it is always necessary to make some assumptions and simplifications. The process of model creation requires judgments about which factors are important and should be included in the model and which factors can be excluded. The result is that models are a simplification of the real system and thus the model results will be approximate. With ever-increasing computational capabilities, the temptation is to build increasingly complex models that include as many mechanisms as possible.

However, added complexity does not necessarily result in improved accuracy. Every new parameter included in a model introduces new uncertainty (Curran and Hammah, 2006).The quote attributed to Einstein comes to mind: ‘Make things as simple as possible, but no simpler’. A model might be overly simplistic if it does not simulate an important mechanism that determines system response. For example, if post-failure deformation of rock plays an important role in support design, a model that does not simulate post-failure mechanics will be insufficient for the intended application. A model should therefore endeavour to capture the essence of the system under consideration without unnecessary complication. Overly complex models can become difficult to understand and costly or impractical to use.

Calibration and verification Once a model has been created it is necessary to conduct calibration and validation studies. The calibration process involves improving the agreement of the model with respect to a chosen set of benchmarks through the adjustment of parameters in the model (Trucano et al., 2006). In rock engineering practice, a model is often calibrated against field monitoring data. During the calibration process some of the less well-defined inputs may be modified to achieve greater agreement with the field-measured deformations or other responses. A well-known calibration parameter that needs to be applied in almost every rock engineering model is a factor to account for the reduction in rock strength with increasing scale (Heuze, 1980; Hoek and Brown, 1980).

Validation addresses the question of whether a model produces correct results for its intended application (Thacker et al., 2004; Trucano et al., 2006). Validation involves comparing calibrated model outputs to experimental or other empirical outcomes. During validation, the range of conditions in which the model can provide accurate predictions can be established. In rock engineering, calibrated models can be validated against field monitoring data from alternative experimental sites that was not used in the calibration stage. However, there is a general lack of large numbers of field experimental results, because of the cost and difficulty of conducting such experiments. An alternative approach is to compare model results to empirically derived relationships that describe the average response of an excavation or structure. For example, empirically derived pillar strength equations can be used to determine the validity

▲

756

OCTOBER 2014

VOLUME 114

off a pillar model (Martin and Maybee, 2000; Lunder, 1994; Esterhuizen, 2006; Roberts et al., 2007).

Modelling approach with limited data In rock engineering, numerical modelling is usually conducted with limited data. The field stresses, material properties, and discontinuities in the rock are poorly described and are variable. When data is available, it usually consists of point representations conducted on a very small volume of the problem domain and the natural variability of the parameters is not known. Starfield and Cundall, (1988) discuss an approach for modelling of data-limited problems.

Under such circumstances, there is no point in constructing large and complicated models. Models should rather be simple and should be used to educate the design engineer by providing insight into the possible mechanisms. The models can be used as an experimental test bench, to aid in design and decision-making, rather than being expected to provide absolute design data.

Models in rock engineering Models have been extensively used since the inception of the rock engineering discipline in the 1960s. For example, photoelastic models and centrifuge models were widely used to understand the distribution of stress and deformations around excavations and to model failure under elevated loading conditions (Bieniawski and van Tonder, 1969; Hudson et al., 1972) In coal mining, large physical models were created to investigate the response of coal measure rocks to mining excavations (Hucke et al., 2006; Yeats et al., 1983), while large physical models of rock flow have been used for the design of caving mines (Kvapil, 1992; Laubscher, 2001; Power, 2004). With the development of high-speed computers, physical models have largely been replaced by numerical models of increasing complexity.

Currently, numerical models are used in every area of rock engineering. Software products are available that are able to model all kinds of rock structures, including rock at the grain scale (Potyondy and Cundall, 2004; Cho et al., 2007) and up to large-scale rock masses that include multiple discontinuities and intact rock fragments (Elmo and Stead, 2010; Mas Ivars et al., 2007).

Empirically based models A special type of model that is sometimes called an ‘empirical model’ has been widely used in rock engineering. These models are usually relatively simple mathematical equations that are based on a conceptual understanding of the system under consideration supplemented by statistical analysis of recorded performance of excavations or structures in rock.

The success of empirical models requires a good understanding of the problem at hand, usually through years of trial-and-error experimentation during which the underlying relationships between variables are discovered.

Alternatively, statistical analyses of large numbers of case histories can be used to determine how the parameters are related.

Examples of statistically based empirical models are the widely used pillar strength equations that were developed during the 1960s (Salamon and Munro, 1966; Bieniawski, 1968). These strength equations are based on the The Journal of The Southern African Institute of Mining and Metallurgy

Extending empirical evidence through numerical modelling in rock engineering design understanding that the strength off rock pillars depends on the strength of the intact rock and the width-to-height ratio of the pillar. Empirical evidence was collected from case histories in the field or by physical experimentation.

Statistical techniques were used to determine the best fit of the proposed strength equation to the empirical data. Other examples of empirically based models are excavation stability charts (Mathews et al., 1981) to support design charts (Barton, 2002). In coal mining in the USA, two empirically based models are currently used in the design and layout of pillar and panels for longwalls and retreat mining sections.

These models are called ‘Analysis of Longwall Pillar Stability’ (ALPS) (Mark, 1992) and ‘Analysis of Retreat Mining Pillar Stability’ (ARMPS) (Mark and Chase, 1997). The models are based on very large databases of case histories that are updated as new information becomes available. These empirically developed models are very powerful because they are rooted in the actual performance of real excavations.

The disadvantages of empirically based models are that they cannot easily be extended beyond their original databases and they do not necessarily capture the mechanics of the system being modelled (Mark, 1999). The need for sufficient case histories to develop the model necessitates that the model output represents the ‘average’ response of all the cases evaluated. It becomes difficult to use these methods when site-specific conditions are not well represented by the averaged conditions.

Numerical models in rock engineering The attractiveness of numerical models for rock engineering applications can be explained by the problem of the scale effect of rock strength. Laboratory-scale rock samples are stronger than larger field-scale rocks, while the rock mass, which can contain many discontinuities, is much weaker than the laboratory-scale rocks (Hoek and Brown, 1980). Physical model tests on small-scale rock samples bear little resemblance to the response of a full-scale rock mass containing many natural discontinuities. It is hardly practical to recreate a rock mass in the laboratory that is sufficiently large to capture its response to external loading. Even if this could be done, the required loads to simulate the stress changes associated with mining or tectonics would be well beyond current laboratory testing capabilities. However, numerical models that simulate large volumes of rock with embedded discontinuities can readily be created, allowing numerical experiments to be conducted on the full-scale rock mass. The recent development of discrete fracture network models (Dershowitz et al., 2004) combined with synthetic rock mass models has allowed investigations into large-scale rock mass properties to be conducted (Cundall et al., 2008) – an endeavour that could not be undertaken in a laboratory.

numerical models are based on the mechanics off materials allows the engineer to study interactions and combinations of variables that are not regularly encountered in the field, but may be important from a safety or hazard assessment point of view. In these cases, numerical models can extend the empirical evidence, enabling improved engineering design and excavation safety.

Case studies Two case studies are described in which numerical models were validated against empirical evidence, followed by investigation of a particular parameter on excavation stability using the models. The first case considers the question of the impact of large angular discontinuities on the stability of pillars in underground limestone mines. The second case study describes how numerical models were used to quantify the relationship between coal mine entry stability and the ratio of horizontal to vertical stress in the surrounding rock.

Case study 1: discontinuity effects on slender pillar strength The room-and-pillar method is used in the underground limestone mines in the USA. The room dimensions are typically in the region of 12 to 15 m while pillars have similar dimensions. The mining height is about 8 to 10 m for singlecut mining and up to 20 m when multiple benches are mined.

The width-to-height ratio of pillars varies between 0.5 and about 4.0. Wide area collapses of a panel of pillars are rare in the limestone mines. One or two cases of panel collapse are known to have occurred, and only one has been adequately described in the literature (Phillipson, 2010). A survey of pillar stability issues on 34 different limestone mines identified 18 cases in which single pillars had failed or showed signs of being overloaded (Esterhuizen et al., 2008).

The failed pillars exhibited one or more of the following characteristics: 1. Collapse of the entire pillar 2. Rib spalling to a rounded hourglass shape with open joints and fractures, shown in Figure 1

Synergy between empirical and numerical models

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 1—Partially benched pillar failing under elevated stress at the bench mining front. Typical hourglass formation indicating overloaded pillar VOLUME 114

OCTOBER 2014

757

▲

There is considerable potential for synergy between empirical and numerical models. All numerical models need to be validated against empirical evidence. Since empirical models usually encapsulate a large database of experience, they can be effectively used to test the validity of numerical models.

Similarly, validated numerical models can be used to investigate specific aspects of a system that are difficult to establish by empirical observations alone. The fact that

Extending empirical evidence through numerical modelling in rock engineering design 3. Shearing along large angular discontinuities (dip 30° to 70°) resulting in loss of pillar integrity, shown in Figure 2.

The failed pillars were typically surrounded by pillars that appeared to be stable, showing minimal signs of disturbance.

The observations led to the conclusion that the failed pillars represent the low end of the distribution of possible pillar strengths, and not the average pillar strength.

Of particular concern was that the failed pillars were impacted by large angular discontinuities. It was estimated that these pillars failed when the average pillar stress was only about 5% of the uniaxial compressive strength of the limestone rock material. Discontinuities are not always readily visible to production staff when developing a pillar, but become apparent only when the pillar becomes fully loaded or when bench mining is carried out around the pillars. Particularly hazardous conditions can result if large angular discontinuities cause unstable blocks to slide or topple from the pillar ribs. Of the eighteen failed pillars identified in the above studies, seven were associated with large angular discontinuities. It was also observed that widely spaced angular discontinuities were present in the majority of limestone mines. Clearly, the effect of large discontinuities has to be taken into account in the design of pillar systems in limestone mines.

Empirical design approach The empirically developed design approach of Roberts et al.

(2007) was adopted for estimating the strength of pillars in limestone mines. Using this approach, the strength of a pillar that is square in plan can be expressed as follows: [1] where w and h are the pillar width and height in metres. The value of k can be expressed in terms of the uniaxial compressive strength (UCS) as follows: [2] This equation predicts the average or expected strength of a pillar and does not explicitly account for large angular discontinuities. For example, a 15 m wide by 15 m high pillar

Figure 2—Example of a pillar bisected by a large angular discontinuity

▲

758

OCTOBER 2014

VOLUME 114

can be expected to have a strength that is 29.6% off the UCS of the rock material. However, field data shows that such a pillar may fail when the stress is only about 5% of the UCS if it is intersected by a large angular discontinuity. The concern was that a collapse can occur if angular discontinuities weaken a large proportion of the pillars in a panel. Therefore, it was decided to introduce an adjustment to the pillar strength equation that accounts for the weakening effect of angular discontinuities. However, the lack of sufficient field cases made it impossible to conduct a statistical analysis of the failed cases to estimate the impact of the angular discontinuities. Numerical models were therefore used to investigate the likely effect of discontinuities on pillar strength.

Modelling analysis The numerical models were designed specifically to determine how the inclination and frequency of large, roof-to-floor discontinuities would affect the strength of the slender pillars found in limestone mines. The two-dimensional finite difference software UDEC (Itasca Consulting Group, 2006) was used to model the pillars.

Model calibration and validation was carried out against the empirically developed pillar design method of Roberts et al. (2007) as well as the Lunder and Pakalnis (1997) pillar strength equation for hard rock pillars (Esterhuizen et al., 2008). Models were created to simulate pillars with width-toheight ratios of 0.5 to 2.0 and the results compared to the empirically developed equations. These models did not contain any explicitly modelled angular discontinuities. The rock mass properties were selected to simulate a good quality rock mass representative of the rock found in limestone mines. Figure 3 shows the comparison between empirical and model results. It was concluded that the modelling method provides a realistic representation of pillar strength over the range of width-to-height ratios shown.

Further modelling was conducted by introducing large angular discontinuities into the pillar models. Various analyses were carried out in which the dip of the discontinuity was varied from 30° to 90° and the strength of the pillar was determined by simulating the gradual compression of the pillar until it reached its peak resistance and started to shed load. Figure 4 shows one of the models, indicating the location of the angular discontinuity and associated rock

Figure 3—Validation of numerical model of a pillar against the empirically derived pillar strength equation of Roberts et al. (2007) The Journal of The Southern African Institute of Mining and Metallurgy

Extending empirical evidence through numerical modelling in rock engineering design Case study 2: stress impacts on coal mine entry stability Effects of horizontal stress on roof stability

Figure 4—Model showing damage to a pillar with a width-to-height ratio of 1.0 when loaded beyond its peak strength. The pillar is partially intersected by an angular discontinuity, with rock damage indicated by coloured zones. Loading is simulated by gradually moving the upper platen downward

failure. A series of curves was fitted to the peak resistance of each modelled pillar (Figure 5). The results show firstly that as the discontinuity dip increases from 30° to about 60°, its impact on the pillar strength increases. When the discontinuity dip is greater than 70°, the effect starts to diminish. A vertical joint through the centre of a pillar is seen to have a relatively small impact on pillar strength. These trends in the relationship between pillar strength and discontinuity dip are similar to the results obtained when testing laboratory specimens with inclined planes of weakness.

The width-to-height ratio is also shown to be a significant factor affecting the impact of large discontinuities.

Figure 5 shows, for example, that a pillar with a width-toheight ratio of 0.5 will suffer a 93% reduction in strength if it is intersected by a 60° dipping joint, while a pillar with a width-to-height ratio of 1.0 would only suffer a 34% reduction in strength. The observation that slender pillars intersected by large angular discontinuities can fail when the average stress is only about 5% of the intact rock strength confirms that these large strength reductions do occur in the field and are similar to those predicted by the numerical models.

Horizontal stress has long been known to be a significant factor contributing to roof instability in bedded rocks in both coal and hard rock mines (Aggson and Curran, 1978; Herget, 1987; Mark and Mucho, 1994; Iannacchione et al., 2003). In coal mines in the USA, horizontal stresses are normally greater than vertical stresses. They can cause compressivetype failures in the bedded roof rocks (Mark and Barczak, 2000) and are a critical factor in the stability of mining excavations. The horizontal stresses are closely related to the global tectonic forces (Mark and Gadde, 2008) in the North American plate. One feature of the horizontal stress field is that the magnitude of the major horizontal stress is about 50% greater than the minor horizontal stress. As a result, excavation performance can be highly dependent on the orientation of the long axis of the entry relative to the major horizontal stress (Mark and Mucho, 1994). This case study has the objective of better quantifying the effect of entry orientation on entry stability.

Roof instability of bedded rocks subject to high horizontal stress can take many forms. Failure is often observed to be preceded by ‘delamination’ of the bedded rock into thin

Figure 5—Impact of large angular discontinuities on the strength of pillars, based on numerical model results

The recommended pillar strength equation for limestone mines in the USA includes a large discontinuity factor (LDF) to account for the impact of large discontinuities on pillar strength (Esterhuizen et al., 2008). The LDF accounts for both the dip angle of discontinuities and the frequency of discontinuities, based on the results shown in Figure 5.

Figure 6 shows an application of the modified design equation, in which the impact of a single large discontinuity dipping at 60° on the strength of a 14 m wide pillar is estimated. The discontinuity causes a much greater reduction in pillar strength as the width-to-height ratio drops below 1.0. This case demonstrates how numerical models were used to develop an understanding of the issue of large angular discontinuity impact on pillar strength in the absence of sufficient field cases to conduct a statistical analysis. The outcome of this work was used to explain the unusually low strength of pillars that collapsed in a marble mine in 2010 (Phillipson, 2010).

Figure 6—Expected impact of a large angular discontinuity dipping at 60° on the strength of a 14 m wide pillar at varying width-to-height ratios using the modified empirical pillar strength equation

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

759

▲

Application to design

Extending empirical evidence through numerical modelling in rock engineering design beams in the rooff or ffloor off an excavation (Colwell and Frith, 2010). The thickness of these delaminated beams is determined by the geological composition of the rock, and can vary from tens of centimetres to less than 1 cm. These individual beds are much weaker than the original combined beam and have an important impact on the strength and failure development within the roof. A second commonly observed failure mode is known as ‘cutter’ or ‘kink’ failure, in which crushing and local buckling of thinly laminated roof beds occurs near the corners of an excavation. The kink band or cutter is progressive and typically forms a near-vertical zone of failed rock (Hill, 1986). Figure 7 shows an example of cutter or kink failure in thinly laminated strata. The nearvertical zone of crushed rock that results from this failure mode can lead to the progressive collapse of the entire roof, as shown in Figure 8.

The stability of critically stressed excavations can be highly dependent on their orientation relative to the major horizontal stress. Many cases have been reported where the ground conditions in a mine were significantly improved by simply orienting the main development direction near-parallel to the major horizontal stress (Mark and Mucho, 1994).

Empirical design approach Modern support practices using rock reinforcement have evolved in US coal mines since the 1960s. Extensive

historical experience exists with successful f and unsuccessful f support systems in a wide variety of ground conditions.

Based on this experience, NIOSH developed an empirical support design procedure, called ‘Analysis of Roof Bolt Systems’ (ARBS) (Mark et al., 2001). The procedure is based on a statistical analysis of roof falls and support performance at 37 coal mines across the USA. The ARBS makes use of the Coal Mine Roof Rating (CMRR) (Molinda and Mark, 1996) to quantify the stability of the roof rocks. The support intensity is expressed by an index parameter called PRSUP which combines the bolt length, spacing, capacity and, entry width into a single parameter as follows: [3] where L is the bolt length in metres, N is the number of bolts per support row, C is the bolt capacity in kilonewton, S is the bolt row spacing in metres and We is the width of the excavation in metres. A discriminant line, which defines the required PRSUP to achieve acceptable entry stability for a given CMRR, was determined by statistical analysis of the case histories of ‘acceptable’ and ‘unacceptable’ support performance. The required support intensity is given by: [4] where H is the depth of cover in metres. The horizontal stress is not explicitly included in the calculation of either the CMRR or the PRSUP, but is indirectly related to the depth of cover.

Consequently, the empirical design approach cannot be used to investigate the impact of various orientations of the entry relative to the major horizontal stress.

Modelling analysis of horizontal stress effects

Figure 7—Failure of bedding laminations in the roof of an entry subject to high horizontal stress. This type of failure is commonly called a ‘cutter’ or ‘kink’. (Photo: Chris Mark, NIOSH)

Figure 8—Roof cavity created by progressive failure of alternating shale and siltstone beds affected by horizontal stress. (Photo: Greg Molinda, NIOSH)

▲

760

OCTOBER 2014

VOLUME 114

A numerical modelling study was conducted in which calibrated numerical models were used to investigate the impact of horizontal stress on entry stability. The first objective of the study was to develop a modelling method that allows the stability of an excavation to be quantified in meaningful manner. Model outputs can be expressed in terms of degree of deformation or volume of failure, but these outputs are indirectly related to stability. The operating engineer is interested in the degree of stability of the excavation. The concept of a factor of safety, which is widely used and accepted in engineering practice (Harr, 1987), was selected to express the stability of the modelled excavations.

For this modelling study, the strength reduction method (SRM) (Zienkiewicz et al., 1975) was selected to calculate the factor of safety of supported coal mines entries (Esterhuizen, 2012). The SRM was originally developed to provide an alternative method of calculating the stability of rock slopes using numerical models. It has since found acceptance in rock slope design (Lorig and Varona, 2000; Diederichs et al., 2007), but has not been widely used for underground excavation stability analysis. The SRM is uniquely suited to the objective of expressing entry stability in a format that is meaningful to operating engineers.

The SRM safety factor is determined by first conducting a stability analysis of an excavation using average rock strength properties. Depending on the outcome, the analysis is repeated using either a decreased or increased strength

The Journal of The Southern African Institute of Mining and Metallurgy

Extending empirical evidence through numerical modelling in rock engineering design

Figure 9—Three-dimensional slice model of an entry supported with four 2.4 m long fully grouted rockbolts. Dark shading indicates severity of rock damage; light red shading indicates dome-shaped roof that is moving downwards. Fully grouted bolts are providing anchorage where coloured red The Journal of The Southern African Institute of Mining and Metallurgy

systematic procedure ffor obtaining model inputs ffrom field f data was developed based on the CMRR (Esterhuizen et al., 2013). Using this approach, the rock mass is divided into units, each unit having similar strength properties. The procedures can be used by support designers to create numerical models in the absence of detailed laboratory test results.

The SRM-calculated stability factors were validated by comparing them to ARBS-calculated stability factors. For this study, a total of 15 different cases were evaluated consisting of entries located at three different depths of overburden with five different roof compositions (Esterhuizen et al., 2013).

The support system consisted of five fully grouted rockbolts in a 6 m wide entry. The bolt row spacing was 1.2 m. The stability factors of the 15 cases were first evaluated using the ARBS method. For the SRM analyses, the 15 cases were evaluated for three different stress scenarios. The stress scenarios were selected to represent the range of likely horizontal stress conditions in US coal mines. The SRM results for the three different horizontal stress scenarios were averaged to allow comparison to the ARBS stability factors.

This was done because the ARBS stability factor is based on a discriminant line that represents the entry performance under averaged horizontal stress conditions. Figure 10 presents the correlation between the stability factors of the two methods at 100 m, 200 m, and 300 m depth of cover.

The excellent correlation between the results of the statistically based ARBS method and the SRM confirms the validity of SRM results.

Horizontal stress effects quantified by numerical models A series of SRM analyses was conducted in which the stability factors of entries in various geological settings and field stress conditions were varied. The entries were located at depths of cover of 100, 200, and 300 m. The ratio of the horizontal to vertical stress in the plane perpendicular to the axis of the entry was used to quantify the horizontal stress effect. This ratio is commonly known as the k-ratio in rock engineering. The resulting stability factors were normalized by the depth in metres, the CMRR, and the strength of the immediate roof layer in MPa. Figure 11 shows the results for a 6 m wide entry supported by four fully-grouted 1.8 m long bolts, with rows 1.2 m apart. There is an inverse relationship between the stability factor and the k-ratio. The relationship

Figure 10—Correlation between stability factors calculated by the empirical ARBS method and the numerical model-based SRM VOLUME 114

OCTOBER 2014

761

▲

until the point off ffailure is satisfactorily f defined. f The safety f factor of the system is simply calculated as the inverse of the strength adjustment factor at the point of collapse of the modelled excavation. For example, if collapse occurs when the strength is reduced by a factor of 0.8, the safety factor would be 1.25.

The state of ‘failure’ being investigated must be clearly defined when using the SRM. For this study, an entry is considered to have failed if roof collapse in the model exceeds the bolt length. The term stability factor is used in this paper because the stability is expressed as a ratio of strengths of the rock mass, rather than the ratio of strength to load, as used in the classic safety factor calculation. The interpretation of stability factors is similar to the interpretation of safety factors. A stability factor of 1.0 indicates a system that is at the point of failure, while less than 1.0 indicates an increased likelihood of failure and greater than 1.0 indicates increased likelihood of stability.

The FLAC3D (Itasca Consulting Group, 2011) finitedifference code was used to conduct the strength reduction analyses. Figure 9 shows one of the models, which simulates a vertical slice through a 6 m wide coal mine entry. The thickness of the slice is equal to the support row spacing, typically 1.2 m. Strata layering is modelled with explicit interfaces between the different lithologies. Rockbolts are modelled using the built-in structural elements available in FLAC3D. The bolts are located along the centre line of the slice. The bedded strata are modelled using the strainsoftening ubiquitous joint constitutive model available in FLAC3D. The strength parameters of the rock matrix and the bedding planes are specified separately in the model. It was found that modelling of the anisotropic strength of the bedded rock was a requirement to achieve realistic rock mass response (Esterhuizen and Bajpayee, 2012).

The model inputs and results were initially calibrated against field instrumentation studies. The calibration studies included rock deformation, bolt loads, and entry stability analysis in a variety of geological conditions encountered in the US coal regions. During the calibration stage, a

Extending empirical evidence through numerical modelling in rock engineering design empirically developed design method did not explicitly account for horizontal stress. Numerical models were used to investigate the impact of the k-ratio on entry stability and produced a relationship that can be used to quantify the kratio effect on stability.

In both cases, the numerical model outputs were validated against empirical observations before detailed analyses of the specific issues were conducted. The numerical model results were then used to extend the usefulness of the empirical models, demonstrating the synergy between the two approaches.

Conclusions

Figure 11—Effect of varying the horizontal-to-vertical stress ratio on the stability of modelled entries in various geological settings and depths of cover

can be expressed as follows: [5] Using this relationship, the change in the stability factor with changes in the k-ratio (k) can be estimated as follows: [6]

Consider a 6 m wide entry at 200 m depth in the eastern USA in which the major horizontal stress is 2.5 times the vertical stress and the minor horizontal stress is 1.5 times the vertical stress. Assume the stability factor of an entry is estimated to be 1.6 when calculated by the empirical ARBS method. This stability factor is representative of the entry performance under the average stress conditions, which is a k-ratio of 2.0. The stability factor in the ‘poor direction’, that is, if the major horizontal stress is perpendicular to the axis of the entry (k-ratio is 2.5), can be calculated to be 1.30 using Equation [6]. The stability factor in the ‘good direction’ (k-ratio is 1.5) is calculated to be 2.08. In this example, the reduced stability factor in the ‘poor direction’ indicates that action will be required to ensure a stable, safe entry, while entries developed in the ‘good direction’ can be expected to be adequately supported. The above relationship can be incorporated in empirically based support design equations to better represent the impact of horizontal stresses in the empirical model.

Discussion The two case studies show how specific stability issues identified during empirical observations can be evaluated using numerical models. In the case of the limestone mine pillars, the impact of large angular discontinuities was identified as a potential safety hazard. However, the lack of sufficient case histories in the field made it impossible to conduct a statistical analysis to quantify their effect on pillar stability. Numerical models were used to investigate a large number of scenarios to develop an understanding of the problem and determine an adjustment to the pillar strength equation that accounts for these structures.

In the second case, empirical observations identified the impact of the k-ratio on excavation stability; however, the

▲

762

OCTOBER 2014

VOLUME 114

Models of various types are used in rock engineering to determine the likely response of the rock mass around excavations. Empirical models, based on the analysis of large numbers of case histories, have found wide acceptance as a tool for engineering design. The application of empirical models is limited by the restriction that they should not be used beyond the limits of the empirical base from which they were developed. Numerical models are based on the mechanics of rock behaviour and can be used to answer questions about rock mass response under given loading conditions and to extend the empirical models.

Numerical models must be validated against empirical data. The process of first calibrating numerical models against specific case histories is required because of the uncertainty associated with rock mass strength parameters.

The validity of the models can then be tested against empirically based models.

Synergy exists between the empirical and numerical modelling approaches. Empirical models can be used to validate numerical models, and numerical models can be used to extend the utility of empirical models.

The two case studies presented show how numerical models were applied to answer questions about specific aspects of excavation stability that existing empirical models were unable to provide.

The combined application of numerical and empirical models can help to improve understanding of causes of stability and instability in excavations, resulting in efficient design and increased safety.

Acknowledgements The contributions of my colleagues at the NIOSH Office of Mine Safety and Health Research in the research presented here are gratefully acknowledged. The willingness of mine staff to share their experience and allow NIOSH researchers to conduct mine site surveys and measurements have provided the much-needed empirical data for calibration and verification of numerical model results.

Disclaimer The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the National Institute for Occupational Safety and Health.

References AGGSON, J.R. and CURRAN, J. 1978. Coal mine ground control problems associated with a high horizontal stress field. AIME Annual Meeting, g Denver, CO. 19 pp.

The Journal of The Southern African Institute of Mining and Metallurgy

Extending empirical evidence through numerical modelling in rock engineering design

ITASCA CONSULTING GROUP. 2012. Fast Lagrangian Analysis of Continua in Three Dimensions (FLAC3D), User’s Guide. Minneapolis MN. BARTON, N. 2002. Some new Q-value correlations to assist in site characterisation and tunnel design. International Journal of Rock Mechanics and Mining Sciences, vol. 39, no. 2. pp. 185–216.

BIENIAWSKI, Z.T. 1968. In situ strength and deformation characteristics of coal.

Engineering Geology, vol. 2, no. 5. pp. 325–340.

BIENIAWSKI, Z.T. and VAN TONDER, C.P.G. 1969. A photoelastic-model study of stress distribution and rock fracture around mining excavations.

Experimental Mechanics, vol. 9, no. 2. pp.75–81.

ChO, N., MARTIN, C.D., and SEGO, D.C. 2007. A clumped particle model for rock.

International Journal of Rock Mechanics and Mining Sciences, vol. 44, no. 7. pp. 997–1010.

COLWELL, M. and FRITH, R. 2010. AMCMRR-an analytical model for coal mine roof reinforcement. Coal Operators' Conference, University of Wollongong, Wollongong, Australia. p. 302.

CUNDALL, P.A., PIERCE, M.E., and MAS IVARS, D. 2008. Quantifying the size effect of rock mass strength. Proceedings of the 1st Southern Hemisphere International Rock Mechanics Symposium, Perth, Western Australia, September 2008. Australian Centre for Geomechanics, Nedlands, Western Australia. vol. 2, pp. 3–15.

CURRAN, J.H. and HAMMAH, R.E. 2006. Keynote Lecture: Seven Lessons of Geomechanics Software Development. Proceedings of the 41st US Symposium on Rock Mechanics: 50 Years of Rock Mechanics Landmarks and Future Challenges, Golden, Colorado, 17–21 June 2006.

American Rock Mechanics Association, Alexandria, VA. 15 pp.

DERSHOWITZ, W.S., LA POINTE, P.R., AND DOE, T.W. 2004. Advances in discrete fracture network modeling. Proceedings of the US EPA/NGWA Fractured Rock Conference: State of the Science and Measuring Success in Remediation, Portland, Maine, 13–15 September 2004. pp. 882-894.

DIEDERICHS, M.S., LATO, M., HAMMAH, R., and QUINN, P. 2007. Shear strength reduction approach for slope stability analyses. Proceedings of the 1st Canada-US Rock Mechanics Symposium, Vancouver, BC, 27–31 May 2007. Eberhardt, E., Stead, D., and Morrison, T. (eds.). Taylor and Francis. pp. 319–327.

ELMO, D. and STEAD, D. 2010. An integrated numerical modelling–discrete fracture network approach applied to the characterisation of rock mass strength of naturally fractured pillars. Rock Mechanics and Rock Engineering, g vol. 43, no. 1. pp. 3–19.

ESTERHUIZEN, G.S., DOLINAR, D.R., and ELLENBERGER, J.L. 2008. Pillar Strength and Design Methodology for Stone Mines. Proceedings of the 27th International Conference on Ground Control in Mining, g West Virginia University, Morgantown WV, 2008, pp 241–253.

ESTERHUIZEN, G.S. 2006. Evaluation of the strength of slender pillars.

Transactions of the Society for Mining Exploration and Geology. vol. 320. pp. 69–76.

ESTERHUIZEN, G.S. 2012. A Stability factor for Supported Mine Entries Based on Numerical Model Analysis. Proceedings of the 31st International Conference on Ground Control in Mining, g West Virginia University, Morgantown, WV, 29–31 July 2014. 9 pp.

ESTERHUIZEN, G.S. and BAJPAYEE, T.S. 2012. Horizontal stress related failure in bedded mine roofs—insight from field observations and numerical models. Proceedings of the 46th US Rock Mechanics/Geomechanics Symposium. Chicago, IL, 24–27 June 2012. Bobet, A. (ed.). American Rock Mechanics Association, Alexandria, VA. Paper no. 137. 10 pp.

ESTERHUIZEN, G.S., BAJPAYEE, T.S., ELLENBERGER, J.L., and MURPHY, M.M. 2013.

Practical estimation of rock properties for modeling bedded coal mine strata using the Coal Mine Roof Rating. Proceedings of the 47th US Rock Mechanics/Geomechanics Symposium, San Francisco, CA, 23–26 June 2013. American Rock Mechanics Association, Alexandria, VA. Paper 13–154. 14 pp.

FRIGG, R. and HARTMANN, S. 2006. Models in Science. The Stanford Encyclopedia of Philosophy. http://stanford.library.usyd.edu.au/archives/spr2009/ entries/models-science/ [Accessed November 2013].

HAMMAH, R.E. and CURRAN, J.H. 2009. It is better to be approximately right than precisely wrong: why simple models work in mining geomechanics.

The Journal of The Southern African Institute of Mining and Metallurgy

International Workshop on Numerical Modeling for Underground Excavation Design. Department of Health and Human Services, National Institute for Occupational Safety and Health, Information Circular IC 9512. pp 55–61.

HARR, M.E. 1987. Reliability Based Design in Civil Engineering. Mc Graw Hill, New York.

HERGET, G. 1987. Stress assumptions for underground excavations in the Canadian Shield. International Journal of Rock Mechanics and Mining Sciences and Geomechanics Abstracts, vol. 24, no. 1. pp. 95–97.

HEUZE, F.E. 1980. Scale effects in the determination of rock mass strength and deformability. Rock Mechanics, vol. 12, no. 3-4. pp. 167–192.

HILL, J.L. 1986. Cutter roof failure: An overview of the causes and methods for control. US Bureau of Mines, Information Circular IC 9094. United States Department of the Interior. 27 pp.

HOEK, E. and BROWN, E.T. 1980. Underground Excavations in Rock. Institute of Mining and Metallurgy, London. 527 pp.

HUCKE, A., STUDENY, A., RUPPEL, U., and WITTHAUS, H. 2006. Advanced prediction methods for roadway behaviour by combining numerical simulation, physical modelling and in-situ monitoring. Proceedings of the 25th International Conference on Ground Control in Mining, g West Virginia University, Morgantown, WV. p. 213.

HUDSON, J.A., BROWN, E.T., and RUMMEL, F. 1972. The controlled failure of rock discs and rings loaded in diametral compression. International Journal of Rock Mechanics and Mining Sciences and Geomechanics Abstracts, vol. 9, no. 2. pp. 241–244.

IANNACCHIONE, A.T., MARSHALL, T.E., BURKE, L., MELVILLE, R., and LITSENBERGER, J. 2003. Safer mine layouts for underground stone mines subjected to excessive levels of horizontal stress. Mining Engineering, g vol. 55, no. 4. pp. 25–31.

KVAPIL, R. 1992. Sublevel caving. SME Mining Engineering Handbook. Society for Mining, Metallurgy and Exploration, Littleton, CO. pp. 1789–1814.

LAUBSCHER, D.H. 2001. Cave mining – state of the art. Underground Mining Methods: Engineering Fundamentals and International Case Studies.

Hustrulid, W.A. and Bullock, R.L. (eds.). Society for Mining, Metallurgy and Exploration, Littleton, CO. pp. 455-464.

LORIG, L. and VARONA, P. 2000. Practical slope stability analysis using finitedifference codes. Slope Stability in Surface Mining. g Hustrulid, W.A., McCarter, M.K., and Van Zyl D.J.A. (eds.). Society for Mining, Metallurgy and Exploration, Littleton, CO. pp. 115-124.

LUNDER, P.J. 1994. Hard rock pillar strength estimation an applied approach. MS thesis, University of British Columbia, Vancouver, BC. LUNDER, P.J. and PAKALNIS, R. 1997. Determining the strength of hard rock mine pillars. Bulletin of the Canadian Institute of Mining and Metallurgy, vol. 90. pp. 51–55.

MARK, C. 1992. Analysis of longwall pillar stability (ALPS): an update.

Proceedings of the Workshop on Coal Pillar Mechanics and Design.

Pittsburgh, PA. US Department of the Interior, Bureau of Mines, Information Circular IC 9315. pp. 238–249.

MARK, C. and CHASE, F.E. 1997. Analysis of retreat mining pillar stability (ARMPS). Proceedings – New Technology for Ground Control in Retreat Mining, g Pittsburgh, PA. US Department of Health and Human Services, Public Health Service, Centers for Disease Control and Prevention.

National Institute for Occupational Safety and Health, DHHS (NIOSH) Publication No. 97–122. pp. 17–34.

MARK, C. and GADDE, M. 2008. Global trends in coal mine horizontal stress measurements. Proceedings of the 27th International Conference on Ground Control in Mining, g West Virginia University, Morgantown, WV, 29–31 July 2008. pp. 319–331.

MARK, C. and BARCZAK, T.M. 2000. Fundamentals of coal mine roof support.

New Technology for Coal Mine Roof Support. Proceedings of the NIOSH Open Industry Briefing. g NIOSH IC, 9453. National Institute for Occupational Safety and Health. pp. 23–42.

MARK, C. 1999. Empirical methods for coal pillar design. Proceedings of the 2nd International Workshop on Coal Pillar Mechanics and Design. National Institute for Occupational Safety and Health, Report IC 9448. pp. 145–54.

MARK, C. and MUCHO, T.P. 1994. Longwall Mine Design for Control of Horizontal Stress. New Technology for Longwall Ground Control. Special Publication 01-94. US Bureau of Mines. pp. 53–76.

VOLUME 114

OCTOBER 2014

763

▲

ITASCA CONSULTING GROUP. 2006. Universal Distinct Element Code (UDEC), User’s Guide. Minneapolis MN.

Extending empirical evidence through numerical modelling in rock engineering design MARK, C., MOLINDA, G.M., and DOLINAR, D.R. 2001. Analysis off rooff bolt systems. Proceedings of the 20th International Conference on Ground Control in Mining, g West Virginia University, Morgantown, WV, 7–9 August 2001. pp. 218–225.

f model ffor coal ROBERTS, D.P. and VAN DER MERWE, J.N. 1994. A strain softening via back analysis of in situ coal pillar tests. Proceedings of the 11th International Conference on Computer Methods and Advances in Geomechanics. pp. 437–480.

MATHEWS, K.E., HOEK, E., WYLLIE, D.C., and STEWART, S. 1981. Prediction of stable excavation spans for mining at depths below 1000 m in hard rock.

CANMET DSS Serial No: 0sQ80-00081. Ottawa.

ROBERTS, D., TOLFREE, D., and MCINTYRE, H. 2007. Using confinement as a means to estimate pillar strength in a room and pillar mine. Proceedings of the 1st Canada-US Rock Mechanics Symposium. Taylor and Francis. pp. 1455–1461.

MARTIN, C.D. and MAYBEE, W.G. 2000. The strength of hard rock pillars.

International Journal of Rock Mechanics and Mining Sciences, vol. 37. pp. 1239–1246.

MAS IVARS, D., DEISMAN, N., PIERCE, M., and FAIRHURST, C. 2007. The synthetic rock mass approach – a step forward in the characterization of jointed rock masses. Proceedings of the 11th Congress of the International Society for Rock Mechanics. Taylor and Francis, Leiden. vol. 1, pp. 485–490.

MOLINDA, G.M. and MARK, C. 1996. Rating the Strength of Coal Mine Roof Rocks. Information Circular IC 9444. US Department of the Interior, Bureau of Mines, Pittsburgh. 36 pp.

SALAMON, M.D.G. and MUNRO, A.H. 1967. A study of the strength of coal pillars.

Journal of the South African Institute of Mining and Metallurgy, vol. 68. pp. 55–67.

STARFIELD, A.M. and CUNDALL, P.A. 1988. Towards a methodology for rock mechanics modelling. International Journal of Rock Mechanics and Mining Sciences and Geomechanics Abstracts, vol. 25, no. 3. pp. 99–106.

THACKER, B.H., DOEBLING, S.W., HEMEZ, F.M., ANDERSON, M.C., PEPIN, J.E., and RODRIGUEZ, E.A. 2004. Concepts of Model Verification and Validation. Los Alamos National Laboratory, Los Alamos, NM. 27 pp.

PHILLIPSON, S.E. 2012. Massive pillar collapse: a room-and-pillar marble mine case study. Proceedings of the 31st International Conference on Ground Control in Mining, g West Virginia University, Morgantown, WV. 9 pp.

TRUCANOA, T.G., SWILERA, L.P., IGUSAB, T., OBERKAMPFC, W.L., and PILCHC, M. 2006. Calibration, validation, and sensitivity analysis: what’s what.

Journal of Reliability Engineering and System Safety, vol. 91. pp. 1331–1357.

POTYONDY, D.O. AND CUNDALL, P.A. 2004. A bonded-particle model for rock.

International Journal of Rock Mechanics and Mining Sciences, vol. 41, no. 8. pp. 1329–1364.

YEATES, R.A., ENEVER, J.R., and HEBBLEWHITE, B.K. 1983. Investigations prior to the introduction of longwall mining. Proceedings of the 5th International Society for Rock Mechanics Congress, Melbourne, Australia. pp. E71–E77.

POWER, G.R. 2004. Modelling granular flow in caving mines: large scale physical modelling and full scale experiments. PhD thesis, University of Queensland, Brisbane, Australia. 250 pp.

ZIENKIEWICZ, O.C., HUMPHESON, C., and LEWIS, R.W. 1975. Associated and nonassociated visco-plasticity and plasticity in soil mechanics. Geotechnique, vol. 25. pp. 617–689. ◆

▲

764

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Time-dependent tensile strengths of Bushveld Complex rocks and implications for rock failure around mining excavations

by D. Nyungu* and T.R. Stacey

Despite observations of spalling and damage of mine excavation wall rock in the Bushveld Complex (BC) over the passage of time, there have been very few time-dependent or creep tests carried out in South Africa on rock, particularly on BC rock types. The research described in this paper deals with the investigation of stress and strain conditions influencing spalling of wall rock in BC mine excavations, and the influence of time on the tensile strength of several BC rock types.

Time-dependent laboratory testing of BC rocks was carried out in indirect tension. The results show that the magnitude of the tensile strength of BC rock types is approximately 5% of their uniaxial compressive strength magnitudes. The average long-term uniaxial compressive strength of the BC rocks, interpreted from the axial stress-volumetric strain graphs, is 56% of the UCS value. The long-term tensile strength is shown to be less than 70% of the normal tensile strength. Extension strains at tensile strength failure ranged between 0.16 and 0.21 millistrain. Values corresponding with the long-term tensile strength are less than 70% of this range, namely, 0.11 to 0.15 millistrain. These results represent new knowledge, since such rock testing and analysis does not appear to have been carried out previously on BC rock types.

Elastic numerical modelling was carried out to illustrate the extents of tensile stress zones and extension zones around of typical BC mine excavations. The models showed that large zones of extension strain can occur around BC excavations, and that the magnitudes of the extension strain can substantially exceed the critical values determined from the laboratory testing. The implication of this is that there are substantial zones surrounding BC mine excavations that will be prone to time-dependent spalling conditions and perhaps more significant failure.

Keywords tensile strength, creep, time-dependent failure, rock strength, Bushveld Complex.

Introduction South Africa is a major mining country and is host to a significant proportion of the world’s mineral resources. Gold resources in South Africa occur mainly in the Witwatersrand Basin, and platinum group metals (PGMs) in the Bushveld Complex (BC). The nation hosts most of the world’s mineral reserves of platinum and palladium, about 75% and 50% respectively (Cawthorn, 1999), and future production potential is estimated to be more than a hundred years. Furthermore, Cawthorn (1999) suggests that, since mining of PGMs has progressed only to an average depth of 2000 m below surface, the proven reserves may easily double as deeper exploration and mining take place.

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

The 2060 Ma Bushveld Complex is an irregular, saucer-shaped massive layered igneous intrusion (Figure 1), with outcrop extremities of approximately 450 km east–westt and 300 km north–south (Simmat et al., 2006). The platinum reserves occur in three horizons: the Merensky Reef, the Upper Group 2 (UG2) Reef, and the Platreef (Cawthorn, 1999; Cawthorn and Boerst, 2006). Below these reef horizons lies the Upper Group 1 (UG1) Reef, the platinum content of which has not yet been widely proven to be economically viable. The continuity of the Merensky and UG2 reefs has been confirmed to 3000 m below surface (Cawthorn, 1999). Potholes, faults, and dykes in the Merensky and UG2 reefs disrupt the otherwise uniformly shallow dipping and narrow tabular reef characteristics peculiar to the BC mines.

The main rock types associated with the Merensky and UG2 reefs are gabbro, norite, anorthosite, and pyroxenite. ‘The [Merensky] reef in its most common form is a pegmatoidal (coarse-grained) feldspathic pyroxenite, generally bounded by thin (approx. 20 mm) chromitite layers. The immediate hangingwall is pyroxenite, 1–5 m thick, which grades upwards through norites to anorthosites. The footwall generally consists of various types of norite and anorthosite, and less commonly feldspathic pyroxenite or harzburgite, which however often forms the immediate footwall of pothole reefs’ (Rangasamy, 2010). Anorthosite is an important rock type in the BC. As indicated by Barnes and Maier (2002),

* Anglo American Platinum Ltd and School of Mining Engineering, University of the Witwatersrand. † School of Mining Engineering, University of the Witwatersrand. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12–14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

765

▲

Synopsis

Time-dependent tensile strengths of Bushveld Complex rocks 2007). Damage in excavation walls in BC mines is exacerbated by the intersection of fractures with naturallyoccurring, shallow-dipping discontinuities and layered rock, resulting in the formation of blocks of rock with high fallout and unravelling potential. Instability problems in inclined shafts and in dip-oriented tunnels have occurred in some mines in the Rustenburg mining environment, as evidenced by observed roof failures and resulting ‘gothic arches’ in tunnels oriented on-dip (see Figure 4).

Figure 1—Geological map of the Bushveld Complex (after Viljoen and Schürmann, 1998)

anorthosite is a brittle rock, and can therefore be expected to be prone to failure under tensile stress and extension conditions. They state, ‘Mottled anorthosite refers to anorthosite in which large areas of inter-cumulus orthopyroxene and/or augite (from 10 mm diameter up to the diameter of tennis balls) form dark mottles in a matrix of pure white or pale grey anorthosite. Spotted anorthosite is anorthosite in which a small percentage of cumulus orthopyroxene gives the effects of dark spots in the pale anorthosite matrix.’

In situ stress conditions are very important influences on the behaviour of mining excavations in the BC. Relevant information from a database of in situ stress measurements across southern Africa (Stacey and Wesseloo, 2004) are summarized in Figure 2. High horizontal to vertical stress ratios (2.5–4 in the pseudo-strike direction) are commonly experienced at shallow depths in BC mines. At greater depths of about 1000 m, the stress ratios are lower, in the region of 1 to 1.5.

Mining operations in the Bushveld Complex Underground mines contribute most of the PGM production in South Africa, the bulk of the ore currently being mined at depths between 500 m and 2000 m below surface. Primary and secondary excavations are mined to access the mineral reserves, and these have to remain open and stable for the life of the mine. In shallow BC mines, the compressive strength of the rock (UCS) is usually much greater than the compressive stress in the excavation walls. In these stress conditions, failure of rock would not be expected. However, stress-induced spalling of rock from walls of mining excavations is frequently observed (Ryder and Jager, 2002).

Figure 3 illustrates this behaviour in a haulage tunnel at a depth of approximately 400 m below surface.

The walls of the haulage were observed to scale with the passage of time, implying time-dependent behaviour of the wallrock. This has been observed in the BC mines, between months and years after excavation, due to fracture initiation and propagation in intact rock. Time-dependent stope closure behaviour in BC stopes has also occurred (Malan et al.,

▲

766

OCTOBER 2014

VOLUME 114

Figure 2—In situ stress data for South African mines (Stacey and Wesseloo, 2004)

Figure 3—Spalling in a haulage tunnel in a platinum mine, western limb, Bushveld Complex

Figure 4—Stress-induced failure in the roof of a dip-oriented tunnel in a platinum mine (Stacey and Wesseloo, 2004) The Journal of The Southern African Institute of Mining and Metallurgy

Time-dependent tensile strengths of Bushveld Complex rocks Since rock strengths appear to be considerably greater than rock stresses, the observations indicate that rock failure might be considered to be somewhat unexpected. However, in other situations, rock in the walls of excavations has been observed to fracture or fail at stress levels well below the UCS, as found by Stacey and Yathavan (2003). They reviewed published information on the development of fractures at low stress levels in rock (Grimstad and Bhasin, 1997; Myrvang et al., 2000). This review showed that stressinduced failure can occur even when the maximum induced stresses are as low as one-quarter to one-half of the rock strength.

Ortlepp (1997) observed a rockburst in a sandstone roof of a shallow coal mine about 20 m below surface. These findings point to a different failure mechanism than the mechanisms commonly assumed in the Mohr-Coulomb and Hoek-Brown failure criteria. Observations of face-parallel slabbing or spalling at low confining stress suggest extension as a fracturing mechanism.

Tensile stresses occur in the immediate sidewalls off the pillar, and in the stope hangingwall. The immediate peripheries of the excavations experience substantial zones of extension. Extension strains that exceed a critical extension strain value can indicate the initiation of fracturing in rock (Stacey, 1981). These fractures form in planes normal to the direction of extension strain, which corresponds with the direction of minimum principal stress (the least compressive principal stress). Importantly, extension can occur in an environment in which all three principal stresses

Stress and strain distributions around typical mining excavations Since failure of rock around mining excavations at shallow depth is commonly observed, it was considered appropriate to carry out stress analyses to evaluate the magnitudes of stress and strain that might theoretically be expected. Numerical analyses of typical mining excavations, a stope and a stope pillar, were therefore carried out using k-ratios of 1 and 2, characteristic of the deeper and shallow mining respectively.

Two mining depths were considered, 500 m and 1000 m below surface. Examples of computed minimum principal stresses (σ3), including orientations, and minimum principal strains, around a typical in-stope pillar are shown in Figures 5 and 6.

Distributions of minimum principal stresses and strains around a typical stope excavation were also determined.

Examples of the minimum principal stress and minimum principal strain distributions are shown in Figures 7 and 8.

Figure 6—Distribution of minimum principal strain around a stope pillar at a depth of 500 m, k-ratio of 2

Figure 7—Minimum principal stress around a stope at a depth of 500 m, k-ratio of 2

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 8—Minimum principal strain around a stope at a depth of 500 m, k-ratio of 2 VOLUME 114

OCTOBER 2014

767

▲

Figure 5—Distribution of minor principal stress (σ σ3) around a pillar at a depth of 500 m, k-ratio of 2

Time-dependent tensile strengths of Bushveld Complex rocks are compressive. It is to be noted that, in the calculation off the minimum principal strain, the value of σ2 used was based on the assumption of plane strain.

Observed hangingwall delamination has been attributed to the layered nature of the rocks. However, the effect of low confinement, as well as tensile and extension conditions, could be a direct or a contributory cause. The models indicate zones of extension strain and principal stress orientations that are compatible with the geometry of spalling observed in the excavation wall rock.

The photograph in Figure 9 shows a platinum mine stope that had been standing for at least 6 months. The hangingwall of the stope delaminated and face-parallel fractures, corresponding with the modelled principal stress orientations, developed slowly at stresses lower than the compressive or tensile strengths of the host rock types. Loose blocks form when the fractures propagate and intersect natural discontinuities, resulting in unravelling around support. Installed support in these conditions curbs the propagation of fractures, and slows down the manifestation of excavation wall damage.

Time-dependent behaviour of rock When an excavation is mined, stress redistribution occurs around the opening, and a change in the stress field can result in significant deformation, or creep, occurring over a relatively long period of time. Creep is defined as increasing strain while the stress is held constant (Rinne, 2008) and is observed mainly in soft rocks, for example salt. However, all types of hard rock also exhibit creep characteristics over long enough time intervals (Critescu and Hunsche, 1998).

Creep commonly consists of three stages: after an instantaneous elastic strain when a constant load is applied, primary (transient) creep occurs; then, with time, secondary or steady state creep occurs; finally, tertiary or accelerating creep will occur, leading to eventual failure. The tertiary stage always terminates in fracture and establishes the link with the phenomenon of time-dependent failure (Wawersik, 1972). Drescher and Handley (2003) observed these creep stages when they carried out uniaxial compression creep tests on Ventersdorp lava and Elsburg quartzite.

According to Ryder and Jager (2002), the long-term strength of rock can be as low as 70% of the UCS. However,

tests on granite and anorthosite by Schmidtke and Lajtai (1985) showed that stresses as low as 50% of the short-term strength almost certainly caused time-dependent stresscorrosion cracking in brittle rocks, severe enough to cause delayed failure. Investigations of crack growth in loaded granite using a scanning electron microscope indicated that new cracks developed continuously under constant load (Kranz, 1976). These findings point towards the importance of including time-dependent behaviour in the design of excavations in rock.

Ryder and Jager (2002) state that the creep rate in rock is dependent on the magnitude of the deviatoric stress (σ1 - σ3) and not the individual magnitudes of σ1 and σ3, (where σ1 and σ3 are the major and minor principal stresses respectively). However, orientations of fracturing due to deviatoric stresses would not correspond with the observed underground spalling or slabbing behaviour. In contrast, observations of failure, and the results of the numerical analyses described above, which show that large zones of rock around excavations are likely to be in a condition of extension, indicate that extension fracturing is a more likely mechanism. An investigation into the time-dependent characteristics of BC rock types was therefore considered to be justified.

Laboratory tests on Bushveld Complex rocks Few studies have been conducted on the time-dependent behaviour of strong brittle rocks. Of significance for South African mining conditions is the testing described by Bieniawski (1967c; 1970), Kovács (1971), Drescher (2002), Drescher and Handley (2003), and Watson et al. (2009).

However, these publications provide very little information on the time-dependent properties of BC rock types. Appropriate testing was therefore necessary to provide data for comparison with the results of the numerical analyses.

A range of laboratory rock strength tests was carried out on several BC rock types: UCS tests to establish the elastic properties and failure characteristics under uniaxial compression and, from the UCS test results, interpretation of long-term strength and stress and strain values observed at failure; Brazilian indirect tensile (BIT) tests (ISRM, 2007) to determine normal tensile strengths; and time-dependent BIT tests.

The following methodology was used to achieve the objectives outlined above: ➤ UCS tests on cylindrical specimens of several BC rock types prepared from different depth sections along a single vertical drill-hole core ➤ Normal BIT tests on several BC rock types ➤ Time-dependent constant hold-load BIT tests on several BC rock types stressed to pre-determined hold-load levels. The times-to-failure for the different test categories at constant load were recorded for the determination of time-dependent characteristics.

Figure 9—Delamination of hangingwall in a platinum mine stope

▲

768

OCTOBER 2014

VOLUME 114

The UCS and BIT test specimens were prepared from the core of a single exploration drill-hole. The vertical borehole gives a good cross-sectional representation of the BC layering. Drill-hole core samples were taken from a zone up to 10 m above and below the hangingwall (HW) and footwall (FW) contacts of the Merensky (MR), Upper Group 1 (UG1). and Upper Group 2 (UG2) reef horizons.

The Journal of The Southern African Institute of Mining and Metallurgy

Time-dependent tensile strengths of Bushveld Complex rocks Nine test specimen categories were identified: f spotted anorthosite, mottled anorthosite (A, D, and I), pyroxenite, norite, anorthositic norite, and spotted anothositic norite (F and H).

UCS and BIT test specimens were cut alternately from the core to provide an unbiased sample representation for the two test methods. No two test specimens for the same test method were cut adjacent to one another. In total, 334 specimens were tested. The majority of test specimens were prepared for the BIT tests, and most of these were used in the time-dependent tests. For each test type, the numbers of specimens that were tested successfully with valid results varied.

All specimens were prepared and tested according to the ISRM Suggested Methods for rock testing (ISRM, 2007). The cores were BQ size with a diameter D = 36.3 mm. Average length to diameter ratios of 2.2 and 0.5 were achieved for the UCS and BIT test specimens respectively. Inspection showed that the specimens did not have visible pre-existing deformities. Circumferential and axial strain gauges were attached to the UCS test specimens to measure strains during the tests. All tests were carried out in anhydrous conditions at room temperature and pressure.

The data from the test results was processed and used to determine UCS values, and to plot stress-strain curves, similar to that shown in Figure 10. The average values of elastic properties of mottled anorthosite (A) are indicated on the plot. After initial nonlinear behaviour (possibly crack closure, Bieniawski, 1967a; 1967b), the plot shows largely linear behaviour up to the peak strength of the specimen.

The specimens failed in the typical fashion observed for brittle rock failure – initial axial extension fractures and ultimately shear, resulting in some cases, in conical endpieces and a completely fractured or crushed middle portion.

Axial stress-volumetric strain plots were used to evaluate the ‘long-term strength’ of the rock types. According to Bieniawski (1967a), the ‘nose’ of the stress-volumetric strain plot marks the ‘long-term strength’ of the specimen. Longterm strengths were determined for each rock type and average values calculated. Variations in the test results are attributed to inherent variability in the rock specimens. A

summary off the UCS test results, together with the ‘long-term strength’ values, is given in Table I. The average value of the ‘long-term strength’ for the nine categories of rock types was 78 MPa, which is 56.4% of the UCS value of the rock types tested. The lowest long-term strength values were recorded for pyroxenite and mottled anorthosite at 44% of their respective UCS values.

Brazilian indirect tensile (BIT) tests A servo-controlled rock testing machine was used to carry out normal and time-dependent BIT tests. A constant loading rate of 2 kN/min was used to load specimens, targeting failure in 3 to 4 minutes, depending on the sample’s tensile strength. A summary of the results of the normal Brazilian tensile strength tests is presented in Table II (with elastic modulus values obtained from the UCS tests). On average, the tensile strength magnitude was found to be 5% (1/20) of the UCS of the same rock type.

Typical strain values at failure for each rock type were calculated based on the average elastic modulus of the rock type, and ranged from 0.16 to 0.21 millistrain, with an average value of 0.18 millistrain. This range agrees with published data; for comparison, the value obtained for norite was 0.173 millistrain (Stacey, 1981).

Figure 10—Example of a stress-strain graph for mottled anorthosite

Table I

Summary of average results from UCS tests

Sample diameter, D (mm) Sample length, L (mm) L/D ratio Sample mass, M (g) Sample density, ρ (kg/m3) Failure load, (kN) UCS, σc (MPa) Elastic modulus, E (GPa) Poisson’s ratio, v Long-term strength (MPa) % of UCS

Mottled anorthosite (A)

Spotted anorthositic (B)

Pyroxenite (C) norite

Mottled anorthosite (D)

Norite (E)

Spotted anorthositic norite (F)

36.30 80.74 2.23 231.41 2769.46 180.60 174.51 44.60 0.20 90.2 57

36.30 84.79 2.34 254.20 2898.71 139.40 134.70 33.32 0.21 61.8 46

36.30 81.66 2.25 270.30 3198.41 129.80 125.42 35.49 0.17 56.5 44

36.30 81.13 2.24 230.93 2750.49 140.50 135.76 39.01 0.28 59.75 44

36.30 83.71 2.29 261.32 3016.40 96.00 92.76 30.90 0.19 53.5 57

36.30 82.87 2.28 248.10 2892.39 154.60 149.38 40.65 0.21 75.6 51.4

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

Anorthositic Spotted Mottled norite anorthosite anorthosite (G) (H) (I) 36.30 80.99 2.23 253.48 2990.22 114.00 110.15 37.90 0.15 83.6 72.4

36.30 81.03 2.23 237.80 2835.12 159.60 154.22 42.64 0.22 103.33 67

OCTOBER 2014

36.30 80.98 2.23 232.32 2772.17 182.20 176.05 45.31 0.19 125.75 68.75

769

▲

Rock type/ Code

Time-dependent tensile strengths of Bushveld Complex rocks Table II

Summary of normal BIT test results Rock type

Sample Sample Sample t/D Sample Average load Average BIT Average elastic Average strain Average timeID diameter thickness, ratio mass, at failure, strength, modulus, at failure, to-failure D (mm) t (mm) M (g) P (kN) Ďƒt (MPa) E (GPa) (millistrain) (s)

Mottled anorthosite (A) Spotted anorthositic norite (B) Pyroxenite (C) Mottled anorthosite (D) Norite (E) Spotted anorthositic norite (F) Anorthositic norite (G) Spotted anorthosite (H) Mottled anorthosite (I)

DBA DBB DBC DBD DBE DBF DBG DBH DBI

36.30 36.30 36.30 36.30 36.30 36.30 36.30 36.30 36.30

19.27 18.88 18.97 18.77 18.36 18.61 17.65 17.59 17.06

0.53 0.52 0.52 0.52 0.51 0.51 0.49 0.48 0.47

54.26 55.46 62.67 53.84 57.67 55.24 56.31 51.29 48.93

8.19 6.84 7.43 6.83 6.92 8.27 7.71 7.11 6.82

7.46 6.35 6.89 6.38 6.62 7.76 7.65 7.10 7.04

46.72 33.32 35.40 39.01 30.90 40.65 37.90 42.64 45.31

0.16 0.19 0.19 0.16 0.21 0.19 0.20 0.17 0.16

220.71 205.72 206.77 138.02 160.35 138.17 203.85 213.83 214.03

Table III

Time-dependent test results Rock type . specimen I.D A B C D E F G H I

Static BIT test load, x% of Pmean (kN) and time-to-failure, T (s)

Mean BIT strength, Pmean (kN)

90%

Time, T (s)

85%

Time, T (s)

80%

Time, T (s)

75%

Time, T (s)

70%

Time, T (s)

8.19 6.84 7.43 6.83 6.92 8.27 7.71 7.11 6.82

7.37 6.16 6.69 6.15 6.23 7.44 6.94 6.40 6.14

268 1587 16830 229 1652 643 4214 6213

6.96 5.81 6.32 5.81 5.88 7.03 6.55 6.04 5.80

972 2200 38695 16328 375 375 147 12291 45191

6.55 5.47 5.94 5.46 5.54 6.62 6.17 5.69 5.46

11658 4578 33705 82700 10327 1705 8168 33038

6.14 5.13 5.57 5.12 5.19 6.20 5.78 5.33 5.12

39831 19837 207423 36306 1679 2698 2483 67109 38993

5.73 4.79 5.20 4.78 4.84 5.79 5.40 4.98 4.77

111480 62226 23605 151889 60706 60709 805 62350 39151

A Mottled anorthosite; B Spotted anorthositic norite; C Pyroxenite; D Mottled anorthosite; E Norite; F Spotted anorthositic norite; G Anorthositic norite; H Spotted anorthosite; I Mottled anorthosite

Time-dependent Brazilian indirect tensile (BIT) strength tests Pre-determined load levels, derived from the tensile strength values determined in the normal BIT tests, were used in the time-dependent BIT strength tests. The load levels used represented 70%, 75%, 80%, 85%, and 90% of the corresponding tensile strength for each rock type: At each load level, tests were conducted on sets of five specimens for each test category. Loading of each test specimen was increased at a rate of 2 kN/min up to the required hold-load.

The time-to-failure, T T(s), was recorded for complete test runs where specimen failure was observed. Owing to limited availability of the testing machine, the time-dependent tests were limited to a maximum of three days. In a few cases, particularly in the low load tests at 70%, failure did not occur within three days, and in some cases the testing machine tripped due to overheating. Valid test results were therefore recorded only if the initial loading build-up was completed to the hold stage, the machine did not trip, and the test was completed within three days.

Some of the tests, particularly those at 90% of tensile strength, resulted in failure during the load application stage, i.e. before reaching the constant load phase. Other test runs with similar premature failure results were attributed to rock material variability.

â–˛

770

OCTOBER 2014

VOLUME 114

The time-dependent test results ffor the nine rock types are summarized in Table III. The individual test results showed scatter in the time-tofailure. A logarithmic trend line was fitted to the results for each type category, as shown in Figure 11. The minimum value indicated by the curve may be taken as the long-term tensile strength of the rock type. Similar graphs were produced for all the rock types to determine the time–tofailure trends.

It may be estimated from these results that the long-term tensile strengths of BC rocks are likely to be between 60% and 70% of the short-term tensile strength normally reported for laboratory tests. Since all mining excavations are longterm in the context of the duration of testing carried out, these lower strength values should be taken into account in design.

From these results the equivalent extension strains at failure were calculated, and these indicate that an average value for the critical extension strain is likely to be approximately 0.12 millistrain.

Summary of laboratory test results From the range of strength tests carried out, the following key outputs have been summarized: a) Average UCS values obtained for the BC rocks tested varied between 93 MPa and 176 MPa The Journal of The Southern African Institute of Mining and Metallurgy

Time-dependent tensile strengths of Bushveld Complex rocks

Figure 11—Time-to-failure plot for mottled anorthosite

The extension strain magnitudes determined in the numerical modelling are significantly greater than those indicated in (e) above, the implication being that extension strain may be a suitable criterion for prediction of fracture and failure around BC mining excavations. Observations made in actual BC mine excavations revealed that fracturing of intact rock occurs over a protracted time, possibly due to the development and propagation of extension fractures, and that the manifestation of such fracturing was curbed by installed support.

Conclusions The research described in this paper has dealt with the investigation of stress and strain conditions influencing the spalling of wallrock in mine excavations in the Bushveld Complex (BC). This involved laboratory testing of BC rocks in uniaxial compression and in indirect tension, including timedependent indirect tension, as well as numerical modelling of typical mine excavations. The following conclusions are drawn: ➤ Observations made in BC mine excavations revealed that fracturing of intact rock occurs over a protracted time period, and that its manifestation is curbed by installed support ➤ There have been very few time-dependent or creep The Journal of The Southern African Institute of Mining and Metallurgy

➤

➤

➤

➤

➤

tests carried out in South Africa f on BC rock types. The laboratory testing reported in this paper has provided new data in this regard The laboratory tests have shown that tensile strength magnitudes of BC rock types are approximately 5% of their compressive strength magnitudes The long-term uniaxial compressive strength of the BC rocks, interpreted from the axial stress-volumetric strain graph from the UCS test, is on average 78 MPa or 56% of the average UCS value The tensile strength of the BC rock types was found to be time-dependent. Failure times for individual test specimens showed large variability, but the general indication is that the time-dependent tensile strengths are between 60% and 70% of the tensile strength normally reported from laboratory testing, and possibly may even be less than 60% Extension strains calculated at tensile strength failure ranged between 0.16 and 0.21 millistrain. Values corresponding with the long-term tensile strength are less than 70% of this range, namely, less than 0.11 to 0.15 millistrain Numerical analyses of BC excavations were carried out, using elastic models and assuming homogeneity of material, to investigate the possible occurrence of zones in which tensile stresses and extension strains occur.

The models showed that large zones of extension strain may occur around BC excavations, and that the magnitudes of the extension strain exceed the critical values determined from the laboratory testing.

Predicted orientations of fracturing from these models correspond with observed geometry of spalling in excavations. The implication is that there are likely to be substantial zones surrounding BC mine excavations that will be prone to spalling conditions and perhaps more significant failure.

Acknowledgements Impala Platinum is thanked for provision of the borehole core and associated information on which the research described in this paper was based. Joseph Muaka is thanked for assistance with numerical analyses specifically for the paper.

The input of the second author is based on research supported in part by the National Research Foundation of South Africa (Grant-specific unique reference number (UID) 85971). The Grantholder acknowledges that opinions, VOLUME 114

OCTOBER 2014

771

▲

b) The average long-term strength off the rocks (Bieniawski, 1967a), interpreted from the volumetric strain curves in the UCS tests, was 78 MPa, which is 56% of the UCS. The lowest long-term strength value obtained was 44% of the corresponding UCS value c) Tensile strength magnitudes of the rocks were found to be between 4% and 7% of the UCS magnitudes (i.e. the tensile strength magnitude is about 1/20 of the UCS magnitude) d) The minimum long-term tensile strengths of the rocks could not be determined owing to the limited testing duration of three days, but are certainly less than 70% of the short-term tensile strength normally reported for laboratory tests e) Extension strain magnitudes at strength failure calculated from the normal tensile strength tests indicate a range of between 0.16 and 0.21 millistrain.

Values corresponding with the long-term tensile strength would therefore be less than 70% of this range, i.e. 0.11 to 0.15 millistrain.

Time-dependent tensile strengths of Bushveld Complex rocks findings, and conclusions or recommendations expressed in any publication generated by the NRF-supported research are that of the author(s), and that the NRF accepts no liability whatsoever in this regard.

Geomechanical Abstracts. vol. 16. pp. 23–35.

MALAN, D.F., NAPIER, J.A.L., and VAN JANSE RENSBURG, A.L. 2007. Stope deformation measurements as a diagnostic measure of rock behaviour: a decade of research. Journal of the Southern African Institute of Mining and Metallurgy, vol 107, no. 11. pp. 743–765.

References BARNES, S-J. and MAIER, W.D. 2002. Platinum-group element distribution in the

MYRVANG, A.M., ALNAES, L., HANSEN, S.E., and DAVIK, K.I. 2000. Heavy spalling

Rustenberg layered suite of the Bushveld Complex South Africa. The

problems in road tunnels in Norway – long time stability and performance

Geology, Geochemistry, Mineralogy and Mineral Beneficiation of

of sprayed concrete as rock support. Proceedings of the International

Platinum-Group Elements. Special Volume 54, Canadian Institute of

Symposium on Rock Support – Applied Solutions for Underground

Mining and Metallurgy, Ottawa, Ontario. pp. 431–458.

Structures, Lillehammer, r Norway. Broch, E., Myrvang, A., and Stjern, G. (eds). Norwegian Society of Chartered Engineers. pp. 751–764.

BIENIAWSKI, Z.T. 1967a. The mechanism of brittle fracture of rock. International Journal of Rock Mechanics and Mining Sciences, vol. 4. pp. 396–435.

ORTLEPP, W.D. 1997. Rock fracture and rockbursts – an illustrative study. South African Institute of Mining and Metallurgy, Johannesburg. 255 pp.

BIENIAWSKI, Z.T. 1967b. Mechanism of Brittle of Rock, Part 2. South African Council for Scientific and Industrial Research (CSIR). Institute –

RANGASAMY, T.R. 2010. IMP03: Geotechnical report, Impala 18#. Unpublished company report, Middindi Consulting (Pty) Ltd.

Technology and Engineering, g Pretoria. p. 452.

BIENIAWSKI, Z.T. 1967c. Mechanism of Brittle of Rock, Part 3. South African Council for Scientific and Industrial Research (CSIR). Institute –

RINNE, M. 2008. Fracture mechanics and subcritical crack growth approach to model time-dependent failure in brittle rock. PhD thesis, Faculty of Engineering and Architecture, Dept. of Civil and Environmental

Technology and Engineering, g Pretoria. p. 452.

Engineering, Helsinki University of Technology, Helsinki.

BIENIAWSKI, Z.T. 1970. Time-dependent behaviour of fractured rock. Rock RYDER, J.A. and JAGER, A.J. 2002. A Textbook on Rock Mechanics for Tabular

Mechanics, vol 2. pp 123–137.

Hard Rock Mines. Safety in Mines Research Advisory Committee CAWTHORN, R.G. 1999. The platinum and palladium resources of the Bushveld

(SIMRAC), Johannesburg.

Complex. South African Journal of Science, vol. 95, no. 11/12. p. 481–489.

SCHMIDTKE, R.H. and LAJTAI, E.Z. 1985. The long-term strength of Lac du Bennet CAWTHORN, R.G. and BOERST, K. 2006. Origin of the pegmatite pyroxenite in the Merensky unit, Bushveld Complex South Africa. Journal of Petrology,

granite. International Journal of Rock Mechanics and Mining Sciences and Geomechanical Abstracts, vol. 22. pp. 461–465.

vol. 47, no. 8. pp. 1509–1510.

SIMMAT, C.M., HERSELMAN, P. LE R., RÜTSCHLIN, M., MASON, I.M., and CLOETE, J.H. CRITESCU, N. D. and HUNSCHE, U. 1998. Time Effects in Rock Mechanics. Series in

2006. Remotely sensing the thickness of the Bushveld Complex UG2

Materials Modeling and Computation. Wiley, Chichester, UK.. ISBN 0471

platinum reef using borehole radar. Journal of Geophysics and

955175.

Engineering, g vol. 3. pp. 43 – 49.

DRESCHER, K. 2002. An investigation into the mechanics of time-dependent

Stacey, T.R. 1981. A simple extension strain criterion for fracture of brittle

deformation of hard rocks. MSc dissertation, Faculty of Engineering,

rock. International Journal of Rock Mechanics and Mining Sciences and

University of Pretoria, Pretoria.

Geomechanical Abstracts, vol. 18, no. 6. pp. 469–474.

DRESCHER, K. and HANDLEY, M.F. 2003. Aspects of time-dependent deformation

STACEY, T.R. and WESSELOO, J. 2004. Updated stress database for South Africa.

in hard rock at great depth. Journal of the South African Institute of

Proceedings of the International Symposium on In-situ Rock Stress,

Mining and Metallurgy, vol. 103, no. 5. pp. 325.

Trondheim, Norway, 19–21 June 2006. Lu, M. Li, C.C., and Dalde, H.

GRIMSTAD, E. and BHASIN, R. 1997. Rock support in hard rock tunnels under high stress. Proceedings of the International Symposium on Rock Support

(eds.) Taylor and Francis, London. pp. 46–471.

STACEY, T.R. and YATHAVAN, K. 2003. Examples of fracturing of rock at very low

– Applied Solutions for Underground Structures, Lillehammer, Norway,

stress levels. ISRM – Technology Roadmap for Rock Mechanics. South

22–25 June 1997. Broch, E., Myrvang, A., and Stjern, G. (eds.).

African Institute of Mining and Metallurgy, Johannesburg.

Norwegian Society of Chartered Engineers. pp 504-513.

ISRM. 2007. The Complete ISRM Suggested Methods for Rock Characterisation, Testing and Monitoring: 1974-2006. Ulusay, R. and Hudson, J.A. (eds).

Compilation arranged by the ISRM Turkish National Group, Ankara, Turkey.

Resources of South Africa. Wilson, M.G.S. and Anhaeusser, C.R. (eds.).

Council for Geoscience, Pretoria. pp. 532–568.

WATSON, B.P., KUIPERS, J.S., HENRY G., PALMER, C.E., and RYDER JA. 2009.

Nonlinear rock behaviour and its implications for deeper level platinum

KOVÁCS, I.K.A. 1971. An investigation of the time-dependent behaviour of solid rock in uniaxial compression. Report MEG 1032. National Mechanical Engineering Research Institute, Council for Scientific and Industrial Research, Pretoria, South Africa.

International Journal of Rock Mechanics and Mining Sciences and

▲

OCTOBER 2014

mining. Journal of the Southern African Institute of Mining and Metallurgy, vol. 109, no. 1. pp. 5–9.

Wawersik, W.R. 1972. Time-dependent rock behaviour in uniaxial compression. Proceedings of the 14th US Symposium on Rock Mechanics

KRANZ, R.L. 1976. Crack growth and development during creep of Barre granite.

772

VILJOEN, M J. and SCHÜRMANN, L.W. 1998. Platinum-group metals. The Mineral

VOLUME 114

(USRMS), University Park, Pennsylvania, 11–14 June, 1972, American Society of Civil Engineers. pp. 85–106.

◆

The Journal of The Southern African Institute of Mining and Metallurgy

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts

by B.G. Tarasov*

This paper proposes the further development of a recently identified shear rupture mechanism (fan mechanism) that elucidates a paradoxical feature of hard rocks – the possibility of shear rupture propagation through a highly confined intact rock mass at shear stresses that can be significantly less than frictional strength. In the fan mechanism, failure is associated with consecutive creation of small slabs (known as ‘domino blocks’) from the intact rock in the rupture tip, driven by a fan-shaped domino structure representing the rupture head. The fan head combines such unique features as extremely low shear resistance, self-sustaining stress intensification, and self-unbalancing conditions. Consequently, the failure process caused by the mechanism is inevitably spontaneous and violent. Physical and mathematical models explain unique and paradoxical features of the mechanism, which can be generated in primary ruptures and segmented faults. The fan mechanism provides a novel point of view for understanding the nature of spontaneous failure processes, including shear rupture rockbursts. The process explains, in particular, features of shear rupture rockbursts such as activation at great depths, generation of new shear ruptures in intact rock mass, nucleation of hypocentres at significant distances from the excavation, shear rupture development at low shear stresses, and abnormal rupture violence.

Keywords rock strength, failure at confined compression, shear rupture mechanism, structure of shear rupture, conditions of instability, physical model, mathematical model, shear rupture rockburst.

Introduction It has been observed in field and laboratory conditions that failure of intact hard rocks at highly confined compression can be accompanied by abnormal violence. Under both of these conditions the failure process is associated with shear rupture development.

David Ortlepp, who acquired more than 40 years of experience in the study of shear rupture rockbursts in deep and ultra-deep South African mines, emphasized this phenomenon (Ortlepp, 1997; Ortlepp et al., 2005): ‘All rockbursts, by definition, involve sudden and often violent displacement of rock.

Occasionally however, larger incidents cause damage of such intense violence that it seems that our knowledge of the mechanism of damage is completely inadequate.’ Special field studies (Gay and Ortlepp, 1979; McGarr et al., 1979) have revealed that shear ruptures causing abnormally violent rockbursts are created in intact rock mass. An important The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

ffeature is that they nucleate in zones off highly confined compression that are some distance away from excavation (on the excavation surface the minor stress is equal to zero). It was shown that these mine tremors and earthquakes share the apparent paradox of failure at low shear stresses, while laboratory measurements indicate high material strengths (McGarr et al., 1979).

Recent laboratory studies of post-peak failure of hard rocks (characterized by uniaxial compressive strength above 250 MPa) at highly confined compression (σ1 > σ2 = σ3 when σ3 > 50 MPa) support Ortlepp’s idea about inadequate understanding of the failure mechanism at these loading conditions (Tarasov, 2008, 2010; Tarasov and Randolph, 2008, 2011). Some observed abnormalities that cannot be explained on the basis of conventional approach are presented in Figures 1 and 2.

Figure 1 shows two sets of generic stressstrain curves for different levels of confining pressure σ3. Figure 1a represents the conventional (well-studied) rock behaviour associated with increasing post-peak ductility with rising σ3. For clarity, the variation of the post-peak curves is indicated by dotted lines. Figure 1b represents the unconventional type of rock behaviour. Here, increasing σ3 can lead to a contradictory variation of post-peak properties.

In fact, rock behaviour can be changed from Class I to extreme Class II and then to Class I again. Class I is characterized by a negative post-peak modulus M = dσ d /d∈ d , and Class II by positive (Wawersik and Fairhurst, 1970). At extreme Class II behaviour, values of postpeak modulus M and elastic modulus E = d /d∈ dσ d can be very close, indicating extremely small post-peak rupture energy (compare

* University of Western Australia. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12–14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

773

▲

Synopsis

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts

Figure 1—Two sets of generic stress-strain curves for different levels of confining pressure σ3 illustrating (a) conventional and (b) unconventional rock behaviour. (c) Typical variation of the post-peak brittleness index K with rising σ3 for rocks exhibiting the conventional and unconventional behaviour

shaded areas in Figures 1a and 1b for σ3 = σ3(4)).

A small post-peak rupture energy in turn indicates high post-peak brittleness. A special brittleness index was developed to characterize unambiguously the post-peak brittleness at any type of rock behaviour (see details in We = (M ( Tarasov and Potvin, 2013). The index K = dW Wr/dW E)//M is based on the ratio between the post-peak rupture We withdrawn from the energy dW Wr and elastic energy dW material during the failure process. The index K characterizes the capability of the rock for self-sustaining failure due to the elastic energy available from the failing material. Figure 1c shows variation of the brittleness index K with rising confining pressure σ3 for rocks exhibiting conventional and unconventional behaviour. In contrast to the conventional behaviour, where increasing σ3 is accompanied by a monotonic decrease in post-peak brittleness, the brittleness variation for unconventional behaviour follows a typical pattern of initially increasing, reaching a maximum, and then ultimately decreasing. The harder the rock, the greater the effect of embrittlement. Experiments (Tarasov, 2010) showed that some rocks at high confinement became hundreds of times more brittle compared to their behaviour under uniaxial compression.

Figure 2 illustrates the abnormal violence of hard rock failure at extreme Class II behaviour. The experiments were conducted on an extremely stiff servo-controlled testing machine based upon the loading principles described in Stavrogin and Tarasov (2001). Figure 2a shows a set of stress-strain curves for dolerite (uniaxial compressive strength 300 MPa) obtained at different levels of σ3. At σ3 < 60 MPa the total post-peak control was provided for both Class I and Class II behaviour. Dotted lines here indicate general orientation of post-peak curves. At σ3 ≥ 60 MPa control was possible only at the start of the post-peak stage, after which spontaneous and violent failure took place.

Dotted lines indicate orientation of post-peak curves at the moment that instability starts. In this case M is close to E, E post-peak rupture energy is vanishingly small, and post-peak brittleness approaches absolute brittleness (extreme Class II).

▲

774

OCTOBER 2014

VOLUME 114

To demonstrate the difference ff in violence at spontaneous failure for Class II at low σ3 (where post-peak control is possible) and for extreme Class II at high σ3 (where postpeak control is impossible), some special experiments were conducted. At low σ3 the spontaneous failure was generated at the peak stress due to the absence of post-peak servocontrolling. During failure at all levels (low and high) of σ3 the differential stress variation with time was recorded by a load cell adjoining the tested specimens (Figure 2b).

Two different modes of rock behaviour were distinguished.

Figure 2c shows a stress-time curve typical for σ3 < 60 MPa, while Figure 1d shows a stress-time curve typical for σ3 ≥ 60 MPa. It should be emphasized that for the two curves obtained at σ3 = 30 and 60 MPa the stress drop Δσ (the difference between the stress of the instability start σA and the residual strength σf) was practically the same: Δσ(30) = 310 MPa and Δσ(60) = 340 MPa. Points of instability are marked by asterisks on the stress-strain curves in Figure 2a and stresstime curves in Figures 2c and 2d. Despite the fact that the elastic energy available from the specimen-loading machine is comparable for both experiments, the shapes of the curves differ dramatically. For σ3 = 30 MPa the failure was followed by a conventional stress oscillation around the residual strength. For σ3 = 60 MPa an extraordinary post-failure stress shock was generated, after which the equilibrium condition was reached at a stress level significantly below the residual (frictional) strength σf. Identical stress shocks were observed in all experiments conducted at σ3 ≥ 60 MPa (Tarasov and Randolph, 2008).

The observed features of hard rocks, such as the dramatic post-peak embrittlement with rising confining pressure σ3; the abnormal failure violence within a certain range of σ3; and the huge after-failure stress-shock cannot be explained according to the basics of common understanding of shear rupture mechanisms. This paper shows that the observed abnormalities are generated by a recently identified shear rupture mechanism (fan mechanism) that is activated in hard rocks (uniaxial compressive strength above 250 MPa) at highly confined compression. In the fan mechanism, the rock

Figure 2— (a) Stress-strain curves for dolerite specimens tested at different levels of confining pressure σ3, (b) schema of a specimen with adjoining load cell, (c) and (d) typical stress-time curves recorded by the load cell during spontaneous failure of rock specimens at low and high levels of σ3 The Journal of The Southern African Institute of Mining and Metallurgy

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts

Fan mechanism in primary ruptures Frictional and fan-hinged shear This section discusses the interrelation between well-known failure mechanisms for rocks at confined compression and the fan mechanism. It is known that rock failure mechanisms are dependent on the level of confining pressure σ3 (e.g.

Kirby and McCormick, 1984). Figures 3a and 3b show variations in failure mechanisms, with σ3 rising from left to right, for rocks exhibiting conventional and unconventional behaviour. Rectangles represent rock specimens exhibiting different failure mechanisms. In brittle rocks, pre-existing defects at loading generate tensile cracks, the ultimate length l of which is a function of σ3, as shown symbolically by dotted lines: the higher the σ3 the shorter l. The length l of tensile cracks in turn determines the macroscopic failure mechanism and the failure pattern.

At confining pressures σ3 < σ3min(shear), shear rupture cannot propagate in its own plane due to the creation in the rupture tip of relatively long tensile cracks preventing the shear rupture development. The tensile cracks grow along the major stress. Two failure mechanisms distinguished at these stress conditions are: (1) splitting by long tensile cracks and (2) failure due to coalescence of distributed micro-cracks accumulated within the material body during loading.

Figure 3—Variation in failure mechanisms and failure patterns with σ3 for rocks exhibiting (a) conventional and (b) unconventional behaviour The Journal of The Southern African Institute of Mining and Metallurgy

f mode is localized shear.

At σ3 ≥ σ3min(shear) the failure Due to high confinement, micro-tensile cracks become sufficiently short to cause shear rupture to propagate in its own plane. Here the dilation of one short micro-crack induces the dilation of a closely spaced neighbouring crack (Reches and Lockner, 1994). Due to the consecutive creation of short tensile cracks in front of the rupture tip, the advancing fault itself induces organized damage that is restricted to its own plane. It is important to note that micro-cracks are generated along the major stress, which is at angle αo ≈ (30°–40°) to the shear rupture plane (Reches and Lockner, 1994; Horii and Nemat-Nasser, 1985). This micro-cracking process creates inclined intercrack blocks (known as domino blocks) which are subjected to rotation at shear displacement of the rupture interfaces (Peng and Johnson, 1972; King and Sammis, 1992; Reches and Lockner, 1994). Two specific shear rupture mechanisms have been distinguished here.

Frictional shear The development of a frictional shear rupture can be controlled on stiff servo-controlled testing machines. It has been observed that intercrack domino blocks generated in the rupture tip are subjected to collapse at rotation caused by shear displacement of the rupture faces, creating frictional structureless medium (gouge) in the shear rupture interface (Peng and Johnson, 1972; King and Sammis, 1992; Reches and Lockner, 1994). Increasing confining pressure (σ3 in this case) increases friction within the total rupture zone (including the rupture head), which causes the increase in post-peak rupture energy according to the conventional rock behaviour shown in Figure 1a.

Fan-hinged shear It should be noted that the three mechanisms discussed are activated in practically all types of rock. The fourth mechanism is generated in hard rocks (characterized by uniaxial compressive strength above 250 MPa) and is responsible for the unconventional behaviour (Tarasov, 2008, 2010). Further increases of the confining pressure above σ3 = σ3min(shear) will continue reducing the length l of tensile cracks (dotted curve in Figure 3b) and, consequently, the length of domino blocks composing the fault structure.

Due to the very strong material and proper geometry of short domino blocks within the range σ3min(hinge) < σ3 < σ3max(hinge) they can withstand rotation caused by the shear displacement of the rupture faces without collapse. In this case the domino blocks behave as hinges, and due to consecutive generation and rotation they create a fan structure representing the shear rupture head. The fan structure has several extraordinary features which will be discussed further.

It should be emphasized that the efficiency of the fan mechanism is variable and determined by how perfect the fault structure is. The solid curve in Figure 3b shows symbolically the variation of the fan mechanism efficiency versus confining pressure, which determines the identical variation of post-peak brittleness of rocks exhibiting unconventional behaviour (see Figure 1c). At the low end of the hinge pressure range, when the relative length (length/thickness) of the rotating blocks is still large, the domino blocks are subjected to partial destruction (buckling) as they rotate (the role of block geometry is discussed in more detail later). At higher σ3, with shorter blocks, this imperfection decreases, rendering the fan mechanism more efficient. The optimal VOLUME 114

OCTOBER 2014

775

▲

ffailure is associated with consecutive creation off slabs (known as ‘domino blocks’) from the intact rock in the rupture tip and is driven by a fan-shaped domino structure representing the rupture head. The fan head combines such unique features as extremely low shear resistance, selfsustaining stress intensification, and self-unbalancing conditions. Consequently the shear rupture can propagate through the medium with negligible resistance, resulting in abnormal violence. Physical and mathematical models of the mechanism presented in this paper explain the unique and paradoxical features of the mechanism. This mechanism can operate in small laboratory specimens and in field conditions, causing shear rupture rockbursts and earthquakes. Natural faults normally have a very complicated multi-hierarchical segmented structure. We will firstly discuss features of the fan mechanism operation in primary ruptures (thin continuous formations), and then in complex faults (Ortlepp shears).

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts efficiency ff takes place at a confining f pressure at which the blocks rotate with minimum destruction. At greater σ3 the efficiency reduces because shorter blocks gradually lose any potential to operate as hinges. Finally, very short blocks lose this capability completely and the rock behaviour returns to the commonly accepted frictional mode.

Physical model of the fan mechanism Self-balancing fan structure As discussed above, primary shear ruptures propagate through rocks due to the consecutive formation of identical domino blocks from the intact material in the rupture tip.

Further rotation of the blocks between the shear rupture faces can lead to frictional shear (at block collapse) or to fanhinged shear (at block rotation without collapse). The mechanism responsible for the creation of identical domino blocks is not considered in the physical and mathematical models presented in this paper. The models discuss the influence of the fan structure formed on the basis of rotating blocks on the rupture process. In the models the domino blocks are considered as ‘predetermined’ and operating with optimal efficiency (without collapse at rotation).

Photographs of the physical model in Figure 4a show different stages of fan formation. At the initial condition (Figure 4a-I) a row of identical domino blocks inclined at angle α0 represents an implicit horizontal shear rupture (fault). Surfaces of neighbouring domino blocks are in full contact, providing a very compact ‘monolithic’ material. To simulate the resistance of domino blocks to tearing-off from the monolithic material (which takes place in real materials) the blocks are bonded to each other. The row of domino blocks is located between two layers of elastic material (elastic connectors) representing the fault interfaces. The upper and lower elastic connectors are fixed to corresponding ends of each domino block. Contact areas between the ends of domino blocks and the interfaces we will call joints. As such, a version of the model with bonded blocks can be treated as representing an intact material. Evenly distributed weight located on the upper layer creates normal stress applied to the simulated fault σn = σp.

Propagation of shear rupture along the potential fault can be initiated by application of a local force F to the elastic connector fixed to the top of the first domino block (Figure 4a-II). The applied force will be transmitted to the next blocks by stretching elastic connectors between them at the consecutive separation of blocks and rotation of them against joints. Due to this, the domino blocks will be sequentially (one by one) torn off from the ‘monolithic’ material, forming a fan-shaped structure. The fan has completed when the front block rotates on the total angle βtot = 180° - 2α0 at shear displacement Δ of the rupture faces. The total number of domino blocks involved in the fan structure of the physical model is about 30, while in real materials it can be thousands (Tarasov and Guzev, 2013). The fan structure with small number of domino blocks in the physical model will allow some features of the fan mechanism to be illustrated more legibly.

Figure 4b reflects experimentally determined variation of the force F applied to the first domino block during fan formation and, consequently, the variation of shear resistance of the fan structure at different stages of its formation. Such variation of the shear resistance is

▲

776

OCTOBER 2014

VOLUME 114

determined by the ffact that elementary fforces N (representing normal stress σn) applied to each domino block create horizontal componentss f N of opposite directions in the first and the second half of the fan structure (Figure 4c). Due to this, the maximum resistance is attained at the completion of the first half of the fan, while the totally completed fan structure represents the self-balancing structure with shear resistance equal to zero (details are given in the mathematical model presented later).

Shear resistance below the frictional strength – pulse-like rupture mode To make the fan structure self-unbalancing, a distributed shear stress τ should be applied to the whole domino row.

The simplest way to apply the distributed shear stress for the physical model is to incline the row by angle γ as shown in Figure 5a. The distributed weight σp in this situation creates shear stress τ = σpsinγ along the whole structure. Under the effect of distributed shear stress the fan propagates along the whole row, sequentially moving the loaded upper face against the lower one by distance Δ.

Experiments on the physical model show that the minimum angle γ at which the fan becomes unstable is about 4°. The distributed shear stress generated at this angle is just sufficient to overcome the shear resistance of the fan structure, which is determined by (1) friction in joints of rotating domino blocks and (2) the reduced resistance associated with the tearing off of each front domino block from the intact material. It was established that the resistance to shear provided by the fan structure is very low. To move the same fault faces at common frictional resistance (without fan structure) the angle γ should be about 40°. This experiment clearly demonstrates that shear resistance of the fan structure τfan can be significantly less than the common frictional strength τf. For this particular model, shear resistance (strength) of the fan structure τfan is less than the frictional strength τf: τfan ≈ 0.1τf by a factor of ten.

It should be emphasized that low shear resistance is encountered within the zone of the moving fan head only. In front of the fan the material is in an intact condition. Behind the fan, shear resistance is equal to friction. Due to this the fan mechanism provides the pulse-like rupture mode: at any given time during rupture propagation, slip occurs over only a narrow band (fan head) along the fault and the fault relocks behind the rupture head. This slip pulse propagates forward as the fault proceeds. Pulse-like rupture mode was observed for earthquakes (Healton, 1990) and in laboratory (Ohnaka et al., 1986; Lykotrafitis et al., 2006).

Figure 4—Physical model of the fan structure formation from domino blocks The Journal of The Southern African Institute of Mining and Metallurgy

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts

Conditions of fan nucleation and propagation Figure 5b illustrates schematically the most important feature of the fan mechanism. Horizontal lines show different strength levels along a hypothetical fault zone: τu – strength of intact material (fracture strength); τf – frictional strength of pre-existing fault; τfan – strength (shear resistance) of the fan structure. The graph on the left reflects the procedure of fan mechanism activation as discussed in Figure 4a. To create the initial fan structure a local stress equal to the fracture strength τu should be applied. The development of the initial fan structure is a stable process. After completion of the fan head, further dynamic propagation of the fault through the intact material can occur at any distributed shear stresses exceeding the fan structure strength τfan. New faults in intact materials can thus be produced by the fan mechanism even at distributed shear stresses that are significantly below the frictional strength τf (this will be discussed in more detail further). At the same time, the higher the distributed shear stress applied the higher the rupture speed and rupture violence.

Mathematical model of the fan mechanism Interaction of domino blocks and self-balancing conditions Further elucidation of the fan mechanism will be provided in this section on the basis of the mathematical model (further development of the model by Tarasov and Guzev, 2013). The mathematical model here represents the simplest description of the fan mechanism and reflects the static balance of forces affecting the domino structure before instability. This model was designed to demonstrate the essence of the fan mechanism, including some extraordinary features (e.g. selfbalancing principle, extremely low shear resistance, selfsustaining stress intensification, self-unbalancing conditions). All simplifications made in the model do not distort the essence of the fan mechanism.

The Journal of The Southern African Institute of Mining and Metallurgy

[1] where fc is the reduced resistance of tearing off the front beam from the basis (the elementary force fc is applied to the front beam only and shown by a green arrow in Figures 6b) k is the number of activated beams δ is the average angle between two neighbouring beams described by Equation [2] (Tarasov and Guzev, 2013).

Figure 6—Interaction of domino blocks in the fan mechanism VOLUME 114

OCTOBER 2014

777

▲

Figure 5—(a) Physical model of the fan propagation along the inclined simulating fault, (b) schema illustrating shear resistance of the fanstructure at the stage of initial formation and at the stage of spontaneous propagation

The domino structure off the mathematical model has the same features discussed for the physical model. For simplicity we will discuss the domino blocks as shown in Figure 6a. All blocks here are represented by beams of length r inclined at the initial angle α0 and distanced from each other by s. The distance is s = w/sinα0, where w is the block width. Both ends of each beam are connected to the upper and lower layers by joints. For ease we assume that the upper layer is represented by an elastic material, while the lower layer and the beams (domino blocks) are stiff. As such, a version of the model with blocks bonded together can be treated as representing the intact material.

Evenly distributed normal σn and shear τ stresses are applied to this construction. The value of τ is less than the shear stress necessary for displacement of the upper layer against the lower. Each beam is loaded by the same normal, N, and horizontal (shear), fτ, elementary forces representing N the evenly distributed normal σn and shear τ stresses applied to the whole structure. Elementary forces N (and their horizontal components) are shown by black arrows.

Elementary forcess fτ are shown by red arrows. The meaning of the elementary forces is as follows: N = σns1; fτ = τs1, where depth (or thickness) of the model is equal to unity. It should be noted that in the completed fan structure, due to stretching of the interface the distance between any two neighbouring blocks on one side of the fan increases (compared to the initial distance s) and the interface acquires a wave-like shape, which should affect the values of stresses (σn, τ) and elementary forces (N, N, fτ). To simplify the model we will consider the applied stresses and elementary forces invariable. This simplification does not affect the equilibrium condition of the completed fan and the extraordinary features of the fan-structure, which we shall demonstrate, thanks to the symmetrical configuration of the fan structure.

The fan structure can be formed if, in addition to the existing evenly distributed stresses, a local supplementary force (or stress) is applied. Let us apply a supplementary horizontal force F to the upper end of the leftmost beam.

Figure 6b shows an intermediate stage of the fan formation.

Equation [1] describes a variation of shear resistance (F (F) during formation of the fan structure (see details in Tarasov and Guzev, 2013).

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts [2] where E is the modulus of elasticity of the elastic connector A is the cross-sectional area of the elastic connector.

An essentially important feature of the fan structure is N tgα gα (see Figure 6b) of each gα the fact that the resistance fN = N/ beam decreases with increasing angle α from α0. It reaches zero at α = 90° and becomes negative at α > 90°. This means that the formation of the first half of the fan structure is accompanied by the increase in shear resistance F F; it reaches a maximum when the half of the fan has completed and then decreases to zero, similar to the experimental graph in Figure 4b.

Self-sustaining stress intensification and conditions of instability Figure 7 illustrates some features of the completed fan structure. In Figure 7a, elementary forces resisting displacement of the top rupture face against the bottom one are directed to the right, while forces assisting displacement are directed to the left. Shear resistance of relative displacement of the rupture faces due to the fan mechanism is determined by the sum of all horizontal forces applied to all beams involved in the completed fan, which is represented by Equation [3].

[3] The first term of Equation [3] reflects the effect of elementary forces fN = N/ N tgα gα and elastic forces associated gα with stretching of the elastic connectors on shear resistance of the fan structure. The elastic forces are represented here by angle δ. At self-balancing conditions the elastic forces are neutralized by forces fN. For the completed fan structure α0 + ktotδ = - α0, which means that the first term of Equation [3] is equal to zero.

The second term of Equation [3] reflects the reduced resistance fc of tearing off the front beam (domino block) from the intact material. The elementary force fc (green) is applied to the front beam only. It will be shown that the contribution of fc to the total resistance of the fan structure in real rocks is negligibly small.

The third term of Equation [3] represents an active united force assisting displacement. It is caused by the evenly distributed shear stress τ applied to the material. It should be emphasized that, unlike fc, the elementary forces fτ (red) are applied to all ktot beams of the fan structure.

When τ = 0 the formation of completed fan structure can be conducted in a stable regime using stiff +servo-controlled loading principles. Dotted beams in Figure 7a correspond to positions of them for the self-balancing fan. The dotted black curve in Figure 7b illustrates the variation of the fan head resistance F k at the fan structure development for the selfbalancing condition. The horizontal axis represents the fan structure length l = sk.

With distributed shear stress (τ > 0) the fan head becomes self-unbalancing and starts propagating through the intact material. Separation (tearing-off) of the front domino

▲

778

OCTOBER 2014

VOLUME 114

blocks sequentially ffrom the intact material is the essence off the failure process created by the fan mechanism. It should be emphasized that the fan structure in this case represents a natural stress intensifier that magnifies stresses in the rupture tip providing the tearing-off process. Elementary horizontal forces fτ (red) applied to each domino block of the fan as shown in Figure 7a are transmitted via the elastic connector (interface) to the rupture tip by the principle shown in Figure 7c. Hence, the separation of each front domino block with the reduced resistance fc (green arrow) from the intact material is caused by the united active force ktott fτ where ktot is the total number of domino blocks in the fan head. Calculations by Tarasov and Guzev (2013) show that the fan structure in natural materials can incorporate thousands of domino blocks Hence separation of the front domino block from the intact material can be caused by very low shear stress applied fτ = fc /kktot. The contribution of fc c to shear resistance of the fan structure is therefore negligible due to intensive stress magnification in the rupture tip caused by the fan mechanism.

The solid curve in Figure 7b shows features of the fan formation for the case when τ > 0. At the initial stage of the fan formation F k increases; it reaches a maximum F k = F max, then decreases to zero F k = 0, and finally becomes negative F k < 0. At the last stage (after point A) the fan formation is inevitably unstable because internal forces within the fan structure are not equalized and the fan head starts moving spontaneously as a wave. Depending on the value of distributed shear stress applied the extreme situation, F k = 0 can be reached at different stages before the completion of the fan structure.

Effect of friction in joints and block collapse on shear resistance of the fan structure The analysis of the idealized fan model (without friction in joints of rotating domino blocks) shows that the shear resistance of the completed fan structure propagating through the intact material is negligibly small. If we take into account friction in joints, we can estimate the real shear resistance of the fan structure. Estimations by Tarasov and Guzev (2013) show that despite friction in joints, shear resistance between

Figure 7—(a) Structure of the completed fan head with elementary forces applied to domino blocks, (b) variation in shear resistance of the developing fan structure for self-balancing (shaded curve) and selfunbalancing conditions, (c) principle of the stress intensification provided by the fan mechanism The Journal of The Southern African Institute of Mining and Metallurgy

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts

[4] Equation [4] shows that increasing the ratio r/w decreases the effect of friction in joints on the fan shear resistance: the higher the r/w the greater the efficiency of the fan mechanism. As discussed previously, the length of domino blocks is a function of confining pressure σ3 (dotted line l versus σ3 in Figure 3b). By analogy, the dotted line in Figure 8a shows symbolically the variation of the ratio r/w versus σ3 within the pressure range between σ3min(shear) and σ3max(hinge). This line indicates the hinge efficiency variation with confining pressure if domino blocks are not collapsed at rotation.

It should be noted that each block in the fan functions similarly to a beam with rotation-free end conditions loaded along the beam axis. Figure 8b illustrates features of the axial loading of a block at the initial (α0 = 300) and vertical positions. It shows that the value of axial force applied to the front domino-block exceeds the elementary force N (associated with the evenly distributed normal stress σn) by more than two times. This force is determined by forces N and fc. It means that any block at the front position is in the most stressed conditions. If the block does not collapse at the start of rotation it will be capable of bearing identical stresses at any stage of its rotation, including the vertical position when it deforms the rupture faces by the value μ. All blocks of the fan in combination create additional normal stresses moving apart the fault faces (wedge effect). We can suppose that during rotation from the initial to the final position, each domino block of the completed fan structure is under approximately the same axial stresses.

In rocks under confined compression, domino blocks are subjected to high loads that can lead to buckling and collapse of the blocks. On the basis of information presented by Megahid et al., (1993) we assume that domino blocks with slenderness ratio r/w ≤ 10 are stable at axial loading. Domino blocks with slenderness ratio r/w > 10 will be subjected to different degrees of destruction, depending on the ratio r/w.

The solid graph in Figure 8a illustrates a possible variation of the fan mechanism efficiency versus confining pressure σ3, taking into account the block destruction. At values r/w < 10 the efficiency varies in accordance with Equation [4]. Within the range 10 < r/w < 20, due to different degrees of destruction (depending on r/w) only a part of each block can maintain stability operating as a hinge. Very long blocks with slenderness ratio r/w > 20 completely collapse at rotation, creating gouge and common friction between the interfaces.

This is a preliminary explanation for the variable fan mechanism efficiency. Further experimental and theoretical studies will allow better understanding of this phenomenon.

The important point is that the fan mechanism is active only within the range of confining pressure between σ3min(hinge) and σ3max(hinge), with optimal efficiency at The Journal of The Southern African Institute of Mining and Metallurgy

σ3opt(hinge). Such variation off the ffan mechanism efficiency ff causes corresponding variation of the unconventional rock behaviour (e.g. brittleness variation shown in Figure 1c). In accordance with Equation [4], the shear resistance (strength) of the fan head for domino blocks characterized by the ratio w/r = 0.1 is one-tenth of the frictional strength: τfan ≈ 0.1τf.

This estimation is consistent with the result obtained on the physical model.

Uncontrollable failure and abnormal violence caused by the fan mechanism In this section we discuss an energy balance at spontaneous failure of rock specimens caused by the fan mechanism. It should be noted that loading conditions to generate the fan mechanism in rock specimens are different from those in the physical model discussed previously. In the physical model the fan structure was generated by a stress applied locally.

Rock specimens tested at highly confined compression are small. In order to generate the fan structure in this situation the whole specimen has to be loaded axially to high stresses that correspond to the material strength. Let us analyse features of the failure process caused by the fan mechanism at such stress conditions.

We will do this on the basis of experimental results obtained on the dolerite specimen tested at confining pressure σ3 = 60 MPa (Figure 2). The same stressdisplacement curve is shown in Figures 9a and 9b. Point A beyond the peak stress on these graphs corresponds to the start of instability. Figure 9c shows an enlarged portion of the post-peak stress-strain curve at stress degradation from ultimate stress σu till σA (this is replicated four times). At this post-peak stage the rupture was easily controllled. However, below σA (points A) control became impossible. To analyse the reason for that we divided the post-peak curves into four stages with equal intervals of differential stress. Each stage is characterized by average values of elastic modulus E (solid lines) and post-peak modulus M (dotted lines). Areas located between the E and M lines indicate the current post-peak Wr: rupture energy dW

Here σo and σe are differential stresses at the onset and the end of each stage.

We can see that the current post-peak rupture energy decreases dramatically with the rupture development from Wr with stress stage 1 to stage 4. The variation of dW

Figure 8—(a) Variation of the fan mechanism efficiency versus confining pressure affected by friction in joints and block collapse, (b) loading conditions of the same domino block at the initial and vertical positions VOLUME 114

OCTOBER 2014

779

▲

two interfaces f separated by the ffan structure off domino blocks operating as hinges is lower than the frictional strength of pre-existing faults with common frictional interfaces by the ratio w/r. Here, w is the width and rk is the length of domino blocks (see Figure 6a). The shear resistance associated with friction in joints represents the total resistance of the fan head propagating through the intact material:

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts degradation ffrom σu to σA is illustrated in Figure 9d. At stage 4 the rupture energy becomes extremely small because modulus M approaches modulus E E. Such variation in the post-peak rupture energy can be caused by the formation of the second half of the fan structure as was shown for the physical model on the graph in Figure 4b. After completion of the fan structure the uncontrollable spontaneous failure starts due to the high distributed shear stress (corresponding to point A on the diagrams in Figures 9a and 9b).

Figure 9e illustrates features of the shear rupture propagation through the specimen after the fan structure has completely formed. Figure 9e (1) shows the situation corresponding to point A, after which the fan head started propagating spontaneously along the dotted line. Figure 9e (2) shows an intermediate situation where the rupture is propagating in the pulse-like mode. Shear resistance and displacement along the fault during the failure process are very irregular. Three specific zones can be distinguished: (1) the fan zone where the failure process and domino block rotation is in progress; (2) the frictional zone located behind the fan head where the blocks have completed their rotation and the full friction is mobilized; and (3) the intact zone in front of the fan head. A load cell and an axial gauge in Figure 9e (1) mounted on the specimen as is commonly used in experiments can measure only the average load-bearing capacity and displacement of the specimen during the failure process. Results obtained on the basis of these gauges do not allow estimating the real energy balance of the failure process. However, the new knowledge about the fan mechanism gives us a chance to derive this inaccessible information.

The red area in Figure 9a represents elastic energy accumulated within the specimen and the machine at point A at the moment of the instability start (AD is the loading stiffness). At point A the bearing capacity of the specimen measured by the load cell (axial stress σA) is generally provided by the intact material located in front of the fan head. The contribution of the fan head to the bearing capacity of the specimen is very small. Starting from this moment, the fan head with resistance σfan crosses the specimen. Axial displacement dfan of the specimen caused by this process is associated with rotation of domino blocks as shown in Figure 9f. For the fault thickness h = 0.1 mm and α0 = 30° the displacement dfan ≈ 0.3 mm. Due to a very small shear resistance of the fan head (assume σfan ≈ 0.1 σf) the postpeak rupture energy Wr(fan) associated with failure and displacement by the distance dfan ≈ 0.3 mm is also very small.

Wr(fan) is represented by the grey area in Figure 9b. After the fan head has crossed the specimen, displacement along the whole fault becomes possible. This displacement is accompanied by violent dynamics caused by a large amount of released energy Wa (yellow area in Figure 9b). Shear resistance of the fault at this stage is determined by frictional strength σf.

Due to the inertia associated with high dynamics, the total displacement along the fault dtot will exceed the coordinate of point B corresponding to frictional (residual) strength σf. The equilibrium conditions will be reached below σf at point B1 (see Figures 9b and 2d). The violent displacement along the fault will pulverize the initial domino structure and create gouge between the fault interface at the final stage of the failure process (Figure 9e (4)).

▲

780

OCTOBER 2014

VOLUME 114

It should be emphasized that during spontaneous failure f after point A, self-unbalancing conditions exist within the fan head at any level of stresses above σfan. This means that stable and controllable failure beyond point A is in principle impossible for the fan mechanism, even on an absolutely stiff and servo-controlled testing machine. This explains the absence of experimental post-peak data for hard rocks at highly confined conditions (Mogi, 2007; Shimada, 2000; Tarasov, 2010). High rupture speed and pulverization of the initial fault structure during the spontaneous failure process make it impossible to directly observe and study the fan mechanism, and this explains why the mechanism has not been detected before.

Figure 10 explains the post-failure stress-shock discussed in Figure 2d by comparing different failure regimes (frictional shear and fan-hinged shear) with sliding a board along a hillside. When friction is constant during the dynamic events as shown in Figure 10a, the failure and sliding processes are not accompanied by stress shock. However, if low friction is suddenly substituted by high friction, the stress shock becomes inevitable (Figure 10b).

Fan mechanism in natural faults Structure and evolution of Ortlepp shears Natural faults normally have very complicated structure.

Before elucidating the role of the fan mechanism in the formation of dynamic natural faults, we will discuss features of complex fault development. Different hypotheses of fault evolution have been proposed (Segall and Pollard, 1980; Sibson, 1982; Scholz, 2002; De Joussineau and Aydin, 2009), but there is still no consensus regarding this process.

This section introduces a possible version of fault evolution based upon studies of: (a) faulting process at highly confined compression in laboratory experiments on calcareous siltstone (Otsuki and Dilov, 2005) and (b) structure of dynamic faults in rockbursts generated by shear rupture in brittle quartzite in deep South African mines (Ortlepp, 1997).

Experiments conducted by Otsuki and Dilov (2005) demonstrated the following features of complex faults: (1) Faults are multi-hierarchical segmented formations (2) Segmentation as a mechanism of fault propagation acts on all hierarchical ranks of complex faults

Figure 9—Features of the failure process governed by the fan mechanism in the dolerite specimen The Journal of The Southern African Institute of Mining and Metallurgy

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts

(3)

(4) (5) (6)

(7) (8)

Segmentation is a result of advanced triggering of a new bilaterally propagating rupture (new segment) in front of the propagating current rupture (current segment) The current and new segments propagating toward each other form a jog (step) where they meet Jogs of a compression type are very common in fault zones regardless of their size Once a number of segments of a given hierarchical rank coalesce, they behave as a single new and longer segment of one higher rank Segment of higher rank can trigger a new segment (rupture) at greater distance The new triggered segment starts as a primary rupture.

All these features are illustrated and discussed briefly below. The series of photographs in Figure 11a shows principles of the fault evolution by the advanced triggering of new segments (modified from Otsuki and Dilov, 2005).

Segments are represented here by white lines. The fault propagates from left to right. The segments are generated one by one and propagate bilaterally. Neighbouring segments are connected by a compressive jog where they meet. Overlap zones of the jogs are subjected to significant irreversible deformation. Figure 11b shows features of compressive jogs formed in brittle quartzite (photographs from Ortlepp, 1997).

The overlap zones of these jogs are represented by a row of domino blocks.

Figure 11c proposes a possible mechanism of the domino structure formation in compressive jogs (Tarasov and Ortlepp, 2007). It shows four steps of linkage between two segments (bold arrows) propagating towards each other in intact rock. Figure 11c (1) indicates the directions of the applied stresses against the propagating segments. In Figure 11c (2) the rock mass surrounding the approaching segments is theoretically divided into two massive blocks (A and B) pressed against each other by the applied stresses. The propagation of the segments decreases shear resistance between the massive blocks along these segments, which The Journal of The Southern African Institute of Mining and Metallurgy

Figure 11—Fault segmentation due to advanced triggering of new segments and creation of domino structure on the basis of compressive jogs VOLUME 114

OCTOBER 2014

781

▲

Figure 10—Explanation of the post-failure stress-shock phenomenon by comparing different failure regimes (frictional shear and fan-hinged shear) with sliding a board along a hillside

overstresses an area located between the tips off the approaching segments (Segall and Pollard, 1980). With further propagation of the segments they confine a zone that is now overstressed (Figure 11c (3)). When the extending overlap zone in hard brittle rocks reaches a critical length it fractures dynamically into an echelon of domino blocks (slabs) due to the shear of the overlap zone between the parallel faces of the massive blocks A and B (Figure 11c (4)).

This process is accompanied by the release of some portion of elastic energy. After that the segments stop propagating. The initial orientation of the tensile cracks separating the overlap zone into domino blocks is parallel to the major stress σ1. By analogy with primary ruptures angle αo ≈ (30°–40°). In contrast to primary ruptures where the domino blocks are generated sequentially in the rupture tip, this mechanism generates a row of parallel domino blocks simultaneously within the overlap zone of compressive jogs.

This mechanism can create cascades of compressive jogs which in combination can represent a fault segment of higher hierarchical rank. Figure 12 illustrates the principle of formation of a multi-segmented fault, a photograph of which is shown on the right. The fault propagates upwards. Open arrows indicate the direction of applied shear stress. At stage I a dynamically propagating primary fracture triggers an advanced fracture. Asterisks indicate centres of initiation of advanced triggered fractures. This new fracture (as well as all further triggered fractures) propagates bilaterally towards the current fracture and in the opposite direction. This fracture in turn triggers the next advanced fracture shown at stage II. At this stage the overlap zone between the two bottom fractures has reached the critical length and divided into a row of domino blocks. Further fault development occurs through repetition of similar stages. In Figure 12 new compressive jogs adjoining the tip of propagating fault are shown in red.

Figure 13 shows the evolution of a multi-hierarchical segmented fault. The fault nucleates as a primary rupture (bottom left corner) because the fan mechanism mobilized in very thin primary ruptures is the most energy-efficient shear rupture mechanism. A new primary rupture can be triggered in front of the current one at a distance xI due to stress transfer. Primary ruptures represent segments of rank I. Once a number of segments of a given hierarchical rank coalesce, they behave as a single new and longer segment of one

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts the block orientation is α0. The ffinal block position at the end of the fan head is (-α α0). The segmented fan structure has the same remarkable features as discussed for primary ruptures.

It should be emphasized that the fan structure can be formed if shear displacement between the fault faces is sufficient for the completed block rotation. Figure 14c shows the initial and final position of domino blocks for two shear ruptures of thickness h1 and h2. The thick rupture requires significantly greater displacement Δ to complete the block rotation. Equation [5] allows estimation of the fault displacement Δfault necessary for creation of the fan structure: [5] Figure 12—Cascade-like combination of compressive jogs formed due to advanced triggering of new segments (photograph from Ortlepp, 1997)

Figure 13—Evolution of a multi-hierarchical segmented fault

higher rank. The rank II segmented rupture can trigger a series of new primary ruptures at different distances with maximum remoteness of xII. The key feature of fault segmentation is the fact that a new segment triggered by the current segment of any rank nucleates as a primary rupture. At its propagation towards the current segment (and in the opposite direction) the new segment will be subjected to similar evolution as the current segment. After linkage of a number of rank II segments the next rank III segment will be formed (shown on a smaller scale ≈ 1:5). Further development of this fault will be accompanied by creation of higher rank segments. Where they meet, segments of each hierarchical rank form compressive jogs and domino structures of corresponding rank.

Taking this into account, we can analyse the possibility off the fan mechanism activation in a complex fault shown in Figure 15 (photograph from Ortlepp, 1997). The structure of this fault is shown symbolically on the left. It includes primary ruptures and higher rank segments formed on the basis of compressive jogs (rank II and rank III). Ortlepp (1997) indicates that the magnitude of the jog (or step) of rank III is hIII = 260 mm. The magnitude of shear displacement along the fault is less than 100 mm. This means that the fan structure cannot be formed on the basis of domino blocks of rank III. However, the development of primary ruptures (rank I) and ruptures of rank II can be governed by the fan mechanism because for these Δfault ≥ Δfan.

Due to this, the self-unbalancing fan structure (red zones in Figure 15) is created predominantly in segments of lower ranks. The fan mechanism generated here is responsible for high dynamics of the failure process. Relatively thin localized zones of very intense destruction can be observed in each dynamic fault. The initial domino structure of these segments is completely destroyed by extensive and violent shear and represented by pulverized gouge or even by pseudotachylytes. This explains why the fan structure has never been seen in nature, unlike domino structures of high-rank segments observed in a myriad of different faults. In highrank segments, domino blocks rotate through low angles without destruction and serve as a dampening mechanism.

Nucleation and propagation of dynamic natural faults The analysis of the physical and mathematical models shows that the initial formation of the fan structure requires a high

Formation of the fan structure in segmented faults The photographs in Figure 14a (from Ortlepp, 1997) demonstrate that domino blocks involved in complex faults can be subjected to rotation by angle β due to shear displacements of the fault faces. Observations show that angle β can be relatively high, exceeding 90°. Taking this fact into account we can suppose that during the fault propagation, domino blocks within a set of compressive jogs representing the fault head can created a fan-shape structure as shown symbolically in Figure 14b due to rotation of domino blocks through different angles β caused by the variable shear displacement along the fault. In the front jog

▲

782

OCTOBER 2014

VOLUME 114

Figure 14—Principle of fan structure formation in segmented faults The Journal of The Southern African Institute of Mining and Metallurgy

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts

Figure 15—The fan mechanism is activated predominantly in fault segments of lower hierarchical ranks where Δfault ≥ Δfan f f

local shear stress corresponding to the material’s fracture strength τu. After that the fan head can propagate dynamically through intact rock mass at shear stresses below the frictional strength. It is known that the field stress in the lithosphere cannot exceed the frictional strength. However, local stresses in the intact rock mass in the vicinity of preexisting discontinuities (e.g. boundaries between tectonic plates, faults, deep mines, etc.) can reach the fracture strength levels τu. According to the new approach, preexisting discontinuities act as stress concentrators creating the starting conditions for the fan mechanism, but instability (e.g. earthquakes) occurs due to the development of new faults in the intact rock mass.

Figure 16 illustrates this feature. Figure 16a shows a fragment of the rock mass with the local zone of high shear stress adjoining a pre-existing discontinuity where the fan structure is generated and a large zone of lower stress where the fan head can easily propagate. In Figure 16b the red graph illustrates shear resistance of the fan head at two stages: nucleation (length of fan head fracture lfan, strength τu) and propagation (length of created shear fracture L >>> lfan, strength τfan). The horizontal dotted line shows the level of frictional strength τf. The horizontal bold line corresponds to the field stress level τ.

Paradoxically, the low strength of intact rock provided by the fan mechanism favours the generation of new faults in an intact rock mass over reactivation of pre-existing faults (Tarasov, 2013). This unique feature of the fan mechanism allows the supposition that the majority of dynamic events in the Earth’s crust result from generation of new faults.

However, the proximity of pre-existing discontinuities to the area of instability caused by the fan mechanism creates the illusion of stick-slip instability on pre-existing faults, thus concealing the real situation.

earthquakes share the apparent paradox off undergoing failure at low shear stresses, while laboratory measurements indicate high material strengths.

Figure 17 shows a cross-section of the Earth’s crust involving an opening. The graph on the left illustrates variation of minor stress σ3 with depth. The fan mechanism can be activated below a critical depth corresponding to the critical level of minor stress σ3min(hinge) in Figure 3b. The zone of fan mechanism activity in Figure 17 is shown by the grey area. The efficiency of the fan mechanism increases with increasing brittleness of the rock with depth. However, around the opening the minor stress is below the critical level σ3 < σ3min(hinge). The dotted areas in Figure 17 show zones where the fan mechanism cannot be generated.

Deep openings similar to any pre-existing discontinuity represent stress concentrators. If locally elevated shear stresses in the grey area reach the level of ultimate stress τu (see graph in Figure 16) the fan head will be generated in the intact rock mass distanced away from the opening. After that the fan head can propagate further spontaneously through the zone of lower shear stress, creating new shear rupture and resulting in a shear rupture rockburst.

The fan mechanism is generally activated in intact rocks.

However, pre-existing faults can also be reactivated by the fan mechanism under special conditions. Seismic data and

Figure 16—Fan mechanism operation in vicinity of discontinuities

Generation of shear rupture rockbursts by the fan mechanism

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 17—Nucleation of shear rupture rockbursts by the fan mechanism VOLUME 114

OCTOBER 2014

783

▲

The fan mechanism could be responsible for some types of man-made earthquakes. Special studies conducted in South African mines (Gay and Ortlepp, 1979; McGarr et al., 1979) show that shear rupture rockbursts, which are seismically indistinguishable from natural earthquakes, are generated in intact hard rock (dry quartzite) in zones of highly confined compression. It was shown that these mine tremors and

Fan-structure shear rupture mechanism as a source of shear rupture rockbursts geological observations suggest that ffaults strengthen (heal) during the inter-seismic period of the earthquake cycle due to metamorphism and intruding igneous rocks. When the structure of the healed fault becomes strong enough for the creation of domino blocks capable of rotating without collapse, it can be reactivated by the fan mechanism. The fan mechanism cannot be generated within fresh faults with unconsolidated layers of weak gouge between the fault faces.

Conclusions This paper presents physical rationales for the recently identified fan mechanism generated in hard rocks at highly confined compression. In the fan mechanism, the rock failure associated with consecutive creation of small slabs (known as ‘domino blocks’) from the intact rock in the rupture tip is driven by a fan-shaped domino structure representing the rupture head. The fan head combines such unique features as: extremely low shear resistance, self-sustaining stress intensification, and self-unbalancing conditions. Due to this the failure process caused by the mechanism is inevitably spontaneous and violent. The mechanism is generated in primary ruptures and in segmented faults.

The physical and mathematical models presented highlight a paradoxical feature of the fan mechanism associated with the possibility of creating new shear ruptures in intact rock masses at shear stress levels that are significantly less than the frictional strength. The fan mechanism is the most energy-efficient shear rupture mechanism for rocks at confined compression. This mechanism causes the unconventional rock behaviour associated with drastic rock embrittlement at highly confined compression. The new mechanism provides a novel point of view for understanding the nature of spontaneous failure processes, including earthquakes and shear rupture rockbursts.

Acknowledgements The author acknowledges the support provided by the Centre for Offshore Foundation Systems (COFS) at the University of Western Australia, which was established under the Australian Research Council’s Special Research Centre scheme and is currently supported as a node of the Australian Research Council Centre of Excellence for Geotechnical Science and Engineering and in partnership with The Lloyd’s Register Educational Trust.

References DE JOUSSINEAU, G. and AYDIN, A. 2009. Segmentation along strike-slip faults revisited. Pure and Applied Geophysics, vol. 166. pp. 1575–1594.

GAY, N.C. AND ORTLEPP, W.D. 1979. Anatomy of a mining induced fault zone.

Geological Society of America Bulletin, Part 1, vol. 90. pp. 47–58.

HEATON, T.H. 1990. Evidence for and implications of self-healing pulses of slip in earthquake rupture. Physics of the Earth and Planetary Interiors, vol. 64, no. 1. pp. 1–20.

HORII, H. and NEMAT-NASSER, S. 1985. Compression-induced micro-crack growth in brittle solids: axial splitting and shear failure. Journal of Geophysical Research, vol. 90. pp. 3105–25.

KING, G.C.P. and SAMMIS, C.G. 1992. The mechanisms of finite brittle strain.

PAGEOPH, H vol. 138, no. 4. pp. 611–640.

KIRBY, S.H. and MCCORMICK, J.W. 1984. Inelastic properties of rocks and minerals: strength and rhelolgy. Handbook of Physical Properties of Rocks. Carmichael, R.S. (ed.). CRC Press, Boca Raton, Florida. vol. 3. pp. 139–280.

LYKOTRAFITIS, G., ROSAKIS, A.J., and RAVICHANDRAN, G. 2006. Self-healing pulselike shear ruptures in the laboratory. Science, vol. 313. pp. 1765–1768.

▲

784

OCTOBER 2014

VOLUME 114

MCGARR, A., SPOTTISWOODE, S.M., GAY, N.C., and ORTLEPP, W.D. 1979.

Observations relevant to seismic driving stress, stress drop, and efficiency.

Journal of Geophysical Research, vol. 84. pp. 2251–2261.

MEGAHID, A.R., SOGHAIR, H., HAGEED, M.A.A., and HAFER, A.M.A.A. 1993.

Strength and deformation capacity of slender RC beams. Proceedings of Fracture and Damage of Concrete and Rock – FDCR-2. Rossmanith, H.P. (ed.). Chapman & Hall, London.

MOGI, K. 2007. Experimental Rock Mechanics. Taylor & Francis, London, New York.

OHNAKA, M., KUWAHARE, Y., YAMAMOTO, K., and HIRASAWA, T. 1986. Dynamic breakdown processes and the generating mechanism for hi-frequency elastic radiation during stick-slip instabilities. Earthquake Source Mechanics, Geophysics. Monograph Series, vol. 37. Das, S., Boatwright, J., and Scholz, C.H. (eds.). American Geophysical Union, Washington, DC pp. 13–24.

ORTLEPP, W.D. 1997. Rock Fracture and Rockbursts – an Illustrative Study.

South African Institute of Mining and Metallurgy, Johannesburg.

ORTLEPP, W.D., ARMSTRONG, R., RYDER, J.A., and O’CONNOR, D. 2005.

Fundamental study of micro-fracturing on the slip surface of mineinduced dynamic brittle shear zones. 6th International Symposium on Rockburst and Seismicity in Mines, Perth, Western Australia, 9–11 March 2005. Potvin, Y. and Hudyma, M. (eds.). Australian Centre for Geomechanics, Perth. pp. 229-237.

OTSUKI, K. and DILOV, T. 2005. Evolution of hierarchical self-similar geometry of experimental fault zones: Implications for seismic nucleation and earthquake size. Journal of Geophysical Research, vol. 110, B03303. doi: 10.1029/204JB003359 PENG, S. and JOHNSON, A.M. 1972. Crack growth and faulting in cylindrical specimens of Chelmsford granite. International Journal of Rock Mechanics and Mining Sciences, vol. 9. pp. 37–86.

RECHES, Z. and LOCKNER, D.A. 1994. Nucleation and growth of faults in brittle rocks. Journal of Geophysical Research, vol. 99, no. B9. pp. 18159–18173.

SCHOLZ, C.H. 2002. The Mechanics of Earthquakes and Faulting. Cambridge University Press, Cambridge.

SEGALL, P. and POLLARD, D.D. 1980. Mechanics of discontinuous faults. Journal of Geophysical Research, vol. 85. pp. 555–568.

SHIMADA, M. 2000. Mechanical Behaviour of Rocks under High Pressure Conditions. Balkema, Rotterdam.

SIBSON, R.H. 1982. Fault zone models, heat flow, and the depth distribution of earthquakes in the continental crust of the United States. Bulletin of the Seismological Society of America, vol. 72. pp. 151–163.

STAVROGIN, A.N. and TARASOV, B.G. 2001. Experimental Physics and Rock Mechanics. Balkema, Rotterdam.

TARASOV, B.G. 2008. Intersonic shear rupture mechanism. International Journal of Rock Mechanics and Mining Sciences, vol. 45, no. 6. pp. 914–928.

TARASOV, B.G. 2010. Superbrittleness of rocks at high confining pressure.

Keynote Address, Fifth International Seminar on Deep and High Stress Mining, g Santiago, Chile. pp. 119–133.

TARASOV, B.G. 2013. Depth distribution of lithospheric strength determined by the self-unbalancing shear rupture mechanism. Proceedings of Eurock 2013, Poland. pp. 165-170.

TARASOV, B.G. and ORTLEPP, W.D. 2007. Shock loading-unloading mechanism in rockburst shear fractures in quartzite causing genesis of polyhedral subparticle in the fault gouge. Proceedings of the 4th International Seminar on Deep and High Stress Mining, g Perth, Australia. pp. 183–192.

TARASOV, B.G. and RANDOLPH, M.F. 2008. Frictionless shear at great depth and other paradoxes of hard rocks. International Journal of Rock Mechanics and Mining Sciences, vol. 45. pp. 316–328.

TARASOV, B.G. AND RANDOLPH, M.F. 2011. Superbrittleness of rocks and earthquake activity. International Journal of Rock Mechanics and Mining Sciences, vol. 48. pp. 888–898.

TARASOV, B.G. and POTVIN, Y. 2012. Universal criteria for rock brittleness estimation under triaxial compression. International Journal of Rock Mechanics and Mining Sciences, vol. 59. pp. 57–69.

TARASOV, B.G. and GUZEV, M.A. 2013. New insight into the nature of size dependence and the lower limit of rock strength. Proceedings of the 8th International Symposium on Rockbursts and Seismicity in Mines, St. Petersburg, Russia, 1–7 September 2013. pp. 31–40.

WAWERSIK, W.R. and FAIRHURST, C. 1970. A study of brittle rock fracture in laboratory compression experiments. International Journal of Rock Mechanics and Mining Sciences, vol. 7. pp. 561–575. ◆ The Journal of The Southern African Institute of Mining and Metallurgy

Unique fall-of-ground prevention strategy implemented at Two Rivers Platinum Mine

by A. Esterhuizen*

Regional geology Since 2005 Two Rivers Platinum Mine has set out on an initiative to actively monitor and control ground conditions on a daily basis, by making use of borehole cameras and pro-actively amending the support and mining strategies based on day-to-day observations of the hangingwall conditions.

Today the borehole camera observations form part of the Rock Engineering Department’s daily function, and the size and frequency of falls of ground and ensuing accident rates have been drastically reduced since implementation of the system. The Two Rivers Platinum fall-of-ground management system aims to support 100% of the possible fallout thickness, based on ongoing data gathering and interpretation, thereby ensuring safety and limiting support cost.

Keywords strata control, fall-of-ground management.

Introduction Two Rivers Platinum Mine, a joint venture between African Rainbow Minerals (ARM, 55% ownership and management) and Impala Platinum (45% ownership and smelting, refining, marketing), is situated on the eastern limb of the Bushveld Complex near the town of Steelpoort in Limpopo Province (Figure 1).

The mine is underlain by rock formations of the Critical Zone (Winterveld NoriteAnorthosite) and the Main Zone (Winnaarshoek Norite-Anorthosite). Two economically significant PGM-bearing horizons, namely the UG2 chromitite and Merensky Reef, are located in the Upper Critical Zone, separated by approximately 140 m of norites and anorthosites. The UG2 chromitite seam is extracted through mechanized bord and pillar mining at a rate of 300 000 t per month at an average ore grade of 4.10 g/t (PGE + Au). The underground workings on the UG2 horizon are currently between 40 m and 800 m below surface, at an average orebody dip of 9 degrees, and are accessed through two decline shafts roughly 3 km apart. The Merensky Reef is extracted on a similar bord and pillar mining layout, currently approximately 30 m below surface. The Merensky Shaft is planned to extract 14 500 t of ore per month in the trial mining phase, and thereafter it will build up to 180 000 t/month.

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

The UG2 chromitite layer has been intersected in over 100 boreholes on the host farm Dwarsrivier. The lithological sequence, from the base upwards is illustrated in Figure 2. The UG2 reef horizon lies within a competent pyroxenite band, which extends 2.5 m into the hangingwall and up to 1 m in the footwall. The Footwall 1 Unit is coarse-grained to pegmatoidal pyroxenite/harzburgite and is approximately 1 m thick.

The UG2 averages approximately 180 cm in thickness, and internal pyroxenite and norite partings may be present. These in some cases have highly angular margins and appear to have been derived from erosion and transport within the UG2 of pre-existing layers. To the south and deep central part of the farm, a large area is characterized by the presence of split reef, whereby a pyroxenite or norite lens up to 6 m thick is situated approximately two-thirds from the base of the UG2.

Disseminated sulphide mineralization is generally present, especially around the margins of internal pyroxenite partings. This mainly comprises pentlandite and chalcopyrite, with lesser pyrrhotite. There is minimal cohesion between certain layers due to the mineral composition of the contacts.

The UG2 is overlain by poikilitic pyroxenite that hosts up to three chromitite ‘leader’ layers (collectively termed the UG2A chromitite layers). The pyroxenite is intersected by three major joint sets; Joint set J1 dips at approximately 85° and has a strike of approximately 050° E of N; Joint set J2 dips at approximately 85° and strikes 120° E of N. Joint set J3 dips at

* Open House Management Solutions (Pty) Ltd. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12–14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

785

▲

Synopsis

Unique fall-of-ground prevention strategy implemented at Two Rivers Platinum Mine The pyroxenite is prone to weathering, especially in the shallower workings of the mine where some water inflow is experienced. The oxidation of specifically the contact between the HW1 pyroxenite and HW 2 anorthosite, referred to as the HW 1/2 contact, is problematic and resulted in numerous large collapses during the mine’s early history. It is the weathered nature of this contact, which is located 2.5 m above top of UG2, and the necessity of observing its condition, that led to the implementation of the borehole camera system.

Mine layout Figure 1—Regional setting of Two Rivers Platinum Mine

The mine is laid out in a regular checkerboard bord and pillar layout, with panel widths ranging between 6 m and 12 m, depending on the rock mass rating. Elastic pillars are designed according to the Hedley and Grant pillar formula and increase in size with increasing depth below surface.

Shaft stability is ensured by means of squat pillars.

Although the orebody dips at 9-11 degrees, the mine is undercutting a mountain, which results in a rapid increase in depth below surface.

Two Rivers Platinum is a fully mechanized mine, which perfectly suits the orebody geometry, with nine half-levels (sections) on the Main Shaft and seven half-levels on the North Shaft. A section consists of at least eight panels and is mined with a fleet consisting of three load haul dumpers

Figure 2—Simplified diagram of the lithological sequence

70° to 80° and strikes 080° to 100° E of N. The jointing extends through the reef into the footwall, but rarely extends into the HW2 anorthosite.

The UG2 reef horizon is divided into two portions, the UG2 proper (1.8 m), and the UG2 Leader reef (30-40 cm) by an internal pyroxenite layer (30-40 cm). The mining property is also affected by numerous features related to late-stage Bushveld intrusions. These include potholes and Fe- and Mgenriched replacement pegmatoid bodies emplaced into the Bushveld cumulate stratigraphy. Generally, thinning of the reef layer, coupled with the steep and erratic dips around the potholes, results in a total ground loss during underground mining operations. Potholes are present on varying scales, from the large regional potholes that affect the Merensky stratigraphy in the northwestern Bushveld Complex to small circular features less than 10 m in diameter. Current information at Two Rivers suggests that potholes affect both the Merensky pyroxenite and the UG2. These potholes are envisaged to be similar in their development to those encountered in the western Bushveld Complex. However, no indications of regional potholes affecting the Merensky stratigraphy have been identified.

Numerous wide dolerite dykes intersect the mining property, striking in NW-SW and NW-SE directions. The dykes are brittle, but generally easily negotiated with few stability-related issues.

▲

786

OCTOBER 2014

VOLUME 114

excavations as well as the boxcuts and ventilation shafts

Figure 4—Diagram showing a typical section layout and mining sequence The Journal of The Southern African Institute of Mining and Metallurgy

Unique fall-of-ground prevention strategy implemented at Two Rivers Platinum Mine (LHDs), one roofbolter, f one drill rig, one utility vehicle, and one emulsion utility vehicle (UV). Currently the mine is producing roughly 300 000 t of ore per month through the two shafts. Each section is equipped with a strike conveyor, which in turn tips on the main belt through an orepass.

Support As regional support, elastic pillars are left on a checkerboard pattern, designed with a factor of safety *1,6 and width to height ratio * 3. Due to the frequency and size distribution of the potholes and other major geological features such as shear zones and dykes that are left unmined, no regional/ barrier pillars are purposely designed.

As in-stope support, 1.5 m long × 18 mm diameter full column grouted resin bolts are installed spaced 1 m × 2 m apart to a diamond pattern. The support is installed using 60-second spin-and-hold resin and is aimed primarily at key block suspension and beam building. Where the need to suspend the entire HW1 beam is identified by means of borehole camera observation, 4 m long, 18 mm diameter (380 kN) pre-tensioned, full column grouted cable anchors are installed. The mine currently installs roughly 22 000 bolts and 3 000 cable anchors per month. An average of 3 000 m2 of shotcrete and thin sprayed liners is also applied per month in friable areas.

Major geological features such as shear zones and dykes are left unmined to prevent off-reef mining and exposure of employees to hazardous ground. Where excavations have to traverse wide shear zones, the sheared and weathered HW1 pyroxenite makes conventional drilling and installation of support impossible. In these conditions the HW1 pyroxenite is removed as the HW2 anorthosite behaves significantly better under these conditions.

➤

➤ ➤

➤ ➤

determination off the physical condition off the contact, and not only the location of the contact, was crucial Due to weathering in certain areas of the mine, beam building was not being achieved by the installed support, as alteration on joint walls, and subsequent reduction in cohesion, allowed blocks to rotate and slide out between support units Total suspension of the pyroxenite beam is required under certain conditions Production personnel were unable to successfully identify low-angle/domed-shape joints, which accounted for the other 50% of collapses There was an urgent need to identify areas where the HW1/2 contact was likely to be oxidized Low-angle joints are scattered randomly across the mining property, but joint mapping showed that these features tend to cluster around large potholes.

Based on the findings of the analyses, a FOG management plan was implemented, based on local experience and observations.

Based on the identified requirements, two strata control observers were employed to assist with daily underground data acquisition and identification of weathered ground conditions and low-angle features. In order to assist with the inspection of the HW1/2 contact, a borehole camera was purchased. The camera was used on an ad-hoc basis where the need was identified and did not form part of the normal daily routine.

Analyses of fall-of-ground data and implementation of a strata control system

➤ As the contact is located mine-wide at a fairly consistent distance above the UG2, it was obvious that The Journal of The Southern African Institute of Mining and Metallurgy

Figure 5—Diagram showing the support layout

Figure 6—FOG heights for 2005 to 2006 VOLUME 114

OCTOBER 2014

787

▲

Shortly after the onset of mining in 2005, massive falls of ground (FOGs) occurred at a rate of 0.8 collapses per month.

Some of these collapses were a result of wedge failure brought about by the interaction of the normal near-vertical joints with flatter low-angle or domed joints. The main cause of most of these collapses was found to be the inability of the inexperienced mining crews to identify and timeously support the structures. Of concern, though, was the high frequency of collapses extending up to the HW1/2 contact, which resulted in fallout thicknesses of up to 2.5 m, which is far above the effective length of the installed primary support. These collapses were unexpected, as the initial geotechnical studies and risk assessment did not identify this particular contact as being of high potential risk. By the end of 2006, 50% of all collapses dislodged from this contact.

Examination of the collapses identified early on that the HW1/2 contact and HW1 pyroxenite are prone to oxidation, which reduces the cohesion and self-supporting ability of the beam. The beam would then collapse between joints and could extend up to 20 m in strike length.

The following conclusions were made based on FOG analyses:

Unique fall-of-ground prevention strategy implemented at Two Rivers Platinum Mine In addition to the modifications f to the Rock Engineering Department structure, the production team was given regular underground training in the identification and treatment of low-angle joints. The original risk assessment was updated and the current support at the time, resin bolts, was supplemented with additional 4 m cable anchors where deemed necessary. A great deal of attention was given to raising awareness amongst mining personnel with regard to lowangle features and the problematic HW1/2 contact.

The changes to the strata control system proved to be a step in the right direction. Although the *2 m high collapses were not entirely prevented, the overall size of the collapses was reduced due to the installation of cable anchors where deemed necessary.

In early 2008 the mine suffered another massive collapse.

This time the collapse occurred in a previously identified and supported area, and trapped a LHD. Investigation showed that the extent of the unstable area was underestimated during the initial support recommendation and, in addition, the installed support spacing was not in accordance with the Rock Engineering recommendation.

Based on the prior success gained from the addition of observers and borehole cameras, it was decided to continue and improve on this system, which at the time was new and not used on a daily basis in South African mines.

The Rock Engineering Department staff complement was extended to include an observer and a borehole camera for each section. The borehole inspections now became part of the daily functions of the rock engineering observers, as well as daily inspections of all available working panels. This information was then relayed back to the strata control officer and mine overseer on a daily basis for recommendation and communication. All installed support was inspected and the support spacing measured on a daily basis, with weekly overinspections of all cable anchors and applied thin-skin liners.

The mine’s support standard was amended to ensure that two dedicated camera holes were drilled with every support round. The holes are 4 m deep and 40 mm in diameter, spaced no more than 4 m apart, and drilled midway between the center line and the sidewalls. After drilling, the holes are clearly demarcated and no support units may be installed in these holes. In addition, the width of excavations mining in, or advancing through, deteriorated ground conditions was reduced from 12 m to 6 m, reducing exposure of weakness planes.

The success obtained through this system is self-evident when one considers the mine’s total FOG history from 2005 to 2011. There has been a clear and significant improvement in the FOG height and overall size since the introduction of the camera system. The fallout size has been reduced to the point where, since 2010, most FOGs are barring-related incidents.

The system incorporated at Two Rivers Platinum Mine, along with detailed daily over-examination of working panels by competent persons and proper training of production personnel, resulted in a proactive system where potentially unstable ground is identified and assessed on a daily basis.

The borehole camera system is quick, simple, and easy to understand. Feedback to mining personnel is immediate and visual, which helps foster trust and appreciation of the system amongst miners, which in turn helps to ensure compliance to the standard.

▲

788

OCTOBER 2014

VOLUME 114

Using the information f obtained ffrom the borehole camera observations and FOG investigations, it became apparent that the mine has four different fallout height zones, and if these zones can be correctly and timeously identified, 100% of the fallout height can be supported 100% of the time, without resorting to a mine-wide blanket worst-case scenario support design. The zones include: A. Stringer collapse (0.5 m height) B. Low-angle joint wedge collapse (1.5 m height) C. HW1/2 beam collapse (2.5 m height) D. Shear zone self-mining (up to 4 m height).

Figure 10 indicates that the 2.5 m high collapses related to the hangingwall anorthosite/pyroxenite contact (zone C) have basically been eradicated. This is a result of the ability to identify and treat the structure where necessary. Also obvious is the continuance of the 1.5 m high collapses (zone B). These collapses are related to low-angled joints and domes, which are randomly spaced and oriented, and harder to identify and treat. Approximately 90% of all low-angled joint/dome collapses on Three Rivers Platinum occur in the unsupported area between the face and the last line of support immediately following or concurrent with the blast.

The lack of these collapses in the supported area pays tribute to the ability of the entry examination team, in conjunction with the relevant service departments, to identify and treat the structures.

Figure 7—Reduction in FOG size from 2005 to 2007 due to the introduction of observers and a borehole camera

Figure 8—Diagram illustrating how the support strategy is adjusted according to the local ground conditions. The green line represents a weathered contact and the yellow bars indicate camera boreholes The Journal of The Southern African Institute of Mining and Metallurgy

Unique fall-of-ground prevention strategy implemented at Two Rivers Platinum Mine All designs and assumptions regarding ffallout thickness are based on a detailed actual FOG database. Every incident is thoroughly inspected, regardless of the size or consequence of the collapse. Despite the thoroughness, only 75 cases were investigated over a period of eight years, representing less than 0.05% of the total mined area. In order to verify the conclusions, and to gain confidence in the assumptions, a synthetic database was created, using the software package JBlock. JBlock allows for the creation of a jointed beam based on given joint set parameters. The resultant key blocks can then be analysed for stability.

Actual vs. synthetic database The mine’s FOG database shows that the number of collapses per square metre mined has steadily decreased since the introduction of the current strata control system. Although this is a desirable result of an effective system, it limits observational ability, and with it, the opportunity to learn and gather information. As with any statistical information, a larger FOG database will provide more accurate output and provide more confidence in the assumptions made during the design process.

Jblock was used to create a synthetic database, in order to supplement the actual database and verify the conclusions.

Using Jblock, 10 845 mining steps (advance blasts) and 1.17 million square metres were simulated, which resulted in the analyses of 20 000 key blocks with 5 iterations per block.

The results obtained were very similar to the actual database, and confirmed the prediction of four fallout height zones.

The results of the synthetic database, shown in Figure 12, were obtained by inserting a hangingwall-parallel contact at the actual location of the anorthosite/pyroxenite interface.

The results indicate that the 95% FOG height is 2.95 m, with a maximum expected apex height of 3.58 m. The actual database concluded that the 95% FOG height is 2.7 m, and that the highest actual collapse was 3.5 m. There were other higher collapses in the actual database, but these were related to self-mining shear zones, and did not dislodge from the hangingwall contact. As the contact location is consistent at 2.5 m above the reef, an associated collapse with a fallout height of 3.58 m is not deemed probable because the jointing does not penetrate above the contact into the anorthosite.

The stringer collapses were analysed in a similar fashion, where the structure was modelled as a hangingwall-parallel

Figure 9—Total mine history (2005 to 2013) for FOG size (tonnage)

Figure 10—Total mine history (2005 to 2013) for FOG height. The analyses clearly indicate four separate fallout zones

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

789

▲

Figure 11—Decrease in FOGs per m2 mined

Unique fall-of-ground prevention strategy implemented at Two Rivers Platinum Mine

Figure 12—Actual and synthetic cumulative fallout frequencies for HW1/2 interface

Figure 13—Actual and synthetic cumulative fallout frequencies for the stringer layer

contact, placed at the known location. Here too the results closely resembled the actual database (Figure 13).

The actual database shows that for the stringer layer, a 95% fallout thickness of 0.4 m should expected, whereas the maximum experienced fallout height was 0.56 m. The synthetic database concluded that the 95% fallout height is expected to be 0.56 m with a maximum expected wedge apex of 0.77 m. This maximum expected apex height for the stringer layer is deemed more probable than that predicted for the HW1/2, as the joints are continuous in the pyroxenite and penetrate through the stringer layer.

Conclusions ➤ Based on the study, synthetic databases can be used to supplement the actual mine fall-of-ground database in order to build a larger database with more accurate design parameters ➤ Based on the results of the study and experience onmine, support design is divided into four zones, each

▲

790

OCTOBER 2014

VOLUME 114

with a specific f support length and spacing in order to support 100% of the maximum fallout height expected for that specific zone ➤ The strategy greatly reduced the mine’s fall-of-ground frequency and size, while effectively controlling support costs.

Acknowledgements The authors wish to thank Mr J.D. Bosman for the technical review, and the Management of Two Rivers Platinum Mine for approval of the release of the information published in this paper. We also thank Mr Q. Grix for his input and data acquisition.

References GEOLOGICAL DESCRIPTION – GEOLOGY DEPARTMENT. Two Rivers Platinum Mine public network, 2013.

FOG DATABASE – ROCK ENGINEERING DEPARTMENT. Two Rivers Platinum Mine OHMS. ◆ The Journal of The Southern African Institute of Mining and Metallurgy

In situ monitoring of primary roofbolts at underground coal mines in the USA

by A.J.S. Spearing* and A. Hyett†

Synopsis Primary roof support represents the first line of defence against rockrelated falls of ground in underground mines, and improper utilization or misunderstanding of the applicability and behaviour of primary support can be costly from a safety standpoint. This is a major concern for underground mines, as roof support is the single most costly expense from a mining operational perspective. This is further backed by the evidence that, in the USA, hundreds of injuries and fatalities still occur each year because of rib, roof, and massive roof falls. Additionally, the fully-grouted passive rebar, fully-grouted tension rebar, and resin-assisted mechanical anchor bolts, which constitute a large portion (89%) of the 68 million bolts installed each year in underground mines in the USA can vary in cost quite dramatically. To mitigate this concern a study was conducted in 2010 by the National Institute of Occupational Safety and Health, in conjunction with Southern Illinois University of Carbondale, to assess the performance of primary roofbolts in underground coal mines for improved safety and cost. This was accomplished using underground roofbolt monitoring solutions, field data, and numerical modelling to better understand the quasi-static behaviour of underground coal mine roofs and the response behaviour of the bolts. In particular, over 170 instrumented extensometers, closure meters, shear meters, fully-grouted passive rebar, fully-grouted tension rebar, and resin-assisted mechanical roofbolts were installed at three coal mines across the USA. Of these three mines, two used the room and pillar extraction method and the other used the longwall extraction method. There was no evidence to indicate a difference in performance of the active primary roofbolts compared with the passive primary roofbolts.

Additionally, in the initial loading phase, the active bolts showed no difference in loading, indicating that tension bleed-off is of more of a concern than originally thought. Lastly, for the initial computer modelling studies, challenges still remain in obtaining a good match to the in situ bolt measurements and replicating the discontinuous roof rock and in situ bolt behaviour over time.

Keywords primary support, roofbolts, in situ monitoring.

➤ Supplementary support that is installed as additional support after the secondary support due to poor localized conditions, and typically consisting of truss bolts, cribs, and standing support.

Primary roof support represents the first line of defence against rock-related falls of ground in underground mines, and improper utilization and/or understanding of primary supports applicability and behaviour can be costly and adversely affect rock-related safety.

This is a major concern for underground mines, as roof support is the single most costly expense from a mining operations perspective and is the production bottleneck. This is further backed by the evidence that, in the USA, fatalities and hundreds of injuries still occur each year because of rib, roof, and massive roof falls, as shown in Figure 1 (Mark, Pappas, and Barczak, 2009).

Additionally, the fully-grouted passive rebar (FGPR), fully-grouted tension rebar (FGTR), and resin-assisted mechanical anchor bolts (RMABs), which constitute a large portion (89%) of the 68 million bolts installed each year in underground mines, can vary in cost quite dramatically. As a rough estimate, the FGPR is the least costly support, whereas the FGTR is roughly 10%, and the RMAB is around 29% greater than the FGPR. (Tadolini and Mazzoni, 2006; Spearing and Gadde, 2011).

To mitigate this concern a study was conducted in 2010, funded by the National Institute of Occupational Safety and Health and undertaken by Southern Illinois University of Carbondale and Peabody Energy, to assess

Introduction

➤ Primary support that is installed oncycle, typically resin-grouted rebar, possibly with straps and/or screen ➤ Secondary support that is installed as additional support after the primary support, mainly in intersections, due to the wider effective span, and consisting typically of cable bolts The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

* Southern Illinois University Carbondale, Illinois, USA. † Yield Point Inc., Ontario, Canada © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12–14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

791

▲

Support in US coal mines is divided into three main categories:

In situ monitoring of primary roofbolts at underground coal mines in the USA

Figure 1—Fall-of-ground injuries in US underground coal mines (Mine Safety and Health Administration, 2010, 2011; Reisterer, 2011)

the performance of primary roofbolts in underground coal mines in the USA for improved safety and cost. This was accomplished using underground roofbolt monitoring solutions, field data, and numerical modelling to better understand the quasi-static behaviour of underground coal mine roofs and the response behaviour of the bolts.

Furthermore, over 170 instrumented extensometers, closure meters, shear meters, fully-grouted passive rebar (FGPR), fully-grouted tension rebar (FGTR), and resinassisted mechanical roofbolts (RMAB) were installed at three coal mines across the USA. Of these three mines, two used the room and pillar extraction method and the other used the longwall extraction method.

Mine sites Three underground coal mines were selected for the project.

The sites needed to have similar immediate rock lithology, and mine management needed to provide the equipment and labour to carefully install the expensive instrumented bolts on-cycle, which adversely affected mine productivity. The actual mine names are not disclosed, but they are referred to as mines A, B, and C hereafter. Mines A and B were both room and pillar coal operations located in southwestern Indiana, and Mine C was a longwall operation in northwestern Colorado (Figure 2).

In the following sections, the layout of the instrumentation sites at each mine site, the local geology of the instrumentation sites, and the instrumentation itself will be discussed.

All three mines used no. 6 (19 mm or 0.75 inch nominal diameter) Grade 60 fully grouted passive rebar as their primary support; however, after slotting of the rebar, which is required for instrumenting the rockbolts, the residual yield capacity of the bolts was well below the requirements set out by the mines’ ground control plans. Therefore a bolt with higher yield and ultimate capacity, (20 mm or 0.804 inch Grade 75 rebar) was chosen for the instrumented bolts. A comparison of the yield and ultimate load capacity of the no.

6 Grade 60 and 20 mm (0.804 inch) Grade 75 bolts is shown in Table I. All of the rockbolts, bolt plates, and resin were donated from the same manufacturer. This was to eliminate vendor-related variability of the materials.

Instrumentation All past studies related to roofbolt monitoring have been conducted mainly through the National Institute of Occupational Safety and Health (NIOSH), by Signer from 1984-1997 (Serbousek and Signer, 1984; Signer, 1988; Signer, Franklin, Mark, and Hendon, 1993; Signer, Cox, and Johnston, 1997). For those studies the instrumented rockbolts were equipped with a short baselength (<25 mm) resistive foil strain gauges. The shortcoming of this technology was that only 10% coverage of the bolt was achieved due to the shortness of gauges. Additionally, the loadings that were obtained were highly localized and the entire axial loading profile of the bolts was not well represented.

In contrast to these past studies, a new technology was utilized in an attempt to better capture the axial loading profile of the rockbolts. For this, rockbolts were fitted with

Rockbolts and instrumentation Rockbolts The rockbolts were installed on-cycle with the production operations as primary support at both room-and-pillar mines.

At the longwall mine, the rocksbolts were supplemental support because primary support had already been installed during panel development. Reiterating, the three bolt systems compared in this project were fully-grouted passive rebar (FGPR), fully-grouted tension rebar (FGTR), and resinassisted mechanical roofbolts (RMAB). The FGPR is considered a passive support because it is not tensioned on installation, whereas the FGTR and RMAB are considered active because they are tensioned on installation. As the names suggest, FGPR and FGTR support are installed with full resin encapsulation, while the RMAB utilize an anchor at the back of the hole with a 1.22 m (4 ft) resin encapsulation.

▲

792

OCTOBER 2014

VOLUME 114

Figure 2—Locations of Mines A, B, and C

Table I

Yield load and ultimate load of primary support and instrumentation (Spearing, et al., 2012) Bolt type #6 Grade 60 forged head 0.804in. Grade 75 threaded 0.804in. Grade 75 bar

Yield load (kN)

Ultimate load (kN)

119.75 (minimum) 184.16 (actual) 183.25 (actual)

179.62 (minimum) 257.31 (actual) 261.27 (actual)

The Journal of The Southern African Institute of Mining and Metallurgy

In situ monitoring of primary roofbolts at underground coal mines in the USA long baselength (200–500 mm) displacement sensors developed by YieldPoint Inc. This technology utilizes an array of sub-micrometre resolution displacement sensors that measure the displacement of the bolt. Collectively, the array of gauges provides an axial loading profile of the bolt over time. A more in-depth discussion of this technology is presented by Spearing et al. (2012), but overall, a greater coverage of the bolt is obtained (75%) as well as an averaged and more representative axial loading profile of the rockbolt.

Once the bolt type, capacity, and technology had been determined, the bolts were machined by slotting along their length for placement of the sensors (Figure 3). YieldPoint Inc. was chosen as the manufacturer because of their competitive costs, their willingness to be present during instrumentation installation at the mine sites, and because they were able to develop all of the rockbolt instrumentation, as well as closure meters, extensometer, tilt meters, and data loggers.

The bolts were slotted to a depth of 3.2 mm (0.126 inch) for placement of six 45.7 cm (18 inch) displacement sensors (three on each side). Six sensors per bolt were chosen to mitigate the total cost per bolt while still obtaining a comprehensive coverage of the bolt. The sensors were placed in an end-to-end arrangement within the machined slots and were held in place by epoxy.

The electronics of the sensors were housed in an extended steel head at the end of the rockbolt. When data logging began, Mine A and Mine C utilized the extended bolt head shown in Figure 4. This bolt head protruded some length below the hangingwall; eventually, due to moving machinery at the face, mainly the continuous miner holing through the crosscuts, several instruments were knocked out and destroyed. Therefore a more adaptive shallow head (shown at the bottom of Figure 4) was used later at Mine B to eliminate this problem.

The sensors on the bolts were arranged in two configurations. Mines B and C utilized a stacked configuration shown in Figure 5a and Mine A utilized a staggered configuration shown in Figure 5b. For the stacked configuration the sensors are placed in a diametrically opposed pattern in the machined slots, and for the staggered configuration the sensors are offset by half the baselength of the sensor. It was felt that the stacked configuration could miss some of the localized shearing loads that could occur between the

sensors, and a staggered arrangement was therefore f considered for comparison purposes.

The data was collected using data loggers, shown in Figure 6. Each data box could store over 30 000 readings and was equipped with four channels (one per rockbolt) that were wired to the instruments. The data loggers were manually set to take readings every hour, which could then be retrieved via a USB connection and custom software also developed by YieldPoint. Routine visits to the instrumentation site were scheduled to download the data. The data, which was recorded in microstrain, was then manipulated to obtain axial load, axial strain, and axial stretch. A conversion factor of 153 μ-strain per ton was used, based upon the crosssectional area of the machined rebar and the elastic modulus of the bolt steel.

Figure 3—Slotted rebar for sensor placement (Kostecki, 2013)

Figure 6—d4 logger data-log box by YieldPoint Inc. (Kostecki, 2013)

Figure 5—Representation of the stacked gauge orientation (a) used at Mine B and Mine C, and the staggered orientation (b) used at Mine A (Spearing, et al., 2012)

VOLUME 114

OCTOBER 2014

793

▲

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 4—Instrumented roofbolt with an extended-length head (top) and shallow head (bottom) (Kostecki, 2013)

In situ monitoring of primary roofbolts at underground coal mines in the USA Instrumentation sites and local geology The selection of the instrumentation sites was vital to the overall scope of this project. There were nine instrumentation sites in total – one for each of the three roofbolt systems (i.e.

FGPR, FGTR, RMAB) for each of the mines. The local geology was investigated by a consulting geologist (Padgett, 2010).

Mine A Instruments at Mine A were installed as primary support oncycle. Figure 7 shows the instrumentation sites. From left to right in Figure 7 the entries were numbered 5, 6, and 7.

These entries were chosen for several reasons. First, instruments were placed in three adjacent entries so that the geology was as similar as possible. Next, the adjacent entry to no. 5 was a belt entry and the adjacent entry to no. 7 was a return-air entry blocked off by a stopping. Finally, because monitoring of the instruments was to last several months, these entries were located in an area such that production would not be hindered over the entire monitoring period.

Instrumented roofbolts were installed in the intersections and mid-pillar regions of each entry. The arrows in Figure 7 show the direction in which the face was advancing during instrument installation, and therefore the mid-pillar was mined first and the mid-pillar instruments were installed a few days prior to the intersection instruments. The patterns at each site were identical. The diagonal pattern shown across the intersections was of particular importance as this represents the longest span, and therefore would offer the greatest chance of capturing the highest displacements and loadings over time. The FGTR instrumented bolts were installed in entry 5, FGPR in entry 6, and RMAB in entry 7.

The non-instrumented support surrounding each test site, shown in Figure 7, was of the same type as the instrumented support in each area. Most importantly, each instrumented bolt was zeroed prior to instrumentation using a d-READER instrument reader provided by YieldPoint Inc. The entries were 6.1 m (20 ft) wide with 24.3 × 24.3 m (80 × 80 ft) centre-to-centre pillars (Spearing, et al., 2011).

Two extensometers were installed at each intersection and one in the mid-pillar. The extensometers were anchored 3.66 m (12 ft) into the roof to measure differential movements. Two tilt-meters (shear meters) were installed in

each entry, one in the mid-pillar and one in the intersection.

Finally, closure meters were placed at each mid-pillar and intersection (Kostecki, 2013).

Each instrument was given an individual identification number to describe the instrument type (i.e. extensometer, closure, tilt, rockbolt) and the mine location. For example, the centre instrumented bolt in the intersection of the no. 5 entry is 100575023. In this case the first 5 denotes the mine site (in this case Mine A). The 7 denotes the instrument type (in this case an instrumented rockbolt) and finally, the 5023 denotes the unique instrument identification number. A similar nomenclature is followed for remaining instruments, except the 7 is replaced by a 9 for extensometers, 13 for tiltmeters, and a 2 for closure meters.

Mine A local geology Mine A is located in southwestern Indiana and exploits the Danville No. 7 seam of the Dugger Formation. Borescopes to 4.27 m (14 ft) above the coal seam were taken at each instrument site and the roof lithology is shown in Figure 8. In the immediate roof, the first 0.7 m (2.3 ft) on average was a medium-grey silty shale. This was overlain by a medium dark gray shale to the top of the 4.26 m (14 ft) borescope.

Hairline separations were present in the bottom 0.3 m (1 ft) of the immediate roof in entries no. 5 and no. 7, with the most discernible separation in the no. 7 mid-pillar, where a 6.35 mm (0.25 inch) -45° hairline separation existed. Within the no. 6 entry, hairline separations existed from the 0.61–0.91 m (2–3 ft) level. The immediate floor was a soft medium-grey claystone common to Illinois Basin Mines. The coal seam thickness was 1.34 m (4.4 ft), on average, and was relatively flat-lying. The average mine height was 2.29 m (7.5 ft) with an overburden of 97.5 m (320 ft).

Cutters existed throughout the mine site, shown in Figure 9, along most of the entries and crosscuts. The most significant was a 15.2–30.5 cm (0.5–1.0 ft) cutter at the eastern corner of the no. 6 intersection. A normal fault also existed in the no. 7 entry. This was not discovered until after the instruments had been installed and was not intended as part of the project design. The normal fault had a strike in the N10°W direction with a dip of 20° and a throw of 5 ft, which completely displaced the coal seam. A diagram of the normal fault is shown in Figure 10.

Figure 7—Instrumentation layout at Mine A, comprising instrumented bolts: multi-point extensometers (#), shear meters, (+) and closure meters (*). Mine A: (o) – normal primary bolts used by mine, (o) –- [FGTR], (o) – [FGPR] and (o) – [RMAB] (Spearing, et al., 2011)

▲

794

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

In situ monitoring of primary roofbolts at underground coal mines in the USA Two extensometers were installed at each intersection and one in the mid-pillar, as well as tiltmeters and closure meters. The instrument identification numbers follow the same scheme as previously discussed for Mine A. The entries were 5.5 m (18 ft) wide with 22.9 × 22.9 m (75 × 75 ft) centre–to-centre pillars (Spearing, et al., 2011).

Figure 8—Borescope logs for Mine A (Spearing and Gadde, 2011)

Mine B Instruments at Mine B were installed as primary support oncycle. The instruments were all installed in the same entry (i.e. entry 3 – Figure 11). This was mainly because of ventilation issues at the mine, and limited access to the adjacent entries at the time of installation. Parallel entries similar to Mine A were initially designed for.

Instrumented roofbolts were again installed in the intersections and mid-pillars regions. The arrows in Figure 11 show the direction in which the face was advancing during instrument installation. The FGPR instrumented bolts are denoted by the red circles up to crosscut 12 in Figure 11, RMAB are blue up to crosscut 13, and the FGTR are green up to crosscut 14. The non-instrumented support surrounding each test site was of the same type as the instrumented support, and each instrument was zeroed prior to instrumentation.

Figure 9—Cutter and fault mapping for Mine A (Spearing and Gadde, 2011)

Figure 10—Mapping of fault located in the no. 7 entry of Mine A (Spearing and Gadde, 2011)

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

795

▲

Figure 11—Instrumentation layout at Mine B, comprising instrumented bolts: multi-point extensometers (#), shear meters (+) and closure meters (*).

Mine B: (o) – normal primary bolts used by mine, (o) – [FGPR], (o) – [RMAB], and (o) – [FGTR] (Spearing, et al., 2011)

In situ monitoring of primary roofbolts at underground coal mines in the USA Mine B local geology Mine B is also located in southwestern Indiana, and exploits the Springfield No. 5 coal seam of the Petersburg Formation.

The depth to the instrument site was 106.7 m (350 ft) on average. Borescopes were taken at Mine B as at Mine A, although only four borescopes were obtained in total. The lithology for Mine B is shown in Figure 12.

The immediate roof comprised a black shale in the first 0.6 1m (2 ft) overlain by a dark grey shale with limestone lenses up to 3.05 m (10 ft) in the most extreme case. This was overlain with a brown to medium grey sandy shale up to the top of the 4.27 m (14 ft) borescope. Hairline separations were found from 0.15 m (0.5 ft) up to almost 1.83 m (6 ft).

The area was again underlain by weak underclay.

Figure 13—Instrumentation site at Mine C relative to advancing longwall. The arrows denote the direction of advance (Reisterer, 2011)

Mine C Instruments at Mine C were installed as supplemental support. Figures 13–17 show the instrumentation sites. The gate roads were already developed and had been supported with primary and secondary support. The instrumented supports were installed in the mid-pillar and intersections and were all located in the same entry. The FGPR bolts were installed in the no. 86 crosscut and mid-pillar, the FGTR bolts were installed in the no. 84 crosscut and mid-pillar, and RMAB bolts were installed in the no. 82 mid-pillar and crosscut. The direction of advance of the longwall is shown in Figure 13.

Two extensometers were installed at each intersection and one in the mid-pillar, as well as tiltmeters and closure meters. The instrument identification numbers followed the same scheme as previously discussed. The gate road was 5.8 m (19 ft) wide with 41.1 × 61.0 m (135 × 200 ft) abutment pillars at a depth of 366 m (1200 ft) (Spearing, et al., 2011).

The primary support at the mine consisted of FGPR bolts spaced at 1.52 m × 1.52 m (5 × 5 ft) spacing along and across the entry, with wire mesh. The secondary support used a similar bolting pattern but added steel straps between the previously installed primary supports. The tailgate support also included 22 inch (56 cm) metal cans or cribbing at 3.05 m (10 ft) intervals (Reisterer, 2011). All in all, the tailgate was very well supported prior to and after installation of the instrumentation.

Figure 14—Generalized view of the instrumented bolt locations for Mine C (Reisterer, 2011)

Figure 15—FGPR at mid-pillar and intersection, Mine C instrumentation site (crosscut 86) (Spearing, et al., 2011)

Figure 12—Roof lithology for Mine B instrumentation site (Spearing, et al., 2011)

▲

796

OCTOBER 2014

VOLUME 114

Figure 16—FGTR at mid-pillar and intersection, Mine C instrumentation site (crosscut 84) (Spearing, et al., 2011) The Journal of The Southern African Institute of Mining and Metallurgy

In situ monitoring of primary roofbolts at underground coal mines in the USA

Figure 17—RMAB at mid-pillar and intersection, Mine C instrumentation site (crosscut 82) (Spearing, et al., 2011)

Mine C local geology Mine C is a longwall coal mine located in Colorado and exploits the Wadge coal seam. Borescopes were conducted at Mine C as at mines A and B. The lithology for Mine C is shown in Figure 18.

The immediate roof consisted of broken shale (Unit A) in the first 0.19 m (0.625 ft) on average. This was overlain by shale (Unit B) up to 0.52 m (1.7 ft) on average. Interbedded sandstone and shale (Unit C) overlaid the shale up to 1.52 m (5.0 ft) in the most extreme case. Sandstone with minor shale interbeds (Unit D) constituted the region 2.59–2.89 m (8.5–9.5 ft). Above this region were interbedded shale and sandstone (Unit E), sandstone and shale (Unit F), and shale (Unit G). Water ingress was also noticed within Unit F and Unit E during scoping (Figure 18).

were not intrinsically safe f (i.e. not rated ffor use in potentially explosive atmospheres) and therefore were not permitted by MSHA law to operate until fresh air had been established in the entries. This caused a significant delay (6–10 days) in the initial bolt readings for mines A and B and a short delay (1–2 days) for Mine C (Spearing, et al., 2011).

The earliest bolt readings are shown in Figure 19, which represents an average of all the initial bolt readings from each bolt type (FGPR, FGTR, and RMAB) from each mine (A, B, and C).

The initial assumption was that the active bolts would show a few tons more load on installation than the passive bolts, as they are tensioned on installation (4–5 t versus 1–2 t), especially since the roofbolter was set to 325 footpounds (441 Nm) torque (Kostecki, 2013). However, the results showed that the initial loads on the bolts were not significantly different. Although the results may be a poor representation of the initial loads at mines A and B, because of the delay in readings, the readings at Mine C were obtained soon after installation, and yet there was no significant difference in loadings.

Results and discussion As described earlier, the data loggers were set to extract readings from the instruments at hourly intervals. These readings were then downloaded to the computer during regular visits to the mine every two to four weeks. Since there were a total of 170 instruments, the data obtained was extensive. For instance, for Mine A alone over 200 000 individual readings were taken from the instrumented bolts.

Detailed analysis all of this data is out of the scope of this paper. Several papers have already been published containing a detailed discussion of results from each mine (Spearing and Gadde, 2011; Spearing, et al., 2011; Reisterer, 2011; Ray, Gadde, and Spearing; 2012; Kostecki, 2013). This paper reports and comments on only the broadest findings.

Initial bolt loads As mentioned previously, the data was obtained using the YieldPoint d4 data-loggers. Unfortunately, these data-loggers

Figure 19—Initial bolt loads from mines A (top), B (middle), and C (bottom), as presented by Spearing, et al., (2011)

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

797

▲

Figure 18—Roof lithology for Mine C instrumentation site (Spearing, et al., 2011)

In situ monitoring of primary roofbolts at underground coal mines in the USA Table II displays a more intuitive representation off the average bolt loads per system and per mine. From this table it can be shown that the time delay for the first readings at mines A and B had a significant effect on the loadings overall, as some mining-induced loadings had occurred. More importantly though, Table II captures little difference in loadings between the bolt systems, as the FGPR averaged 2.68 t, FGTR 2.71 t, and the RMAB 2.62 t. Several mechanisms have been proposed to account for the lack of difference in loadings:

Modelling Fast Lagrangian Analysis of Continua FLAC3D (Itasca, 2010) was utilized in an attempt to calibrate a numerical model to the bolt loadings obtained from all three mines. The results from all three mines and bolts systems were generally the same, therefore only results from the passive (FGPR) bolts at Mine A will be discussed.

The roof lithology (Figure 8) and the floor lithology for Mine A were generated into a global grid, shown in Figure 20. This represented a portion of an entire panel at Mine A. The objective was to simulate the mining-induced loadings generated in the bolts as the panel progressed to two points in time – when the first readings were acquired and the readings at the beginning of the next month. These two points and times correspond to the relative face positions shown in Figure 21. Also shown in Figure 21 are the locations of the no. 5, 6, and 7 entries where the FGTR, FGPR, and RMAB were installed.

The panel was a large region, therefore to limit the model run-time, the zones were kept relatively large throughout (10 × 10 ft; 3.05 × 3.05 m), except in the area of interest. In this case the no. 6 mid-pillar and intersection were of interest because this was the location of the passive instrumented bolts. Therefore the zone sizes were ’densified’ via a FISH subroutine, which broke the zones down into smaller 0.3 m × 0.3 m (1 × 1 ft) zones. An excavation sequence, which made a 12.2 m (40 ft) cut of coal and then placed a pattern of bolts, was then modelled. The excavation sequence progressed until the face positions were matched to those shown in Figure 21.

At this point the loads on the instrumented bolts were extracted from the model and processed in a similar manner as the actual instrumented bolt data. Figure 22 shows the same global grid as in Figure 20 with the lithology above the coal pillars removed to show the excavations generated. In

➤ The first mechanism, which was proposed by Spearing et al. (2011), is that when upthrust from the bolter is applied to the bolt during installation a significant reduction in tension on the active bolts can be lost, as opposed to active bolts installed with zero thrust. This can be further justified by observing the bolts at Mine C. Although only the averages of all the bolts at Mine C are shown in Figure 19, some individual bolt readings showed negative axial tension. This was not observed at mines A and B, probably because of the delay in obtaining the initial readings, and enough time had passed for mining-induced loadings to the bolts. This behaviour has also been observed by several other researchers (Karabin and Debevec, 1976; Mahyera Kempen, Conway, and Jones, 1981; Mazzoni, Karabin, and Cybulski, 1981) ➤ Another likely reason could be resin creep occurring soon after installation. This would be particularly evident in the active bolts, as very small displacements could result in the load loss of the bolts.

Overall, no significant difference was found in the initial loading phase at all three mines. Additionally, more detailed statistical analysis found that there was no significant difference in loadings over time either (Kostecki, 2013). In particular, Kostecki (2013) observed that, for the instrumented bolts in this study, if 70% of the bolt yield was assumed to be applied during bolt tensioning, then only 0.156 inches of bolt displacement would be needed to either retain or lose tension. Seventy per cent of the bolt yield was chosen because the torque-to-tension ratio was not known for the bolts and this was the recommended value given by a leading bolt manufacturer. Considering this, it was determined that when bolts are installed where the immediate roof is prone to weathering, such as the case at Mine A, this small amount of displacement could ’release’ the tension in the active bolts over time – that is, if the load was not already lost soon after installation. A similar observation was made

by Unrug, Padgett, and Campoli (2004).

Figure 20—Global grid for Mine A

Table II

Initial bolt loads (Reisterer, 2011) Time between installation and 1st reading

FGPR, t

FGTR, t

RMAB, t

Average, t

Mine A

6–9 days

1.52

2.92

3.25

2.56

Mine B

10 days

5.83

4.41

3.74

4.66

Mine C

1–2 days

0.68

0.8

0.88

0.79

2.68

2.71

2.62

Average

▲

798

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

In situ monitoring of primary roofbolts at underground coal mines in the USA Conclusions After comparison of the in situ data of the three most popular primary roofbolt systems, there was no evidence to indicate a difference in performance of the active primary roofbolts versus the passive primary roofbolts on any of the three mines. Additionally, in the initial loading phase, the active bolts showed no difference in loading, indicating that tension bleed-off is more of a concern than originally thought. For the initial computer modelling studies, of replicating in situ primary roofbolt loading mechanisms, challenges still remain in obtaining a good match and replicating the discontinuous

Figure 21—Map of the relative face positions for Mine A

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 22—The global grid showing the excavation sequence and the densified region of the FGPR bolts

Figure 23—Actual and modelled elastic-perfectly plastic Mohr Coulomb (First Scan Plastic and June 30 Plastic) loadings for the 100575073 FGPR bolt (top) and the 100575074 FGPR bolt (bottom) VOLUME 114

OCTOBER 2014

799

▲

the case of Figure 22, the face positions match the exact face position of the ’Start of July Cuts’ shown in Figure 21. Also shown is the densified no. 6 mid-pillar and intersection where the FGPR bolts were located. The model was assumed to undergo elastic-perfectly plastic behaviour and followed the Mohr-Coulomb failure criteria.

Bolts 100575073 and 100575074 were arbitrarily chosen for discussion because they best represent the challenges still remaining in the modelling portion of this study. The actual results from the instrumented bolts at two points in time (i.e.

First Scan Actual and June 30 Actual) and the results of an elastic-perfectly plastic Mohr Coulomb model at two points during the excavation (i.e. First Scan Plastic and June 30 Plastic) are shown in Figure 23. The actual bolt loadings of the FGPR bolts tend to show peak loadings nearest to the excavation hangingwall, shown by point A in Figure 23. This was to be expected, as over time FGPR bolts are loaded by the downward movement of rock, which should transfer down the length of the rebar to the head of bolt. For the modelled bolts, the peak loadings tend to be somewhat near the head of the bolt but not nearly as distinctly as the actual loadings.

Additionally, the modelled loads tend to be generally lower and less pronounced with no real peaks in loading, as is the case for bolt 100575074 (point B of Figure 23). In the actual bolts, the peaks are generated because of some discontinuity driving the loading at that point (e.g. dilation or shearing of the strata). For example, from point A in Figure 23, it is unclear what phenomena are driving this peak load, but these peak loadings were not re-created within the model.

It was therefore concluded that although the modelling shows potential to recreate the in situ loadings, challenges still remain. A particular challenge is replication of the bearing plate at the head of the bolt, which could be the reason behind the lack of peak modelled loads nearest to the hangingwall. Challenges also remain in modelling the geological structure on a local scale. For instance, dilation and shearing of the immediate roof over time could account for the absence of peak loadings shown by point B in Figure 23.

In situ monitoring of primary roofbolts at underground coal mines in the USA rooff rock and in situ bolt behaviour over time. In particular, simulation of the bolt bearing plate and the recreation of the geological structure seem to be the greatest challenges at this time. This reinforces the idea that in situ measurements are still needed for design and to improve current support practices. The in situ monitoring technology from YieldPoint worked well over the entire monitoring period. The increased coverage of the long baselength displacement sensor technology improves measurement of the overall load profile of the rockbolts. Due to the length of the long baselength sensors, and the end-to-end arrangement of sensors, a more averaged loading profile is obtained at the expense of localized loads captured by short baselength strain gauges used by previous in situ bolt studies.

Secretary ffor Mine Safety f and Health. http://www.msha.gov/Stats/Part50/ Yearly IR's/2011/Coal Injury Experience-2011.pdf PADGETT, J. 2010. Personal communication (private consulting geologist).

RAY, A., GADDE, M., and SPEARING, A.J.S. 2012. Comparison of the performance of active and passive roof bolts in an Illinois Basin coal mine. Proceedings of the 31st International Conference on Ground Control in Mining. g West Virginia University, Morgantown, West Virginia.

REISTERER, J.R. 2011. The interaction of active or passive roof bolts, stress conditions, and the immediate roof strata in a longwall mine in the United States. Master’s thesis, Southern Illinois University Carbondale, Carbondale, IL. SERBOUSEK, M.O. and SIGNER, S.P. 1984. Load transfer mechanics in fully-

Acknowledgements The funding and support provided by NIOSH (under BAA number 2008-N-10989) is greatly acknowledged, as well as the considerable support given by Peabody Energy. Special thanks to Jennmar Corporation, Minova USA, and Dr A.J. Hyett from YieldPoint Inc. for providing the instrumentation for the project and for their assistance during the instrumentation installation.

grouted roof bolts. Proceedings of the 4th International Conference on Ground Control in Mining. g West Virginia University, Morgantown, WV. pp 32–40.

SIGNER, S. 1988. Comparative studies in the mechanics of grouted roof bolts.

Proceedings of the 7th International Conference on Ground Control in Mining. g West Virginia University, Morgantown, WV. pp 282–288.

SIGNER, S.P., COX, D.J., AND JOHNSTON J.L. 1997. A method for the selection of

References

rock support based on bolt loading measurements. Proceedings of the

ITASCA CONSULTING GROUP, Inc. 2010. Fast Lagrangian Analysis of Continua in 3

16th International Conference on Ground Control in Mining. g West Virginia University, Morgantown, WV. pp. 183–190.

Dimensions, version 4.0. Minneapolis, MN.

KOSTECKI, T. 2013. The instrumentation of primary roof bolts in a room-andpillar mine and the modeling of their performance. Master’s thesis, Southern Illinois University Carbondale, Carbondale, IL.

KARABIN, G J. and DEBEVEC, W.J. 1976. Comparative evaluation of conventional

SIGNER, S.P., FRANKLIN, G., MARK, C., and HENDON, G. 1993. Comparison of active versus passive bolts in a bedded mine roof. Proceedings of the 12th Conference on Ground Control in Mining. g Lakeview Resort and Conference Center, Morgantown, WV. pp. 16–23.

SPEARING, A.J.S. and GADDE, M.M. 2011. Final report on NISOH funded project

and resin bolting systems. Informational Report 1003. Mine Enforcement

Improving underground safety by understanding the interaction between

and Safety Administration, Arlington, VA.

primary rock bolts and the immediate roof strata. NIOSH Project, BAA no.

2008-N-10989. National Institute for Occupational Safety and Health,

MAHYERA, A., KEMPEN, C.J.H.B., CONWAY, H.P., and JONES, A.H. 1981. Controlled

Atlanta, GA.

thrust and torque placement of mechanical anchor bolts and their relationship to improved roof control. Proceedings of the 1st International Conference on Ground Control in Mining, g Morgantown, WV. pp. 98–105.

SPEARING, A.J.S., GADDE, M., RAY, A., and LEE, S. 2011. The initial performance of commonly used primary support on US coal mines. Proceedings of the 30th International Conference on Ground Control in Mining. g Lakeview

MAZZONI, R.A., KARABIN, G.J., and NS CYBULSKI, J.A. 1981. A trouble-shooting guide for roof support systems. Informational Report 1237. Mine Safety and Health Administration, Arlington, VA..

Scanticon Resort & Conference Center, Morgantown, WV, 26–28 July.

SPEARING, A.J.S., HYETT, A.J., KOSTECKI, T., and GADDE, M. 2012. New technology for measuring the in- situ performance of rock bolts. International Journal

MARK, C., PAPPAS, D.M., and BARCZAK T.M. 2009. Current trends in reducing

of Rock Mechanics and Mining Sciences, vol. 57. pp. 153–166. doi: 10.1016/j.ijrmms.2012.07.027

groundfall accidents in U.S. coal mines. Proceedings of the SME Annual Meeting and Exhibit. t Society for Mining, Metallurgy, and Exploration, Inc., Littleton, CO. pp. 1-5.

TADOLINI, S.C. and MAZZONI, R.A. 2006. Twenty-four conferences: more than one-hundred seventy papers; understanding roof bolt selection and design still remains priceless. Proceedings of the 25th International Conference

MINE SAFETY AND HEALTH ADMINISTRATION. 2012. Injury experience in coal mining, 2010. (IR 1354). Department of Labor, Assistant Secretary for Mine Safety

on Ground Control in Mining. g Lakeview Scanticon Resort & Conference Center, Morgantown, WV. pp. 382–389.

and Health. http://www.msha.gov/Stats/Part50/Yearly%20IR% UNRUG, K., PADGEET, P., and CAMPOLI, A. 2004. Coal mine primary support

27s/2010/Coal%20Mining%202010.pdf

selection: tension versus non-tensioned roof bolt systems. Proceedings of MINE SAFETY AND HEALTH ADMINISTRATION. 2012. Injury experience in coal mining, 2011. Informational Report 1359. Department of Labor, Assistant

▲

800

OCTOBER 2014

VOLUME 114

the 23rd International Conference on Ground Control in Mining. g Lakeview Scanticon Resort & Conference Center, Morgantown, WV. pp. 258–263. ◆ The Journal of The Southern African Institute of Mining and Metallurgy

Pillar behaviour and seismicity in platinum mines

by S.M. Spottiswoode and M. Drummond*

Crush pillars are widely used in mine workings on the Merensky Reef in the Bushveld Complex to prevent panel collapses. Crush pillars are expected to fail as or soon after they emerge from the face and failure should occur non-violently. Unfortunately, violent failure occurs frequently and is said to be the main cause of seismicity associated with mining of the Merensky Reef.

Recent work by Napier, Malan, and du Plessis has shown that limitequilibrium quasi-static models are able to simulate pillar failure using three states of strength of rock in pillars, namely intact, residual after failure, and decayed strength after later time-dependent (viscous) weakening. We have previously introduced an additional state of strength to account for the dynamic failure that results in seismic events, and found that this approach could be used to generate synthetic seismic catalogues similar to observed seismicity for deep-level gold mines, where seismicity takes place mostly on advancing faces. The less brittle seam material of the Merensky Reef, compared to the brittle quartzites and lavas of the Witwatersrand reefs, results in little or no face bursting and is modelled with an assumed plastic strain of some 0.005 over an effective stope width of 2 m before failure. When this plastic yield is surpassed, we allow the reef to fail ‘seismically’.

We show that synthetic seismic catalogues modelled in this way have some of the features of observed seismicity. Analysis is greatly facilitated using our custom-built software that reads the mine’s survey data into a database and presents results in an interactive graphical form.

Keywords seismicity, dynamic failure, crush pillars, pillar behaviour, pillar failure, simulation, numerical modelling.

Introduction One of the hazards of deep-level Witwatersrand gold mines is rockbursting associated with mine seismic events. This hazard is addressed principally through the designs for stoping layouts and support (e.g. g Jager and Ryder, 1999). Three broad factors are considered for layout design, namely regional support in the form of rock pillars, local stability through correct sequencing and face shapes, and the presence of geological discontinuities and intrusions. Backfill is widely used to provide a dual function of regional and local support. Spottiswoode et al.

(2008) studied dip pillar mining at two mines and found that seismicity was proportional to elastic strain energy release and the dip pillars were performing stably as required.

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

* Drummond Technical Services. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12–14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

801

▲

Synopsis

While rockbursts are much less common in the mines of the Bushveld Complex, they are considered to be an increasing hazard (Ledwaba et al., 2012). Essrich et al. (2011) recommended that similar hazard control measures to those applied for deep gold mines be applied for mining of the platinum reefs, in particular the Merensky Reef. These measures include avoidance of highly-stressed remnants, ‘improved sequencing’, prevention of long abutments, adherence to stopping lines between raise lines, limiting and controlling leads and lags, and ‘cutting pillars to the correct width and length dimensions’. These ‘suggested changes’ are arguably too vague to be readily implemented without more studies of the relationship between mining geometry and sequencing and associated seismicity. We hope to show in this initial study that the modelling and back-analysis methodologies that we present for the first time can be applied to answer some of the questions implicit in the broad recommendations of Essrich et al.

(2011).

In this paper we simulate possible pillar failure and further deformation using a limit equilibrium model for pillar strength. Limit equilibrium modelling was suggested by Brummer (1987) using a high ‘effective’ stope width (mining height) as a way to explain the vertical extent of fracturing around deep-level Witwatersrand stopes. Napier and Malan (2007) applied limit equilibrium modelling to simulate deformations of in-stope pillars using two measures of reef strength, namely ‘intact’ and ‘residual’ (post-failure). The same authors (Malan and Napier, 2011) referred to uncertainties in pillar strength and loading

Pillar behaviour and seismicity in platinum mines stiffness ff as reasons ffor moving beyond the sole use off empirical formulae for pillar design towards combining monitoring and numerical modelling to obtain the best insights into design problems.

They extended the work further (Napier and Malan, 2012) by considering a third measure of strength to account for time-dependency in the fractured rock mass, and introducing a ‘decayed strength’ that is reached exponentially and is described in terms of a half-time (for example 20 days) in a manner similar to radioactive decay. Du Plessis et al. (2011) presented modelled stope closure data and concluded that Napier and Malan’s (2012) time-dependent model was not adequate to explain observed stope closures.

They plan further fieldwork to guide changes to their model.

The work by Napier, Malan, and du Plessis mentioned above followed a considerable amount of fieldwork and detailed numerical modelling by Watson and others (e.g Watson et al., 2008; 2010) to provide some measures of intact and post-failure strength.

The modelling work presented here uses the limit equilibrium approach of Napier, Malan, and du Plessis with a fourth value of strength, namely ‘dynamic strength’, the stress held by an element of reef when it initially loses its intact strength or when its short-term post-failure strength is exceeded. These four measures of strength were introduced

by Spottiswoode (2001) for studies of seismicity associated with the Witwatersrand deep gold mines. We assess the extended model by comparing catalogues of synthetic seismicity with observed seismicity.

Source mechanisms of Bushveld Complex mine seismicity Impala Platinum mines the Merensky and UG2 reefs in the western limb of the Bushveld Complex. Seismicity has been increasing (Ledwaba et al., 2012) and is considered to be an increasing hazard. Most of the seismicity, as indicated by rockburst damage, is associated with the in-stope pillars, which are nominally 6 m by 3 m in size but may differ substantially for various reasons.

Whereas seismic source mechanisms on deep-level gold mines are generally compatible with shear failure (e.g Hoffmann et al., 2013), source mechanisms of events at Impala (Spottiswoode et al., 2006) and other mines in the area are compatible with pillar failure and accompanying stope closure (Malovichko et al., 2012).

U is the UCS, as experienced at the fface c is the cohesive strength against shear μ is the coefficient of friction S

T = (Eww) is a measure of the slenderness of each small (T<<1) limit equilibrium element, with Sw being the stope width (or height in layman’s terms) and Ew the element width.

Eliminating τi and σi from Equations [1], [2], and [3] we have an equation for the increase of horizontal stress ahead of the face: [4] where constants e and f are given by [5] and [6] and then the vertical limit equilibrium stress is given by Equation [1].

Note that Equation [4] does not guarantee that the confining stress increases with increasing distance from the face, as a value of element width Ew ≥ μmSw causes the horizontal confining stress, and hence the limit equilibrium strength, to become infinite (when T = μm) or negative. This is clearly incorrect, as Napier and Malan (2011) and others have shown that the limit equilibrium strength increases exponentially with distance from the face. This mathematical problem disappears in the limit as the element width tends to infinity (T → ∞ or Ew → 0). We therefore take care to choose Sw values of element size Ew << μm within practical limits.

Identifying and measuring synthetic seismic events Constructing a synthetic seismic event As seismicity on the Bushveld Complex platinum mines appears to be predominantly associated with in-stope pillars, we cumulated convergence associated with seismic stress drop within unmined areas to form synthetic seismic events.

The ‘location’ of a synthetic event is given as the centre of gravity of the convergence on the elements. Most events locate on in-stope pillars and, for most pillar shapes, plot within the pillar. Plots of seismicity and mining within a finite time window of may incorrectly appear to have occurred in mined ground, as will be seen in Figure 5.

A limit equilibrium model for pillar failure Following the lead of Malan and Napier (2011), we model pillar strength in terms of a limit equilibrium model. Figure 1 is a discretized approximation of reef strength in terms of normal and normal stresses ahead of a face or into an abutment or pillar.

[1] [2] [3] where m is the strengthening factor with confining stress s

▲

802

OCTOBER 2014

VOLUME 114

Figure 1—Forces and dimensions for limit equilibrium model The Journal of The Southern African Institute of Mining and Metallurgy

Pillar behaviour and seismicity in platinum mines Calculating the size (moment-magnitude) of synthetic events Sudden loss of reef strength and accompanying convergence of the roof and floor of the stope and across the failed reef constitutes a seismic event in this paper. The ‘size’ of the event is given by the volume of convergence multiplied by an elastic constant. In general, the volume change attributed to a pillar cannot be calculated by direct summation as associated stope convergence might be affected by deformations in other areas of reef. We take an energy approach to calculate the volume of convergence.

Consider three states of a pillar-stope configuration at each on-reef element, namely before any mining, and the states immediately before and after a seismic failure, as in Figure 2.

The work done to move from the previous to the current configuration can be written either directly from step 1 to 2 or from the difference in work done from virgin (pre-mining) conditions to steps 1 and 2 (see Table I): [7]

This simplifies to [8] Separating the reef (R) and stope (S) components and setting the zero values from Table I gives [9] The change in volume around the failed reef and associated stoping is then [10] and is called ΔV here.

We compare observed seismicity to mining in any area and time period using

database to store both mine plans and solutions. The main reason for employing a database was encapsulation (i.e. avoidance of an otherwise messy file system) but other database features were attractive with respect to efficient data retrieval (e.g indexing).

MinX essentially employs a pipeline architecture, as pipelines accommodate later modifications and extensions with a low amount of disruption (Garlan and Shaw, 1993). It is hoped that as the software reaches maturity, the pipeline aspects may be exposed via the user interface to allow users to reproduce or create new workflows as well as modify existing processes.

MinX process An advantage of MinX is that surveyed digital mining geometries are employed. Monthly mining polygons are imported in ESRI Shapefile format (an open format) – .dgn (Bentley Microstation format) and other proprietary formats are easily converted to Shapefile format via the mine survey department’s software. The authors attempt to employ only open, published formats to allow for easy sharing and access of data – e.g the ESRI Shapefile format, which has become a de-facto standard for spatial data (ESRI, 1997) Mining outlines are stored within the MS Access database in Shape binary format, though it is intended to employ the WKB format (Well Known Binary format of the Open Geospatial Consortium) in future versions (Open GIS Consortium, 1999). Queries are executed based on both mining date ranges and geometric intersections between the area of interest and bounding boxes of mined polygons.

Optional on-screen digitizing allows users to make corrections to mining layouts. This has proved to be particularly useful in the elimination of ‘slivers’ (thin unmined areas between adjacent polygons due to poor digitizing) that may, during rasterization, be interpreted incorrectly as pillars.

The digitizing facility also allows users to reconcile the boundaries of early unsurveyed layouts with later surveyed polygons. Polygons representing mined areas can be created for any particular mining step. Users can also elect to overwrite mined polygons with their own unmined polygons.

User-created polygons reside in a layer of their own and do not affect the imported surveyed polygons – they are only combined in the raster created during the scan process.

[11] where χ is an elastic modulus taken as (λ+2G)/2.

Table I

MinX MinX is a software environment for creating, running, and visualizing MinFT solutions. MinX employs an MS AccessTM

Symbols used to describe the stresses and displacement discontinuities on unmined and mined ground Mining step

0. No mining 1. Previous step 2. Current step

Figure 2—Sketch of pillar and stope convergence associated with pillar failure The Journal of The Southern African Institute of Mining and Metallurgy

Stress

Displacement discontinuity

Stope

Reef

Stope

Reef

σV √

σV √

0

0

0.

σ1iR √

D1i ?

S

D1i √

0.

σ2Ri

S D2i

D2i √

√

?

R

R

√: values provided directly as part of the simulation ?: unknown values VOLUME 114

OCTOBER 2014

803

▲

Computer program development

Pillar behaviour and seismicity in platinum mines Sheets are created and positioned interactively by the user on the mine-plan. Dip and dip direction are calculated automatically on placement, based on local geometry, but can be over-ridden if desired. All sheet-related parameters and MinFT run-flags are easily set through dialogs (brief descriptions, default values, and ranges of allowed values are also provided for run-flags).

Mine plan polygons are scanned (per mining step) into sheet images (comprising a raster of percentage-mined blocks). The creation of MinFT input files, launching of the MinFT solver, and subsequent loading of the solution back into the database are automated.

every time a solution is run and loaded. Whether it will be beneficial to provide the databases to MinX through links in an essentially empty master database will have to be determined through experiment. It remains to be seen, now that the initial desired single database model is no longer feasible, whether multiple linked databases would prove to have any advantage over multiple files (the situation the developers were trying to avoid). Certainly, the single database model would work well with a shared ‘industrialstrength’ database (e.g MS SQL Server) and the additional spatial functionality would be welcome, but this would impact severely on portability.

MinX issues

MinX interface

Not long after the first release, it became apparent that MinX and MinFT were too tightly ‘coupled’ – most obviously when problems were encountered updating both the database and the MinX software to accommodate changes and extensions made to MinFT. Given the advantages of decoupling (upgrades and extensions could be made to MinFT without corresponding changes to MinX or the database), it was decided to expend considerable effort on the problem.

Currently an xml file is employed to communicate to MinX all relevant metadata with respect to both MinFT variables (e.g name description, units) and MinFT run-flags (e.g name, description, valid values/range, default values). This has proved highly effective in allowing autonomous and asynchronous development of the MinFT solution engine.

Further problems arose as solutions for larger and larger sheets caused the 2 GB size limit of MS Access to be exceeded. At first it was thought that with automated maintenance of the database (i.e. scripted compress and repair operations) this could be avoided, but it is now evident that to accommodate very large sheets it will be necessary to provide a database per sheet anyway. This move will actually provide more decoupling as there will now be one database providing mining geometry and several sheet databases.

Maintenance of the sheet databases will in future become irrelevant as a new database will be created programmatically

In line with modern interface design, interface panes are configurable by the user and can be re-sized and either docked or floated. This allows users to configure the interface to best suit the format and resolution of the monitor(s) in use.

An overview pane ensures that there is always a view of the entire mine plan. The actual area being worked in is highlighted on this overview. It is envisaged that this pane could be made interactive to allow for fast repositioning to different areas of the mine plan.

A manager pane shows all sheets available and allows the user to select or create a new sheet to work with. As mentioned previously, when new sheets are created a dialog prompts users and assists to complete the sheet definition process. Selection of a sheet causes the surface projection window to zoom to the extents of the selected sheet. Further activation triggers the opening of an on-sheet projection window for the creation and/or viewing of a sheet solution.

Solutions are viewed in an on-sheet projection window.

The on-sheet projection window is layer-based and, as well as sheet variables, allows for visual comparison of observed vs. modelled seismic events. A glyph provides sheet orientation information (dip and dip direction), while legends provide information about the mining step and activevariable ranges.

Figure 3—The MinX interface. The small panes on the left are (from top to bottom) ‘Manager’, ‘Info’, and ‘Overview’ panes. The sheet is shown in Sheet Projection view. To the immediate right of the Sheet Projection window is the ‘Sheet-Data’ pane. Pillar histories are visible at far right in the ‘Histories’ pane. The ‘Layer-Order’ pane is visible at bottom right

▲

804

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Pillar behaviour and seismicity in platinum mines A Layer-Order pane provides a means off determining the rendering order of layers, also providing the ability to switch layer visibility on or off as desired.

The Sheet-Data pane allows selection of mining step and variable. Variables are currently refined into three groups, namely ‘Dependent’, ‘Difference’, and ‘Independent’.

‘Dependent’ variables represent the default type – values exist for every mining step – they are listed in the variables window in black font. ‘Difference’ variables are calculated from ‘Dependent’ variables and show the difference in values between the current step and the previous step – listed in blue font. ‘Independent’ variables do not change with mining step (i.e. they occur only once and are stored with a mining step value of -1) – they are listed in green font.

Pillar history plots (stress, convergence, and stress vs. convergence) are linked to pillar selection in the on-sheet projection window. Conversely, selection of an individual history in any of the plot windows will highlight the respective pillar in the on-sheet projection window.

In order to retain a sense of how the model is defined, it was decided to visualize the sheet elements as discrete grid cells devoid of interpolation (as opposed to e.g. g contours or methods employing values interpolated from cell centres to cell vertexes). The value of each cell is mapped to a colour value through a lookup table. Image manipulation methods (provided by the application framework) allowed for relatively fast rendering times. Seismic events are visualized as circles (centred on event origins and scaled to magnitude) – modelled events appear with white outlines, whereas recorded events appear with black outlines. Again, there has been no problem with rendering times – although should this at some stage prove to be a bottleneck, the circle geometries will be reduced to far simpler primitives (e.g. g squares, diamonds).

the MinX environment; to exploit the MinX graphical interface; and to accommodate the particular mining geometry used in the platinum mines.

As shown above, the limit equilibrium model requires small element sizes. The strength values are built up element by element from open stopes into the solid using the strength parameters and values in Table II. The values were based partly on those used by Malan and Napier (2012) with the ‘dynamic’ strength equal to the ultimate strength. We have not ‘tuned’ the values through back-analysis for this paper.

Mining outlines are made available on a monthly basis. A face advance of 20 m or a typical month’s mining could result in actual stress increase and decrease of over 100 MPa. It was necessary therefore to interpolate the month’s advance (called major steps) in small increments (minor steps) to track stress changes on pillars. A typical day’s face advance is about 1 m and we used an element size of 0.8 m in this study. Solutions consisted of 117 major steps (close to 10 years) extending up to September 2013 and typically over 1000 minor steps, with each step allocated to a date to allow comparison between observed and modelled seismicity.

Platinum mine seismicity and mining geometry are strongly influenced by pillars, and in-stope pillars in particular. Pillars are left either for regional support or for local hangingwall support (grid pillars). Surveying around small pillars is difficult, especially if they have started to fracture and a substantial amount of failed rock has been (correctly) left in place. The hand-digitizing mentioned above may still leave small mistakes, and some of these can be (and are) rectified automatically, such as isolated mined or unmined elements.

In MINFT, pillars are defined as areas of reef that consist of adjacent (contiguous) elements that are unmined during the last step. In most cases, there are many more in-stope (or grid) pillars than regional support pillars.

Integration with the MinX environment has required some customization of reading input and writing output files.

However, the MinX graphical interface gives many opportunities for the display of values for an assortment of parameters that describe the mining geometry and program solutions. Parameters are either fixed over the duration of the solution or change from month to month. Some time-varying parameters can be displayed in terms of the difference from month to month. The parameters were chosen during code development to assist in visualizing and debugging. We hope that users of the MinX suite will use these displays to gain

MINFT MINFT calculates stresses and displacement around tabular mining excavations using the displacement discontinuity method (DDM). The most calculation-intensive part of the DDM is calculating stresses from displacement discontinuities (elastic stope convergences). In MINFT this is done with the aid of Fourier transforms (Peirce et al., 1992).

Extensive changes to the code used by Spottiswoode (2001) were made for the platinum version of the MINFT program. Changes were necessary for introduction of the limit equilibrium model described above; to integrate MINFT with

Table II

Strength and time decay parameters U, MPa

M

C, MPa

μ

Intact Residual Ultimate Dynamic Stope width Element size Young’s modulus Poisson's ratio Reef deformation at intact strength before failure Residual strength half-life

50 5 1 1 2.0 m 0.8 m 70 GPa 0.28 10 mm 20 days

2.0 2.0 2.0 2.0

5.0 2.0 0.5 0.5

0.6 0.6 0.6 0.6

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

e 53.9 7.69 1.69 1.69

OCTOBER 2014

f 2.85 2.85 2.85 2.85

805

▲

State

Pillar behaviour and seismicity in platinum mines understanding off the how the model and data analysis works. Some of the available parameters are listed in Table III. The PANG parameter shows groups of panels that were mined ‘close’ to one another in space and time. We chose ‘close’ to be 60 m, as it seemed to be approximately the seismic location error as judged by the scatter of locations around active faces and their adjacent pillars.

As the entire stress-deformation history of each individual pillar is of interest, average stress and average displacements for every minor mining step and every pillar are written to a file for plotting by MinX. The stress field in the platinum mines appears to be approximately hydrostatic, with k-ratios (horizontal / vertical stress) of around 1.0 (Handley, 2013). For the current study we assumed k = 1.0 and hydrostatic stress. This has the advantage of eliminating the need to solve for on-reef shear stresses and ride components, thereby minimizing solution times.

Platinum mine seismicity is considered to occur on pillars (e.g. g Essrich et al., 2011). Identification of modelled seismic events is therefore generally straightforward once pillars have been identified, as is the case with MINFT. Seismic moment is calculated for sudden failure of pillars using Equation [11].

Table III

Some geometry or solution parameters displayed by MINX. ‘Type’ is ‘fixed for not time-variable, ‘vary’ for time variable, and ‘v+d’ to display time-variable parameters and their difference from step to step.

The meaning of PANG is explained in the text Short code

Type

Description

VZZ

fixed

Virgin stress normal to reef, MPa

PATT

fixed

Mining step

NAPS

fixed

Pillar number

PANG

special

Global panel number

v+d

Roof-to-floor convergence, stoping and reef deformation, mm

DZZ DZZF

v+d

Portion of DZZ on pillars

SZZ

v+d

Stress normal to reef, MPa

APS DZZS

v

Average pillar stress, MPa

v+d

Portion of DZZ that takes place ‘seismically’

Analysis Superficial analysis of a 410 m by 410 m area of Impala mine is shown here mostly in terms of screen dumps and X-Y plots. Figure 4 is a screen dump of stope convergence with pillar deformations for the study area. One in-stope pillar has been selected and its stress-convergence-time plots are highlighted in red.

Figure 5 illustrates observed and modelled seismicity.

Most modelled seismicity takes place on the pillars – some seismicity takes place ahead of the advancing faces. There would be considerably more face seismicity if we did not allow some (10 mm in this case) reef deformation at the intact strength before allowing the reef to fail. The in-stope seismicity results from failure of the face, associated in part

Figure 5—Observed (black circles) and modelled (white circles) seismicity between 15 May and 15 June 2011. Light blue through to red shading shows intensity of ‘seismic’ deformation

Figure 4—Screen dump of face outlines and DZZ (convergence) of the study area in September 2013. Dark blue is unmined and unfailed reef. The marked pillar with its history of stress-convergence-time is shown

▲

806

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Pillar behaviour and seismicity in platinum mines

Figure 6—Comparison between observed and modelled seismicity. Left: cumulated observed and modelled seismic moment as a function of cumulated area mined

The Journal of The Southern African Institute of Mining and Metallurgy

One might intuitively expect that the rate off stress drop is a factor that controls the burst potential of a pillar. To study this, we show the loading stiffness that drove each event on each pillar with a size of less than 50 m2 as a function of pillar size in Figure 9. Note the large range of stiffness values. For comparison of the estimates from the individual event loading stiffness values, we created two special simulations: 1. The stiffest loading is expected to be for the case of failure of each pillar when surrounded by unmined ground stabilized only by large pillars, as shown by the ‘Mined first’ symbols in Figure 9 2. The softest loading might intuitively be expected to occur for the pillars to be intact until they are allowed to fail completely only after all mining has place. This is shown by ‘Mined last’ symbols based on a twostage simulation: namely, all final mining followed by complete failure (evaporation) of the smaller (< 50

Figure 7—Order of mining (left) and numbered ‘global’ panel numbers (right), numbered in order in which they were mined. The panels in box 2 were started from a later raise and therefore lagged behind shallower panels and panels in box 1

Figure 8—Cumulated seismic volume as a function of cumulated area of influence for the two global panels (1 and 2). Solid lines indicate data for observed (‘o’) and dashed lines for modelled (‘m’) values VOLUME 114

OCTOBER 2014

807

▲

with the lagging configuration of panels. As expected, the observed seismicity shows considerably more scatter than the modelled seismicity.

The total amount of modelled seismicity mimics the observed seismicity quite well (Figure 6) considering all the assumptions and uncertainties and the fact that we have made very few attempts at ‘tuning’ strength values to match the modelled results to the observed seismicity. The lower number of small observed events compared to the number of modelled events is a result, in part, of network sensitivity.

We now compare modelled to observed seismicity associated with two identified areas of mining (Figure 7), based on the global panel numbers (PANG, above). This was facilitated by using a purpose-built CSV file written by MINFT. This file lists the total amount of observed and modelled seismic moment associated with each panel grouping for each month. Two well-grouped large areas are labelled 1 and 2 in Figure 7. Global panels were not always as clearly grouped, as can be seen by the wide range of colours on the right of Figure 7. Poor grouping occurred for complicated mining situations, such as when re-raising was needed to re-establish panels that had become difficult to mine.

For each area, we plot the cumulated modelled and observed seismicity expressed as inferred volume of coseismic convergence (Equation [10]) as a function of the area of influence for capturing the events (Figure 8).

As expected, the lagging panels of group 2 were more active, both for the modelled and the observed seismicity, than the previous group 1 panels. The modelled seismicity matched the observed seismicity well for the lagging faces (group 2), but over-estimated the seismicity in group 1.

Data for other global panels has not been presented here, as interpretation would arguably be best served by comparison with similar situations elsewhere on the mine within a wider study than this initial work.

The slope of the graphs in Figure 8 is analogous to γE (normalized seismic deformation) presented by Spottiswoode et al. (2008) for two much larger case studies in deep gold mines. The observed values for two global panels in Figure 8 are 0.018 and 0.068 for panels 1 and 2 respectively. This is substantially less than the values of 0.25 and 0.19 reported

by Spottiswoode et al. (2008) for gold mines.

One of the features of both the observed and modelled seismicity is that there are more events than pillars: each pillar might fail seismically many times. This is illustrated by the multiple stages of stress drop in the highlighted stressconvergence plot on the bottom right of Figure 4.

Pillar behaviour and seismicity in platinum mines that extensive back-analysis work be done to study situations that might have been considered to have been stable, but which did generate large events.

Proactive

Figure 9—Loading stiffness driving synthetic seismic events (‘Event’) as a function of pillar size for smaller pillars. Values for ‘Mined first’ were obtained by mining the pillars only while leaving the stopes unmined.

‘Mined last’ was the stiffness for each pillar by not allowing any failure of these pillars until the very last step

Under ideal geological conditions, a regular grid of in-stope pillars and regional pillars might be designed. However, the geometry of the mining in the study area contains sufficient evidence to indicate that odd-shaped pillars are likely to be used. It is essential that regional pillars, of whatever origin, do not fail entirely, nor that pillars that are meant to fail do so at a later stage. On the other hand, in-stope pillars do need to fail in or close to the face and then provide enough support resistance.

Conclusions m2) pillars. This was not the case, as much of the softest loading takes place with some elements holding some of the stress drop of failing elements. This could be important for understanding the damage potential of pillar events.

Discussion Most work to date on numerical modelling and seismic studies in South African mines has focussed on the deep gold mines (e.g. g Jager and Ryder, 1999). Much of the research has focused on shear slip on geological discontinuities (e.g. g Hoffmann et al., 2013), as large seismic events are commonly attributed to previously mapped faults or dykes.

On the other hand, seismicity is not always dominated by the mapped discontinuities, and has been found to be broadly proportional to elastic strain energy released in two mines in the Carletonville gold field (Spottiswoode et al., 2009). Elastic strain energy release for flat-dipping reefs is a weak function of the horizontal stress. Handley (2013) has pointed out that the apparently greater success with pillars aligned with the dip direction compared to pillars along strike could be due to the horizontal stress component along strike generally being greater than the dip-aligned horizontal stress. Narrow deeplevel stopes are surrounded by fractures that extend many metres from the reef. A limit equilibrium approach for Witwatersrand gold mines would perhaps need to consider several layers extending into the hangingwall and footwall.

Horizontal stress would play an important role, as would backfill in limiting the bulking of the face area.

The work presented here is mostly a description of processes for modelling deformations around a typical platinum mining geometry and sequence. The aim has been more as an introduction to what MINX / MINFT can do than how it should be used for mine planning. Application on a mine is briefly summarized in Figure 10.

This paper reports the first results from a new suite of programs to model pillars in platinum mines. The ultimate aim of this modelling is to provide a tool for on-mine rock engineers to interpret current and planned mining geometry by extrapolating comparison of historical modelled and observed seismicity into the future for better and safer mining. Further work is planned in collaboration with mine staff, and it is hoped to expand the work to other mining operations on the Merensky Reef.

Acknowledgements We thank Michael du Plessis, John Napier, Francois Malan, and Jan Kuijpers for useful discussions. The work would not have been possible without many discussions with, and assistance, from Rock Engineering staff of Impala Platinum.

Les Gardner drove the discussion that led to Figure 10. The work has been supported in part by Impala Platinum, who have also given permission to publish this paper.

References BRUMMER, R.K. 1987. Modelling the non-linear behaviour of fractured seams in deep gold mines. APCOM 87. Proceedings of the Twentieth International Symposium on the Applications of Computers and Mathematics in the Mineral Industries, Johannesburg, South Africa, 19-23 October 1987.

South African Institute of Mining and Metallurgy, Johannesburg. vol. 1, pp. 21–32.

DU PLESSIS, M., MALAN, D.F., and NAPIER, J.A.L. 2011. Evaluation of a limit equilibrium model to simulate crush pillar behaviour, Journal of the Southern African Institute of Mining and Metallurgy. vol. 111, no. 12. pp.

875–885.

ENVIRONMENTAL SCIENCES RESEARCH INSTITUTE INC. 1997. ESRI Shapefile Technical Description. ESRI White Paper. Environmental Sciences Research Institute, Redlands CA.

Reactive The work in this paper is of a reactive nature based on backanalysis of a single area. The general term ‘pillar stability’ can be divided into two components, namely well-behaved yield of grid (in-stope) pillars and large pillars maintaining a core of unfailed ground that provides regional stability by being capable of maintaining a stress an order of magnitude larger than the virgin stress. It is of utmost importance that a nominally stable pillar does not fail unexpectedly. We suggest

▲

808

OCTOBER 2014

VOLUME 114

Figure 10—Diagram showing an initial proposal for application for analysis and planning software presented here The Journal of The Southern African Institute of Mining and Metallurgy

Pillar behaviour and seismicity in platinum mines

The Journal of The Southern African Institute of Mining and Metallurgy

NAPIER, J.A.L and MALAN, D.F. 2012. Simulation off time-dependent crush pillar behaviour in tabular platinum mines. Journal of the Southern African Institute of Mining and Metallurgy, vol. 112, no. 8. pp. 711–719 OPEN GIS CONSORTIUM. 1999. Open GIS Simple Features Specification for SQL. PEIRCE, A.P., SPOTTISWOODE, S.M., AND NAPIER, J.A.L. 1992. The spectral boundary element method: a new window on boundary elements in rock mechanics.

International Journal of Rock Mechanics and Mining Sciences and Geomechanical Abstracts, vol 29, no. 4. pp. 379–400.

SPOTTISWOODE, S.M. 2001. Synthetic seismicity mimics observed seismicity in deep tabular mines. Keynote address: 5th International Symposium on Rockbursts and Seismicity in Mines. South African Institute of Mining and d Metallurgy, Johannesburg. pp. 371-378.

SPOTTISWOODE, S.M, SCHEEPERS, J., and LEDWABA, L. 2006. Pillar seismicity in the Bushveld Complex. Proceedings of SANIRE 2006: Facing the Challenges.

South African National Institute of Rock Engineering. pp. 140-158.

SPOTTISWOODE, S.M., LINZER, L.M., and MAJIET, S. 2008. Energy and stiffness of mine models and seismicity. 1st Southern Hemisphere International Rock Mechanics Symposium, Perth, Western Australia, 16-19 September 2008.

Australian Centre for Geomechanics. pp 693–707.

SPOTTISWOODE, S.M., MILEV, A.M., LINZER, L.M., and MAJIET, S. 2009. Evaluation of the design criteria of regularly spaced dip pillars (RSDP) based on their in-situ performance. Draft Final Project Report SIM 04 03 01. Safety in Mines Research Advisory Committee, Johannesburg. http://stevespot.yolasite.com/resources/RSDP.pdf WATSON, B.P., RYDER, J.A., KATAKA, M.O., KUIJPERS, J.S., and LETEANE, F.P. 2008.

Merensky pillar strength formulae based on back-analysis of pillar failures at Impala Platinum. Journal of the Southern African Institute of Mining and Metallurgy, vol. 108, pp. 449–461.

WATSON, B.P., KUIJPERS, J.S., and STACEY, T.R. 2010. Design of Merensky Reef crush pillars. Journal of the Southern African Institute of Mining and Metallurgy, vol. 110, no. 10. pp. 581–591. ◆

VOLUME 114

OCTOBER 2014

809

▲

ESSRICH, F., HANEKOM, J.W.L., STANKIEWICZ, T.B.A., and RANGASAMY, T. 2011.

Minimising the increasing seismic risk in the platinum sector. Safety in Draft Final Project Report. Project number: SIM100301. Safety in Mines Research Advisory Committee, Johannesburg. 368 pp.

GARLAN, D. and SHAW, M. 1993. An introduction to software architecture.

Advances in Software Engineering and Knowledge Engineering, g vol. I. World Scientific Publishing, New Jersey. pp. 6–8 HANDLEY, M.F. 2013. Pre-mining stress model for subsurface excavations in southern Africa. Journal of the Southern African Institute of Mining and Metallurgy, vol. 113, no.6. pp. 449–471.

HOFFMANN, G., MURPHY, S., SCHEEPERS, L., and VAN ASWEGEN, G. 2013. Surface stress modelling of some shear slip seismic events that occurred in Anglogold Ashanti’s tabular mines. 8th International Symposium on Rockbursts and Seismicity in Mines, St Petersburg and Moscow, 1–7 September 2013. Geophysical Survey of Russian Academy of Sciences, pp. 219–231.

JAGER, A.J. and RYDER, J.A. 1999. A Handbook on Rock Engineering Practice for Tabular Hardrock Mines. SIMRAC, Johannesburg.

LEDWABA, L.S., SCHEEPERS, J., DURRHEIM, R.J., and SPOTTISWOODE, S.M. 2012.

Seismic damage Mechanism at Impala Platinum Mine. SHIRMS 2012, Second Southern Hemisphere International Rock Mechanics Symposium, Sun City, South Africa, 14–17 May 2012. Southern African Institute of Mining and Metallurgy, Johannesburg. pp. 367-386.

MALOVICHKO, D., VAN ASWEGEN, G., and CLARK, R. 2012. Mechanisms of large seismic events in platinum mines of the Bushveld Complex (South Africa).

Journal of the Southern African Institute of Mining and Metallurgy, vol. 112, no. 6. pp. 419–429.

NAPIER, J.A.L. and MALAN, D.F. 2007. The computational analysis of shallow depth tabular mining problems. Journal of the Southern African Institute of Mining and Metallurgy, vol. 107, no. 11. pp. 725–742.

NAPIER, J.A.L AND MALAN, D.F. 2011. The design of stable pillars in the Bushveld Complex mines: a problem solved? Journal of the Southern African Institute of Mining and Metallurgy, vol. 111, no. 12. pp. 821–836.

Management of the Nkomati Mine crusher slope failure

by R. Armstrong* and K. Moletsane†

Due to limited available level ground, Nkomati Nickel Mine cut a weathered rock slope at the base of a mountain spur in order to create a platform for construction of the primary crusher plant and run-of-mine stockpiles. As space is limited around the mining area, ore processing at Nkomati is based on a high reliability of flow of material through the crusher plant, with minimal usage of other and larger designed stockpiles. Evidently, any crusher plant shutdown will render the mine as a whole unproductive and put excessive strain on the medium- to long-term large ore stockpiles, the deposition rates for which are restricted by founding material consolidation requirements.

At the onset of the 2012 rainy season, movement was identified on the slope monitoring system and cracks developed on the slope. After a minor failure on the crusher slope an assessment of the slope stability was conducted and a slope management plan recommended, which included deployment of real-time monitoring. An evaluation of the conditions leading to instability was conducted and the likely causes for the failure identified. A full evaluation of the slope monitoring, rainfall, and mining conditions was undertaken and movement triggers were determined. This paper describes the events leading to the development of the failure and the evaluation of the monitoring data to determine a management plan for the failure that allowed for minimal shutdowns of the primary crusher.

Keywords slope stability, slope management, slope monitoring.

Introduction Nkomati Mine is a joint venture between African Rainbow Minerals ARM (50%) and Norilsk Africa (50%), who jointly manage the mine and project. It is situated between Badplaas and Nelspruit in the Mpumalanga Province of South Africa, approximately 300 km east of Johannesburg (Figure 1). The nickel deposit is situated in a steep-sided valley with limited flat ground for mining infrastructure development, as illustrated in Figure 2.

Due to limited available level ground, a weathered rock slope was cut at the base of a mountain spur in order to create a flat platform for construction of the primary crusher plant and run-of-mine stockpiles (Figure 3). As space is limited around the mining area, ore processing at Nkomati is based on a high reliability of flow of material through the crusher plant, with minimal usage of other and The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

Historic slope performance In early 2010 a small toe failure and cracks developed on the slope above the primary crusher. A review of the design was undertaken, which included detailed mapping of the geology and laboratory testing by means of of Gradings and Atterburg Limit tests (Dlokweni and Terbrugge, 2010). The analysis of the slope indicated that the slope stability was sensitive to groundwater and slope angle, even at the relatively flat design angles of 23°.

Three recommendations were made: to flatten the slope to 19°, backfill the slope with compacted waste material, or dewater the slope and manage the risk. Following consideration of the benefits and costs associated with each option, it was recommended that the slope should remain unchanged and that survey monitoring, depressurization, and groundwater monitoring systems be installed.

An automated survey monitoring system was installed and piezometer holes were drilled. Survey monitoring has been ongoing; however, as the piezometer holes were dry on drilling, the piezometers were not installed and no further dewatering measures were taken.

No further records related to groundwater monitoring were available at the time of this study.

* SRK Consulting. † Nkomati Nickel Mine © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12–14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

811

▲

Synopsis

larger designed stockpiles. Evidently, any crusher plant shutdown will render the mine as a whole unproductive and put excessive strain on the medium- to long-term large ore stockpiles, the deposition rates for which are restricted by founding material consolidation requirements.

Management of the Nkomati Mine crusher slope failure

Figure 1—Location of Nkomati Mine

the safety f off the personnel working at the primary crusher area, and also prevent unnecessary downtime for the crusher.

Leading up to the review, the mine redeployed the open pit slope stability radar to augment the total station monitoring and dumped a waste rock buttress at the toe of the failure.

Geology Nkomati Mine is located within the Uitkomst Complex, which is intruded into the lower Transvaal Supergroup. The mafic magma intruded vertically into the Transvaal host rocks and formed into what has been described as an ‘anvil–shaped’ body.

The lower levels of the valley expose basement granite, followed by Black Reef Formation (the base of the Transvaal Supergroup) through to the Timeball Hill formation higher up. The Transvaal sediments have been intruded by disconFigure 2—Gladdespruit valley and Nkomati Mine

Figure 3—Location of the primary crusher in the mine complex

Cracks were observed early in the rainy season of 2012 high up on the slope together with an increase in displacement measured on the survey monitoring system (Figure 4). This prompted a review of the slope instability and the requirement of a management plan that would ensure

▲

812

OCTOBER 2014

VOLUME 114

Figure 4—Location of cracks on the primary crusher slope The Journal of The Southern African Institute of Mining and Metallurgy

Management of the Nkomati Mine crusher slope failure tinuous diabase sills. Geological investigations were undertaken involving mapping the slope (Dlokweni and Terbrugge, 2010) and interrogation of historic borehole cores and logs. These indicated that the slope consists of an unweathered to slightly weathered dolomite bedrock, including chert horizons along the bedding, capped by a completely weathered diabase sill. Depth of weathering was uncertain over the extent of the slope; however, the cores drilled approximately halfway up the slope indicated that the base of weathering was at 22–35 m.

Slope monitoring Following the recommendations in the 2010 review, Nkomati installed an Optron robotic total station to the south-east of the slope together with 15 reflective prisms which were measured several times a day (Figure 5). Measurements taken in the early (pre-sunrise) morning were used for regular surveillance of the slope. Following the detection of slope movement, the Groundprobe slope stability radar was redeployed from the open pit and stationed to the south-west of the slope to supply pseudo-real time slope monitoring.

Groundwater and rainfall Nkomati is in a summer rainfall area. The first rainfall in the 2012 season occurred in September with occasional showers, but the rains started in earnest during November. It is

important to note that when the piezometers holes were drilled following the 2010 recommendations and found to be dry, it was during the dry winter months, following which there is no groundwater information. On inspection of the site in December 2012 after the initiation of the slope failure, it was found that one of the old piezometers holes (located between CP23 and CP7 in Figure 5) contained standing water. Several weeks later (January 2013), the hole was dry.

This indicates that the groundwater on the crusher slope is meteoric. During the rainy season groundwater builds up along the weathered/bedrock surface. Due to the low permeability of the weathered rock at the base of the slope the water egress is slower than ingress, allowing for a pore pressure build-up that will dissipate with time.

Slope failure triggers Observations of the slope indicated that degradation of the toe had occurred resulting in undercutting of the slope. At the time of inspection (December 2012), waste rock was in the process of being dumped at the slope toe in an attempt to buttress the failure. Figure 6 shows a comparison of the toe of the slope in 2010 (left) and December 2012 (right). Based on the height of the TR100 dump truck the slope was undercut by 5 m (vertically) above the buttress. The undercutting of the toe resulted in an increase of the overall slope angle from 23° to 25°, the 2° increase being well within the sensitivity of the slope as determined in the 2010 report.

Figure 5–Location of the primary crusher slope monitoring

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

813

▲

Figure 6—Comparison of the slope toe in 2010 (left) and in December 2012 (right)

Management of the Nkomati Mine crusher slope failure

Figure 7—Survey monitoring results compared to daily rainfall

A comparison off the survey monitoring results and measured rainfall is presented as Figure 7. It is evident that the movement of the slope is related to major rainfall events.

The combination of the in situ groundwater and the oversteepening of the slope is considered to be the driving factor for the slope instability.

Remedial action and failure management plan Observations of rainfall and slope monitoring records indicate that there is a significant increase in slope movement following large rainfall events. It was recommended that any rainfall event greater than 20 mm should be considered a ‘trigger’ event and monitoring data should be reviewed and the area evacuated if necessary.

The inverse velocity technique (Rose and Hungr, 2007) was also recommended as a tool for analysing monitoring.

Since the velocity of movement becomes asymptotic as failure is approached, it stands to reason that the inverse of the velocity will trend to zero. Using this method, the inverse velocity graph can be extrapolated to an estimated time of failure. It is important to note that the method leads to an estimation, and the techniques used to extrapolate the inverse velocity trend can affect the predicted failure time. A daily average velocity and inverse velocity graph for CP2 is presented in Figure 7. The accelerations in the velocity graph can be linked to rainfall events up to two days prior to the spike. Furthermore, these accelerations can be correlated with a decrease in the inverse velocity. It should be noted that at the time of the site visit the inverse velocity had a slight upward trend. It is assumed this reduction in velocity (and increase in inverse velocity) is related to the dumping of the buttress at the slope toe.

Proposal for future remediation Following the above assessment, a stability investigation was undertaken to determine the risk to the primary crusher from

▲

814

OCTOBER 2014

VOLUME 114

further instability. This programme is currently underway, and has involved the installation of inclinometers into the slope to determine the depth of the failure surface, testing undisturbed soil samples to define the material properties, geophysical tests to determine the depth of the weathered surface, back-analysing the slope to determine the overall strength parameters, and an exercise to design a long-term remedial measure to mitigate the risk to the primary crusher.

Conclusions Following the review of the slope failure at Nkomati, the following conclusions can be drawn: ➤ The failure initiated following the undercutting of the slope and the onset of the rainy season ➤ The failure responded to buttressing the toe by a reduction in slope movement ➤ Piezometers are required to determine the actual pore pressure that builds up during the rainy season ➤ Rainfall events greater than 20 mm were considered trigger events for slope movement ➤ Following the remedial measures taken by the mine and management of the failure, the primary crusher is able to operate under the failure until permanent mitigation measures can be put in place.

Acknowledgements The authors would like to thank Nkomati Mine for the permission to publish the results of this study.

References DLOKWENI, T. and TERBRUGGE, P.J. 2010. Review of the stability of the Nkomati Primary Crusher Slope. Report no. 413130. SRK Consulting, Johannesburg, South Africa.

ROSE, N.D. and HUNGR, O. 2007. Forecasting potential rock slope failure in open pit mines using the inverse-velocity method. International Journal of Rock Mechanics and Mining Sciences, vol. 44, no. 2. pp. 308–320. ◆ The Journal of The Southern African Institute of Mining and Metallurgy

Grid-based analysis of seismic data

by J. Wesseloo*, K. Woodward*, and J. Pereira*

Sweden, local rock engineers tend to ffocus on the overall rock mass response to mining based on accurate source location and the analysis of populations of seismic events with magnitude of completeness as small as ML-2.

A grid-based spatial analysis of seismic data was developed to improve and simplify quantitative seismological interpretation within this environment. The methods, however, have broader application.

Funk et al. (1997) presented work on the visualization of seismicity that resulted in the systems for generating contours and isosurfaces of seismic parameters. The work presented here generalizes and extends the concepts used in that work.

Synopsis Quantitative seismology is an important tool for investigating mineinduced seismicity and the quantification of the seismic rock mass response. The spatio-temporal interpretation of seismic data within an ever-changing three-dimensional mining environment provides some challenges to the interpretation of the rock mass response. A grid-based approach for the interpretation of spatial variation of the rock mass response provides some benefits compared with approaches based on spatial filters. This paper discusses a grid-based interpretation of seismic data. The basic methods employed in the evaluation of the parameter values through space are discussed and examples of applications to different mine sites given.

Keywords quantitative seismology, mine induced seismicity, rock mass response, spatial evaluation.

Analysis with spatial filters

Seismographs were first installed to monitor mining-related seismicity in South Africa in 1910. The first in-mine seismic systems were installed in the 1960s, mainly for research purposes. Intensive research, development, and commercialization during the 1990s led to the widespread implementation of real-time digital monitoring systems and quantitative methods of analysis (Durrheim, 2010; Mendecki et al., 1997).

Since then, the understanding of mineinduced seismicity and the quantification of the seismic rock mass response has formed the basis of seismological analysis and seismic hazard mitigation in South Africa. This principle of quantifying and understanding the rock mass deformation and failure mechanism has been introduced in other countries in different forms and focused on different mining environments. Potvin and Wesseloo (2013) point out that the mines in Australia, Canada, and Sweden tend to have more complex three-dimensional orebodies and are generally smaller and much more contained compared to the seismically active South African mines. For that reason it is easier to install a more sensitive three-dimensional array to cover the mine volume. With the more sensitive systems in Australia, Canada, and The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

Seismic events tend to cluster at the locations of active seismic sources where some form of dynamic failure process occurs. The identification and understanding of seismic sources is important in seismic risk management in that they may be, or become in the future, the cause of significant seismic hazard. In particular, small events may start to form clusters at an early stage of extraction, with a relatively small stress change. When the seismic system is sensitive enough to capture small events, this can assist in the timely identification of seismic sources and allow for the tracking of how these sources respond to mining and, more specifically, how seismic hazard related to these sources evolves as extraction progresses.

* Australian Centre for Geomechanics, The University of Western Australia, Australia. Š The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12â&#x20AC;&#x201C;14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

815

â&#x2013;˛

Introduction

Grid-based analysis of seismic data Interpretation off the character off seismic sources in space and time is generally done by analysis of some spatially subfiltered data-set. This sub-filtering is often simply based on a three-dimensional volume in space (polygon). In Australia, Canada, and Sweden, the data is often sub-filtered on clusters.

The most widely used clustering method is probably that implemented in the software mXrap (formerly known as msrap) based on the Comprehensive Seismic Event Clustering (CSEC) technique described by Hudyma (2008). The CSEC method is a semi-automated two-pass approach, which was developed on the basis of generic clustering techniques: CLINK and SLINK (Jain et al., 1999, Romesburg, 2004). The first pass of clustering using CLINK is totally automated, and clusters together events based exclusively on spatial criteria.

The CLINK clusters are then submitted to a second pass of processing, where clusters are selectively grouped into ’cluster groups’ representing individual seismic sources. This cluster grouping is a manual process that requires interpretation of the likely seismic sources at the mine and a sound knowledge of the geology and the induced stress conditions.

Cluster grouping is generally based on the similarity of source parameters, the spatial proximity of clusters, and on the correlation of the location with known geological or geometric features.

The cluster grouping process can be seen as building a seismic source model by using the generated clusters as basic building blocks. In this sense it is similar to analysing an area with the use of polygons, as the polygons become the basic units within the seismic source model. The use of a spatial filter (cluster groups or polygons) to provide a basic spatial unit for quantitative seismic analysis is a common and practical approach. This, however, introduces a bias of interpretation towards the pre-defined polygon as the polygon is originally chosen by the analyst based on preconceived ideas. This process is, per se, subjective, and the value depends to a large degree on the understanding and training of the person performing this analysis. Subjectivity, however, is part of geotechnical engineering and attempting to eliminate subjectivity from geotechnical analysis is a futile exercise. With the grid-based approach, however, we can aim to reduce interpretation bias by providing a spatial interpretation of the data that is independent of any chosen spatial filter.

Having said this, one has to recognize that the grid-based interpretation is not free of user influence as it also is influenced by the chosen analysis parameters. It is our conviction, though, that the nature of the analysis parameters, and the ease of testing the influence on the analysis parameters on the results, leads to a systematic reduction in personal bias.

These differences ff can be related to the difference ff in the source mechanism in these areas; the higher b-values correspond with stress fracturing seismicity, and the lower b-values relate to a shear mechanism.

The same plot is combined with a spatial distribution of apparent stress in Figure 2. The colouring of each grid point in space is the same as that in Figure 1, but in Figure 2 each grid-point marker is scaled by the geometric mean of the apparent stress. The apparent stress is proportional to the mean shear stress at the source of the event (McGarr, 1994).

The areas of the mine showing higher b-values generally show a low apparent stress, while the lower abutment area shows a high apparent stress state corresponding with low b-values. In this example, the distribution of b-values through space can be obtained without a pre-defined model of rock mass response.

Figure 1—Grid based interpretation of the spatial distribution of bvalues in a mining area at an Australian mine. Blue and red highlighted areas enclose the stoping and abutment areas respectively

Grid-based analysis of seismic data In the grid-based approach, the seismic source parameters are assessed through space by interpolating the source parameters. This approach allows for anomalies to be identified without prior selection of groups or polygons. This is illustrated in Figure 1 and Figure 2.

Figure 1 shows the spatial distribution of b-values at an Australian mine. High values occur around the stoping volumes while low b-values occur at the lower abutment.

▲

816

OCTOBER 2014

VOLUME 114

Figure 2—Grid-based interpretation of seismicity showing the distribution of b-value by colour. The size of the grid point markers are scaled by apparent stress. Blue and red highlighted areas enclose the stoping and abutment areas respectively The Journal of The Southern African Institute of Mining and Metallurgy

Grid-based analysis of seismic data General principle

Obtaining a grid-based interpretation of b-value

The general principle behind the grid-based approach for quantitative mine seismology can be summarized as follows: Assign a representative p seismic parameter value to grid points in space based on the events in its neighbourhood g in order to extract information from the variation of these different parameter values in space.

Added value is obtained from this approach when the different parameters are retained on each grid point, which allows for the analysis of these different parameters together as shown later.

What can be regarded as ‘representative’ and ‘the neighbourhood’ depends on the purpose of the analysis, the density and quality of the seismic data, and the type of parameter interpolation that is performed i.e. obtaining the cumulative parameter assessment, interpolation of the mean value, or obtaining the b-value.

The neighbourhood is defined by assigning a maximum influence distance, with the addition of two more quality checks. For a grid point to be assigned a representative seismic parameter, the grid point events around it must be both close enough and dense enough. This is illustrated in Figure 3. The condition that events must be dense enough in the vicinity of the grid-point prevents the transfer of parameter values from areas further away, with dense data, to a grid point where the density of the events close to the grid point does not warrant the calculation of a parameter value.

The b-value of the frequency magnitude distribution is proportional to the mean of the magnitude and, as such, is simply a statistical parameter. The appropriate b-value cannot be obtained without knowledge of the magnitude of completeness of each subset of data. For this reason, the method for obtaining the spatial distribution of the b-value is quite involved and is treated separately in the sister paper published in this volume (Wesseloo, 2014).

For current purposes it will suffice to summarize the process as follows. For each grid point:

Grid-based analysis of seismic data The grid-based interpretation of seismic data requires different approaches for different types of parameters, whether it is the mean or cumulative value of a parameter that is of interest, or the b-value.

It is important to note that in all the different approaches, the gridding process involves some level of smearing, the degree of which is dependent on the analysis parameters that are discussed in this paper. It is important that the resolution of the interpretation should match the resolution of the original input data. Sparse data-sets would require more smearing than high-resolution dense data-sets. A simple sensitivity analysis should be performed to test the sensitivity of the outcomes to chosen input parameters.

➤ Obtain the closest N events ➤ For these events obtain the mmin and associated bvalue ➤ Retain b-values for grid points passing the quality tests.

Mean value of all parameters In order to obtain the mean value of parameters, the geometric mean of the parameters of the closest events to each grid point is calculated. The process can be summarized as follows. For each grid point: ➤ Find all the events (if ≥ N) within a distance Rmin or the closest N events within a seach distance of Rmax ➤ Calculate the geometric/arithmetic mean for the parameter ➤ Perform quality checks (see above) and retain calculated values for only the grid points passing the quality tests.

This approach is used for obtaining spatial distribution of parameters like energy index (EI), apparent stress (AS), and time of day (TOD).

Smearing cumulative parameters In the cases where one is interested in the cumulative effect of different events, for example to obtain an event density or the cumulative apparent volume, a smearing process is used.

In contrast to the method used to obtain the mean parameter of neighbouring events (previous section), the parameter value (intensity, I ) of each event is distributed to (or ‘smeared onto’) grid points within its zone of influence.

This procedure can be summarized as follows: ➤ For each event – Find all grid points within its influence zone – Distribute a portion of its value to every grid point ➤ For every grid point – Sum all the portions received from each event – Perform quality checks.

This is performed with a variable smoothing where the kernel bandwidth is linked to the event source size.

The distribution of the events’ parameter is performed with an inverse distance weighting. The cumulative parameter at each grid point is obtained as the sum of all the values of all the events registered to that grid point. This can be expressed as follows: [1]

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

817

▲

Figure 3—Illustration of the concept of data neighbourhood

Grid-based analysis of seismic data [2] [3]

[4] where pj is the cumulative parameter at grid point j, and s are the location vector of the event and grid points. The intensity value, I, is the parameter value of interest and K(θ ) is the kernel function. The bandwidth function h defines the influence zone of each event, i.

F is a correction factor that ensures that no errors are introduced due to the discretization of the volume space, i.e that the following condition is met: [5] In other words, summing the parameter values over all events must equal the sum of the parameter value associated with each of the grid points.

This process is graphically illustrated in Figure 4. The circles in Figure 4a represent different events with different source sizes. The parameter values of each of these curves are distributed to the grid points in the influence zone with the kernel function (Figure 4b). These values for each grid point sum to the final spatial distribution of the parameter value (Figure 4c).

In the smearing process, described in the following section, each event has an influence zone. Our current approach is to define an influence based on the event source radius, as defined by Brune (1970). A lower cut-off value equal to the grid spacing is imposed to ensure stability of the method for coarser grid discretization. A limiting ceiling value is also introduced for numerical efficiency and stability. The results are not sensitive to the ceiling value.

It is important to note that the results off the smearing process are not sensitive to the influence of large events, as the parameter values of these large events are distributed to more grid points within the larger influence zone.

Grid-based quantitative analysis This section provides a short discussion on some of the parameters used in the grid-based analysis and examples to illustrate the application of the method. The use of these methods is not limited to these parameters.

Energy stress and apparent stress Apparent stress is generally calculated as: [6] where G is the shear stiffness of the rock mass, and E and Mo are the total radiated energy and average moment for an event, respectively. As indicated by its name, the definition of apparent stress relates to the stress state in the rock mass at the occurrence of the event. The apparent stress is proportional to the mean shear stress at the source of the event, (McGarr, 1994), and is defined as follows: [7] where η is the seismic efficiency and τ and τr are the peak and residual shear stress, respectively.

As η is unknown, the absolute shear stress is also unknown. The value of apparent stress is, however, a good indicator of the relative stress state. This concept is refined by Mendecki (1993), who showed that for a given slope of the log(E)-log(M) relation, the intercept value relates to the stress level. A simpler way to express this relative value of the intercept is the log(EII), as defined by van Aswegen and Butler (1993).

Obtaining the log(E)-log(M) trendline introduces some difficulties with unsatisfactory ‘best-fit’ lines. This problem can easily be overcome. For the sake of maintaining the focus of this paper, this will be discussed further in Appendix A. The results of a grid-based analysis of the log(EII) at an Australian mine are shown in Figure 5. The colour scale of the grid points reflects the values of log(EII). The transparency of the grid points is also scaled with log(EII).

The upper 50% of the log(EII) values are more solid while the lower 50% are more transparent.

In this particular case, low log(EII) values occur at the centre of the volume with surrounding higher log(EII) values.

This corresponds with lower stress areas in the immediate vicinity of the mined stopes, while the more competent rock further away from the stopes is under a higher stress state.

The particular shape of the grid cloud is determined by the location of seismic data, as grid points are generated only where seismic data exists.

Time of day

Figure 4—Illustration of the smearing of parameters onto grid points

▲

818

OCTOBER 2014

VOLUME 114

The time-of-day parameter (TOD) is a measure of the temporal differences in the seismic response. It is defined as the ratio of the rate of seismicity occurring within a specified time window(s) to the rate of seismicity occurring outside of The Journal of The Southern African Institute of Mining and Metallurgy

Grid-based analysis of seismic data those time windows. As an example, where the rock mass reacts strongly (in a purely temporal sense) to blasts, the TOD value will be high for a time window around blast time.

Orepass noise, on the other hand, results in TOD values of less than 1 for time windows around shift change, as no orepass activity occurs during shift change.

Grid points with a low TOD were isolated ffor an Australian mine. These grid points all located in a couple of distinct areas. The diurnal chart for the events reporting to these grid points is shown in Figure 7. From this chart it is clear that the events are small and inversely correlated to the shift change, indicating that these events are human-induced noise, in this case orepass noise. The fact that these events are human-induced noise is highlighted by the fact that the time of lunch breaks during the two shifts is visible in the diurnal chart.

Figure 8 shows the b-value and TOD plots for three adjacent areas in a mine. Area (1) has an unnaturally high b-value with a very low TOD and is the result of crusher noise. Area (2) has a very high b-value and a higher TOD and is the result of a raise bore experiencing some dogearing.

Area (3) has a lower, but still fairly high, b-value, with a very high TOD. The seismicity in this area generally relates to stress fracturing around development blasting temporally concentrated around blast time. The diurnal charts for these three areas area shown in Figure 9.

Cumulative damage

Figure 5—Grid-based interpretation of log(EI ( I). Both colour and transparency reflect the log(EI ( I)

Figure 6—TOD definition

It is generally accepted that there is a correlation between historical seismic activity in an area and the damage accumulated in the rock mass. Despite the work of Falmagne (2001), Cai et al. (2001), Coulson and Bawden (2008), and Pfitzner et al. (2010), there appears to be no accepted way to quantify this damage accumulation from seismic data. Until these difficulties are solved, we propose the use of apparent volume and the cube root of moment as proxies for damage.

Apparent volume has been linked to the amount of coseismic strain (Mendecki, 1997), while the cube root of moment is proportional to the maximum displacement at the seismic source (McGarr and Fletcher, 2003).

Figure 10 shows the results of a grid-based analysis at an Australian mine. In this example, log(EII) as a proxy for stress and the cube root of moment as a proxy for damage are combined. Log(EII) is represented by the colour scale and the transparency varies with damage. The mean log(EII) is calculated for a 6-month data period, while the damage is accumulated over the whole history of the mine.

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

819

▲

Figure 7—TOD definition

Grid-based analysis of seismic data

Figure 10—Analysis results, combining log(EI ( I) results in colour with cumulative ‘damage’ results plotted with varying transparency

Figure 8—b-value and TOD plots for three adjacent areas in an Australian mine

The destressed area in this case corresponds with the area of high historical damage, with some high-stress areas on the abutments, where similar levels of damage have accumulated. It should be noted that the destressing did not take place due to the damage alone but is a result of the mining voids. As one would expect, the area of high damage is concentrated close to the stopes.

Event density Event density is conceptually a very simple parameter and is very easy to interpret as the number of events occurring per unit volume. From a mathematical viewpoint, event density is a cumulative parameter with the intensity value, I, equal to 1 (refer to Equation [5]) and for this reason is calculated with the same method as is used for the cumulative damage.

Figure 11 shows examples from the Tasmania Mine for different time periods during the life of the mine.

Calibrating numerical models

Figure 9—Diurnal charts for the events shown in the three areas in Figure 8

▲

820

OCTOBER 2014

VOLUME 114

In mining geomechanics, the need for calibrating or constraining the models with physical observations is well recognized. Seismic data provides a valuable source of information on the rock mass response to mining, and for this reason has been used by some investigators to provide calibration data for their models.

Often the calibration of numerical models with seismic data is limited to a visual correlation between the event location and strain or stress contours from the models, or the visual correlation of event density with areas of higher plastic strain. More recently, correlations between the energy release monitored within a specified cell and the modelled plastic strain energy have been used (Levkovitch et al., 2013; Arndt et al., 2013). Both of these groups limit themselves to energy and perform basic grid calculations, simply summing the energy of all monitored events located within a specific grid cell.

The Journal of The Southern African Institute of Mining and Metallurgy

Grid-based analysis of seismic data tation should match the resolution off the original input data.

A sparse data-set would require more smearing than highresolution dense data-sets. A simple sensitivity analysis should be performed to test the sensitivity of the outcomes to chosen input parameters.

The grid-based analysis approach is well suited to compare with results from numerical modelling approaches.

Acknowledgements Our sincere appreciation to William Joughin and Professor Ray Durrheim for reviewing the paper. The following organizations provided funding for this research through the Mine Seismicity and Rockburst Risk Management project: Barrick Gold of Australia, BHP Billiton Nickel West, BHP Billiton Olympic Dam, Independence Gold (Lightning Nickel), LKAB, Perilya Limited (Broken Hill Mine), Vale Inc., Agnico-Eagle Canada, Gold Fields, Hecla USA, Kirkland Lake Gold, MMG Golden Grove, Newcrest Mining, Xstrata Copper (Kidd Mine), Xstrata Nickel Rim, and The Minerals Research Institute of Western Australia.

References ARNDT, S., LOUCHNIKOV, V., WELLER, S., and O'HARE, A. 2013. Forcasting mining induced seismicity from modelled energy release in high stress stope extraction. 8th International Symposium on Rockbursts and Seismicity in Mines, St. Petersburg and Moscow, Russia, 1–7 September 2013. pp. 267–272.

BRUNE, J. 1970. Tectonic stress and the spectra of seismic shear waves from earthquakes. Journal of Geophysical Research, vol. 75. pp. 4997–5009.

CAI, M., KAISER, P., and MARTIN, C.D. 2001. Quantification of rock mass damage in underground excavations from microseismic event monitoring.

Figure 11—Event density results for different time periods at the Tasmania Mine. The mine layout is representative of the mine during the stage represented in (d)

International Journal of Rock Mechanics and Mining Sciences, vol. 38. pp. 1135–1145.

COULSON, A. and BAWDEN, W. 2008. Observation of the spatial and temporal changes of microseismic source parameters and locations, used to identify the state of the rock mass in relation to the peak and post-peak strength

We are of the conviction that the grid-based approach to the evaluation of seismic data presented here provides the opportunity to better utilize the seismic data to constrain numerical models.

As the grid-based interpretation can be continually updated, this provides further opportunity to continually test predicted rock mass behaviour against experienced behaviour, with the possibility of flagging deviation from the predicted behaviour.

conditions. 42nd US Rock Mechanics Symposium, San Francisco, CA, 29 June – 2 July 2008. American Rock Mechanics Association, Alexandria, VA. DURRHEIM, R.J. 2010, Mitigating the rockburst risk in the deep hard rock South African mines: 100 years of research. Extracting the Science: a Century of Mining Research. Brune, J. (ed.). Society for Mining, Metallurgy, and Exploration, Inc. pp. 156–171. ISBN 978-0-87335-322-9, FUNK, C.W., VAN ASWEGEN, G., and BROWN, B. 1997. Visualisation of seismicity.

Proceedings of the 4th International Symposium on Rockbursts and

Concluding remarks

Seismicity in Mines, Kraków, Poland, 1–14 August 1997. Gibowicz, S.J.

A grid-based interpretation of seismic data has been discussed and some examples of results obtained with the method presented. A grid-based interpretation allows the spatial variation of seismic source parameters to be evaluated without predetermined analysis volumes. As such, it provides some buffer against biasing of interpretations towards preconceived ideas.

The gridding process involves some level of smearing, the degree of which is dependent on the analysis parameters discussed. It is important that the resolution of the interpre-

and Lasocki, S. (eds). CRC Press, Boca Raton, FL.

seismic monitoring and applications for mine design. PhD thesis, Queen's University, Ontario.

HUDYMA, M. 2008. Analysis and interpretation of clusters of seismic events in mines. PhD thesis, University of Western Australia, Perth.

JAIN, A.K., MURTY, M.N., and FLYNN, P.J. 1999. Data clustering: a review. ACM Computing Surveys (CSUR), vol. 31. pp. 264–323.

VOLUME 114

OCTOBER 2014

821

▲

The Journal of The Southern African Institute of Mining and Metallurgy

FALMAGNE, V. 2001. Quantification of rock mass degradation using micro-

Grid-based analysis of seismic data LEVKOVITCH, V., BECK, D., and REUSCH, F. 2013. Numerical simulation off the released energy in strain-softening rock materials and its application in estimating seismic hazards in mines. 8th International Symposium on Rockbursts and Seismicity in Mines, St. Petersburg and Moscow, Russia, 1–7 September 2013. pp. 259-266.

MCGARR, A. 1994. Some comparisons between mining-induced and laboratory earthquakes. Pure and Applied Geophysics, vol. 142. pp. 467–489.

MCGARR, A. and FLETCHER, J. 2003. Maximum slip in earthquake fault zones, apparent stress, and stick-slip friction. Bulletin of the Seismological Society of America, vol. 93. pp. 2355–2362.

MENDECKI, A. 1993. Real time quantitative seismology in mines. Keynote Address: 3rd International Symposium on Rockbursts and Seismicity in Mines, Kingston, Ontario, 16-18 August 1993. Young, R. Paul (ed.).

Barnes & Noble. pp. 287–295.

MENDECKI, A. 2013. Characteristics of seismic hazard in mines. Keynote Lecture: 8th International Symposium on Rockbursts and Seismicity in Mines, St. Petersburg and Moscow, Russia, 1–7 September 2013. pp. 275–292.

MENDECKI, A.J. 1997. Seismic Monitoring in Mines. Chapman & Hall, London; New York.

When evaluating the log(E)-log(M) relationship we are not interested in obtaining log(E) as a function of log(M) but in the general statistical relationship between these parameters.

Wesseloo and Potvin (2012) suggested the use of the log(E)-log(M) quantile-quantile (QQ) relationship for this purpose, which is the method implemented in the software mXrap (Figure 13).

It should be noted that this approach, although much simpler, is equivalent to the approach suggested by Mendecki (2013).

The QQ method for obtaining the log(E)-log(M) relationship is performed as follows. ➤ Independently sort E and M, both in ascending order ➤ Plot log(Ei) against log(Mi) for every value of i. Note that the actual values of log(Ei) and log(Mi) are not from the same event and the only link between them is i the fact that they both represent the N quantile of the two different sets log(E) and log(M) ➤ For practical use a best-fit line can be fitted to the QQ relationship. ◆

PFITZNER, M., WESTMAN, E., MORGAN, M., FINN, D., AND BECK, D. 2010. Estimation of rock mass changes induced by hydraulic fracturing and cave mining by double difference passive tomography. 2nd International Symposium on Block and Sublevel Caving, g Perth, Australia. Potvin, Y. (ed.). Australian Centre for Geomechanics, Perth, WA. POTVIN, Y. and WESSELOO, J. 2013. Improving seismic risk management in hardrock mines. 8th International Symposium on Rockbursts and Seismicity in Mines, St. Petersburg and Moscow, Russia, 1–7 September 2013. pp. 371–386.

ROMESBURG, H.C. 2004. Cluster Analysis for Researchers. Lifetime Learning Publications, Belmont, CA. 334 pp.

VAN ASWEGEN, G. and BUTLER, A.G. 1993. Application of quantitative seismology in SA gold mines. 3rd International Symposium on Rockbursts and Seismicity in Mines, Kingston, Ontario, 16–18 August 1993. Young, R.P.(ed.). Balkema, Rotterdam. pp. 261–266.

WESSELOO, J. And POTVIN, Y. 2012. Advancing the Strategic Use of Seismic Data in Mines. Minerals and Energy Research Institute of Western Australia,

Figure 12—Least-squares ‘best fit’ to the log(E (E) log((M) data

East Perth, WA. WESSELOO, J. 2014. Evaluation of the spatial variation of b-value. Journal of the Southern African Institute of Mining and Metallurgy, vol. 114, no. 10. pp.823–828.

Appendix A Obtaining the log(E)-log(M) trendline It should be recognized that a least-squares ‘best-fit’ approach assumes the total radiated energy (E ) to be a variable dependent on the total moment (M). This assumption is incorrect as M and E are two independent parameters and as such the use of the least-squares best-fit method is invalid.

Applying the general least-squares best-fit approach to the Log(E)-Log(M) relation underestimates the slope of the trendline and overestimates the intercept (Figure 12).

▲

822

OCTOBER 2014

VOLUME 114

Figure 13—‘Best fit’ to the QQ plot of log(E (E)-log((M) data The Journal of The Southern African Institute of Mining and Metallurgy

Evaluation of the spatial variation of bvalue

by J. Wesseloo*

The estimation of the spatial variation of the b-value of the GutenbergRichter relationship is important for both the general interpretation of the mechanism of rock mass response and seismic hazard assessment. The interpretation of b-value as a parameter of rock mass response is discussed in this paper. The methods applied to evaluate the spatial variation of b-value and the algorithm for obtaining the magnitude of completeness and b-value for subsets of data are presented with some verification analyses.

The algorithms presented enable the automation of a spatial evaluation of b-value.

Keywords mine induced seismicity, seismic hazard assessment, Guttenberg-Richter relationship, b-value, spatial evaluation.

Introduction One of the cornerstones of seismic data interpretation and hazard assessment is the Gutenberg-Richter (GR) relationship in the frequency-magnitude plot. Estimation of the spatial variation of the b-value is useful for both the general interpretation of the mechanism of rock mass response as well as seismic hazard assessment (Wesseloo 2013).

The assessment of the spatial variation of seismic parameters in general is discussed in a sister paper in this volume (Wesseloo et al., 2014). This paper focuses specifically on obtaining the b-value, as it is quite involved and warrants a separate discussion.

To estimate the b-value requires that the magnitude of completeness, mmin, is known.

As both these parameters vary spatially and temporally, it is necessary to automatically obtain the most likely mmin and b-value for every spatial subset of data.

The paper discusses the algorithm for spatially sub-sampling the data as well as the algorithm for obtaining the mmin and b-value for every spatial sub-sample.

b-value The frequency-magnitude relationship (inverse cumulative distribution) of seismic event magnitude generally follows a power law relationship which is often described by the well-known GR relationship. The b-value The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

* Australian Centre for Geomechanics, The University of Western Australia, Australia. Š The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12â&#x20AC;&#x201C;14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

823

â&#x2013;˛

Synopsis

describes the ffrequency distribution off magnitudes occurring in a given seismic dataset and, as such, is a key component in any seismic hazard assessment. Assessing the spatial variation in the b-value forms one of the key components of any seismic hazard map.

Apart from the obvious importance of the b-value for hazard assessment, it is also a valuable parameter for interpreting the rock mass deformation and failure mechanism.

Several studies in both seismology and the mining environment support this notion.

Wyss and his co-workers (Wyss et al., 1997) pointed out that mapping the b-value is equivalent to mean magnitude and assumes that this is proportional to the mean crack length. They also point out that along fault zones the low b-values seem to correspond with asperities (Amelung and King, 1997; Wiemer et al., 1998; Wiemer and Wyss, 1997), while high b-values correspond with creeping sections of faults. High b-values seem to be a characteristic of active magma chambers (Wiemer et al., 1998; Wyss et al., 1997) where seismicity is dominated by the creation of new fractures under stress build-up.

Mogi (1962) noted that increasing material heterogeneity results in a high b-value, while others have pointed out that an increase in applied shear stress (Scholz, 1968; Urbancic et al., 1992; Wyss et al., 1997) or an increase in effective stress decreases the b-value (Wyss, 1973; Wyss et al., 1997).

In the mining environment different b-values have been associated with different rock mass failure mechanisms. Legge and

Evaluation of the spatial variation of b-value Spottiswoode (1987) pointed out that higher b-values may be expected to result from seismic events occurring on different planes dispersed in a three-dimensional volume, while lower b-values will be associated with events distributed uniformly in two dimensions, as on a single plane. Further along the scale, very low b-values can be associated with one-dimensional linear distributions, which could result from the interaction between the mining-induced stress change and extensive planar discontinuities.

The b-value has in some cases been linked to a fractal dimension, D. Aki (1981) related the b-value and the fractal dimension as D = 2b. The conclusions of Spottiswoode and Legge are consistent with a fractal interpretation of the b-value, where a b-value of 0.7 would relate to a D of 1.5, which could be interpreted as planar spatial distribution of events; and a b-value of 1.5 relates to a D of 3, which could be associated with a three-dimensional distribution of events.

The spatial assessment of the b-value is, therefore, valuable for both hazard assessment and the interpretation of the rock mass response to mining.

A generalized and simplified summary of the literature on the interpretation of the b-value is provided in Table I.

value which depends on the resolution off the data, the data density, and the purpose of the analysis.

With this method, only the maximum search distance is specified and the real search distance is determined by the distance to the N Nth neighbouring event. Each grid point therefore has a unique search distance. This is illustrated in Figure 1, where the sizes of the spheres illustrate the search volume with radius Rmax.

A user-defined Rmax value limits the analysis to be performed on grid points with N or more values within a radius smaller or equal to Rmax, the purpose of which is to restrict the grid point b-value to local data.

Several quality checks are built into the analysis, which are discussed in the sister paper (Wesseloo et al., 2014). In addition to these, the following checks are also implemented in relation to the b-value assessment: ➤ The value of mmin must be within reasonable expected bounds ➤ The number of events with magnitude greater than mmin must exceed a set threshold value.

Evaluation of the spatial variation of b-value In order to evaluate the spatial variation of b-value, a grid is generated over the volume of interest and the b-value obtained for every grid point. Several crustal seismology studies were performed where the spatial variation in b-value was evaluated (e.g. Wiemer et al., 1998; Wiemer and Wyss, 1997; Wyss et al., 1997). These studies used the methods developed by Wyss and co-workers that are incorporated into a Matlab library (Zmap) (Wyss et al., 1997). The methodology presented here is, in general terms, similar to the approach used by Wyss and his co-workers.

The method used in this study can be summarized as follows: ➤ Events with magnitudes much smaller than the estimate of the overall sensitivity based on the whole data-set (mmin - Δ) are excluded from the analysis. This is done to speed up the calculations by excluding very small events that do not contribute to obtaining the b-value. Including these very small events also has a negative impact on the overall algorithm performance as it reduces the number of events useful for b-value calculations within the search distance Rmax from each grid point ➤ For each grid point, the mmin and b-value are obtained from the closest N points and calculated with the search radius limited to a value Rmax. Rmax is a user-defined

Figure 1—Illustration of the search distance and associated b-value for an Australian mine

Table I

Typical interpretation of b-value in terms of rock mass deformation mechanism Low b-value

High b-value

Events localize on a plane Corresponds with asperities on a fault

Events spread out within the volume

Legge and Spottiswoode (1987)

Correspond with creeping sections of faults

Amelung and King (1997); Wiemer and Wyss (1997); Wyss et al. (2004)

Increasing material heterogeneity Increase in applied shear stress

▲

824

OCTOBER 2014

References

Mogi (1962) Scholz (1968); Urbancic et al. (1992)

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Evaluation of the spatial variation of b-value This added quality check is necessary to ensure that reasonably stable b-values are obtained, as the standard deviation of the b-value is inversely proportional to the square root of the number of events used in the b-value calculation (the number of events above mmin) (Kijko and Funk, 1994), i.e.: [1]

➤ The minimum value off k = 10 is an arbitrarily chosen practical lower limit and can be set higher. The purpose of this minimum value is simply to ignore erratic behaviour for the very high tail end of the distribution ➤ For each subset k, obtain the Kolmogorov-Smirnov goodness of fit parameter, KS. This parameter is not sensitive to the exponential tail end of the distribution and is, therefore, well suited for the stable estimation of mmin ➤ The decision parameter, r C, is defined as follows:

Algorithm for finding mmin and b for a given subset of

[3]

data In order to obtain the grid-based spatial distribution of the b-value, I developed an algorithm to automatically obtain the mmin and b-value for the subset of data associated with every grid point. Examples of the results obtained by the algorithm are shown in Figure 2.

The algorithm aims to maximize the number of data points included above mmin, while minimizing the deviation from the log-linear GR relationship. This described process is performed for any data-set for which the mmin and b-value need to be obtained.

The process for obtaining mmin can be described as follows: ➤ Sort the data-set of magnitudes in descending order ➤ Calculate the b-value for subsets of the data, where each subset of data is defined as consisting of data points 1 to k where k varies between 10 and the full number of data points in the data set. The b-value for each subset, i, is therefore defined as follows:

The decision parameter has the form (A ( · B ) · (C ), with each of these components combined to provide a good and stable estimate of mmin.

A gives more weight to steeper b-values. This component mitigates against the search algorithm overshooting the true mmin value as a result of a balancing effect of residual values on both sides of the best-fit relationship near the mmin value.

B gives more weight to more data included above the mmin value and works against local minimum values for small values of k.

C defines the goodness of fit, with larger values defining a better fit ➤ The value of k for which Ck is a maximum defines the number of data points in the full data-set. That is: [4]

Independent check for the algorithm [2]

Check on the second moment Figure 3a shows the result of the mmin algorithm applied to

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

825

▲

mmin and b-value using the proposed algorithm

Evaluation of the spatial variation of b-value data ffrom an Australian mine. The dashed lines show the GR relationship for b ± σb. The change in the decision parameter is shown in Figure 3b. Figure 3c shows the first and second moment values of the distribution with mmin assumed to be at each of the event magnitude values. This graph serves as an independent check of the algorithm.

Figure 3c shows the 1st and 2nd moment values of the distribution with mmin assumed to be at each of the event magnitude values. For a negative exponential distribution, the first and second moments (mean and standard deviation of the distribution) are equal.

It should be recognized that the open-ended GR relationship is the negative exponential distribution of the translated values of (Magnitude - mmin).

For the GR relationship, therefore, the translated first and second moments should be equal. The change in the calculated first and second moments of the data-set is shown in Figure 3c. In this case, the algorithm estimate of mmin is at the smallest magnitude before the values of the first and second moments start to deviate from each other. That is the smallest value of mmin at which the distribution exhibits the characteristics of a negative exponential distribution. As the second moment is not used in the mmin algorithm, this provides independent support for the reliability of the algorithm.

to determine which events will be detected by the system. The resulting data-set provides one data sample for testing the proposed algorithm.

Algorithm verification using synthetic data-sets In order to verify the developed algorithm, synthetic data-sets were generated. These enable the known values of the b-value and mmin to be compared with those obtained automatically through the use of the proposed algorithm.

To generate the synthetic data-set, random deviate sampling was performed from a specified Truncated GR relationship. This data-set was randomly distributed in a rectangular volume with an arbitrary chosen array of sensors.

The distance between the event and sensor location was used

Figure 3—Results from mmin-b-value algorithm applied to data from an Australian mine; (a) provides the frequency-magnitude relationship, while (b) and (c) plot the change in the value of the decision parameter and the first and second moment of the magnitude, with magnitude change, respectively

Figure 4—Example results of the verification tests performed on synthetic data-sets

▲

826

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Evaluation of the spatial variation of b-value The abovementioned process was repeated several hundred times to obtain the distributions of mmin and b-values. Two examples are shown in Figure 4.

In the figure, a histogram of the estimated mmin values is shown in blue on the chart on the right side. Also shown in the right hand chart is one of the several hundred sampled data-sets in red, together with its corresponding mmin value as a black dot on the curve and the fitted GR relationship as a solid blue line.

The distribution of mmin values follows a fairly narrow distribution around the true value.

On the left side in Figure 4, a histogram of estimated b-values is shown in green. The distributions of b-values follow a normal distribution around the specified b-value. It should be noted that a variation in the estimated b-value will result from the fact that each of the samples provides only a small portion of the true population. This can be seen in Figure 5. The solid green histogram shows the variation in the estimated b-value obtained from the data-set for which the mmin value is estimated, while the open blue histogram shows the distribution of results for which the true mmin value is specified.

The similarity in the distributions confirms the reliability of the method for obtaining a good GR relationship for the data-set. Note that it is not implied that same b-value is always obtained for the two cases, but that the inherent uncertainty in the b-value is not increased by applying the algorithm to calculate mmin.

Figure 5â&#x20AC;&#x201D;Example results of the verification tests performed on synthetic data-sets

Results of evaluation of the spatial variation of b-value The method described previously was applied to the database of Tasmania Mine (formerly known as Beaconsfield).

Figure 6 shows the isosurfaces enveloping the higher, intermediate, and lower b-values for different dates during the history of the mine. The green volume indicates b > 1.2, orange indicates 0.8 < b < 1.2, and red b < 0.81. The volume covers changes over time as mining progresses and the seismically active volume increases.

This method clearly demonstrate that the b-value changed over time for different areas. Some areas show a high b-value which reduces as mining progresses, and later increases with further progressing of the mining. The original high b-value is associated with fracturing taking place as stress change occurs. With further mining deformation mechanism is dominated by larger structures with an associated lower b-value. At later stages in the mining, the stresses on these structures are released and the seismicity takes the form of continued fracturing in the hangingwall.

This is similar to the results reported for a deep-level South African environment by Legge and Spottiswoode (1987). In contrast to some of the other areas, the western (left) abutment continues to exhibit a low b-value throughout the mining history. This implies that seismic deformation in this area is typically concentrated on major structures.

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 6â&#x20AC;&#x201D;Examples of the spatial variation of the b-value at Tasmania Mine at different stages throughout the mine life VOLUME 114

OCTOBER 2014

827

â&#x2013;˛

1Note that these isosurface values were chosen for the local magnitude scale used on site, which tends to give lower b-values than for moment magnitude

Evaluation of the spatial variation of b-value Concluding remarks The b-value of seismicity has been linked to the seismic deformation mechanism of the rock mass and as such is important for quantitative seismology in mines. The spatial and temporal evaluation of the b-value is important for the evaluation of the rock mass response to mining and changes in seismic hazard as a result of mining.

The evaluation of the spatial variation of the b-value was successfully implemented on a grid basis with an automated method for obtaining the mmin and b-value for subsets of data.

Acknowledgements My sincere appreciation to Mr William Joughin and Professor Ray Durrheim for reviewing the paper. I thank the following organizations who provided funding for this research through the Mine Seismicity and Rockburst Risk Management project: Barrick Gold of Australia, BHP Billiton Nickel West, BHP Billiton Olympic Dam, Independence Gold (Lightning Nickel), LKAB, Perilya Limited (Broken Hill Mine), Vale Inc., AgnicoEagle Canada, Gold Fields, Hecla USA, Kirkland Lake Gold, MMG Golden Grove, Newcrest Mining, Xstrata Copper (Kidd Mine), Xstrata Nickel Rim, The Minerals Research Institute of Western Australia.

References AKI, K. 1981. A probabilistic synthesis of precursory phenomena. Earthquake Prediction, an International Review. Maurice Ewing Series, vol. 4.

Simpson, D.W. and Richards, P.G. (eds.). American Geophysical Union, Washington. pp. 566–574.

AMELUNG, F. and KING, G. 1997. Earthquake scaling laws for creeping and noncreeping faults. Geophysical Research Letters, vol. 24. pp. 507–510.

▲

828

OCTOBER 2014

VOLUME 114

KIJKO, A. and FUNK, C. 1994. The assessment off seismic hazards in mines.

Journal of the South African Institute of Mining and Metallurgy, vol. 94, no. 7. pp. 179–185.

LEGGE, N. and SPOTTISWOODE, S. 1987. Fracturing and microseismicity ahead of a deep gold mine stope in the pre-remnant and remnant stages of mining.

6th ISRM Congress, Montreal, Canada. Balkema, Rotterdam. pp. 1071–1077.

MOGI, K. 1962. Magnitude-frequency relation for elastic shock accompanying fractures of various materials and some related problems in earthquakes.

Bulletin of the Earthquake Research Institute, University of Tokyo, vol. 40. pp. 831–853.

SAMMIS, C., NADEAU, R., WIEMER, S., and WYSS, M. 2004. Fractal dimension and b-value on creeping and locked patches of the San Andreas Fault near Parkfield, California. Bulletin of the Seismological Society of America, vol. 94. pp. 410–421.

SCHOLZ, C. 1968. The frequency-magnitude relation of microfracturing in rock and its relation to earthquakes. Bulletin of the Seismological Society of America, vol. 58. pp. 399–415.

URBANCIC, T.I., TRIFU, C.I., LONG, J.M., and YOUNG, R.P. 1992. Space-time correlations of b-values with stress release. Pure and Applied Geophysics, vol. 139. pp. 449–462.

WESSELOO, J. 2013. Towards real-time probabilistic hazard assessment of the current hazard state for mines. Proceedings of the 8th International Symposium on Rockbursts and Seismicity in Mines. Saint-Petersburg Moscow, pp. 307–312.

WESSELOO, J., WOODWARD, K., and PEREIRA, J. 2014. Grid based analysis of seismic data. Journal of the Southern African Institute of Mining and Metallurgy, vol. 114, no. 10. pp. 815-822.

WIEMER, S., MCNUTT, S., and WYSS, M. 1998. Temporal and three-dimensional spatial analyses of the frequency-magnitude distribution near Long Valley Caldera, California. Geophysical Journal International, vol. 134. pp. 1–13.

WIEMER, S. and WYSS, M. 1997. Mapping the frequency-magnitude distribution in asperities: An improved technique to calculate recurrence times? Journall of Geophysical Research, vol. 102. pp. 15115–15128.

WYSS, M. 1973. Towards a physical understanding of the earthquake frequency distribution. Geophysical Journal of the Royal Astronomical Society, vol. 31. pp. 341–359.

WYSS, M., SHIMAZAKI, K., and WIEMER, S. 1997. Mapping active magma chambers by b-values beneath the off-Ito volcano, Japan. Journal of Geophysical Research, vol. 102. pp. 20413–20422. ◆

The Journal of The Southern African Institute of Mining and Metallurgy

Testing tendon support units under a combination loading scenario

by N.L. Ayres* and L.J. Gardner*

Tendon support systems have been successfully used to stabilize excavations. Tendon support systems are routinely designed using the axial load-bearing capacity of tendons, namely the tensile strength. To attain tensile strength the tendon must be loaded along its length, which often does not occur in practice. Tendons should optimally be installed at 90° to the surface of the excavation to achieve maximum penetration depth, yet this is often not physically or practically possible, and installations at angles less than 90° occur.

Furthermore, the intersection of geological features within the rock mass frequently results in complex loading situations on tendons. The position and angle at which loading occurs results in different combinations of tensile and shear forces acting on the tendon, which can impact on the support performance of each unit and ultimately the whole system. All factors that influence the support system should be understood and taken into account to ensure a sound support design.

Combination loading situations are further investigated and tested to obtain a better understanding of the mechanisms involved and the effects on tendon load-bearing capacity. Tendon support units were tested at different installation angles to establish the tendon performance, mechanical behaviour, and load capacity during these loading situations.

The results and outcomes are aimed at providing rock engineers with additional data and improved understanding of how tendons could perform under certain conditions.

Keywords tendon support, combination loading, shear strength, tensile strength.

Introduction Tendon support units are primarily tested for tensile and shear strength and performance.

Hangingwall support designs are frequently based on the axial load-bearing capacity of the tendons (i.e. the pure tensile strength of the tendon), rather than the tendon’s load-bearing capability in shear (which is usually significantly lower than the tensile strength). The load-bearing capacity and support performance of tendons under different combination loading situations and installation angles should be investigated and quantified to ensure this information can be incorporated into the fundamentals of the support design.

This becomes even more significant when geological structures are known to be present.

In an ideal world, tendons would be installed at 90° to the surface of the excavation, yet this is sometimes not physically or practically achievable. Across the The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

industry, accepted installation angles typically vary between 70° and 90° to the hangingwall, with even lower angles being specified (and/or measured) on occasion. Justification and support performance information for these different installation angles should be available for the support design. At all installation angles (including 90°) the varied nature of geological features (including joints) can result in various combination loading situations and different loading angles on a tendon. This does not create pure tensile or shear loading, but results in various combinations of concurrent tensile and shear loading on the tendon, influencing the performance characteristics used for the design.

Limited information and test results are available for tendon performance under combination loading. A testing programme of combination loading on tendons was therefore conducted. By sharing the outcomes and difficulties encountered in performing combination load testing on friction tendon support units, the authors hope to assist in advancing the combination load testing method towards a standardized international test method for all types of support tendons.

Such a standardized combination test method will deliver more appropriate support capacity data to better address varied geological influences, allow for better support designs, and assist with back-analysis after failures. This would also provide manufactures with a standard to test their products against, and rock engineers with a constant base for comparison between support units, as well as more reliable criteria for selection of the appropriate support unit type for a specific rock mass environment.

* Impala Platinum Limited. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12–14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

OCTOBER 2014

829

▲

Synopsis

Testing tendon support units under a combination loading scenario Process for determining testing method for combination loading Owing to the limited literature dealing with combination loading, together with the lack of a formal testing method, the testing jigs and test method for combination load testing were developed by trial and error. Several factors were investigated to determine the basis of the testing method, including: ➤ Typical installation angles and how these influence tendon loading ➤ Geological and mining-induced structures/discontinuities within the rock mass and how these influence tendon loading, particularly when combined with a range of installation angles ➤ The type of tendon being tested ➤ Characteristics of the testing machine ➤ Simulating how the behaviour of an installed tendon is affected by the in situ rock mass environment, particularly with regard to the loading forces and directions.

Installation angle In Figure 1, a number of installation angles are illustrated, ranging from -90° to +90° from the horizontal. All installation angles between these extremes are possible. It can be seen that, due to the layout of an excavation, a number of different installation angles are required. In certain situations, some installation angles are not achievable in practice due to limitations of excavation size and/or orientation. In particular, it should be noted that in a 35° inclined excavation the typical tendon installation of 90° to the hangingwall results in an installation angle of less than 90° to the horizontal (as illustrated in Figure 1, this results in a 55° installation angle).

Geological structures and rock blocks Most rock masses are divided into blocks of various sizes, shapes, and orientations by joints and other geological structures. All blocks are subjected to the influence of gravity, which always acts vertically. The position and orientation at which the excavation intersects the blocks, together with the orientation of the structures, results in various loading directions as illustrated in Figure 2.

Figure 2—Geological structure orientations creating different loading directions

Figure 1—Illustration of tendon installation angles

▲

830

OCTOBER 2014

Tendon A in Figure 2 is installed vertically and at 90° to the hangingwall, and should be subjected to a pure tensile load under the influence of gravity. However, due to the orientation of the geological structures, the resultant loading force is not vertical and a combination of tensile and shear loading now acts on the tendon. This, together with the range of tendon installation angles, leads to an infinite number of possible loading situations.

Testing of all installation angles and joint intersections is neither practical nor possible, and therefore only selected angle intervals were tested. Results can be interpolated between testing intervals. It should be noted that in many instances during mining, induced stresses can influence excavations from any direction and thereby further complex loading situations are created in a three-dimensional environment. Combination load testing is a simplification as only a two-dimensional loading situation is simulated.

Different failure modes such as shear, dilation, cantilever, and toppling are possible, although only the shear and dilatory failures are discussed in this paper.

Movement along geological structures will occur as either shearing (where lateral movement occurs along the joint planes) or dilation (where the joint planes move away from each other and open up). Pure tensile and pure shear loading occur at 0° and 90° respectively, as illustrated in Figures 3 and 4. Combination loading occurs between 0° and 90° – this loading is either predominantly tensile (tendon is extended) or predominantly compressional loading (tendon is shortened). The combination of the tensile and shear load components varies as the loading angle changes in relation to the tendon’s long axis.

Where the force acts parallel to the failure plane, shearing occurs (as illustrated in Figure 3). Loading will then be by either tensile or compressional shear. In Figure 3, the tensile shear zone (where the tendon extends) is to the right of 0°, and the compressional shear zone, where the tendon will be compressed along the failure plane, to the left of 0°. At 45°, the shear, tension, and compression components are equal.

This should be the inflection point at which the mechanical performance of the tendon can change.

Where the forces do not act in parallel, but rather at an angle across the failure plane, dilation occurs. Figure 4 illustrates the combination loading situations where the force

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Testing tendon support units under a combination loading scenario

Figure 3—Combinations of shear loading

friction tendon. The paper discussed the shear stress and decoupling along the tendon during loading and how this affects the axial load. This provides a better understanding of the loading in a jointed rock mass and how several axial stress peaks can exist along the tendon.

The HydraboltTM and XpandaboltTM are manufactured in the same way; the difference lies in the pressure remaining in the tubular tendon after pre-stressing. The tendons are prestressed with water and the XpandaboltTM releases the water after pre-stressing, while the HydraboltTM retains the water in the tube, resulting in a stiffer support unit and a shorter critical bond length. Both tendons have a roll or ‘valley’ along the length of the tendon which creates a sectional profile resembling a horseshoe shape. For the tests, the ‘valley’ was always placed in the same direction to ensure the tested profile was constant, to counteract any possible effect of the tendon profile on the load performance.

Testing machine

Shear loading, g, cause by y lateral movement Combinations of loading g created on a vertical failure p plane under the influence of g gravity y at various intersection angles g

Figure 4—Combinations of tensile loading

acts perpendicular to a horizontal failure plane. To the left and the right of 0°, the effect on the tendon will be the same.

As the load angle on the tendon changes from 0° to 90° the tensile component will decrease from maximum to zero while the shear component increases from zero to maximum.

Figure 5 further illustrates all these loading conditions occurring in situ within the rock mass. The loose block will load the tendon and any movement on the joint plane will cause the load to be concentrated at the point where the tendon intersects the structure. This will be true for friction tendons and full column grouted tendons, as they behave in the same manner. Mechanically end-anchored tendons will take up load along a greater portion or the entire length of the tendon, depending on the loading direction.

As the tendons must be tested at different angles, a jig is required to hold the tendons in the required positions. A twopart steel jig (as seen in Figure 6) houses the tendon and creates the required failure plane for testing. Examination of the testing machine revealed that limited space was available for the tendon and the test jig; therefore the test jigs could not be very bulky. The stroke (i.e. the distance over which the machine can create a loading force) was limited. The clevis attachment points on the machine were aligned vertically. The jig attachment flanges and holes, as well as the failure plane, had to intersect this vertical line so that forces were transmitted to the tendon and not to the test jig.

Any influence from the test jig would skew the results of the tendon performance. The jig attachment flanges and holes had to be correctly aligned so that the jig portions, failure plane, tendon, and the loading direction all lined up.

Attachment flanges tended to be large, to enable lining up the holes and failure plane, yet they needed to be kept as small as possible to utilize the stroke of the testing machine.

As the angle of installation tended toward 90°, the holes in the attachment flanges grew further apart; this was much more pronounced in the vertical failure plane testing. Many off the tendons in the 70° and 80° range fitted into the machine but too short a stroke was available to test the tendons successfully; so the test set was incomplete.

Tendon type

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 5—Different combination loading situations due to geological structures VOLUME 114

OCTOBER 2014

831

▲

The type of tendon used for the testing programme was the hydraulically pre-stressed friction bolt manufactured by New Concept Mining (Pty) Ltd. For a friction tendon there is direct contact between the entire length of the tendon and the rock mass and therefore loading will be concentrated at the intersection of a structure, as previously explained. Li and Stillborg (1999) propose the concepts of ‘neutral point’, ‘pickup length’, and ‘anchor length’ which describe the mechanical coupling at the interface between the rock and a

Testing tendon support units under a combination loading scenario Testing off the tendons at different ff installation angles, and therefore under different combination loading situations or different combinations of tensile and shear load, was made possible by creating failure planes (one vertical and one horizontal) that intersected the tendon axis at different angles, as shown in Figure 6 (installation angles of 45° (left) and 70° (right) are shown). This allowed the force or load to act at different angles across the tendon axis and thereby produced different combinations of tensile and shear force components. A tendon installed in the test machine at 60° installation angle with a horizontal failure plane is shown in Figure 7.

The test machine generates either a compressional or tensile force in a vertical direction. To achieve shear or lateral movement the force must act parallel to the failure plane and the failure plane must therefore be vertical, as seen in Figure 6. All tests where the failure plane was vertical are referred to as vertical tests. The vertical failure plane intersected the tendon axis at different angles and this in turn related to different installation angles. All possible configurations could be achieved with this set-up. The tensile force component increased as the intersection angle between the failure plane and the tendon axis decreased from 90° towards 0°. Jig configurations allowed for the testing of tensile shear only – no compressional shear was tested in the investigation.

To achieve dilation or opening up of joints, where forces act perpendicular to the failure plane, the failure plane had to be horizontal, as seen in Figure 6; therefore these tests are referred to as horizontal tests. The force acted in the gravitational direction and this configuration resulted in predomi-

nantly tensile fforces. The ffailure plane intersected the tendon axis at different angles and this in turn related to different installation angles. The limited extent of the stroke on the machine did not allow for the testing of high (i.e. nearvertical) installation angles as the tendons were too long. The shear force component increased as the intersection angle of the failure plane to the tendon axis decreased from 90° towards 0°.

Bending moment and rotation have a large influence during the combination loading. In the investigation, only forces that acted parallel or perpendicular to the failure plane were investigated. Numerous other configurations can exist where the force acts at angles less than 90° to the failure plane. This can occur where the joint or failure plane is not horizontal and loading occurs under the force of gravity or the applied force direction is not vertical i.e. when a rotation or cantilever occurs. The results of tests where the applied forces act either parallel to, or perpendicularly across the joint / failure plane could possibly be interpolated to represent such situations.

The test jigs were prepared in such a manner that the applied force acted either parallel or perpendicular to the failure plane. Observations during the tests revealed that the vertical test configuration created more than just a combination of tensile and shear forces. Compression across the axis of the tendon was generated in the area of the failure plane, which increased the circumferential or radial forces due to the decrease in the circumference. During horizontal tests, a number of couples and moments occurred, which resulted in rotation.

To limit rotation of the test specimen during testing, two tendons were used to attach the jig to each clevis. Tendons were installed upside-down in the machine. The machine was zeroed at the lowest position to allow for the maximum stroke length. The upper section of the machine was raised vertically to represent a purely gravitational load. The jig was then pulled until the tendon failed, and the strength of the tendon material determined. In the case of the 90° test, the loading force acted axially along the tendon, creating a purely tensile force on the tendon.

Figure 6—Test jigs used for testing tendons at 45° (left) and at 70° (right)

Simulating in situ conditions

Figure 7—Test set-up for a tendon installed at 60° with a horizontal failure plane

▲

832

OCTOBER 2014

VOLUME 114

Simulation of the in situ conditions of the installed tendon and the surrounding rock mass was investigated to reproduce the loading conditions to which tendons are subjected. As the tendons being tested were friction bolts, testing was carried out at the same diameter and profile that the tendons would be when loaded underground. A jig was required to pressurize the tendons to the correct diameter. As the tendons could not be properly secured in the testing machine at different installation angles, several jigs were required.

Simulation of all in situ conditions was not possible for all components for a number of reasons, including: ➤ All jigs were constructed from steel tubing. The friction contact plane was thus steel (tendon) on steel (jig). The frictional load from steel on steel is much lower than that of steel on rock. Where possible, the tendon was locked into the jig to prevent slipping to test the load performance of the tendon, rather than the pullout strength The Journal of The Southern African Institute of Mining and Metallurgy

Testing tendon support units under a combination loading scenario ➤ Testing was carried out with all fforces acting in the vertical plane as per gravity ➤ The test jigs were set up so that the centre of gravity for each test was in the centre of the failure plane.

Testing did not take into account other loading situations, such as cantilevering or where the centre of gravity could be offset from the tendon axis ➤ The steel jigs created sharp edges along the failure planes, which would form prominent shearing/cutting edges in the vertical tests. ➤ The rigidity of the jigs did not allow for any breakout in the areas where the main force concentrations occur (breakout could occur here in the in situ conditions), although some deformation did occur at these points.

The test jigs were constructed to form a stiff testing system, which aimed to represent reality. During the initial tests, the rotation and deformation of the jigs revealed that the testing system was not stiff enough. A second set of jigs was constructed, using thicker steel tubing and with gussets welded along the length of the jig. Two holes (instead of one) were cut into the attachment flanges and clevises to prevent rotation.

Despite these improvements, rotation of the test tendons still occurred, indicating that the system was possibly still too soft. The progressive rotation during testing of a tendon and 30° test jig is shown in Figure 8. Deformation and failure of a second generation 65° vertical test jig with gussets is shown in Figure 9. The two attachment tendons at the attachment flange onto each clevis (as shown) failed to prevent rotation on the jig about the end of the gusset position, resulting in the test jig tearing along the attachment flange.

The results for the combination loading tests and pure tensile test are illustrated in Figure 10. This shows all the test results with the maximum loads achieved during every combination load and tensile test. The data includes the maximum loads achieved during slipping of tendons and jig failures, which introduces a large degree of variability into the data for the combination loading and tensile tests. Some of the loads therefore appear to be low and trend lines cannot be established as the data is skewed by the affected data. The

data-set is not complete, and ffurther testing is required to adequately describe the performance, yet the general trend of performance is shown.

From the failure curves for the vertical and horizontal tests respectively at the different tendon installation angles, it was noted that the curve profiles for both the HydraboltTM and X-PandaboltTM for each test type (i.e. the vertical and the horizontal tests) are similar. This indicates that the test results are comparable and that both tendon types behave similarly, but with different degrees of stiffness and load capacities.

Conclusions Testing process ➤ Combination load testing is a challenging process that requires testing jigs to be prepared for different intersection angles. This can be time-consuming and costly ➤ Thought must be given to fitting the test jig and tendon into the testing machine and achieving the correct loading direction, so as to test the tendon performance without interference from the test jig itself ➤ Simulation of in situ installed tendon and rock mass conditions is difficult, but must be taken into consideration

Figure 9—A test jig with 65° installation angle and vertical failure plane, where failure has occurred on the test jig itself

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 10—Plot of maximum loads achieved for all tests VOLUME 114

OCTOBER 2014

833

▲

Figure 8—A series of photographs from a 30° horizontal test showing the rotation that occurs during testing

Testing tendon support units under a combination loading scenario ➤ Robust test jigs are required, as the torque component loads affect both the test jig and testing machine adversely. A test jig machined from a solid block of metal would be more appropriate, although the resilience of such a jig for conducting multiple tests would have to be confirmed ➤ In a testing programme, the testing machine and a limited number of jigs should be made and tested in a first phase. Unexpected defects in the jig design and test outcomes may require modification and remanufacture of the test jigs ➤ Anchoring the tendon into the jig securely and preventing slippage is problematic – this requires further experimentation for each type of tendon to be tested ➤ Testing at higher installation angles requires larger testing machines with longer strokes (over 1 m) to accommodate vertical alignment of the attachment points ➤ Photographs and videos taken during the testing process represent very valuable evidence, as they record actions that cannot be seen with the naked eye and can be revisited numerous times after testing ➤ If possible, physical investigation (and re-investigation) of the failed units can offer clues to the mechanism of failure and are valuable records.

bolt – a laboratory investigation. Jurnal Kejuruteraan Awam, vol. 16, no. 1. pp. 1–12.

BRADY, B.H.G., and BROWN, E.T. 2004. Rock Mechanics for Underground Mining. 3rd edn. Kluwer Academic Press, Dordrecht.

BLANCO, M.L., HADJ-HASSEN, F., TIJANI, M., and NOIRET, A. 2011. A new experimental and analytical study of grouted roofbolts. 45th US Rock Mechanics/Geomechanics Symposium, San Francisco, CA, 26-29 June 2011.

CROMPTON, B. 2003. Shear, bending and combined load testing on Hydrabolts, X-panda bolts and other support systems. New Concept Mining, Internal Report, 26 November 2003. Reference HB/0002, DAEHNKE, A., VAN ZYL, M., and ROBERTS, M.K.C. 2001. Review and application of stope support design criterion. Journal of the South African Institute of Mining and Metallurgy, g May/June. pp. 135–150.

DSI DYWIDAG-SYSTEMS INTERNATIONAL P/L. Evaluation development in roofbolting. Technology in Australian Coal Mining. DSI DYWIDAGSystems International P/L. 1–13.

FERNANDES, N. 2007. Changing from X-Panda bolts to Hydrabolts in the ASG’s, Re/Pre development. Internal report, Impala Platinum Limited. pp. 1–4.

GARDNER, L.J. 2004. Feedback on rock engineering issues assessed during recent visit to Ngezi Mine. Internal report, Impala Platinum Limited.

GAUDREAU, D., AUBERTIN, M., AND SIMON, R. 2004. Performance assessment of tendon support systems submitted to dynamic loading. Proceedings of the Fifth International Symposium on Ground Support, t Perth, Australia,

Test results

28–30 September 2004. Villaescusa, E. and Potvin, Y. (eds.). Taylor &

➤ Combination loading of tendons commonly occurs in any rock mass, due to the variety of jointing orientations and tendon installation angles ➤ The intersection angle of the tendon and the geological structure/discontinuity, together with the position and orientation of the load, will determine the ratio of tensile component versus shear component ➤ Generally, the tendon failure mode tends towards tensile failure, where the tensile component is higher than the shear component ➤ Where the shear component is higher than the tensile component, the failure loads are much higher than for pure shear ➤ A component of rotation is involved in cases where the tendons tend towards tensile failure, and this can aid in wedging blocks of rock in place and preventing failures.

Finally, this testing programme represents a mere starting point for understanding the combination loading on tendons.

Further testing of all types of tendons is required to reach a better understanding of the effects of combination loading on tendon support units and systems.

Francis, London.

HUTCHINSON, J.D. and DIEDERICHS, M.S. 1996. Cablebolting in Underground Mines. BioTech Publishers, Richmond, BC, Canada. pp. 28–33, 120–125.

IMPALA PLATINUM. Review of Impala Platinum Mine’s tendon support strategy with specific reference to protruding tendons. Internal report.

JOUGHIN, W.C., NEZOMBA, E., RWODZI, L., and JAGER, A. 2011. Rockfall elimination Track B, a risk based approach to enhancing support design in Bushveld underground mines, Volume 1. MHSC, Safety in Mines Research Advisory Committee SIM 060201 Track B, Research agency SRK Consulting. July 2011. pp. 76–110.

LI, C. and HAKANSSON, U. 1999. Performance of the Swellex bolt in hard and soft rocks. Rock Support and Reinforcement Practice in Mining. g Villaescusa, E., Windsor, C.R., and Thompson, A.G. (eds.). Balkema, Rotterdam. pp. 103–108.

LUO, J.L. 1999. A new rock bolt design criterion and knowledge-based expert system for stratified roof. Phd dissertation, Faculty of the Virginia Polytechnic Institute and State University. pp. 4–20.

MAHONY, L., HAGAN, P., HEBBLEWHITE, B., and HARTMAN, W. 2005. Development of a laboratory facility for testing shear performance of installed rock reinforcement tendons. Proceedings of the 24th International Conference on Ground control in Mining. g University of West Virginia, Morgantown,

Acknowledgements The authors wish to express their gratitude to the management of Impala Platinum Limited for assistance in performing the research and permission to publish the work.

The authors acknowledge the contribution of New Concept Mining (Pty) Ltd for the supply of tendons, testing facilities, and assistance during testing.

MCHUGH, E. and SIGNER, S. Roof bolt response to shear stress: laboratory analysis. National Institute for Occupational Health and Safety Spokane Research Laboratory Spokane, WA.

Reference LI, C. and STILLBORG, B. 1999. Analytical models for rock bolts. International

Bibliography

Journal of Rock Mechanics and Mining Sciences, vol. 36.

AMIN, M., SIANG, K.K., and CHON, C.H. 2004. Reinforcement mechanisms of rock

▲

834

August 2005.

OCTOBER 2014

VOLUME 114

pp. 1013-1029.

◆

The Journal of The Southern African Institute of Mining and Metallurgy

Estimation of future ground vibration levels in Malmberget town due to mining-induced seismic activity

by T. Wettainen* and J. Martinsson*

Malmberget town (Figure 1) is located in northern Sweden, approximately 70 km north of the Arctic Circle.

The Malmberget iron-ore deposits have been known for centuries. The ore was initially transported from the mine by horses and reindeer, until the railway to the coastal city of Luleå was completed in 1888 and large-scale production began. Today most of the refined iron ore products from Malmberget such as fines and pellets are transported to Luleå for

* LKAB, Sweden. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. This paper was first presented at the, 6th Southern African Rock Engineering Symposium SARES 2014, 12–14 May 2014, Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, Muldersdrift.

Synopsis Malmberget town is located in northern Sweden, approximately 70 km north of the Arctic Circle. Parts of the town overlie more than 20 iron orebodies, consisting mainly of magnetite with smaller quantities of haematite. The mine is operated by the mining company LKAB. Mining started in the 17th century, but not until the railway to the coastal city of Luleå was completed in 1888 did large-scale production commence. Around 1920, mining proceeded underground and today sublevel caving is the only mining method used. Sublevel caving causes subsidence of the ground surface, and buildings and residential areas have been relocated due to the mining activities for more than 50 years. The number of seismic events accompanied by strong ground vibrations is now increasing. In 2008 the mine received a permit from the Environmental Court of Sweden to increase production to 20 Mt of crude ore per year. A prerequisite for the permit was that the mine conducts a number of investigations regarding the environmental impact on the residents of Malmberget. One of these investigations concerned how seismicity will change as production increases and what measures could be taken to reduce inconvenience to the town residents.

Today the mine possesses an extensive seismic monitoring system with more than 180 underground and surface geophones. For this study, eleven seismically active volumes in Malmberget mine were identified, and for each of them, a yearly future maximum magnitude interval was estimated based on the current production plan. Relationships between historical seismic events and measured ground vibrations in the town of Malmberget were established, and future ground vibrations caused by expected seismic events were estimated using a probabilistic approach. The outcome was the number of intervals of expected ground vibration per year and per monitoring point. Possible measures to reduce inconvenience for the town residents include blast restrictions, sequencing, and possibly preconditioning. The ultimate long-term solution is an almost complete relocation of Malmberget town. This process has recently been formalized and LKAB is taking an active part in realizing this goal.

Keywords sublevel caving, mining-induced seismicity, surface vibrations, future estimations, environmental impact.

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

835

▲

Introduction

use in steel mills mainly within the Baltic Sea region. The mine consists of around 20 orebodies of varying sizes and shapes (Figure 2).

Production in 2012 was 14 Mt of crude ore from 12 of the orebodies mined to a maximum depth of approximately 850 m below surface.

The main mineral is magnetite, but there are also smaller quantities of haematite. The mine is operated by Luossavaara-Kiirunavaara AB (LKAB). The town of Malmberget has existed in close relationship with the mine for well over a hundred years. The first houses were built by miners who wanted to live close to their place of work. More miners followed and soon a shanty town resembling those found in Klondike during the gold rush era was established on the hill slopes of Malmberget (Figure 3).

The rich ore deposits were not depleted, and as the town grew the old shacks were eventually replaced by modern buildings.

Mining started in open pits, but by the 1920s more than 90% of all mining was performed underground (Forsström, 1973). Many orebodies dip to the southwest, which gradually brings mining activities closer to the town itself. Since the 1960s sublevel caving has been the predominant mining method.

This is a large-scale method, in which different activities take place on several levels simultaneously (Figure 4).

This method makes it possible to efficiently mine deposits at great depth at competitive cost. The main disadvantages are subsidence of the ground surface and, for Malmberget town, vibrations caused by seismic events due mainly to the caving of hangingwall areas (Figure 5).

Estimation of future ground vibration levels in Malmberget town

Figure 5—Environmental impact of sublevel caving Figure 1—Aerial view of Malmberget

Figure 2—Orebodies in Malmberget, metric scale, plan view

Buildings, residential areas, and infrastructure have been relocated due to the presence of mining areas for more than 50 years, and the acceptance of this process is generally high among the residents of Malmberget. Inconvenience caused by seismic events is, however a relatively new problem.

Vibrations from blasts have always been a part of everyday life in Malmberget, but vibrations from seismic events are perceived differently. Not only do they occur randomly rather than at specific hours, but the ground motion is also different due to the frequency and characteristics of the source. In 2008 the mine received a permit from the Environmental Court of Sweden to increase production to 20 Mt crude ore per year. A prerequisite to the permit was a number of investigations to be conducted by the mine regarding the environmental impact on the residents of Malmberget. One of these investigations concerned how seismicity will change as production increases and what measures could be taken to reduce inconvenience for the town residents.

Seismic monitoring system

Figure 3—Malmberget main street 1895 (LKAB archives)

The mine possesses an extensive seismic monitoring system, provided mainly by the Institute of Mine Seismology (IMS).

The first sensors were installed in 2005, around the time when vibrations of unknown origin started to occur in Malmberget. The system has been expanded stepwise and by the end of 2013 there were around 180 sensors operational in the mine. The array consists of a few 1 Hz geophones but mainly 4.5 Hz (1/3) and 14 Hz (2/3) geophones. The 1 Hz geophones are installed on concrete slabs. Five are located on the surface in a ring formation surrounding the mine. An additional 1 Hz geophone is located underground in order to improve the vertical location accuracy of seismic events. Data from the 1 Hz array, which is stored in a separate database, was not used in this project, since the array had only recently been installed. The other types of geophones are all installed underground, from around 100 m below surface to the deepest level of the mine. Approximately 500 actual seismic events are currently processed daily. The classification of seismic events is done manually. LKAB uses the same local magnitude scale (ML), which is based on seismic energy (E) and moment (M), as many South African mines [1] In addition to the geophones installed in the mine, there are also eight 8 geophones bolted to foundations of residential buildings around town to serve as vibration monitoring points. The incoming vibration is measured according to Swedish Standard SS 460 48 66. Figure 6

Figure 4—Sublevel caving

▲

836

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Estimation of future ground vibration levels in Malmberget town

Figure 6â&#x20AC;&#x201D;Vibration monitoring points in Malmberget town. Blue line is the industrial fence, and black line is recent fence advancement

shows these monitoring points on surface, f ffor which ffuture vibration levels have been estimated. The amplification effects on the upper floors of buildings have not been considered, since each building responds differently.

Estimation of future ground vibrations Seismic sources The first step is to define seismically active volumes or sources in the mine. As new seismic events occur, they are automatically placed in spatial clusters by the analysis software mXrap (formerly MS-RAP), developed by the Australian Centre for Geomechanics. Clusters are then sorted into larger groups by mine personnel on a daily basis.

Seismically active volumes were identified by combining groups together. The criteria for creating a volume were (i) to include most events with a local magnitude of 1 or higher and (ii) to obtain a good visual fit to the Gutenberg-Richter curve. This curve represents a frequency-magnitude distribution and is commonly used for seismic analysis (Gutenberg and Richter, 1944; 1954). It shows the activity rate (number of events) and the relation between small and large events for a given timespan and population. The observed magnitudes are assumed to be exponentially distributed within the sensitivity of the array. Eleven seismically active volumes were identified in the mine (Figure 7).

Only seismic events associated with the seven major orebodies were considered in the current analysis. The other orebodies are smaller and mined at shallower depths. They have not yet caused significant seismic events and are not expected to do so within the foreseeable future.

predictions are valid only under the condition that, given the known covariates, the estimated parameters in the distribution will have the same characteristics in the future. The Environmental Court asked LKAB how seismicity will change when production increases. For this reason, the historical relationship between activity and monthly production was modelled for each of the seven major orebodies. All seismic events, regardless of size, recorded by the seismic system and spatially associated with the individual orebodies were considered. A linear regression model was used to describe the relationship between mining-induced activity as a function of production and mining depth during the same time period. For a given production rate and mining depth, the activity (number of events per month), is assumed to be lognormally distributed, and the maximum likelihood estimation (MLE) of the parameters is considered under the condition that the parameters are positive.

Mining depth was included as a parameter but proved to be insignificant for the data considered in this study. Data from 2010 until late 2012 was used, and the number of new mining levels established during this time is limited. The activity-production relationships were then used to predict

Production and activity rate

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 7â&#x20AC;&#x201D;Isometric view of events in seismically active volumes. Black dots represent the surface vibration monitoring points in Figure 6 VOLUME 114

OCTOBER 2014

837

â&#x2013;˛

Using a well-defined population of seismic events, it is possible to statistically model the distribution of the magnitude of events using the Gutenberg-Richter relationship, and from that, estimate the distribution of the maximum event for a given period of time. However,

Estimation of future ground vibration levels in Malmberget town future activity based on the current production plan (see examples in Figure 8–10). In this study we restrict ourselves to considering the orebodies individually, given their monthly production and their monthly average mining depth. This means that possible interactions between the orebodies are neglected in order to reduce the dimensionality of the estimation problem. On a mine-wide level they are dependent, but to capture this dependency we need to resort to multilevel or hierarchical models (Gelman et al., 2004; Gelman and Hill 2007) and treat the estimation problem jointly. However, this will increase the dimensionality by a factor equal to the number of orebodies; often together with additional hyper-parameters describing the mine-wide prior distributions for the model parameters.

different ff seismic sources within one single volume. The sum of exponential distributions is also multiplied by a sensitivity function for the seismic measurement system. The entire PDF is given by

Future magnitudes

describes the sensitivity function where h(.) denotes the Heavyside function. The term

The probability density function (PDF), p(m), of the magnitude m is modelled as a sum of exponential distributions in order to account for the possible presence of several

[2] where [3] is the exponential tail and [4]

[5]

Figure 8—Modelled historical activity (top) and predicted future activity (bottom) for Fabian orebody

Figure 9—Modelled historical activity (top) and predicted future activity (bottom) for Dennewitz orebody

▲

838

OCTOBER 2014

VOLUME 114

depends only on the model parameters and makes the entire distribution in Equation [2] integrate to unity. The parameter vector for the entire PDF is θ = [μ, τ, wT, βT]T, where w = [w0,…,wk-1]T is the weights under the condition that ||w||1=1 and the elements in β = [β0,…,βKK-1]T are the parameters for the exponentials in Equation [3]. Figure 11 shows an example of the density in Equation [2] and its components in Equations [3] and [4] for K=1 and θ = [-1, 1/2, 1, 2]T. The number of mixtures to use is determined by the Bayesian information criterion (BIC) to avoid over-parameterization (Stoica and Selen, 2004; Pintelon and Schoukens, 2001). Examples of estimation results using Equation [2] from the different orebodies can be seen in Figure 12-14.

Note the model's ability to capture complex features, such as the data shown in Figure 13 exhibiting an additional activity at m=-1 and the distinct mode in Figure 14.

Figure 10—Modelled historical activity (top) and predicted future activity (bottom) for Kapten orebody The Journal of The Southern African Institute of Mining and Metallurgy

Estimation of future ground vibration levels in Malmberget town

Figure 12—Estimated probability density function (PDF) of the magnitudes in Dennewitz volume

Figure 13—Estimated probability density function (PDF) of the magnitudes in Fabian volume The Journal of The Southern African Institute of Mining and Metallurgy

Figure 14 – Estimated probability density function (PDF) of the magnitudes in Printzsköld 3 volume

The advantages of introducing a sensitivity function (Equation [4]) is to avoid a hard truncation of the data at a specific sensitivity threshold using the more common magnitude distributions such as the Gutenberg-Richter relation for magnitudes or the open-ended power law (Gutenberg and Richter, 1944; Ishimoto and Iida, 1939), the upper-truncated power law (Page, 1968) or gamma type distributions (Saito et al., 1973).

Truncating data in general means that our statistical model is valid only in a specific region of the data. Truncation also means that we need to estimate the region for which our model is valid, and there is a risk of either throwing away valuable data if we truncate too much or jeopardizing the validity of our model if we expand this region too much. This choice will affect the uncertainty of our parameter estimates and may also contribute to bias estimates using improper models (see e.g. Pintelon and Schoukens, 2001; Kay, 1993; Scharf, 1990). For example, the lower truncation limit for using a translated exponential distribution (i.e. the openended power law) would be in the region containing most of our data and the estimation effects will be even more severe.

The difficulties of applying some of the common magnitude distributions mentioned above can be seen e.g. in Mendecki (2008). Lasocki and Orlecka-Sikora (2008) also argue that the use of these common distribution models brings about an unacceptable and systematic over- or underestimation of the seismic hazard parameters.

Introducing a sensitivity function in Equation [4] means that the statistical model is valid for the entire data-set at the cost of two additional parameters (μ,τ ) describing the sensitivity of the system. Considering a mixture of exponentials means that we can also model more complex behaviour discussed above and allow the collected data, together with an information criterion, to determine the shape of the distribution in the volume of interest.

The other alternative for modelling a complex size distribution is to consider a non-parametric kernel estimator of the density (see e.g. Lasocki and Orlecka-Sikora, 2008; Kijko et al., 2001). This method has the ability to capture complex shapes and behaviour, but comes at the cost of choosing an appropriate kernel function and corresponding kernel VOLUME 114

OCTOBER 2014

839

▲

Figure 11—An example of the probability function in Equation [2] and its components (Equations [3] and [4]) for the parameter vector θ = [-1, 1/2, 1, 2]T

Estimation of future ground vibration levels in Malmberget town bandwidth (or scale). The main reason ffor not taking this route in this study is that the number of kernel parameters is proportional to the number of events, resulting in an unmanageable density in the sequential Monte Carlo analysis. Also, the largest magnitudes observed here still have an exponential (or power law) behaviour that we like to preserve.

Instead, we have focused on using only a few kernels in our mixture model. The kernel includes the sensitivity of the measurement system in order to avoid the truncation problems discussed above. It can handle complex shapes caused by multiple sources (Figure 12-14) and we have preserved the exponential tail (i.e. the power law behaviour).

The cost of a parametric mixture model is that we have to estimate the appropriate number of mixtures (or kernels) to use by applying some cross-validation techniques (Ljung, 1987) or some well-established information measure (Stoica and Selen, 2004).

Gutenberg-Richter curves for each volume were plotted and modelled using measured data from 2010 until late 2012. Since the sensitivity of the system is limited, small events are more difficult to detect and the number of events below the sensitivity limit is less, as shown in Figures 12-14 and the top plot of Figure 15. The vertical axis in latter plot represents probability of occurrence but could also show number of events. Probability 100% (10° or 1) corresponds to the total number of events in the data-set. We expect all detectable events to be larger than approximately magnitude -2, which is roughly the detection limit of the seismic system in this particular area of the mine. The curves were normalized to one year and annual activity rates were used to obtain future magnitudes. In this way, maximum expected magnitude intervals were estimated for each volume and year.

The model of the survival function, representing the blue curve in the top plot in Figure 15, is given by

P(m) denote the probability that the magnitude off an event is larger than m. The intersection of the horizontal red line at m=mlow (Figure 15) gives us probability A-1 of occurring. The magnitude density of the largest event occurring is approximated by [7] where mh is the magnitude of the largest expected event that year and p(m) denotes the PDF of the magnitude of an event.

The shape is given by a translated exponential distribution starting from mlow where the term A makes it integrate to unity. A comparison between the approximation and true distribution (P(m))A for a specific activity A can be seen in Figure 16.

However, as the estimated activity and the model SF(m) are associated with uncertainties, shown by the dashed red and blue lines in the top of Figure 8, the intersection is also uncertain. Monte Carlo integration is applied to obtain an average distribution p(mh) for K possible intersections mlow(k), k = 1,...,K, taking these uncertainties into account.

This part of the analysis is where the approximation of (P(m))A in Equation [7] is useful. The effect of the approximation errors using Equation [7] (Figure 16) is small in comparison to the uncertainties from the possible intersections shown in Figure 15. Note also that the assumption of independent identically distributed magnitudes leading to (P(m))A is often violated in mining conditions. The resulting average distribution of K possible intersections is

[6] where P(m) is the cumulative distribution function (CDF) of the magnitude describing the measured data represented by the black dots. The survival function simply gives the probability that the magnitude of an event is larger than m.

The complementary cumulative histogram of the magnitude data, represented by the black dots, is shown with relative frequencies as opposed to the number of events in each histogram bin to obtain an estimated probability. Using relative frequencies means that we can plot the complementary cumulative histogram of the magnitudes against the probability given by the survival function.

For a specific activity (number of events per year), denoted by A, we can estimate the magnitude distribution of the largest event that year. If the cumulative probability distribution P(m) of the magnitude m is given, then for a given specific activity A the probability that m is the maximum value is (P(m))A, assuming independent identically distributed magnitudes (see e.g. g Gumbel (1967), and consequently 1-(P(m))A is the probability that m may be exceeded. To simplify the sequential Monte Carlo analysis described below, this distribution is approximated by a translated exponential distribution describing the dominated terms in the tail of the distribution (P(m))A. Let SF(m)=1-

â&#x2013;˛

840

OCTOBER 2014

VOLUME 114

utions for Fabian volume

Figure 16â&#x20AC;&#x201D;Comparison between the approximation in Equation [7] and the true distribution for A=10 000 The Journal of The Southern African Institute of Mining and Metallurgy

Estimation of future ground vibration levels in Malmberget town represented by the blue curve in the bottom plot in Figure 15 together with the histogram of the possible intersections mlow(k) in red. The green histogram mmedian(k) shows the distribution of the median magnitude of each p(mh|mlow(k)).

To summarize, we approximate the largest event per year to have a magnitude density according to the blue curve.

One of the 11 identified seismic volumes includes a caving void. The Printzsköld orebody was not initially mined from surface, but mining started underground. This created an empty room that eventually will cave up to the ground surface. It seems that the Gutenberg-Richter approach overestimates the magnitudes in this case. The measured data has a poor fit to the Gutenberg-Richter model (Figure 17).

One possible explanation could be that there is a different ratio between small and large magnitudes in this particular case. Caving rooms in Malmberget have been monitored before and experience shows that the magnitudes associated with this process are not significant. Instead, the following equation, presented in Mendecki (2008), was used to estimate future maximum magnitudes for the caving volume.

estimation errors ffrom previous events is used to describe the attenuation caused by wave propagation from specific locations to each sensor. The calibration reduces the estimation errors by 50% when it is evaluated using crossvalidation. 275 historical seismic events with a total of 1027 corresponding surface vibration values were used for calibration. Only vertical vibration components were considered since this data is more abundant. Cross-validation was performed and the root mean square deviation (lgPGV) is 0.22. Figure 18 shows measured and predicted peak vibrations from 90 seismic events at a specific monitoring point.

Using the surface vibration relations and expected magnitude distributions, future vibrations at the monitoring points were estimated probabilistically. Monte Carlo simulation was used to sample hypocentre, magnitude, and

[8] where log gPmax is equal to the largest observed seismic potency plus the largest jump between two record potencies in the observed series. log gP was replaced with local magnitude ML in this analysis.

Table I shows Mmax for each seismic volume and year.

The median magnitude of each Mmax interval is considered as Mmax here for illustrative purposes, except in column Printzsköld 1 alt. where Equation [8] was used and the actual Mmax is obtained. Printzsköld 1 is the volume with the caving room and the magnitude difference between the two methods vary between ML 1.1 and ML 1.4. Even though Equation [8] does not take production increase into account, the magnitude it provides is considered reasonable in this context.

Figure 17—Gutenberg-Richter curve, model for caving volume Printzsköld 1

Surface vibrations Relationships between magnitudes, spatial locations, and measured surface vibrations were established for each surface monitoring point. A regression model is used to estimate the vibration as a function of the magnitude of the event, the hypocentre location of the event, and the sensor coordinate on the surface. The vibration is assumed lognormally distributed and MLE of the parameter is considered under the condition that the parameters are positive. A calibration function which is retrieved through

Figure 18—Measured vs predicted peak vibrations, linear scale (top) and log scale (bottom) at monitoring point Sveavägen 7

Table I

Mmax by volume and year Alliansen Dennewitz Parta 1 Parta 2 ViRi Kapten-Fabian Kapten Fabian Printzsköld 1 Printzsköld 1 alt. Printzsköld 2 alt. Printzsköld 3 alt.

1.9 1.9 2.0 2.0 2.0 2.1 2.0 2.1

2.1 2.3 1.8 2.3 2.2 2.3 2.2 2.2

2.4 2.5 2.4 2.4 2.3 2.3 2.3 2.2

2.7 2.8 2.7 2.6 2.6 2.6 2.6 2.4

1.8 1.8 1.9 1.8 1.8 1.8 1.8 1.7

1.4 1.6 1.6 1.5 1.6 1.7 1.7 1.7

The Journal of The Southern African Institute of Mining and Metallurgy

0.5 1.1 1.1 0.5 1.0 0.9 1.0 1.0

2.0 1.9 2.0 2.0 2.0 2.1 2.2 2.2

3.1 3.1 3.2 3.2 3.2 2.9 3.1 3.0

VOLUME 114

1.8 1.8 1.8 1.8 1.8 1.8 1.8 1.8

2.2 2.2 2.2 2.2 2.3 2.0 2.1 2.0

OCTOBER 2014

2.4 2.4 2.5 2.5 2.5 2.2 2.4 2.3

841

▲

2013 2014 2015 2016 2017 2018 2019 2020

Estimation of future ground vibration levels in Malmberget town corresponding surface f vibrations ffor each seismic volume.

Figure 19 shows maximum expected peak vibrations for each year at a specific monitoring point.

Table II summarizes the median of the maximum expected peak vibration values for each monitoring point.

The critical year for each point is highlighted. No predictions have been made beyond 2020 because of large uncertainties in the production plan. Every change in production input will result in different vibration levels. The maximum measured vertical vibration ever within populated areas has been 19.5 mm/s, although the absolute majority is below 5 mm/s.

Measures to reduce inconvenience LKAB has identified a number of measures that could be taken in order to reduce vibrations from seismic events in populated areas.

clear to LKAB that the largest orebodies expand with depth, a long-term action plan for community relocation became necessary. In 2012 an agreement was reached with the municipal council stipulating LKAB’s responsibilities as well as obligations by local authorities. The agreement contains a time plan for relocation of residential areas in Malmberget (Figure 20). This is dependent on successful completion of corresponding expansion plans by the municipality in Malmberget’s twin town Gällivare, some 5 km to the south. The relocation plan does not include one area in eastern Malmberget or housing areas owned by LKAB, where efforts will be made to maintain good living conditions.

Conclusions

➤ Blast restrictions are already applied in the mine and can be used in specific areas based on previous experience. A numerical stress model can also be used to identify potentially critical areas or orebodies. The number of orebodies is decreasing with depth as some of the smaller orebodies will become narrower with depth while the larger ones will increase in size. Blast restrictions will have an increasingly negative effect on production ➤ Sequencing can be used to avoid stress concentrations in known structures or weak zones. Currently, sequencing is used primarily to ensure underground safety. Using it also with the intention to reduce surface vibrations will most likely adversely affect production ➤ Pre-conditioning methods such as hydraulic fracturing have been used in other mines to reduce large seismic events in production areas (Quinteiro, 2012). LKAB will investigate this option further, but the events that cause disturbances on the surface are usually located some distance into the hangingwalls. It may be difficult to access these areas with pre-conditioning equipment ➤ Relocation of Malmberget town is the ultimate solution.

This is also the measure that will provide the desired results with high confidence. Malmberget has continuously been affected by the mine piece by piece for over 50 years. Buildings have been demolished or moved and roads have been closed. Since it became

LKAB was asked by the authorities to assess the probable levels of strong ground motion from future seismic events and their impact on residents of Malmberget. LKAB used a well-known method based on Gutenberg-Richter curves and observed historical data to estimate the largest expected seismic magnitudes. To estimate the impact of an increase in production, the curves were adjusted for different activity rates. This modification has to our knowledge not been done or published before, but is a way to try to answer the question from the Environmental Court. The final outcome was intervals of expected future vibrations under given conditions and probabilities. One advantage in a mining environment is that the driving force of seismic events, mining itself, can be controlled to some degree. Measures can be taken in order reduce seismic activity, but how fast and to what extent results will be achieved is impossible to say.

Many measures could also lead to production losses and

Figure 19—Estimated maximum vertical vibration levels at monitoring point Sveavägen 7

Table II

Estimated vertical peak vibration (mm/s per year) and monitoring point Hermelinsbacken 7

Lövberga

Konsum

Bolagskontoret

Hertiggatan 18

Murgatan 1

Sveavägen 7

Malmstavägen 11

2013

10.0

2014

11.1

6.1

9.3

18.9

4.5

7.4

13.1

8.2

6.8

10.4

21.1

5.5

8.3

14.6

2015

9.1

9.7

6.0

9.2

18.7

5.7

7.3

12.9

8.1

2016

8.7

5.2

8.0

16.4

4.4

6.4

11.3

7.0

2017

9.1

4.7

7.2

14.8

6.0

5.7

10.2

6.3

2018

7.4

4.7

7.2

14.7

6.2

5.9

10.1

6.3

2019

7.5

4.7

7.2

14.7

7.1

6.9

10.1

6.3

2020

6.1

3.7

5.6

11.6

7.0

6.6

8.0

5.0

▲

842

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Estimation of future ground vibration levels in Malmberget town GELMAN, A. and HILL, J. 2007. Data Analysis using Regression and Multilevel/Hierarchical Models. Cambridge University Press, New York.

GUMBEL, E.J. 1967. Statistics of Extremes. Colombia University Press, New York.

GUTENBERG, B. and RICHTER, C.F. 1944. Frequency of earthquakes in California.

Bulletin of the Seismological Society of America, vol. 34. pp. 185–188.

GUTENBERG, B. and RICHTER, C.F. 1954. Seismicity of the Earth and Associate Phenomena, 2nd edn. Princeton University Press, Princeton, NJ. ISHIMOTO, M. and IIDA, K. 1939. Observations of earthquakes registered with the microseismograph constructed recently. Bulletin of the Earthquake Institute of Tokyo University, vol. 17. pp. 443–478 (in Japanese).

KAY, S.M. 1993. Fundamentals of Statistical Signal Processing: Estimation Theory, vol. 1, Prentice-Hall. New Jersey.

KIJKO, A., LASOCKI, S., and GRAHAM, G. 2001. Non-parametric seismic hazard in mines. Pure and Applied Geophysics, vol. 158. pp. 1655–1675.

Figure 20—Time plan for relocation of Malmberget

Lasocki, S. and Orlecka-Sikora, B. 2008. Seismic hazard assessment under complex source size distribution of mining-induces seismicity.

Tectonophysics, vol. 456. pp. 28–37.

Ljung, L. 1987. System Identification: Theory for the User. Prentice-Hall, New Jersey.

LKAB depends on large-scale production in order to remain competitive. The only way to completely avoid complaints from residents near a seismically active mine is not to have any neighbours at all. It has long been a principle of LKAB not to mine directly under residential areas, even though ground subsidence caused by mining of shallow-dipping orebodies would not occur in many years, if ever. The relocation of Malmberget is partly to ensure the comfort of residents rather than being a safety issue.

References

Mendecki, A. 2008. Forecasting seismic hazard in mines. Keynote address: First Southern Hemisphere International Rock Mechanics Symposium, Perth, Western Australia, September 2008.

Page, R. 1968. Aftershocks and microaftershocks of the great Alaska earthquake of 1964. Bulletin of the Seismological Society of America, vol. 58, no. 3. pp. 1131–1168.

PINTELON, R. and SCHOUKENS, J. 2001. System Identification: a Frequency Domain Approach. IEEE Press, New York.

QUINTEIRO, C. 2012. Report from mine visit at El Teniente, Chile. Internal LKAB document.

SCHARF, L.L. 1990. Statistical Signal Processing: Detection, Estimation and Time Series Analysis, Vol. 1. Addison-Wesley, New York.

STOICA, P. and SELEN, Y. 2004. Model-order selection: a review of information criterion rules. IEEE Signal Processing Magazine, vol. 21, no. 4. pp. 36–47.

GelmAN, A. CARLIN, J.B., STERN, H.S., and RUBIN, D.B. 2004. Bayesian Data Analysis, 2nd edn. Chapman & Hall/CRC.

WETTAINEN, T., MARTINSSON, J., and PERMAN, F. 2014. Investigation U6, Seismic activity Malmberget. LKAB report 14-806 (in Swedish). ◆

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

843

▲

FORSSTRÖM, G. 1973. Malmberget. Department of Human Geography, University of Stockholm. ISBN 91-85336-54-8 (in Swedish).

Ceramite

®

+,99 Ø !:D Ø 97&$19 &/ 3-9 973,9.9 Stronger than conventional steel linings. Stronger than traditional wear resistant tiles. Ceramite® stands up best to the heat and erosion challenges of heavy industry.

Free flow Ceramite®, containing bauxite and/or silicon carbide aggregate, cast with a minimum thickness of 12mm, has a extremely high wear resistance, and an operational life much longer than traditional components in this harsh environment. Ceramite® provides far greater reliability and lower maintenance costs.

Ceramite® has been used extensively in the power, cement, ferro-alloy and platinum industries, and has been remarkably successful with platinum smelter granulated slag slurries, pulverised coal pneumatic conveying, furnace ore feed chutes and many more applications. With a minimum of 48 hour workshop turn-around installation, there is no quicker and stronger method to protect your pipes and other industrial challenges.

Furthermore, Ceramite® can be precast into absolutely any shape to meet your needs. Ceramite® may be applied to existing component designs or Process Consultants will work with Refraline to assess your unique requirements, and design and install a cost-saving protection measure that delivers reduced downtime, less maintenance, lower overall capital costs, and higher production levels.

In tough operating conditions; where your assets, your people, your product quality and your output levels need protection; only Ceramite® shapes up. Put it to work for you.

FREE FLOW CERMITE PROVIDES EXTREME WEAR RESISTANCE IN PIPE LININGS - 48 HOUR MINIMUM WORKSHOP TURN-AROUND

Manufacturers of Ceramite®

Installation Contractor

Visit www.processconsultants.co.za for data sheets and case studies or contact us on 011 803 2013, 082 657 2610 or processconsultants@mweb.co.za

PROCESS CONSULTANTS CONSULTANTS TO THE METALLURGICAL INDUSTRY

Manufacturing and Distribution Licensee

Outsourcing in the mining industry: decision-making framework and critical success factors

by C.J.H. Steenkamp* and E. van der Lingen*

Theoretically, the main driver behind a mining operations’ sourcing decision should differ from company to company, and within a company from project to project, but in reality it often relates to cost. Research confirms that there are a number of factors, including cost, to consider when choosing between in-house and outsourced mining. While the literature is rife with factors to consider, little information exists on how to apply these in practice and the relative importance of the different factors to be considered.

A study was conducted to determine whether mining is truly a core competency for a mid-tier commodity specialist mining company.

Furthermore, a decision-making framework for mining operations sourcing was developed, and the critical success factors that should be adhered to if outsourced mining is chosen were determined.

The research showed that owner mining is not a core competency for the mining company investigated. A decision-making framework was developed using the order winner/order qualifier structure, which can be used to facilitate the mining sourcing decision. The most important tools at the disposal of a mine owner’s team to manage a contractor miner are the social and output control mechanisms, according to the critical success factors study.

Keywords insourcing, outsourcing, contractor, mining, decision-making, critical success factors.

Introduction Outsourcing strategies have been viable in organizations since the 1980s (Hätönen and Ericksson, 2009), and gained significant momentum during the 1990s (Morgan, 1999; Corbet, 2004). Fill and Visser (2000) consider outsourcing as one of the most widely adopted practices followed by firms. Firms started to outsource functions that were not considered their area of expertise in order to become more cost-efficient (Porter, 1996). A number of authors have argued whether or not outsourcing can be viewed as strategy. In the publication ‘What is Strategy?’ Porter (1996) strongly protests that outsourcing is a tool and not, in and of itself, a strategy. In contrast to this, outsourcing is defined as a strategic decision (Embleton and Wright, 1998; Gottschalk and Solli-Saether, 2005). McIvor (2000) goes as far as to say that outsourcing can succeed only if carried out from a strategic perspective, and by fully integrating it with an The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

* Department of Engineering and Technology Management, University of Pretoria, South Africa. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. Paper received r May 2014; revised paper received Aug. 2014.

OCTOBER 2014

845

▲

Synopsis

organization’s larger corporate and operational strategy. Regardless of the different opinions in the literature, most authors agree that outsourcing is an important tool for implementing rapid strategic change.

Outsourcing has further become an international phenomenon in order to provide businesses with a competitive edge in a global market. Several publications report on various aspects of international outsourcing, e.g. expansion and development path (Mol et al., 2004), partnership model (Kedia and Lahiri, 2007), drivers of offshore business processes (Kshetri, 2007), skills-intensive tasks and wage inequality (Anwar et al., 2013).

Although international outsourcing related at some stage mainly to the manufacturing sector, such as changing the production location to obtain a capital-labour trade-off, it progressed to service-based and knowledgebased outsourcing, such as advanced information technology design, legal services, medical diagnostics, etc. (Parkhe, 2007).

Recent publications indicate a tendency to return manufacturing to home countries due to rising labour costs in originally preferred countries, such as China, India, and Mexico (Eichler, 2012; Pearce, 2014; Kazmer, 2014).

Outsourcing is prevalent in the retail and manufacturing industries (Bryce and Useem, 1998; Kazmer, 2014), whereas mining is one of the industries with the lowest propensity for outsourcing (Embleton and Wright, 1998). The supposition can be made that this is due to the fact that, historically, mining has been quite a protected industry, unlike manufacturing and retail where the fierce competitive environment has forced companies to be innovative with regard to their business models. In a commodity-based business such as mining, outsourcing has become a potential solution to overcome two main challenges, namely cost

Outsourcing in the mining industry: decision-making framework and critical success factors and the acquisition and retention off skilled people (Deloitte Management Consulting, 2012). In apparent contradiction with conventional outsourcing theory, which dictates that companies should focus outsourcing efforts on non-core competencies, many mining companies have considered outsourcing their mining operations i.e. drilling, blasting, loading and hauling – the very core of their business (Quelin and Duhamel, 2003).

Table I shows a summary of the literature on outsourcing in general (Jiang and Qureshi, 2005). The literature is dominated by research focusing on the outsourcing process with the outsourcing decision, the focus of this research, and outsourcing results lagging behind. Popular research methodologies include surveys and conceptual framework, which will form the basis of the research methodology of this study as well.

While the literature is rife with long lists of factors to consider, little to no information exists on how to apply these in practice and the relative importance of the different factors to be considered (Quelin and Duhamel, 2003). The opportunity therefore exists to develop a framework for deciding between outsourcing and insourcing of core mining operations.

A study was conducted to determine whether mining is truly a core competency for a specific mid-tier geographic commodity specialist mining company and evaluate this against the perceptions among management. A decisionmaking framework for mining operations sourcing was developed. The study further set out to determine the critical success factors (CSFs) that should be adhered to if outsourced mining is the decision. The objectives of this study are to: ➤ Determine whether mining is truly a core competency for a mid-tier geographical commodity specialist and evaluate this against the perceptions among management in such a company ➤ Develop a decision-making framework for mining operations sourcing for future mining projects, which includes a prioritized list of factors to consider ➤ Determine the CSFs that should be adhered to if outsourced mining is the chosen option.

Strategic outsourcing decision factors

(2000) proposes a couple off ffactors that should fform the framework within which a company should make outsourcing decisions, e.g. cost analysis, associated risks, supplier influences, and strategic intent. Recent research by Freytag et al. (2012) defines three major categories that should be used when evaluating outsourcing decisions, namely (i) cost-based, (ii) competence-based, and (iii) relationship-based. Quelin and Duhamel (2003) do not place as much emphasis on relationship-based factors, and split cost considerations into two categories, namely (i) operational costs and (ii) effective use of capital. They also add flexibility-based factors as a separate category. The approaches of Freytag et al. (2012) and Quelin and Duhamel (2003) were combined for the purposes of this study, as shown schematically in Figure 1.

Operational cost-based factors Transactional cost theory and the drive for efficiency has long been the dominant reason for outsourcing (Holcomb and Hitt, 2007). Any organization, but particularly commodity organizations like mining companies, strives to minimize production and transaction costs. Companies sometimes outsource a function to convert a fixed-cost operation into a variable-cost operation (Freytag et al., 2012), thereby minimizing the risk of a negative profit margin under low production volumes. Often the decision on project issues is not made on the best fit, but more on the sensitivity of the project business case. In order to consider cost comparisons, one must take care to evaluate outsourcing on par with internal capabilities and their associated costs (Embleton and Wright, 1998). The one area of concern, according to Kirk (2010), is the redundancy in overheads – as both the owner and contractor will require some management and administrative functions, even under a fully outsourced mining model.

Capital efficiency-based factors In addition to minimizing low volume risk, companies prefer to outsource functions that depend heavily on fixed investment in order to avoid spending large amounts of capital (Quelin and Duhamel, 2003). In mining operations, for example, conducting the function in-house means that the

The context within which outsourcing decisions are made is extremely important, and making the decision on cost alone is dangerous and short-sighted (Fill and Visser 2000). McIvor

Table I

Number and focus of publications on outsourcing (Jiang and Qureshi, 2005) Research type Case study Survey Conceptual framework Mathematical modelling Financial data analysis Total [%]

▲

846

Decision framework

Outsourcing process

Outsourcing results

34 31 24 7 1 97 36.9%

54 28 19 13 0 114 43.3%

15 14 8 11 4 52 19.8%

OCTOBER 2014

VOLUME 114

Figure 1—Outsourcing decision factors to consider The Journal of The Southern African Institute of Mining and Metallurgy

Outsourcing in the mining industry: decision-making framework and critical success factors company needs to invest capital at start-up in order to acquire a mining fleet (trucks, excavators, dozers, etc.) and then periodically replace these assets as they age, which requires additional capital. A mining contractor will model this, and build the capital requirement into their variable rate; thereby smoothing cash flow and converting capital spent into a variable operational expenditure. The work by Kirk (2010) substantiates this further, saying that for mining companies, the main consideration from a corporate perspective is the availability, accessibility, and cost of capital. The key difference between owner and contractor mining rests in the former being heavily capital-intensive initially, but potentially lower cost in terms of operational expenditure over the life-of-mine.

entity, and this is considered under the relationship-based grouping of factors. Outsourcing, if applied correctly, can create bonds and network an organization in such a way as to increase productivity (Freytag et al., 2012). The strategic relationship between client and vendor becomes quite important. Holcomb and Hitt (2007) indicate that goal congruence, or the degree of overlap between the two parties’ strategic and operational objectives, must be considered. In order for a win-win relationship to exist and be sustained, objectives must be aligned (or more specifically, alignable).

Managing capabilities, even if not under the direct operational control of the managing party, is in itself also a capability (Loasby, 1998). A company considering outsourcing a function should evaluate its own ability (as a core competency) to manage such an arrangement.

Flexibility-based factors Mining is an industry with a number of variable influences, from geology and labour conditions to the seasons and commodity prices. Flexibility-based factors historically come second only to cost when outsourced mining is motivated, with various aspects to be analysed. Each mining project is unique, and presents its own unique complexities and challenges. Kirk (2010) suggests that projects with a relatively short life-of-mine (five years or less) and with widely varying mining rates will be suitable candidates for outsourced mining. Holcomb and Hitt (2007) indicated that flexibility-based factors become a higher priority in organizations operating in a market where technology is the basis of competitive advantage, and with significant technological uncertainty. In the mining industry, the tendency is to move from labour-dependent, largely manual technologies to automated methods, resulting in mining contractors partnering with equipment suppliers to enable them to access new technologies.

Competence-based factors From a competence-based perspective, various factors need to be considered. A company should look for opportunities to (i) protect and develop its core competencies internally, even at a slightly higher transactional cost, and (ii) look to balance and sharpen its competitive edge by outsourcing non-core competencies to best-in-class service providers (Freytag et al., 2012). By accessing more efficient and potentially more value-creating capabilities a firm can fundamentally change its competitiveness in the marketplace. Holcomb and Hitt (2007) support the contention that firms should form alignments with outsourcing partners in order to gain access to complementary capabilities or competencies.

Complementary competencies are defined as those that not only supplement a company’s internal core competencies, but have the ability to enhance them as well. During times of industry-wide shortages of skills, outsourcing can further alleviate the need to invest in expensive individuals, (Kirk, 2010). Mining is one of the industries where the attraction and retention of talent is one of the main operational challenges.

Critical success factors for contractor mining The benefits of contractor mining are not always achieved, even when the model suits the project perfectly. Dunn (1998) provides examples where, after substantial periods under the contractor mining model, companies have concluded that owner mining is significantly cheaper, has marginally lower risk, and in general is better value for money. In a survey conducted by Deloitte Consulting (2012) 48% of respondents have at some stage terminated an outsourcing contract, and in a third of these cases ended up insourcing the function. It is thus believed that the benefits of contractor mining are not a given, but are heavily dependent on choosing the right contractor, setting the appropriate incentives through contracting, and implementing the business solution correctly.

According to Kang et al. (2012), the appropriate control mechanisms should be in place in order to ensure the realization of outsourcing benefits. They group these into three categories, namely (i) process control mechanisms, (ii) output control mechanisms, and (iii) social control mechanisms. However, the foundation of a successful outsourcing arrangement starts with the development of a comprehensive and intelligent contract (Webb and Laborde, 2005). In this study, contract-based factors will be added as an amendment to the Kang et al. (2012) mechanisms (Figure 2).

Relationship-based factors

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 2—Critical success factor categories for contractor mining VOLUME 114

OCTOBER 2014

847

▲

Outsourcing decisions are not made on cognitive reasons alone (Webb and Laborde, 2005). No company is truly a sole

Outsourcing in the mining industry: decision-making framework and critical success factors Contract-based mechanisms According to Kirk (2010), contracts between owners and mining contractors have become more detailed. This is a result of both parties acquiring additional experience, and in a number of occasions learning some expensive lessons in how to manage a win-win contractor mining relationship.

CSFs under this category (Webb and Laborde, 2005) include: ➤ A fair and mutually beneficial contract ➤ Adequate incentive schemes, both gains and penalties – and how these will be shared between owner and service provider ➤ Flexibility in the contract to allow for changes to scope and conditions.

service provider should ffit into the culture off the outsourcer (Webb and Laborde, 2005) ➤ Vendor relationship management is often stated to be the single most important factor to any outsourcing arrangement (Parsa and Lankford, 1999). The longevity and success of an outsourcing model is dependent on the success or failure of the client/vendor relationship (Webb and Laborde, 2005) ➤ Strong communication channels, both formal and informal. Clients appreciate when service providers communicate with them in a proactive manner, and vice versa (Webb and Laborde, 2005) ➤ Senior management support and continuous involvement.

There is, however, still widespread disagreement on the best practice in structuring these contracts, with some owners still employing the traditional schedule-of-rates contract and others moving towards a more cost-plus-profit arrangement, and in some extreme cases even offering equity in the mining owner company. In addition to this, various incentive and penalty schemes have also been employed, with varying success.

Quelin and Duhamel (2003) state that frequently cited outsourcing benefits cannot be divorced from the efforts and investments required to continuously monitor and control the service, and as such all four control mechanisms discussed above will be considered as part of this research.

Process control mechanisms

The first proposition evaluates mining as a core competency using the criteria of Quinn and Hilmer (1994). They use the following dimensions to evaluate whether a function is core:

Process control mechanisms focus on the vendor’s method, i.e. the process through which the service provider delivers (Kang et al., 2012). CSFs under this category include: ➤ Standard operating procedures ➤ Formalization of roles and responsibilities on all positions ➤ Training on the outsourcer’s processes, procedures, methodologies, and policies ➤ Extensive reporting from service provider to owner on standards and performance ➤ Support (formalized within the employee performance contract) of internal functional employees – without which no outsourcing arrangement can thrive (Webb and Laborde, 2005). Not having this in place often leads to employees of the outsourcer feeling threatened.

Research propositions and hypotheses Mining tested as core competency

(i)

(ii)

(iii)

(iv)

Output control mechanisms Output control mechanisms are focused on the goals and objectives of the outsourced process (Kang et al., 2012). CSFs under this category include: ➤ The establishment of goals and objectives. It is important to focus on measurable objectives to ensure clarity of expectations (Webb and Laborde, 2005) ➤ Recovery plans when outcomes are at risk ➤ Regular reviews of contractor performance – including a detailed reporting schedule – on all required levels ➤ A strong group of contract management specialists to manage deviations from the contract (Embleton and Wright, 1998).

Social control mechanisms Social control is most prevalent in outsourcing arrangements driven by the need to innovate (Kang et al., 2012). CSFs under this category include: ➤ Shared values and beliefs, driven by a mutual respect and cultural fit between the contractor and owner. The

▲

848

OCTOBER 2014

VOLUME 114

(v)

(vi)

(vii)

Skill or knowledge sets, not products or functions: Based more on intellectual property and knowledge than on physical assets Flexible long-term platforms capable of adaptation or evolution: This competency evolves over time, and creates flexibility, rather than inhibiting it Limited in number: Most successful organizations target only two or three core competencies. For mining companies it has been suggested that only two core competencies are required for success – financing and management (Hamel and Prahalad, 1996) Unique sources of leverage in the value chain: Core competencies often fill gaps in the industry where severe knowledge deficiencies exist, for which the company has been specifically positioned through investment Areas where the company can dominate: Typically the areas where an organization can significantly outperform its peers. According to Stacey et al.

(1999), core competencies can be identified by asking the following questions: What does the organization do better than anyone else? What does the organization do so well that it will be able to sell this as a service to other companies? Where does a company achieve best-in-class status?

Elements important to customers in the long run: Companies should ask themselves if their customers and shareholders care that they perform this function Embedded in the organization’s systems: Core competencies are not determined by a couple of particularly talented employees, but rather by systems and practices that are standard and that surpass the employment history of these talented employees.

The Journal of The Southern African Institute of Mining and Metallurgy

Outsourcing in the mining industry: decision-making framework and critical success factors H1: The tested reality will show mining operations as a non-core competency for the company under investigation H2: There will be a difference in perception and tested reality – miners will believe that mining is a core competency.

Choosing between owner and contractor mining The second research proposition relates to the development of a decision-making framework, using the outcomes of an order-winner/order-qualifier analysis. This framework shows under which conditions owner mining has the upper hand in the mining sourcing decision, and under which scenarios it will not. Generic decision frameworks can be misleading, as the reason behind an outsourcing decision will very much depend on what an organization is trying to achieve. The following hypotheses are made with regard to the mining sourcing decision: H3: Operational cost-based factors will emerge as order qualifiers rather than order winners for a specific model (insourced or outsourced mining) H4: Capital efficiency-based factors will be an order winner for contractor mining H5: Flexibility-based factors will be an order winner for contractor mining H6: Competence-based factors will emerge as order qualifiers rather than order winners for a specific model (insourced or outsourced mining) H7: Relationship-based factors will be an order winner for owner mining.

Critical success factors for contractor mining The third research proposition is to determine and prioritize the four CSFs in the event that outsourced mining is the chosen alternative. The top-priority CSFs should be evaluated against a mining company’s internal capabilities to ensure that if contractor mining is chosen as the preferred scenario, the internal workings of the firm support the success of this strategy.

H8: Contract-based mechanisms will be high-priority critical success factors H9: Social mechanisms will emerge as high-priority critical success factors H100: Output mechanisms will be prioritized relative to process control mechanisms.

Research methodology All research techniques have inherent inadequacies, and as such are best applied in conjunction with other techniques in order to counter these shortcomings, also known as triangulation. The methods chosen for this research includes findings from literature, the sample survey, and judgement task or nominal group (Barry et al., 2009) methods.

The objectives for the judgement task as part of this study were: ➤ Discuss and group the various factors (potential order winners and order qualifiers) that should be considered by a mine owner when choosing between owner and contractor mining ➤ Discuss and group the various CSFs that should be in place in order to ensure that contractor mining achieves its theoretical benefits.

The findings from the literature, as well as the outcomes of the judgement task conducted, were consolidated in a survey, structured as a combination of a questionnaire and a structured interview. The questionnaire consisted of four sections, namely one for the respondent’s details and one on each of the three research propositions. This research instrument was then used in gathering data from 80% of the relevant heads of departments and general management of a mid-tier geographic commodity specialist mining company.

The sample was stratified on three dimensions of the organization to ensure a complete data-set is gathered: ➤ Mining type—managers e from both underground and open cast operations ➤ Departments—respondents s from various departments, including mining, beneficiation, finance, and engineering ➤ Seniority—heads y of departments (middle management) and general management.

Results Respondents included managers from both underground (28%) and opencast (72%) operations. This is also representative of the ratio of the number of operations of this mid-tier mining group. A good balance between the different functional backgrounds was accomplished and the percentage feedback from the different departments consists of mining (22%), finance (22%), plant (17%), engineering (6%), and general management (33%). The ratio of senior management

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

OCTOBER 2014

849

▲

Figure 3—Mining tested as a core competency using the Quinn and Hilmer model

Outsourcing in the mining industry: decision-making framework and critical success factors to heads off departments was 11:14. It is believed that the stratified sample adheres to the requirements of the research methodology and can thus be considered as a fair representation of the population of the whole mid-tier mining company. However, care should be taken not to assume the validity of these results for all mining companies. The results of the investigation are discussed along the lines of the three research propositions.

Mining tested as core competency Figure 3 shows a consolidated summary of the findings of research proposition 1; testing mining as a core competency.

The majority (72%) of respondents indicated that they believe that owner-operated mining is a core competency for the mid-tier geographic commodity specialist mining company. This number was then used as the cut-off point for the testing of this fact through the Quinn and Hilmer model.

It can be seen that on all seven dimensions of the Quinn and Hilmer model, the response frequency is biased towards owner-operated mining as a non-core competency (as measured against the 72% cut-off). The average of the responses on the 7 dimensions is 35% in favour of core competency, i.e. 65% in support of the fact that mining is a non-core competency – a clear conflict with the management teams’ articulated perceptions.

By way of example, Quinn and Hilmer state that a company’s core competency must be important to the customers in the long run; in other words, companies should ask themselves if their customers and shareholders care that they perform this function. Only 11% of respondents indicated that this is the case for owner-operated mining (see dimension titled ‘Customer Impact (i) Yes or (ii) No’ in Figure 3), and as such this question presents strong evidence that mining is a non-core competency.

H1 and H2 are discussed in the light of the research findings on proposition 1, summarized in Figure 3.

H1: The tested reality will show H1 is accepted mining operations as a non-core competency for a mid-tier mining company.

Not a single dimension from the Quinn and Hilmer model tested above the 72% cut-off level. In fact, five of the seven dimensions did not even receive a 50% response rate in favour of mining as a core competency. This supports what was found in the literature, in particular the work of Hamel

and Prahalad (1996) that signifies f that often f business functions (such as management, financing, and resource acquisition) rather than technical functions (such as mining, geology, surveying, etc.) are core competencies of mining corporates.

The literature showed that not all non-core competencies should be outsourced, but rather evaluated for outsourcing (Quinn and Hilmer, 1994; Leavy, 2004; Stacey, 1999). This, together with the fact that Hypothesis 1 is true, renders research proposition 2 relevant, i.e. mining is proved to be a non-core competency, and it should be evaluated as a candidate for outsourcing, necessitating the development of a make-or-buy decision framework.

H2: There will be a difference in H2 is accepted perception and tested reality – miners will believe that mining is a core competency.

There is a clear conflict between the respondents’ articulated opinion of owner-operated mining as a core competency (72% of respondents) and the evidence tested against the Quinn and Hilmer model (35% over the seven dimensions). This is to be expected, as miners will believe that the owner’s ability to operate a mining operation should be the core competency of a mining company. However, the prevalence of so many mining companies already employing a model of full or partial outsourced mining clearly proves that this is not always the case. The research conducted on proposition 1 in this study supports this fact conclusively.

Choosing between owner and contractor mining Figure 4 shows a consolidated summary of the finding of research proposition 2; choosing between owner and contractor mining. The data has been ranked from factors most strongly supporting contractor mining (order winners for contractor mining) to factors most strongly supporting owner mining (order winners for owner mining). The proportion of respondents that listed the factor in question as indecisive towards a particular model (titled ‘It Depends’ on the questionnaire) is also indicated under the category ‘Unsure (Order Qualifier)’. Figure 5 shows the same data as in Figure 4, but grouped according to decision factor category (Figure 1), as developed by combining the work of Freytag et al. (2012) and Quelin and Duhamel (2003).

A decision-making framework was developed using the order winner / order qualifier structure. Table II summarizes

Figure 4—Proportion of respondents classifying the decision factors as order winners for contractor and owner mining

▲

850

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Outsourcing in the mining industry: decision-making framework and critical success factors

Figure 5—Decision factors grouped according to category

Table II

Mining insourcing versus outsourcing decision model Decision making factors

Owner mining winner

Qualifier

Operational cost

• Composite unit cost of production • Lowest fixed cost / variable cost ratio

X

Capital efficiency

• Managing potential variability in cash flow • Enabling efficient allocation of capital investment

X X

• Adapting with variability in mining rate • Adapting to rapid change in mining technology

X

Flexibility Competence Relationships

X

X

• Industry-wide shortage of mining skills • Value chain integration

X

• Industrial relations and union / community • Alignment to changes in larger corporate strategy

X X

the ffindings on research proposition 2 into a decision model.

Care should be taken in the extrapolation of these findings to a different time and/or context, as order qualifiers and order winners are highly dependent on market context, and will change over time.

Hypotheses 3 to 7 are now discussed in the light of the research findings on proposition 2, as summarized in Figures 4 and 5 and Table II. H3: Operational cost-based factors will emerge as order qualifiers rather than order winners for a specific model (insourced or outsourced mining).

H3 is rejected

Elements of the hypothesis were proved true, but there was not sufficient evidence from the data to determine if operational cost-based factors will in all instances be an order qualifier. With regard to the ‘Unit Cost’ t factor, the hypothesis is supported to be true, i.e. the response rates are not biased towards a particular mining sourcing model so as to indicate the factor as an order winner for that model. However, with regard to the ‘Optimal Fixed/Variable Cost Ratio’, contractor mining was shown to have this factor as an order winner.

The finding on ‘Optimal Fixed/Variable Cost Ratio’ as an order winner for contractor mining can be explained as follows. For small to medium-sized mines the ‘Optimal Fixed/Variable Cost Ratio’ leans towards more variable costs, as it is difficult to dilute fixed cost on a low-volume operation. Clearly, contractor mining has a higher proportion The Journal of The Southern African Institute of Mining and Metallurgy

Contractor mining winner

X

off variable cost, and this benefits f small/medium sized operations in this regard. However, large to mega-sized mining operations can effectively dilute fixed cost, which will lead to higher profit margins. Under these circumstances, the ‘Optimal Fixed/Variable Cost Ratio’ leans to having more fixed costs, with owner mining thus a better option.

All the respondents came from mines that can be classified as small/medium sized operations (producing 4 Mt or less per annum, with a LOM less than 5 years), which would be the reason why these respondents listed ‘Optimal Fixed/Variable Cost Ratio’ as an order winner for contractor mining. Keeping this in mind, ‘Unit Cost’ t is included as an order qualifier in the decision-making model. ‘Optimal Fixed/Variable Cost Ratio’ is included as an order winner for contractor mining, specifically in the context within which this study was conducted.

H4: Capital efficiency-based factors H4 iis rejected will be an order winner for contractor mining.

The data shows a very balanced result, with a large proportion of respondents listing these factors as order qualifiers, and the rest split between order winners in favour of the two mining sourcing models. This finding can be substantiated as follows: contractor mining requires the mine owner to spend less capital, because the contractor will build the capital requirements into its rates, thereby effectively converting capital expenditure into operational expenditure.

Over the life of the mining project, this will not necessarily VOLUME 114

OCTOBER 2014

851

▲

Category from the lliterature

Outsourcing in the mining industry: decision-making framework and critical success factors have any impact on the net present value (NPV) off the project, unless there is a substantial difference in the costs structure of the contractor, compared to that of the mine owner.

Furthermore, the decrease in capital expenditure does not guarantee that an alternative investment exists with a better business case, which would lead to an increase in capital efficiency. When taking a portfolio view, spending capital on an owned fleet might be a better investment than spending it on a different project in the mine owner’s portfolio.

It is therefore deduced that capital efficiency-based factors are order qualifiers in the mining insourcing versus outsourcing decision. This finding contradicts the work by Quelin and Duhamel (2003) and Kirk (2010), both of whom advocate the decrease in capital expenditure as a strong argument, i.e. order winner, for outsourcing.

H5: Flexibility-based factors will be H5 is rejected an order winner for contractor mining.

H5 could not conclusively be proved, i.e. there was not sufficient evidence from the data to determine if flexibilitybased cost factors will in all instances be an order winner in favour of contractor mining. With regard to the ‘Variable Mining Rate’ factor, the hypothesis is supported to be true, i.e. the results show that contractor mining acquired a high frequency of positive responses – enough to classify the factor as an order winner for that model. This was to be expected, and is supported by Kirk (2010), who listed variability in mining rate required as a key argument in favour of contractor mining. However, with regard to ‘Mining Technology Change’, contractor mining has a slight advantage over owner mining according to the research respondents, but not by a large enough margin to prove that model has a clear advantage in absolute terms.

Changing technology will in the future become progressively more important in the mining industry, which according to Holcomb and Hitt (2007) will increase the importance of this decision factor. Currently, however, technology in the mining industry is fairly standard, due to the industry being highly averse to change. Under these conditions, it is to be expected that such a factor will be an order qualifier (Hill, 2000), as was found in this study. It is concluded that ‘Variable Mining Rate’ is an order winner for contractor mining, and that ‘Mining Technology Change’ is an order qualifier.

owners can take to overcome the industry-wide shortage off skills. It is believed that the reason for this lies in the difference in context within which the work was done – Kirk’s study was conducted on the contractor mining industry in Australia, which has a much more mature professional services sector than South Africa.

Owner mining did, however, emerge as having an advantage in terms of ‘Value Chain Integration’. The respondents believed that the integration of a mining contractor into the larger resources extraction value chain will not happen as smoothly as in the case of owner mining.

Reasons cited for this include differences in incentives between mine owner and contractor, as well as sub-optimal communication channels. These will again be discussed as CSFs (under research proposition 3). It is concluded that ‘Value Chain Integration’ is an order winner for owner mining, and that ‘Shortage of Skills’ is an order qualifier. eH7 is accepted H7: Relationship-based factors will beH an order winner for owner mining.

A large majority of respondents believe that an owneroperated mining operation is better equipped to manage its relationship with the community in which it operates, as well as its relationship with the corporate to which it reports. From this finding, as well as the focus group that was conducted as verification for the research instrument, it is concluded that the importance of the relationship-based factors cannot be overstated, especially in the South African context. It is, however, a topic that is often overlooked, as was done by Quelin and Duhamel (2003), who focused mostly on operational cost- and capital efficiency-based factors. It is concluded that ‘Union/ Community Considerations’, as well as ‘‘Alignment with Corporate Objectives/Strategy’ are order winners for owner mining.

Critical success factors for contractor mining Figure 6 shows the consolidated findings on research proposition 3 – CSFs for contractor mining. The different mechanisms behind the CSFs (see Figure 2) are indicated using a colour scheme.

Figure 6 has been divided into four different regions:

H6: Competence-based factors will H6 is rejected emerge as order qualifiers rather than order winners for a specific model (insourced or outsourced mining).

H6 was proved partly true. The mining industry faces extreme difficulties with regard to attracting and retaining talent. Mine owners and contractors experience this difficulty more or less to the same degree; they compete in the same labour market and offer similar employee value propositions.

It is thus no surprise that the respondents were divided in their opinion on the ‘Shortage of Skills’ factor, with 33% listing this factor as an outright order qualifier. This finding is in contradiction with the work by Kirk (2010), who advocates contractor mining as a mitigation action that mine

▲

852

OCTOBER 2014

VOLUME 114

Figure 6—Critical success factors for contractor mining grouped by mechanism The Journal of The Southern African Institute of Mining and Metallurgy

Outsourcing in the mining industry: decision-making framework and critical success factors

It was found that the most important tools at the disposal of a mine owner’s team to manage a contractor miner are the social and output control mechanisms. Firstly, a mine owner must understand the objectives and strategic intent behind outsourcing the mining operation. If the objectives of such an operational strategy are not understood, it is likely that they will not be achieved. It is believed that this is currently not the case in the mid-tier geographic commodity specialist mining company, as indicated by the conflict found regarding research proposition 1. Secondly, communication between owner and contractor, as well as ongoing senior management involvement, should be a priority. The mine owner should focus on the output of the contractor, and leave the process by which that output is produced to the contractor to manage.

H8: Contract-based mechanisms will H8 is rejected be high-priority critical success factors.

It was found that two contract-based mechanisms plotted in the ‘Priority 3’ category, and the remaining one in ‘Priority 2’. This is an extremely interesting and counter-intuitive finding. From follow-up discussions with some of the respondents, the following justification is put forward in support of this finding: The respondents believe that contracts with contractor miners have become standardized to such an extent that it has become very difficult to negotiate any contract that falls outside this standard. For this reason, the level of effort required is deemed to be so high relative to the other CSFs that this mechanism is deprioritized. This does not mean that the contract between mine owner and mining contractor is not important, in fact the ‘Fair and Mutually Beneficial Contract’ t critical success factor ranked second on the benefits ranking scale. It does, however, show that the industry has cemented the legal aspects of an outsourced arrangement to such an extent that it has become unfeasible to negotiate customized incentives, penalties, and flexibility into contracts.

H9: Social mechanisms will emerge H9 is accepted as high-priority critical success factors.

Two of the CSFs under this mechanism plotted under ‘Priority 1’. One can understand that mining contractors perform better under a spirit of partnership between contractor and client. Communication between owner and contractor on all levels (including senior management) is paramount to a successful relationship. The one CSF that stands out here is ‘Shared Values and Beliefs’, which is plotted as ‘Priority 3’. From follow-up discussions with The Journal of The Southern African Institute of Mining and Metallurgy

research respondents, the ffollowing justification f is presented for this anomaly.

The ‘Shared Values and Beliefs’ CSF is believed to be the single most difficult CSF to put in context, both because a cultural match is difficult to assess during the tender process, and because values and beliefs are virtually impossible to manipulate once the contractor has been chosen. This CSF is therefore deprioritized; not because it is not important, but rather because it is extremely difficult to influence.

H100: Output mechanisms will be H10 is accepted prioritized relative to process control mechanisms.

Output-based mechanisms had two CSFs under ‘Priority 1’ and the third under ‘Priority 2’, while the CSFs pertaining to process control mechanisms are all plotted under ‘Priority 2’. One of the benefits most frequently cited for outsourcing (Harland et al., 2005; Kedia and Lahiri, 2007) is that it enables organizations to focus on core competencies, i.e. to sharpen their strategic focus. Mine owners therefore does not wish to micro-manage the contractor, i.e. control how they conduct their business, as long as they deliver results to the level that was agreed.

It was expected that ‘Clear Goals and Objectives’ would feature high on the priority list. That is exactly what happened, with almost 80% of respondents listing this CSF under their top four with regard to potential benefits. This echoes with findings from the literature, with Embleton and Wright (1998), Gottschalk and Solli-Saether (2005), and McIvor (2000) all emphazing that the outsourcing arrangement is likely to fail if the party outsourcing the function does not understand what it aims to achieve through such an action.

Conclusions and recommendations A decision-making framework was developed to facilitate the decision between owner and contractor mining. This contributes in a number of ways.

Firstly, most of the literature deals with outsourcing in the manufacturing, services, or retail environment – this study provides some theory and application relevant to the mining industry.

Secondly, most of the literature lists factors to be considered in the decision-making process without detailing how these factors should be assessed or prioritized. By applying the order winner/order qualifier framework, a model is developed that can be used to structure the make-or-buy decision specifically for mining.

Finally, although numerous studies list critical success factor (CSFs) for outsourcing, none could be found that prioritize these to assist management in the development of a focused implementation roadmap. The CSFs for the outsourced mining application were prioritized using a benefit/effort matrix that highlighted the critical elements that a mine owner management team should focus on to mitigate the risks of outsourcing a mining operation, thereby increasing the probability of success.

It is recommended that the decision-making model developed under research proposition 2 (the mining sourcing decision) be used to facilitate the make-or-buy decision process for mining operations. It is believed that this will VOLUME 114

OCTOBER 2014

853

▲

➤ Priority 1—CSFs with high benefits f requiring little effort to implement. These should be the focus of a management team entering a contractor mining arrangement ➤ Priority 2 (two regions)—CSFs with high benefits but requiring a high-level effort to implement, as well as CSFs with low benefits but requiring a low-level effort to implement. These should be a secondary focus of a management team entering a contractor mining arrangement ➤ Priority 3—CSFs with low benefits requiring a high level of effort to implement. These can be deprioritized for implementation purposes.

Outsourcing in the mining industry: decision-making framework and critical success factors result in a more structured approach and a holistic view compared to making the decision on a cost analysis only.

This will result in a higher probability of making the correct decision, especially on greenfield mining projects.

The opportunity exists to expand the study to a wider population. This could include other mining companies, commodities, geographical contexts, and scales of operation.

Also, since this study focused on the mining operation only. and excluded other core and peripheral activities in the commodity value chain such as beneficiation and logistics, the opportunity exists to conduct a similar study to develop make-or-buy decision models for these activities.

HÄTÖNEN, J. and ERICKSSON, T. 2009. 30+ years off research and practice on outsourcing – exploring the past and anticipating the future. Journal of International Management, t vol. 15. pp. 142–155.

HOLCOMB, T.R. and HITT, M.A. 2007. Toward a model of strategic outsourcing.

Journal of Operations Management, t vol. 25. pp. 464–481.

JIANG, B. and QURESHI, A. 2005. Research on outsourcing results: current literature and future opportunities. Management Decision, vol. 44, no.1. pp. 44–55.

KANG, M., WU, X., HONG, P., and PARK, Y. 2012. Aligning organizational control practices with competitive outsourcing performance. Journal of Business Research, vol. 65, no. 8, pp. 1195–1201.

KAZMER, D.O. 2014. Manufacturing outsourcing, onshoring, and global equilibrium. Business Horizons (in press).

Acknowledgement The authors would like to thank the Graduate School of Technology Management, University of Pretoria, for the opportunity to publish the results.

KEDIA, B.L. and LAHIRI, S. 2007. International outsourcing of services: a partnership model. Journal of International Management, t vol. 13. pp. 22–37.

KIRK, L.J. 2010. Owner versus contract mining. Mine Planning and Equipment Selection, vol. 1. pp. 437–442.

References ANWAR, S., SUN, S., and VALADKHANI, A. 2013. International outsourcing of skill intensive tasks and wage inequality. Economic Modelling, g vol. 31. pp. 590–597.

BARRY, M.L., STEYN, H., and BRENT, A. 2009. The use of the focus group technique in management research: the example of renewable energy technology selection in Africa. Journal of Contemporary Management, t

KSHETRI, N. 2007. Institutional factors affecting offshore business process and information technology outsourcing. Journal of International Management, t vol. 13, no. 1. pp. 38–56.

LOASBY, B.J. 1998. The organisation of capabilities. Journal of Economic Behaviour and Organisations, vol. 35, no. 2. pp. 139–160.

MCIVOR, R. 2000. A practical framework for understanding the outsourcing process. Supply Chain Management, t vol. 5, no. 1. pp. 22–36.

vol. 6. pp. 229–240.

BRYCE, D. and USEEM, M. 1998. The impact of corporate outsourcing on company value. European Management Journal, vol. 16. pp. 635–643.

CORBET, M.F. 2004. Outsourcing Revolution: Why it makes Sense and How to do it Right. Dearborn Trade Publishing, Chicago.

MOL, M.J., PAUWELS, P., MATTHYSSENS, P., and QUINTENS, L. 2004. A technological contingency perspective on the depth and scope of international outsourcing. Journal of International Management, t vol. 10, no. 2. pp. 287–305.

MORGAN, J. 1999. Purchasing at 100: where it’s been, where it’s headed?

DELOITTE CONSULTING. 2012. Global Outsourcing and Insourcing Survey Executive Summary, 2012. http://www.deloitte.com/assets/DcomUnitedStates/Local%20Assets/Documents/IMOs/Shared%20Services/us_s dt_2012Global OutsourcingandInsourcingSurveyExecutiveSummary_

Purchasing, g vol. 127, no. 8. pp. 72–94.

PARKHE, A. 2007. International outsourcing of services: Introduction to the special issue. Journal of International Management, t vol. 13. pp. 3–6.

PARSA, F. and LANKFORD, M.W. 1999. Outsourcing: a primer. Management

050112.pdf [Accessed 20 March 2013].

DUNN, S. 1998. Evaluating the use of contractors as a cost-cutting measure.

Decision, vol. 37, no. 4. pp. 310–316.

PEARCE, J.A. 2014. Why domestic outsourcing is leading America’s re-

PriceWaterhouseCoopers, Shine Meeting.

emergence in global manufacturing. Business Horizons, vol. 57.

EICHLER, A. 2012. April 20. Manufacturing companies considering moving jobs back to the U.S. from China, survey finds. Huffington Post. www.huffingtonpost.com/2012/04/20/manufacturing-companies-move-us-china_n_ 1441587.html [Accessed 5 June 2013].

EMBLETON, P.R. and WRIGHT, P.C. 1998. A practical guide to successful outsourcing. Empowerment in Organisations, vol. 6, no. 3. pp. 94–106.

FILL, C. and VISSER, E. 2000. The outsourcing dilemma: a composite approach to make or buy decision. Management Decision, vol. 38, no. 1. pp. 43–50.

FREYTAG, P.V., CLARKE, A.H. and EVALD, M.R. 2012. Reconsidering outsourcing solutions. European Management Journal, vol. 30. pp. 99–110.

GOTTSCHALK, P. and SOLLI-SAETHER, H. 2005. Critical success factors from IT outsourcing theories: an empirical study. Industrial Management and Data Systems, vol. 105, no. 6. pp. 685–702.

pp. 27–36.

PORTER, M.E. 1996. What is strategy? Harvard Business Review, vol. 74, no. 6. pp. 61–79.

QUELIN, B. and DUHAMEL, F. 2003. Bringing together strategic outsourcing and corporate strategy: outsourcing motives and risks. European Management Journal, vol. 21, no. 5. pp. 647–661.

QUINN, J.B. and HILMER, F.G. 1994. Strategic outsourcing. Sloan Management Review, vol. 35, no. 4. pp. 43–55.

SOUTH AFRICAN CHAMBER OF MINES. 2012. Putting South Africa First: Mining’s contribution to South Africa.: http://chamberofmines.org.za/mediaroom/mining-publications [Accessed 12 September 2014].

STACEY, T.R., STEFFEN, O.K.H, and BARRETT, A.J. 1999. Outsourcing of professional services. Journal of the South African Institute of Mining and

HAMEL, G. and PRAHALAD, C.K. 1996. Competing for the Future. Harvard Business Review Press, Boston, MA.

Metallurgy, vol. 99, no. 4. pp. 181–184.

HILL, T. 2000. Manufacturing Strategy: Text and Cases. 3rd edn.

HARLAND, C., KNIGHT, L., LAMMING, R., AND WALKER, H. 2005. Outsourcing: assessing the risks and benefits for organisations, sectors and nations.

Irwin/McGraw-Hill, Burr Ridge, IL. WEBB, L. and LABORDE, J. 2005. Crafting a successful outsourcing vendor/

International Journal of Operations and Production Management, t vol. 25,

client relationship. Business Process Management, t vol. 11, no. 5.

no. 9. pp. 831–850.

pp. 437–443.

▲

854

OCTOBER 2014

VOLUME 114

◆

The Journal of The Southern African Institute of Mining and Metallurgy

Focal depths of South African earthquakes and mine events

by M.B.C. Brandt*

Focal depths of 15 tectonic earthquakes and 9 mine-related events were determined for South Africa using data recorded by the South African National Seismograph Network. These earthquakes and events were relocated by means of the Hypocenter program using direct P-waves ((P Pg) , critically refracted P-waves (P (Pn) , and first-arrival S-waves for the magnitude range 3.6 ) ML ) 4.4. Focal depths were first determined by means of the minimum root mean square (RMS) of the differences between the measured travel times and those predicted using the velocity model.

The depths for tectonic earthquakes had a 2 km ) D ) 10 km range and an average depth and standard deviation of 6.9 ± 2.3 km. Depths for minerelated events ranged over 0 km ) D ) 7 km, averaging 3 ± 2.3 km. Next, arrival times for the additional regional depth phases sP Pn , PmP, P sPmP, P and SmP were measured. Focal depths were re-determined for the relocated epicentres, with the minimum variance (i.e. spread) of the differences between the measured travel times and travel times predicted by means of the Wentzel, Kramer, Brillouin, and Jeffreys (WKBJ) method for synthetic waveform modelling. Depth ranges were 4 km ) D ) 7 km (average 5.9 ± 1.2 km) and 1 km ) D ) 4 km (average 2.4 ± 1.2 km) for tectonic and mine-related events, respectively. The derived depths were verified for one tectonic earthquake with synthetic-to-recorded-waveform fits using the WKBJ synthetic seismogram software for the abovementioned regional phases. The focal mechanism parameters for this earthquake source were obtained from the National Earthquake Information Centre. Focal depths were estimated for nine stations by visually comparing synthetic waveform phases with recorded waveforms, ranging from 5 km to 8 km.

Keywords focal depth, earthquake location, regional depth phases, waveform modelling

Introduction Almost all continental earthquakes are confined to a crustal layer that varies in thickness between 10 km and 40 km, measured from the surface. Hence, continental earthquakes do not occur in the mantle (Maggi et al., 2000). In many stable continental regions, focal depths occur as a bimodal distribution within the upper third (0 km to 10 km) and/or lower third of the crust (20 km to 35 km), while the middle crust (10 km to 20 km) tends to be aseismic (Klose and Seeber, 2007). The distribution can vary in terms of bimodal depth and strength, with some stable continental regions showing very well-developed bimodal distributions (e.g. the The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

* Council for Geoscience, Seismology Unit, Pretoria, South Africa. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. Paper received r Apr. 2014; revised paper received Jun. 2014.

OCTOBER 2014

855

▲

Synopsis

North Alpine fforeland basin in Europe), while others show only weak or no bimodal distribution (e.g. New Madrid seismic zone in the central USA). Klose and Seeber’s study (2007) found that many large earthquakes with magnitudes Mw from 4.5 to 8.0 have a focal depth of less than 5 km, with almost 80% of the seismic moment density of shallow stable continental region ruptures being released in the uppermost 7 km of the crust. It is noteworthy that accurate focal depths have important implications for seismic hazard estimations, specifically in respect of ground motion predictions.

In a previous case study in North America, Ma and Atkinson (2006) found only weak support for a bimodal distribution. For southern Ontario, Canada, and northern New York, USA, focal depths range from 2 km to 15 km. In parts of western Quebec and along the Ottawa River Valley in Ontario, focal depths range from 2 km to 25 km. These authors noted that more than half of the earthquakes, in a cluster of activity near Maniwaki in western Quebec, Canada, are deeper than 20 km. To the author’s knowledge, only one such study has been performed for tectonic earthquakes in South Africa: Mangangolo et al. (2014) applied synthetic-to-recorded-waveform fits to data recorded by the stations of the temporary Kaapvaal Craton array. The 50 broadband stations of this array were installed on a 1° × 1° grid at 80 sites in central southern Africa (c.f. Nguuri et al., 2001). The study found shallow hypocentres ranging between depths of 5.6 km and 18.6 km, with an error in depth of approximately 3 km.

Focal depth is the most difficult parameter to determine for regional earthquakes and mine-related events when recorded by means

Focal depths of South African earthquakes and mine events off a sparse national seismograph network, unless there are stations very close to the epicentre (e.g. Havskov and Ottemöller, 2010b). This difficulty is attributable to the large epicentral distances (approximately 100 km to 1 000 km) versus the small focal depths (approximately 2 km to 10 km), causing a near-zero change in the observed travel times of direct seismic waves for the different focal depths. The practice of the South African National Seismograph Network is to measure only the first-arrival seismic phases. When locating the events and earthquakes, focal depths are fixed as follows: explosions to 0 km, mine-related events to 2 km, and tectonic earthquakes to either 5 km or 10 km, depending on which of these depths gives the best fit to the travel time data (Saunders et al., 2008). However, by following this procedure it may be surmised that both mine-related events and tectonic earthquakes are shallow and that South African seismicity does not follow the bimodal depth distribution observed in other stable continental regions.

The purpose of this study was to evaluate South African hypocentre depths. Figure 1a summarizes the epicentres and recording stations of the data-set. Focal depths were estimated by stations (triangles with station codes) that recorded waves originating from 15 tectonic earthquakes and 9 mine-related events (stars) with a magnitude range 3.6 ) ML ) 4.4. Mine-related events were identified from their location within a gold mining area. The focal depth obtained by means of seismic phases that had travelled along ray paths (dashed lines) originating from the mine-related event in the Klerksdorp gold mining area, shown in Figures 2a and 2b, was determined by stations situated at Parys (PRYS), Kloof Mine (KLOOF), Koster (KSR), Schweizer-Reneke (SWZ), Silverton, Tshwane (SLR), Boshof (BOSA), Lobatse, Botswana (LBTB), Kokstad (KSD), Pongola (POGA), Upington (UPI), Mussina (MSNA), Somerset East (SOE), Calvinia (CVNA), and Ceres (CER). The focal depths obtained using waveform modelling for waves that had travelled along ray paths (dashed lines) originating from the tectonic earthquake near Augrabies, shown in Figures 7a and 7b, were determined using stations situated at Komaggas

(KOMG) and Calvinia (CVNA). Tectonic epicentres located near Leeu-Gamka, Pofadder, and Augrabies; mine-related events located in the Free State (FS), Klerksdorp (KLE), and Far West Rand (FWR) gold mining areas. Details of the analyses performed in Figures 2a, 2b, 7a, and 7b follow.

To estimate focal depths of earthquakes and mine-related events, this study applies arrival times for various waves.

These include: ➤ Direct P-waves (P Pg) ➤ Critically refracted P-waves (P (Pn) ➤ The ascending S-wave converted at the surface to a critically refracted P-wave (sP sPn) ➤ Reflected P-wave at the Moho discontinuity (PmP ( P) ➤ Ascending S-wave converted at the surface to a reflected P-wave at the Moho discontinuity (sPmP) P ➤ A descending S-wave converted to a P-wave when reflected at the Moho discontinuity (SmP).

P (Figure 1b, with velocity structure in Table I). The author used the SEISAN earthquake analysis software (Havskov and Ottemöller, 2010a) to first pick the arrival times of direct Pwaves, critically refracted P-waves, and first-arrival S-waves.

The events were re-located using the Hypocenter software (Lienert et al., 1986). Focal depths were determined by means of the root mean square (RMS) of the differences between the measured travel times and those predicted using the velocity model. Next, arrival times for additional regional P sPmP, P and SmP were measured. depth phases sP Pn , PmP, Focal depths were re-determined for the re-located epicentres, with the minimum variance (i.e. spread) of the differences between the measured travel times and travel times predicted by means of the synthetic seismogram software introduced

by Chapman (1978), which uses the Wentzel, Kramer, Brillouin, and Jeffreys (WKBJ) method. The focal mechanism parameters for one tectonic earthquake source had already been determined by the National Earthquake Information Centre. Hence, the derived depths for this earthquake were verified by means of synthetic-to-recorded-waveform fits using the WKBJ synthetic seismogram software for the abovementioned regional phases.

Figure 1a—Map of epicentres and recording stations of the data-set. For details of map symbols, see text in the introduction

▲

856

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Focal depths of South African earthquakes and mine events where v is the medium velocity. Partial derivatives ffor x, y, and z can be estimated from Equation [1] to make the corrections needed during the iterations. If the stations are at the surface (zi = 0), the derivative for depth is: [2]

Note that for the sparse national network, z << x-xi and y-yi; hence changes in focal depth lead to a near-zero

Figure 1b—Diagram of Pg, Pn, sP Pn, PmP, sPmP, and SmP rays travelling through the crust from the hypocentre (star) to the station (triangle).

Symbols C1, C2, and M1 refer to the velocity layers in Table I. This study applied the arrival times of these rays to estimate the focal depths of earthquakes and mine-related events

Table I

Velocity structure of the diagram in Figure 1b used in earthquake location by the South African National Seismograph Network Layer

Layer thickness (km)

P-wave velocity (km/s

S-wave velocity (km/s)

20 18 22

5.800 6.500 8.040

3.353 3.757 4.647

C1 C2 M1

correction in the travel times.

At distances of more than about 90 km, where both direct P-wave, Pg and/or the critically refracted wave, Pn arrive at a station, the location algorithm has some sensitivity to depth owing to the steeply descending Pn rays, although clear Pn phases can usually be identified only at distances of more than about 130 km (Figure 2a). The author re-measured the Pg phase arrival times, added the Pn phase arrival times, and re-located all the earthquakes and events using the Hypocenter software. The crustal velocity model (Table I) was used throughout this study. This model is a simplified version of the structure determined by Wright et al. (2002), who derived a high-quality multi-layer model using the temporary Kaapvaal Craton array. This study (and routine practice with the South African National Seismograph Network) avoids a multi-layered velocity model, specifically at the expected depths of hypocenters between 0 km and 10 km. A multi-layered model results in clustering of focal depths at layer boundaries, which is caused by the discontinuities in the travel-time curves of the direct phase, Pg, as a

The P-wave to S-wave velocity ratio is 1:33, 3 as is commonly found (or assumed).

Method Re-location

[1]

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 2a—Example of a Pg – Pn phase analysis. This mine-related event that occurred on 18 April 2009 at 02:38 GMT with epicentre in the Klerksdorp gold mining area was re-located to determine the focal depth. A diagram of Pg and Pn rays traveling through the crust from the hypocentre (star) to the station (triangle) is shown at the bottom right.

Waveforms (black and blue traces) bandpass-filtered between 0.8 Hz and 8 Hz recorded by stations listed to the right of the data are overlain

by Pg and Pn travel time curves (green and red lines with symbols) predicted by the velocity model in Table I for a focal depth of 2 km VOLUME 114

OCTOBER 2014

857

▲

Earthquake location is defined by the hypocentre (x x0, y0, z0), with x0 the longitude, y0 the latitude, and z0 the focal depth and origin time t0 (c.f. Havskov and Ottemöller, 2010b). To calculate the location, the Hypocenter software applies an iterative method where the location problem is linearized (Lienert et al., 1986). The method is optimized for the best general epicentre, depth, and origin time accuracy. First, a guess is made in respect of the hypocentre (x, y, zz) and origin time, tt. If this guess is close enough to the true hypocentre, the travel time residuals at the trial hypocentre are a linear function of the correction required in respect of hypocentral distance. Successive iterations converge to the local minimum provided that the problem is well conditioned.

The simplest case for earthquake location is a homogeneous medium with direct waves, where the calculated travel times, titra, at the iith station (xi, yi, zi) are (e.g. Havskov and Ottemöller, 2010b):

Focal depths of South African earthquakes and mine events f function off depth at layer boundaries. The Pg travel time suddenly decreases when the hypocentre crosses a boundary (for example, the discontinuity at 20 km depth in Table I) since a larger part of the ray is suddenly in a higher velocity layer, while the Pn travel time continuously decreases as the depth increases (Havskov and Ottemöller, 2010b).

The phase analysis shown in Figure 2b is an example for the signals recorded by selected stations of the South African National Seismograph Network for the mine-related event displayed in Figure 2a. In Figures 2a and 2b(1), the Pn phase overtakes the Pg phase at a distance of approximately 150 km. It is noteworthy that, while the Pn phase arrival times (i.e. travel time curves) are sensitive to the different focal depths of 2 km, 5 km, and 10 km, Pg is insensitive to changes in focal depth at these large hypocentral distances.

Focal depth determination depends on accurate Pg and Pn travel times and therefore small uncertainties in the epicentre, origin time, and crustal velocity model may result in inaccurate depth estimates. These inaccuracies show up as misalignments between the phase arrival measurements and travel time curves in Figure 2b. Misalignments are approximately 2 to 10 km and ½ to 1 second, which is a typical epicentre and origin time uncertainty for locations determined by means of the South African National Seismograph Network. To ensure a reliable depth estimate, several Pg and Pn phase arrival measurements must be available.

After f the initial determination off the epicentre off a minerelated event or tectonic earthquake at a fixed depth of 2 km or 5 km, respectively, the depth parameter is set free but no weighting can be applied to the depth parameter. The RMS of the differences (residuals) between the measured travel times and those predicted by means of the velocity model in Table I is calculated as a function of depth (Figure 3). Focal depth is measured at the minimum RMS. Depths, D, for tectonic earthquakes have a range of 2 km ) D ) 10 km and an average depth with a standard deviation of 6.9 ± 2.3 km.

Depths for mine-related events range over 0 km ) D ) 7 km, averaging 3 ± 2.3 km. One tectonic earthquake and two mine-related events were rejected from the original data-set because their smallest values were reached at a depth of 0 km without what looked like a minimum. Also note that the WKBJ synthetic seismogram software applied below cannot accommodate a hypocentre at the surface.

Figure 2b(1) – Pg – Pn phase analysis of stations Koster (KSR), Schweizer-Reneke (SWZ), and Silverton, Tshwane (SLR) from Figure 2a.

Recorded waveforms (thin traces) bandpass-filtered between 0.8 Hz and 8 Hz are overlain by Pg and Pn travel time curves predicted by the velocity model in Table I for focal depths of 2 km (dashed line), 5 km (solid line), and 10 km (dash-dot line). Note that Pg travel time curves plot on top of one another. Phase arrival measurements are indicated by vertical lines

Figure 2b(2)—P Pn (left) – Pg (right) phase analysis of stations Lobatse, Botswana (LBTB), Pongola (POGA), and Upington (UPI) from Figure 2a.

Recorded waveforms (thin traces) bandpass-filtered between 0.8 Hz and 8 Hz are overlain by Pg and Pn travel time curves predicted by the velocity model in Table I for focal depths of 2 km (dashed line), 5 km (solid line), and 10 km (dash-dot line). Note that Pg travel time curves plot on top of one another. Phase arrival measurements are indicated by vertical lines

▲

858

OCTOBER 2014

VOLUME 114

Travel times with regional depth phases Focal depth estimation may be further improved by including so-called ‘depth phases’, which are sensitive to changes in the focal depth. In this study the author measured additional arrival times for the ascending S-wave converted at the surface to a critically refracted P-wave, sP Pn; reflected P-wave at the Moho discontinuity, PmP; P ascending S-wave converted at the surface to a reflected P-wave at the Moho discontinuity, sPmP; P and a descending S-wave converted to a P-wave when reflected at the Moho discontinuity, SmP (Figure 1b). Figure 4a shows synthetic waveforms and

The Journal of The Southern African Institute of Mining and Metallurgy

Focal depths of South African earthquakes and mine events

Figure 4a—Synthetic waveforms (black and blue traces) at hypocentral distances from 25 km to 800 km bandpass-filtered between 0.5 Hz and 1.5 Hz overlain by Pg, Pn, sPn, PmP, sPmP, and SmP travel time curves (coloured solid and dotted lines with symbols) predicted by the velocity model in Table I for a focal depth of 7 km. A diagram of the rays travelling through the crust from the hypocentre (star) to the station (triangle) similar to Figure 1b is shown at the bottom right The Journal of The Southern African Institute of Mining and Metallurgy

Figure 4b—Synthetic waveforms (black and blue traces) at a hypocentral distance of 325 km bandpass-filtered between 0.5 Hz and 1.5 Hz overlain by Pg, Pn, sPn, PmP, sPmP, and SmP travel time curves (coloured solid and dotted lines with symbols) predicted by the velocity model in Table I for focal depth of 1 km to 15 km VOLUME 114

OCTOBER 2014

859

▲

Figure 3—Root mean square (RMS) of the differences between the measured travel times and those predicted using the velocity model in Table I, calculated as a function on of depth for mine-related events (left) and tectonic earthquakes (right). Focal depth is measured at the minimum RMS. The average focal depths are indicated by dash-dot lines and standard deviations by dashed lines

corresponding travel time curves ffor these regional depth phases at hypocentral distances between 25 km and 800 km for a focal depth of 7 km. The waveforms and travel time curves were generated by means of the WKBJ synthetic seismogram software (Chapman, 1978) for the velocity model in Table I. Note that the travel time differences between Pn and sP Pn as well as PmP and sPmP are nearly constant over the whole hypocentral distance range for which these phases can be measured. In Figure 4b, synthetic waveforms and travel time curves are shown for a hypocentral distance of 325 km and focal depths that vary from 1 km to 15 km. As before, focal depth is insensitive to Pg arrival times, but is sensitive to the arrival times of all the other phases. Note that focal depth is especially sensitive to the relative travel time differences between Pn and sP Pn and the PmP and sPmP phases.

An example of a depth phase analysis is shown in Figure 5 for the signals recorded by selected stations of the South African National Seismograph Network for the minerelated event depicted in Figure 2a. The predicted arrival times were calculated using the WKBJ synthetic seismogram software. The number of available phases to analyse as well as the misalignments between the measured and predicted phases had increased. This may be ascribed to an increasing uncertainty for the crustal velocity model, e.g. depth of the Moho for the reflected PmP phase and S-waves velocity 1 (which is assumed to be 33 3 of the P-wave velocity) for the converted SmP phase. An additional uncertainty was introduced by the phase measurements: Pg and Pn phases show up best on a signal bandpass filtered between 0.8 Hz and 8 Hz; hence this filter was used to re-locate the events and earthquakes in the previous section. Although phases sP Pn and PmP are usually clearer after they have been filtered

Focal depths of South African earthquakes and mine events between 0.8 Hz and 3 Hz, this is not always the case. For example, the ‘incorrectly measured’ PmP phase arrival in Figure 5(1) for station PRYS was analysed using the 0.8 Hz to 8 Hz filter, which showed a clearer onset than for the lower filter. It should be noted that the Pn phase onset for station LBTB was not clear for the lower filter (Figure 5(2)).

To determine the focal depth, earthquake hypocentres were first calculated (i.e. re-located) for the depth range 1 km to 15 km using the Hypocenter software. The corresponding predicted arrival times of the depth phases were then calculated using the WKBJ synthetic seismogram software.

The introduction of additional measured phases, but with larger errors, required a more robust cost function than the RMS to estimate the focal depth. Hence the author applied variance as the cost function to this regional depth phase travel time investigation (e.g. Steyn et al., 1999). The variance of the travel time residuals estimates the spread of the measured and predicted travel time differences. A zero variance means that all the measured and predicted travel times are identical. A sample variance may be applied to estimate the variance of a continuous distribution from a sample of that distribution: this is an unbiased estimator of the variance of the population from which the sample variance is drawn provided that the range consists of independent, identically distributed samples. The minimum sample variance (of the sample variances calculated for different depths) of the travel time residuals of Pg and depth phases Pn, sP Pn, PmP, P sPmP, P and SmP would give an unbiased estimate of the focal depth. The requirement is that the selection of depth phases must be large and representative enough so that it is possible to discard uncertainties in respect of epicentre, origin time, and crustal properties.

Figure 5(1)—Depth phase analysis of stations Parys (PRYS), SchweizerReneke (SWZ), and Boshof (BOSA) from Figure 2a. Theoretical phases Pg, Pn, sPn, and PmP predicted by the velocity model in Table I for a focal depth of 2 km are indicated by dashed vertical lines, and the corresponding measured phases by solid vertical lines. The waveforms were bandpass-filtered between 0.8 Hz and 3 Hz

▲

860

OCTOBER 2014

VOLUME 114

Figure 6 shows sample variance as a ffunction off depth. Focal depth is measured at the minimum sample variance. Depth ranges and averages are 4 km ) D ) 7 km with 5.9 ± 1.2 km and 1 km ) D ) 4 km with 2.4 ± 1.2 km for tectonic and mine-related events, respectively.

Waveform modelling The author verified the focal depths of 8 km (re-location) and 5 km (travel times with regional depth phases) for one tectonic earthquake that occurred on 18 December 2011 at 18:07 GMT with its epicentre in the Augrabies area.

Synthetic-to-recorded-waveform fits were applied using the WKBJ synthetic seismogram software for the abovementioned depth phases (c.f. Ma, 2012). The focal mechanism parameters for this earthquake source needed to generate the synthetic waveforms were obtained from the National Earthquake Information Centre. Unfortunately, no focal mechanism parameters could be calculated for any of the other earthquakes or events. Brandt and Saunders (2011) had obtained regional moment tensors in a previous study using the dense Kaapvaal Craton array, but the National Network is too sparse to obtain reliable fault plane solutions for single, large earthquakes. Before depth phase modelling can begin, each synthetic seismogram is time-shifted with respect to its recorded counterpart until the first-arrival recorded and synthetic phases (P Pg or Pn) are aligned. In this way, uncertainties in the epicentre, origin time, and crustal velocity model are smoothed out.

Examples of seismograms illustrating the method of recorded-to-synthetic-waveform fits are shown in Figures 7a and b. The focal depth is obtained by visually comparing

Figure 5(2)—Depth phase analysis of stations Lobatse, Botswana (LBTB), Pongola (POGA), and Musina (MSNA) from Figure 2a.

Theoretical phases Pg, Pn, sPn, PmP, and SmP predicted by the velocity model in Table I for a focal depth of 2 km are indicated by dashed vertical lines and the corresponding measured phases by solid vertical lines. The waveforms were bandpass-filtered between 0.8 Hz and 3 Hz The Journal of The Southern African Institute of Mining and Metallurgy

Focal depths of South African earthquakes and mine events

absolute and relative arrival times of the synthetic phase signals with recorded phases. (Also see Figures 4a and 4b).

All the phases could be identified after some additional filtering on the seismogram recorded at station KOMG (Figure 7a). Station CVNA at Calvinia, at almost the same hypocentral distance but situated south of the epicentre (Figure 1a), recorded only clear Pn and sP sPn phases; the other phases could not be identified (Figure 7b). This may be attributed to the double couple source mechanism that generates directional seismic waves. Ma and Atkinson (2012) found that usually only one or two stations (among a range of stations) can be used to determine focal depth.

Useful stations usually also record only either clear Pn and sPn or PmP and sPmP phases, although Ma and Atkinson (2012) did not apply different filters to their data as has been done in this study. The author was able to identify Pn and sPn phases on seismograms recorded by eight stations (e.g. for CVNA in Figure 7b) with modelled focal depths of 5 km or 6 km using relative arrival times. This study obtained a depth of 8 km for station KOMG with an overall, general best fit for the absolute arrival times of all the phases and relative times for phases Pn and sP Pn as well as PmP and sPmP (Figure 7a).

Conclusions The results confirm the assumption that focal depths of South African earthquakes and mine-related events are shallow — within the upper third (0 km to 10 km) of the crust.

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 7b—Example of a synthetic-to-recorded-waveform fit at station Calvinia (CVNA) at a distance of 318 km for a tectonic earthquake that occurred on 18 December 2011 at 18:07 GMT with epicentre in the Augrabies area. Recorded waveforms (thick black trace) and synthetic waveforms (thin black and blue traces) bandpass-filtered between 0.5 Hz and 1.5 Hz are overlain by Pg, Pn, sPn, PmP, sPmP, and SmP travel time curves (coloured solid and dotted lines with symbols) predicted by the velocity model in Table I for a focal depth of 6 km. The synthetic seismograms are shifted by -1.47 seconds to align the recorded and synthetic Pn phases VOLUME 114

OCTOBER 2014

861

▲

Figure 6—Variance of the differences between the measured travel times and those predicted using the velocity model in Table I calculated as a function of depth for mine-related events (left) and tectonic earthquakes (right). Focal depth is measured at the minimum variance.

The average focal depths are indicated by dash-dot lines, and standard deviations by dashed lines

Figure 7a—Example of a synthetic-to-recorded-waveform fit at station Komaggas (KOMG) at a distance of 311 km for a tectonic earthquake that occurred on 18 December 2011 at 18:07 GMT with epicentre in the Augrabies area. Recorded waveforms (thick black trace) and synthetic waveforms (thin black and blue traces) bandpass-filtered between 0.5 Hz and 1.5 Hz are overlain by Pg, Pn, sPn, PmP, sPmP, and SmP travel time curves (coloured solid and dotted lines with symbols) predicted by the velocity model in Table I for a focal depth of 8 km. Recorded signals around the Pn and SmP phases, bandpass-filtered between 2 Hz and 4 Hz (thick black dotted traces) are plotted on top of the waveforms.

Synthetic seismograms are shifted by -0.71 seconds to align the recorded and synthetic Pn phases

Focal depths of South African earthquakes and mine events Mine-related events are classified f by operators off local underground mine networks as fracture-dominated rupture events or development blast and friction-dominated slip events or mining-induced events. Development blast events typically occur within 100 m of active mine tunnels, whereas mining-induced events may occur up to 200 m from tunnels (e.g. Spottiswoode and Linzer, 2005). A detailed underground inspection of fault zones associated with the magnitude 4.6 Matjhabeng earthquake that occurred in 1999 near Eland shaft at Welkom, Free State gold mining area, reported that the active faults could be accessed and slip and displacement measured (Dor et al., 2001). Gold mining is now reaching depths of around 4 km and mine events are thought to be associated with active mining. This confirms the derived depth range of foci between 1 km ) D ) 4 km.

No evidence was found to support the hypothesis that earthquake focal depths in South Africa follow a bimodal distribution, with deeper hypocentres in the lower third of the crust (20 km to 35 km), as had been determined for other stable continental regions. However, this result should be interpreted with circumspection, because of the small tectonic data-set, and the fact that most of the earthquakes occurred in the Augrabies area.

The study justifies the routine practice used by the South African National Seismograph Network of fixing the depth of mine-related events to 2 km and tectonic earthquakes to either 5 km or 10 km when locating these events by means of first-arrival phases.

Focal depths determined when re-locating earthquakes with Pn and Pg phases are similar to the depths obtained using the more advanced techniques of travel times with regional depth phases and waveform modelling. Given the minimal effort involved in measuring additional Pn and Pg phases and re-locating an earthquake once the epicentre and fixed depth have been routinely determined (by means of the first-arrival phases), this technique may be applied regularly to major earthquakes in the future.

Magaliesburg, South Africa. f Southern African f Institute off Mining and Metallurgy, Johannesburg. pp. 109–112.

HAVSKOV, J. and OTTEMÖLLER, L. 2010a. SEISAN earthquake analysis software for Windows, Solaris, Linux and Macosx. Ver. 8.3. University of Bergen, Norway.

HAVSKOV, J. and OTTEMÖLLER, L. 2010b. Routine data processing in earthquake seismology. Springer Science + Business Media. 347 pp. doi: 10.1007/978-90-481-8697-6 KLOSE, C.D. and SEEBER, L. 2007. Shallow seismicity in stable continental regions. Seismological Research Letters, vol. 78. pp. 554–562 doi: 10.1785/gssrl.78.5.554 LIENERT, B.R.E., BERG, E., and FRAZER, L.N. 1986. Hypocenter: an earthquake location method using cantered, scaled, and adaptively least squares.

Bulletin of the Seismological Society of America, vol. 76. pp. 771–783.

MA, S. 2012. Focal depth determination for moderate and small earthquakes by modeling regional depth phases sPg, g sPmP, P and sPn. Earthquake Research and Analysis — Seismology, Seismotectonic and Earthquake Geology.

D'Amico, S. (ed.). InTech, USA. ISBN: 978-953-307-991-2.

MA, S. and ATKINSON, G.M. 2006. Focal depths for small to moderate earthquakes (mN * 2.8) in western Quebec, southern Ontario, and northern New York. Bulletin of the Seismological Society of America, vol. 96. pp. 609–623 doi: 10.1785/0120040192 MAGGI, A., JACKSON, J.A., MCKENZIE, D., and PRIESTLEY, K. 2000. Earthquake focal depths, effective elastic thickness, and the strength of the continental lithosphere. Geology, vol. 28. pp. 495–498.

MANGONGOLO, A., STRASSER, F.O., and SAUNDERS, I. 2014. Depths of South African earthquakes. Seismological Research Letters (in press).

NGUURI, T.K., GORE, J., JAMES, D.E., WEBB, S.J., WRIGHT, C., ZENGENI, T.G., GWAVAVA, O., SNOKE, J.A., and KAAPVAAL SEISMIC GROUP. 2001. Crustal

Acknowledgements This research was funded as part of the operation and data analysis of the South African National Seismograph Network.

Two anonymous reviewers made thoughtful suggestions to improve the article. The author wishes to thank the Council for Geoscience for permission to publish the results. Zahn Nel undertook the language editing.

References

structure beneath Southern Africa and its implication for the formation and evolution of the Kaapvaal and Zimbabwe cratons. Geophysical Research Letters, vol. 28. pp. 2501–2504.

SAUNDERS, I., BRANDT, M.B.C., STEYN, J., ROBLIN, D.L., and KIJKO, A. 2008. The South African National Seismograph Network. Seismological Research Letters, vol. 79. pp. 203–210. doi: 10.1785/gssrl.79.2.203 SPOTTISWOODE, S.M. and LINZER, L.M. 2005. A hybrid location methodology.

BRANDT, M.B.C. and SAUNDERS, I. 2011. New regional moment tensors in South Africa. Seismological Research Letters, vol. 82. pp. 69–80. doi:

Journal of the Southern African Institute of Mining Metallurgy, vol. 105. pp. 417–426.

10.1785/gssrl.82.1.69 STEYN, A.G.W., SMITH, C.F., DU TOIT, S.H.C., and STRASHEIM, C. 1999. Modern CHAPMAN, C.H. 1978. A new method for computing synthetic seismograms.

Statistics in Practice, 1st edn. Van Schaik, Pretoria, South Africa. 764 pp.

Geophysical Journal of the Royal Astronomical Society, vol. 5.

WRIGHT, C., KWADIBA, M.T.O., KGASWANE, E.M., and SIMON, R.E. 2002. The

pp. 481–518.

structure of the crust and upper mantle to depths of 320 km beneath the DOR, O., RECHES, Z., and VAN ASWEGEN, G. 2001. Fault zones associated with

Kaapvaal craton, from P-wave arrivals generated by regional earthquakes

Matjhabeng earthquake, 1999, Eland shaft, Welkom. Proceedings of the

and mining-induced tremors. Journal of African Earth Sciences, vol. 35.

5th International Symposium in Rockbursts and Seismicity in Mines,

pp. 477–488.

▲

862

OCTOBER 2014

VOLUME 114

◆

The Journal of The Southern African Institute of Mining and Metallurgy

Design and positive financial impact of crush pillars on mechanized deep-level mining at South Deep Gold Mine

by B.P. Watson*, W. Pretorius*, P. Mpunzi†, M. du Plooy*, K. Matthysen*, and J.S. Kuijpers‡

Description of the mine

Crush pillars have been incorporated into a mechanized, low-profile trackless system at South Deep Gold Mine. These pillars had to be designed to fail near the face and to ensure that pillar failure is contained within the pillar, to avoid bursting and the risk of high loads being generated during a seismic event, respectively. PoweRite backfill bags were recommended to maintain the integrity of the pillars; except in the main access drives, where the sidewalls were to be supported on 5.6 mm diameter weldmesh and yielding anchors. The results of a trial site investigation exceeded expectations, showing a residual pillar strength of about 37 MPa for a newly formed pillar and 8 MPa for a pillar subjected to seismicity and a closure of more than 300 mm. The introduction of these pillars has improved the rock mass conditions because of the active nature of the support, compared to the previous passive backfill method. Importantly, the pillars have increased mining efficiencies and improved face availability. A potential cost saving to the mine of R140.9 million could be realized over a period of 10 years.

Keywords crush pillars, de-stress mining, backfill.

Introduction Crush pillars were introduced to the destress stopes of South Deep Gold Mine (South Deep) to improve efficiencies and face availability.

The behaviour of crush pillars has been studied previously on the platinum mines (Watson, 2010), and the findings were adapted to the different environment at South Deep. One of the most significant environmental differences was the larger closure rate with occasional rapid loading, typical of a deep mining environment. The quartzite rocks of the Witwatersrand gold mines are also more brittle than rocks of the platinum mines, which needed to be incorporated in the design philosophy. This paper describes the design methodology as well as measurements conducted to verify the design. Finally, the results from a trial site are described.

The old and new mining methods are compared to provide the reader with a sense of the positive financial impact that the pillars can have on the mine.

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

* † ‡ ©

Gold Fields Ltd. SRK Consulting (South Africa) (Pty) Ltd. Centre for Mining Innovation CSIR. The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. Paper received r Nov. 2013; revised paper received Jun. 2014.

OCTOBER 2014

863

▲

Synopsis

South Deep is a mechanized mine, extracting a wide reef at depth. Such deep-level, wide reefs have not been mined elsewhere in the world and the production layouts developed at the mine are therefore unique. South Deep is situated approximately 45 km west of Johannesburg and 20 km south of Randfontein in the West Witwatersrand mining region (Figure 1). The entire mining area covers 4 232 ha and extends for 9.5 km north-south and 4.5 km east-west at its widest points.

The South Deep orebody lies within the Central Rand Group of the Witwatersrand Supergroup and is overlain by the Ventersdorp lavas (Figure 2). The Ventersdorp Contact Reef (VCR) and Upper Elsburg reefs are of economic importance.

The Upper Elsburg reefs subcrop against the base of the VCR, which is a major stratigraphic unconformity (Figure 3). Towards the east the orebody diverges and thickens up to about 130 m at the eastern extremity of the mine boundary, with an increasing percentage of unpay quartzite middlings in the thicker regions. The dip and strike of the orebody vary across the mine, but it generally dips to the south at between 10° and 14°. This dip angle is too steep for normal mechanized equipment.

The orebody is currently being mined at depths of between 2500 m and 2700 m, and future mining is planned at 3400 m below surface. The virgin vertical stress is high and will become higher as the depth of the overburden increases.

Most of the conglomerate layers within the Upper Elsburgs are extremely strong, brittle rocks. In high-stress environments, these rocks store strain energy that can be released

Design and positive financial impact of crush pillars on mechanized deep-level mining

Figure 1—Locality map showing South Deep in relation to adjacent mines

Figure 2—Simplified 3D section showing the stratigraphy around the strategic reefs

narrow cut was backfilled f in the past to limit bedding separation in the hangingwall and to restrict rockburst damage, due to its stiffness, areal support, and energy absorption capacity. The new concept involves a combination of crush pillars and backfill to achieve the same objectives.

Early elastic numerical modelling work (Smallbone, James, and Isaac, 1993) indicated that an optimum conventional destressing cut could be achieved with backfill in a stoping width of 1.5 m to 2.0 m. The panels in the destress cut could generally be kept below the mining industry’s accepted average energy release rate (ERR) criterion of 30 MJ/m2 (Jager and Ryder, 1999) if the span was limited to 250 m, by appropriately spaced stability pillars (Joughin and Pethö, 2007). It was shown that the destress mining could ‘cheat’ gravity by reducing the major stress component from around 70 MPa to about 30 MPa in the shadow area of the destress stope (Figure 4). The stress conditions in the area marked as the ‘window of massive mining opportunity’, in the figure is similar to what could be expected at a depth of 1 200 m. The major stress is horizontal in this scenario.

Simple elastic modelling showed that the vertical stress could be reduced to between 10 MPa and 20 MPa for a distance of 30 m above and below the stope. At a middling distance of 60 m, this stress increased to about 30 MPa. The horizontal stress in the north-south direction was shown to be the highest, increasing from 20 MPa near the destress stope to 50 MPa at 50 m above and below this horizon. The virgin stress tensor used in the modelling was determined from stress measurements using CSIR cells, carried out at a depth of 2 650 m below surface (Smallbone, James, and Isaac, 1993).

The horizontal mechanized method of mining that is employed at the mine was developed at South Deep. It involves layered horizontal destress cuts that overlap and destress the large orebody target horizons (Figure 5). Access to the destress cut is through a spiral decline, which is optimally sited beneath a previously mined destressed area.

Access drives are developed horizontally from the spiral decline to each horizontal destress horizon. Each horizon is mined as a series of 5.0 m wide by 2.2 m high drives, which need to be sequenced optimally to mitigate rockburst and stress damage. This destress cut is subsequently integrated

Figure 3—Generalized east-west section showing the stratigraphy of the orebody

violently in the form of rockbursts. Experience has shown that the higher the walls of an excavation, the greater the tendency for buckling and violent failure.

Destressing philosophy It was found that a narrow tabular cut could sufficiently destress the orebody to allow normal massive mining techniques to be conducted above and below the cut. This

▲

864

OCTOBER 2014

VOLUME 114

Figure 4—Diagram showing a sectional view of the destress concept.

The red and white lines show the vertical and horizontal stress trajectories, respectively The Journal of The Southern African Institute of Mining and Metallurgy

Design and positive financial impact of crush pillars on mechanized deep-level mining with the longhole stoping by utilizing the destress cut drives as longhole drives (Figure 5). To achieve this objective, the dimensions of the destress drives are increased to 5 m × 5 m behind the destress face by sliping the footwall. This layout allows for selective mining in the massives, where there may be sub-economic quartzite middlings.

The layout of the current destress mining cut is shown in Figure 6. Main access drives (MADs) are generally developed in the dip direction. Stope drives (SDs) are mined adjacent to the MADs in a staggered configuration to maintain industryacceptable lead-lag distances (Jager and Ryder, 1999). Stope access drives (SADs) are created every 15 m by cutting 5 m holings through the crush pillars at appropriate locations.

This ensures that cross-fracturing is avoided in the SAD hangingwall. Separate cuts are spaced about 17 m vertically, dictated by the average width of the target grade.

Backfill bags from Reatile TimRite (PoweRite bags) are installed along the edge of the crush pillars to provide confinement to the pillars and to stop pillar disintegration at large closures. The sidewalls of the MAD are supported on 5.6 mm diameter weldmesh and yielding tendons.

repeated every time the MAD was advanced, and an SD could be mined only after the adjacent SD or MAD was backfilled (Figurs 9).

PoweRite bags were used as bulkheads and to ensure

Figure 7—Plan view of the original destress method showing the backfill requirement

Original destress method In the past, a typical mining sequence would start by advancing the MAD by 15 m and the first SAD (TOP SAD) by 5 m on either side of the MAD (Figure 7). The first 10 m of the MAD would subsequently be backfilled, before the SDs on either side of the MAD could be developed. The SDs would then be backfilled as indicated in Figure 8 and the backfill in the MAD mined out. The process described above would be

Figure 8—Plan view of the original destress method showing the first phase of development (backfill mined back out of the MAD)

Figure 5—Schematic section view of the horizontal destress layout

The Journal of The Southern African Institute of Mining and Metallurgy

Figurs 9—Plan view of the original horizontal destress layout after several cycles of installing and removing backfill in the MAD VOLUME 114

OCTOBER 2014

865

▲

Figure 6—Plan view of the current horizontal destress layout with crush pillars

Design and positive financial impact of crush pillars on mechanized deep-level mining tight ffilling adjacent to the MAD. However, the backfill f behind the bags was rarely tight-filled due to the horizontal inclination of the excavation.

The new layout involves the use of crush pillars with PoweRite bags, as shown in Figure 6.

Advantages of the crush pillar The destress layout with crush pillars is depicted in Figure 6.

The pillars allow mining to take place in the adjacent SDs without first having to backfill the previously mined excavation. The implications are: ➤ No backfill required in the MAD, improving mining efficiencies ➤ Face availability is improved because SDs can be mined at an earlier stage in the process ➤ Reduced backfill dilution during longhole stoping.

is 59% (Figure 12). A comparison off the three designs after f a period of three months is provided in Figure 13.

While destress mining alone controls how much of the reserves are prepared (destressed) for stoping, other components such as capital infrastructure to supply air and rock-handling capabilities all combine to form a plan. This plan dictates the steady-state volumes that can be handled.

In the above case study, Design 1, Design 2, and Design 3 reach steady states by weeks 29, 21, and 19, respectively (Figure 12). The effect of the ’free square metres’ from the crush pillars in Design 3 can be seen in the difference between steady-state production outputs. The average effective square metres achieved in Design 3 is 18% greater at steady state than for the other two designs.

In addition, the crush pillars are an active support and inhibit hangingwall unravelling, which is often observed where the reef is replaced by backfill. If properly designed, these pillars do not create additional fractures in the hangingwall and overall stability is improved. Subsequent to the destressing process, the crush pillars will be mined out with the longhole stopes. Therefore, the extraction efficiency is not affected by leaving crush pillars.

Method comparison Three design options were compared to determine the financial impact of crush pillars at South Deep (Figure 10).

These options are described as:

Figure 11—Isometric oblique view of the design comparisons after four weeks

➤ The original design – MADs that are initially mined, backfilled, and mined out a second time (backfill mined out) once the neighbouring SDs are mined and backfilled (Design 1) ➤ Crush pillars on the MADs only – here, time is saved as the MADs do not require re-mining (Design 2) ➤ Total crush pillar design – all MADs and neighbouring SDs are separated by crush pillars (Design 3).

After four weeks, Design 3 achieves 44% more square metres than Design 1 (Figure 11). During this stage, Design 1 and Design 2 achieve similar square metres, as Design 2 is not constrained by backfill yet. The variance between Design 3 and Design 1 increases until week 19, when the difference

Figure 12—Progressive square metres destressed per week

Figure 10—Isometric oblique view of the three design options

▲

866

OCTOBER 2014

VOLUME 114

Figure 13—Oblique view of the design comparison after three months The Journal of The Southern African Institute of Mining and Metallurgy

Design and positive financial impact of crush pillars on mechanized deep-level mining Financial impact of crush pillars Savings associated with the crush pillar design can be separated into three components: ➤ Reduction in backfill requirement ➤ Reduced destress mining ➤ Productivity improvements.

seismic events is an important consideration in the deep-level gold mining environment. Fracturing should therefore be contained within the pillar so that high load-generation and hangingwall deterioration is avoided during such an event.

The investigations to determine a suitable pillar size included back-analyses, numerical modelling, and underground instrumentation.

The requirement for backfill is reduced in two areas: a) In the establishment of a new destress cut, all MADs had to be filled and subsequently mined again due to a mining span constraint. With the advent of the crush pillar method this is no longer a requirement.

The savings associated with this activity reduce the amount of backfill placed in the destress mining by 4.3% or 62 400 m3 in 10 years at a cost of R299 per square metre.* This saves R18.7 million over the next 10 years (Table I) b) Less backfill is required as the crush pillar volume is no longer backfilled. This reduces the area to be filled by 15.4% at a saving of R66.9 million for the next 10 years (Table II).

There is a reduction in the planned low-profile mining of 15.4% (crush pillar area). This reduces the destress mining requirement by 90 100 t or R55.3 million at R614 per ton* over a 10-year period (Table I).

Productivity is improved through the increased speed at which a new destress cut is established through the reduced volume of mining required.

Backfill adds approximately 18% to the direct cost of mining the destress. The total cost savings in backfill of 23% in the destress reduces the total direct cost of mining the destress (including backfill) by 3.5% (from R725 per ton to R699 per ton).

Stable crush pillar loading environment An opportunity to determine a limit on loading stiffness was provided when the width of a pillar/peninsular between two destress panels was inadvertently reduced. Violent failure occurred when a 2.2 m high face was advanced towards the north (Figure 14), reducing the pillar to a width of 5 m and length of 6 m. The incident was modelled using MAP3D (map3D.com, 2013) to determine the loading stiffness of the strata when the violent failure occurred. The results of the investigation are compared to a more comprehensive investigation performed on the platinum mines (Watson, Kuijpers, and Stacey, 2010) in Figure 15. In both instances (platinum mines and South Deep), the laboratory-determined elastic constants of the hangingwall and footwall rock types were used in the numerical analysis. In the case of the South Deep model, the Young's Modulus and Poisson's Ratio were 70 GPa and 0.25 respectively.

Crush pillar design considerations The main objective of crush pillar design was to ensure that the residual strength of the crush pillars is sufficient to maintain hangingwall stability, and limit the demand on the tendon support. Pillar size was therefore designed with the residual strength in mind. However, pillar bursts show that the peak strength and loading environment also need to be considered in the design. Larger pillars have a higher peak strength. Therefore, a pillar that is cut too large is likely to fail violently in the back areas where the loading environment is soft. In addition, dynamic loading during

Figure 14—Plan view showing the burst pillar that was modelled using MAP3D to determine an unacceptable loading stiffness

Table I

Cost savings associated with the crush-pillar method Savings R million

Backfill – no mining of backfill in MAD Backfill – reduced requirement in crush pillar Mining – Reduced destress volume to be mined

18.7 66.9 55.3

Total cost saving (for 10 years)

140.9

*2014 first Quarter actual costs used in the estimated savings The Journal of The Southern African Institute of Mining and Metallurgy

Figure 15—Acceptable and unacceptable strata stiffness of the hangingwall and footwall. The dotted line is parallel to the ‘Platinum unacceptable’ for comparison VOLUME 114

OCTOBER 2014

867

▲

Description of savings

Design and positive financial impact of crush pillars on mechanized deep-level mining The unacceptable strata stiffness ff ffor the platinum mines was determined from extensive underground measurements, which are described by Watson (2010). The black loading line (Figure 15) describes the unacceptable stiffness for South Deep, which was determined from the single pillar burst shown in Figure 14; where the pillar dimensions and face positions were reliably determined. The softness of the loading strata at South Deep was calculated to be about 4.7 mm/GN. This unloading environment is slightly stiffer than the critical 5.0 mm/GN limit for the platinum mines. The investigation suggests that the more brittle conglomerate pillars at South Deep may require a slightly stiffer loading environment, for stable failure, than the platinum mines. A numerical sensitivity analysis on pillars cut in a typical stope configuration (Figure 16), using the same elastic constants as the South Deep line in Figure 15, suggests that stable pillar failure can occur only very near, or at, the face (Figure 17).

The residual strengths of the surrounding crush pillars were assumed to be zero in the model.

Material strength Uniaxial and triaxial compressive strength tests (UCS and TCS) were conducted on the reef material at the trial site. A representative example of the results is shown in Figure 18.

The stress-strain curves are typical of Witwatersrand quartzites, with little nonlinear behaviour and sudden failure which is characteristic of brittle rocks. The triaxial postfailure results, in the same figure, confirm the brittle nature.

A typical set of Merensky Reef test results is shown for

Figure 16—Plan view of the stope (red) and modelled ‘crush’ pillars

Figure 17—Loading environment with distance from an advancing face

▲

868

OCTOBER 2014

VOLUME 114

comparison in Figure 19. In this ffigure the post-failure f results of the triaxial tests indicate a comparatively more ductile material. The differences in behaviour and strength show that the work done previously on the Merensky crush pillars (Watson, 2010) cannot be directly applied to South Deep.

Further research was necessary.

Pillar behaviour The crush pillars at South Deep are 10 m long, 1.5 m wide, and 2.2 m high. Although these pillars are not truly 2D, the ‘perimeter rule’ (Equation [1]), described by Wagner (1974), may be used to compensate for pillars with finite length. The equation suggests that the pillars at South Deep have an effective width (we) that is only about 10% less than a full 2D model.

[1] Equation [1] accounts for square and rectangular pillars by taking cognisance of pillar length (L). In essence, the equation compensates for the effects of the fracture zone around a pillar.

A series of 2D plane strain FLAC (Itasca Consulting Group, Inc., 1993) models were constructed to investigate pillar behaviour in the context of a realistic loading environment. This environment included the pillar foundations that could sustain damage. For the purposes of

Figure 18—Laboratory test results on reef material from South Deep Gold Mine. Triaxial confinement shown in the explanation

Figure 19—Laboratory test results on Merensky Reef from the Bushveld Complex mines. Triaxial confinement shown in the explanation The Journal of The Southern African Institute of Mining and Metallurgy

Design and positive financial impact of crush pillars on mechanized deep-level mining the model, the ffoundation material properties were assumed to be the same as the pillar, which was a reasonable approximation for all the instrumented pillars. The model included the pillar itself and the immediate hangingwall and footwall.

One important parameter that needed to be considered for pillar behaviour was material brittleness. In the models, brittleness is defined as the rate of stress decrease after failure. In these models, post-failure behaviour is controlled by cohesion loss. Therefore, a direct relation between cohesion softening (strain softening) and material brittleness can be quantified. The internal friction angle and the dilation angle were not varied so as to avoid additional complications.

Figure 20 shows the Mohr-Coulomb parameters that were used in the models. These parameters were calibrated from underground measurements of pillar stress and strain in the platinum mines (Watson, Kuijpers, and Stacey, 2010).

Unfortunately, such measurements were not available for the South Deep pillars.

Boundary conditions play an important role in the failure mechanism, as they affect horizontal confinement. In the models (Figure 21), the vertical boundaries were not allowed to move in a horizontal direction (thus simulating a fully replicated set of pillars). The presence of discontinuities such

as bedding planes, ffaults, and joints should also affect ff failure, but this was not investigated in the models.

The models were used to evaluate the effect of pillar width on strength. Stope span was about five times the pillar width in the model (extraction ratio approx. 83%), and the model height was more than eight times the pillar width. The model results for the Merensky Reef material (Figure 22) (Watson, Kuijpers, and Stacey, 2010) compared favourably to the peak pillar strength formula determined for Merensky pillars (Watson et al., 2008).

The numerical modelling showed that very little pillar strengthening occurs below a width to height (w/h) ratio of 0.75. In addition, pillar punching and foundation failure is likely to initiate only at a w/h ratio of about 1.25 for Merensky Reef pillars. Generally, the larger crush pillars on the platinum mines do not fail throughout, but the centres of these pillars punch into the surrounding strata. In essence, the system fails as shown in Figure 22.

Pillars at South deep are more likely to follow a more brittle profile (as discussed previously). A subsequent sensitivity analysis on brittleness showed that the w/h ratio at which punching initiates increases with brittleness (Figure 23). Thus failure would be safely contained within a pillar if w/h ratios were kept below 1.25. To ensure that the pillars did not develop solid centres, a w/h ratio range between 0.68 and 0.91 was recommended for these pillars.

Figure 22 suggests that pillars with w/h ratios in this range would have a strength in line with the material UCS, i.e.

Figure 20—FLAC model material properties

Figure 22—Comparison between the FLAC modelling (red curves), Merensky Reef strength database (blue lines), and South Deep backanalysis (black line)

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 23—Sensitivity analysis on brittleness, FLAC modelling VOLUME 114

OCTOBER 2014

869

▲

Figure 21—Diagram showing the double symmetry FLAC model used in the pillar and foundation investigations. The model was loaded along the bottom edge

Design and positive financial impact of crush pillars on mechanized deep-level mining approximately 200 MPa. Numerical modelling showed that there would be sufficient stress to ensure failure/crushing at the face, even at relatively small spans, at the recommended w/h ratios.

The pillar used in the South Deep strength back-analysis (Figure 14) had a w/h ratio of 2.3 and was 2.1 times the strength of a similar Merensky pillar (Figure 22). This ratio corresponds favourably to the UCS strength ratio between the two materials, suggesting a reasonable correlation between the model results and the South Deep pillars. Figure 24 shows the APS values in the pillar and foundation curve in Figure 22, multiplied by a factor of 2.1. The curve provides a probable relationship between peak pillar strength and w/h ratio for the pillars at South Deep.

Underground investigations Having established the optimum pillar dimensions, a trial site was created to monitor the pillar behaviour underground (Figure 25). Visual observations formed an important part of the investigations. A section of a pillar was removed to expose the fracturing in the pillar (Figure 26). It was clear that the pillars had failed properly as the fracturing extended throughout. The hangingwalls were also no more severely fractured than where crush pillars were not used. These observations suggested that the foundations were not being damaged by the pillars. No violent pillar failures occurred when the pillars were cut to the stipulated w/h ratios, even though some dynamic closure occurred during seismic events. However, small strain bursts occurred when the pillars were cut at widths in excess of 3 m (w/h of 1.4) and a pillar burst occurred at a w/h ratio of 2.3 (Figure 14).

Close examination off ffracturing within a properly crushed pillar showed curved fractures extending at least 1 m into the pillar sidewall (Figure 27) on the side of the advanced MAD or SD. These fractures tend to create an hourglass-shape sidewall on that side of the pillar, while the lagging face created a more square sidewall. The implication being that the pillars are failing right at the lagging face, as designed.

The effectiveness of the crush pillar lies in its ability to carry the load required to promote stability and break the span between two adjacent excavations. It was therefore necessary to establish the residual strength of the pillars immediately after formation and after large deformations had taken place. In addition, the effects of dynamic closure that occurs during a seismic event needed to be assessed. Three pillars were selected for stress and closure measurements.

Two of these pillars are described in the paper: one newly formed, and one subjected to closure in excess of 300 mm (including dynamic deformation during a seismic event).

It was necessary to establish an optimum height above the pillars where a point measurement could provide a reasonable estimate of the average pillar stress (APS). In

Figure 26—Narrow edge of a crush pillar showing the fracturing in the pillar. A PoweRite backfill bag is shown on the right of the pillar

Figure 24—Pillar and foundation curve in Figure 22 multiplied by a factor of 2.1 and compared to the South Deep back-analysed pillar strengths

Figure 27—Typical fracturing in a pillar. Black lines highlight the fractures formed by the advanced face

Figure 25—Panoramic view of a crush pillar layout

▲

870

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Design and positive financial impact of crush pillars on mechanized deep-level mining order to determine this optimum height, use has been made of Boussinesq equation ([2]). This equation quantifies the effect of a point load on a free surface (half space). The effect of a pillar, with its particular stress distribution, can be quantified by a number of Boussinesq equations, each representing a specific area of the pillar. Since the application of the Boussinesq equation is restricted to a free surface, care must be taken that neighbouring pillars do not affect the stress levels. This can be achieved by limiting the distance above or below the pillar where stresses are to be evaluated.

Numerical modelling has shown that a distance of less than 3 m results in an error of less than 1% for the crush pillar spans that are used at South Deep. Boussinesq equations can thus be used to provide a unique relationship between the APS of a pillar and the stresses measured at a point above or below that pillar.

[2] where: σzz = stress at a point in space Ai = area of the grid ’i’ pzi = vertical stress carried by the grid ’i’.

A matrix of Boussinesq equations (Figure 28) was used to evaluate typical stress profiles of solid and failed pillars.

The profiles used in the analysis are shown in Figure 29, and the results of the investigation are provided in Figure 30.

The blue curve in Figure 30 shows that for a 1 m wide pillar and a measurement height of 2 m, the difference in point-stresses between pillars with different ‘realistic’ stress profiles is negligible. However, the red and green curves suggest that at the same height only about 20% of the APS could be measured. It was also established that the optimum measurement height varies proportionately with the width of

the pillar. A good compromise between a reasonably large measurement and an inordinate error resulting from an unknown stress profile for a 1.5 m wide pillar is probably between heights of 1.5 m (1.0 in Figure 30) and 3 m (2.0 in Figure 30).

The first residual measurement was made at 2.71 m above a 2.5 m to 3.0 m wide and 5.7 m long pillar (Figure 31). This pillar had been newly formed with minimal closure at the time of the measurements; and the face position was about 10 m from the pillar. The analysis in Figure 30 suggested an 11% error in the residual strength evaluations, but the point stress would be about 50% of the APS if the cell was installed exactly over the centre of the pillar. However, the actual position was off-centre and required a dedicated matrix of Boussinesq equations to calculate the residual strength.

The stress profile of the pillar in Figure 31 was estimated from the single point-stress measurement, using a similar matrix to Figure 28 and Equation [2]. Since there was only one reliable measurement, a trial-and-error approach was applied to likely stress profiles, until the measured stress level was simulated. The residual strength of Pillar 1 was calculated from the stress profile in Figure 32 to be about 37 MPa.

Figure 30—Optimum height of a point-stress measurement above a pillar to determine average pillar stress Figure 28—Grid used in the matrix of Boussinesq equations

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 31—Pillar 1, which was instrumented soon after formation. No dynamic closure had occurred VOLUME 114

OCTOBER 2014

871

▲

Figure 29—Typical stress profiles for solid and crush pillars, used in the analysis shown in Figure 30

Design and positive financial impact of crush pillars on mechanized deep-level mining A series off FLAC models were run to substantiate the analytical solution results. Several curves with different dilation angles were modelled, and the results are provided in Figure 37 for comparison. The figure shows a reasonable correlation between the underground measurements (Pillar 2), analytical solution, and the FLAC models.

Figure 32—Estimated pillar stress profile from a single point-stress measurement over Pillar 1

The second pillar was between 1.4 m and 2 m wide and 7.9 m long (Figure 33). Stress measurements were conducted more than six months after pillar formation and more than 300 mm closure had taken place. Some of the above closure was recorded during a nearby seismic event. The residual strength of the pillar was determined to be about 8 MPa, from a point measurement made at 2.5 m above the pillar.

A stress profile was estimated for Pillar 2 using a similar approach to that used for Pillar 1. The profile is shown in Figure 34.

Closure measurements were made adjacent to Pillar 2 in the MAD and SD. Unfortunately, the MAD was required for production and only the first part of the closure curve shown in Figure 35 was monitored here. A cubby was created adjacent to the pillar, i.e. the backfill was mined out, to allow the remaining measurements. It should be noted that the measurements shown in the graph may be slightly overstated because the backfill, normally adjacent to the crush pillars, was not present to carry some of the load or to confine the pillar. The pillar failure/crushing period can clearly be seen in the data measured from the MAD. In addition, a nearby event caused about 26 mm ’dynamic’ closure on the pillar. The location of the event that is most likely to have caused this dynamic closure is shown in Figure 36.

Figure 33—Pillar 2, stress measurements were conducted six months after formation. More than 300 mm closure, including some dynamic closure, had taken place

Analytical solution for residual strength A relatively complicated analytical solution was derived by Salamon (Ryder and Jager, 2002) to describe the stress distribution in a plastic pillar, based on a simple limit equilibrium model. Equation [3] provides a relationship between APS and w/h ratio, assuming a friction angle of 30°. (The equation applies to the stress values across the centre of the pillar.) A reasonable correlation between the measured residual strength of Pillar 2 and the equation was obtained if a cohesion of 1.6 MPa was assumed for the failed pillar material (Figure 37). A similar high cohesion was suggested by the research conducted on Merensky crush pillars (Watson, Kuijpers, and Stacey, 2010). The high residual strength measured over Pillar 1 was probably because the small closure had not caused the pillar to reach its final residual strength at the time of the measurement.

Figure 34—Estimated pillar stress profile from a single point-stress measurement over Pillar 2

[3] Figure 35—Closure measurements adjacent to Pillar 2

▲

872

OCTOBER 2014

VOLUME 114

The Journal of The Southern African Institute of Mining and Metallurgy

Design and positive financial impact of crush pillars on mechanized deep-level mining

Figure 36—Seismic events that occurred between 14 December 2012 and 18 December 2012

designed. Their residual strength and behaviour has exceeded expectations, both under quasi-static and dynamic loading conditions. The concept has the potential to improve mining efficiencies in the destress stopes significantly. In addition, the establishment of a new stope is much faster than previously accomplished. Potentially the system could save the mine R140.9 million over a 10-year period, without considering the quicker build-up value.

Crush pillars have improved hangingwall conditions because full relaxation of the strata does not take place over the pillars, as in the backfill paddocks. In addition, distances between pillars are limited to the cut-span, without the larger spans that develop in backfilled areas due to poor filling practice.

The only disadvantage of the crush pillar system is that disciplined mining is essential. There is little tolerance for off-line mining, and pillar bursting is a threat if pillars are cut too large.

Subsequent to the investigations described in this paper, crush pillars have been rolled out across the mine.

Acknowledgements Gold Fields is acknowledged for facilitating the success of the research work described in this paper. In particular, the management of South Deep are thanked for their assistance.

References ITASCA CONSULTING GROUP, INC. 1993. Fast Lagrangian Analysis of Continua

Figure 37—Relationship between w/h ratio and residual strength.

Cohesion = 1.6 MPa, angle of internal friction = 30°

(FLAC), Vers. 3.2. Minneapolis Minnesota USA. JAGER, A.J. and RYDER, J.A. 1999. A Handbook on Rock Engineering Practice for Tabular Hard Rock Mines. Safety in Mines Research Advisory Committee (SIMRAC), Johannesburg, South Africa.

JOUGHIN, W.C. AND PETHÖ, S.Z. 2007. South Deep regional pillar modelling Part I - Design of regional pillars at South Deep Gold Mine. Challenges in Deep

Previous investigations using FLAC models and underground measurements on the platinum mines (Watson, Kuijpers, and Stacey, 2010) showed little increase in residual strength above a w/h ratio of about 2.5. This is apparently due to the degree of foundation fracturing that occurs above this w/h ratio during pillar failure.

The residual strength estimate from Equation [3] and Figure 37 will need to be downrated slightly to account for its finite length.

and High Stress Mining. g Potvin, Y., Hadjigeorgiou, J., and Stacey, T.R (eds). Australian Centre for Geomechanics, Perth, Western Australia.

MAP3D. 2013. www.map3D.com.

RYDER, J.A. and JAGER, A.J. 2002. A Textbook on Rock Mechanics for Tabular Hard Rock Mines, Safety in Mines Research Advisory Committee (SIMRAC), Johannesburg, South Africa. pp. 174-278.

SMALLBONE, P.R., JAMES, J.V., and ISAAC, A.K. 1993. In situ stress measurements and use in the design of a deep gold mine. Innovative Mine Design for the 21st century. Bawden, W.F. and Archibald, J.F. (eds.). AA Balkema,

The use of crush pillars in a brittle quartzite environment requires disciplined mining. There is little tolerance for offline mining, and pillars cut too wide are risky, with a propensity for bursting. However, if cut properly, the pillars effectively break the span between individual SDs and between the MAD and SDs. This allows for more efficient mining since backfill does not need to be re-handled and faces become available quicker than with the previous mining method.

Conclusions The measurements and visual observations show that the crush pillars at South Deep Gold Mine were properly The Journal of The Southern African Institute of Mining and Metallurgy

Rotterdam.

WAGNER, H. 1974. Determination of the complete load-deformation characteristics of coal pillars. Proceedings of the 3rd International Congress on Rock Mechanics, ISRM, Denver. vol. 2B. pp 1076–1082.

WATSON, B.P. 2010. Rock behaviour of the Bushveld Merensky Reef and the design of crush pillars. PhD thesis, School of Mining Engineering, University of Witwatersrand, Johannesburg, South Africa.

WATSON, B.P., KUIJPERS, J.S., and STACEY, T.R. 2010. Design of Merensky Reef crush pillars. Journal of the Southern African Institute of Mining and Metallurgy, vol. 110, no. 10. pp. 581–591.

WATSON, B.P., RYDER, J.A., KATAKA, M.O., KUIJPERS, J.S., and LETEANE, F.P. 2008.

Merensky pillar strength formulae based on back-analysis of pillar failures at Impala Platinum. Journal of the Southern African Institute of Mining and Metallurgy, vol. 108. pp. 449–461.

VOLUME 114

◆

OCTOBER 2014

873

▲

Discussion

S13924

A trusted industry leader For 90 years, SCAW, a South African industry leader, has partnered with leading mining houses to ensure safety critical standards and productivity targets are consistently met. Whether it’s hoisting, grinding or excavating, Scaw produces an extensive range of products that drive mining safety and productivity. %DFNHG XS E\ WHDPV RI TXDOL¿ HG HQJLQHHUV ZLWK H[WHQVLYH H[SHULHQFH LQ DOO DVSHFWV RI JULQGLQJ media, steel wire rope, chain and cast products, Scaw’s product specialists are available to advise on the design, selection, installation and maintenance of safety critical and mining products manufactured by Scaw.

www.scaw.co.za

Experience Matters.

The application of geophysics in South African coal mining and exploration

by M. van Schoor* and C.J.S. Fourie†

Delineation of old workings

Synopsis Coal remains South Africa’s most abundant and cheapest source of energy, and there is an ever-increasing necessity for optimal and safe extraction of the remaining reserves. Increasing focus on cost-effective mining and zero harm to the environment and miners has resulted in a shift in attitude towards the application of geophysics in local coal mining and exploration.

Furthermore, technological advances have contributed to geophysics being embraced more readily by the coal mining industry, compared to a decade or two ago. Predictably, the growing interest in geophysical technologies has also created a need for education and training in the basic principles and application of geophysical methods, as local coal mining companies generally do not have in-house geophysicists. Consequently, the Coaltech Research Organisation’s Geology and Geophysics working group forum compiled a textbook aimed at addressing this need: to produce a guide for applying geophysics to coal mining problems in South Africa. The target audience for such a book would be coal geologists, mine surveyors, mine planners, and other mining staff with limited or no geophysics background. This paper provides a very brief overview of the book by summarizing key sections and selected examples. In doing so, the value of geophysics to solving a range of coal mining and exploration problems is highlighted.

Keywords geophysics, coal, old workings, dykes, sills, faults.

In areas where mining encroaches on a previously mined area, the historic mine plans do not always provide accurate information regarding the extent of historic and existing developments. Mining into old workings may result in hazards such as flooding of advancing workings. The development of infrastructure over old workings may also be undesirable, especially in areas that are susceptible to surface subsidence. There is therefore a need for a non-invasive technology that can accurately delineate old workings.

Near-surface cavity detection Old workings in previously mined areas commonly deteriorate over time and roof support pillars or beams fail. This may lead to subsidence or caving (also known as ratholing) above the old workings. This caving may ultimately break through to the surface, or it could form undetected, near-surface cavities; either way, presenting a clear mining hazard.

Detection of intrusive dykes and sills

Safe and efficient coal extraction is often compromised by a variety of geological and mining-induced problems. In this paper, we will focus on the five most commonly occurring problems and explain how the application of selected geophysical methods can play a key role in addressing these problems. The problems considered here are as follows: ➤ Delineation of old workings ➤ Near-surface cavity detection and evaluation of surface depressions ➤ Detection of intrusive dykes and sills (magnetic and non-magnetic) ➤ Structural and in-seam continuity disruptions (faults, lenses) ➤ Dolomitic pinnacles/uneven basement.

The Journal of The Southern African Institute of Mining and Metallurgy

VOLUME 114

* CSIR, Natural Resources and the Environment, Pretoria, South Africa. † Environmental Water and Geological Sciences Department, Faculty of Science, Tshwane Univerisity of Technology , Pretoria, South Africa. © The Southern African Institute of Mining and Metallurgy, 2014. ISSN 2225-6253. Paper received Jun. 2013; revised paper received May 2014.

OCTOBER 2014

875

▲

Overview of coal mining problems

Intrusive dykes and sills impact adversely on the operational cost of coal extraction. These geological structures typically disrupt the continuity of coal seams and may also adversely affect the properties of coal seams that occur in close proximity to the intrusions (Du Plessis, 2008). Intrusive bodies typically constitute a relatively tough geological material, such as dolerite, which presents a great challenge for mining machines and can

The application of geophysics in South African coal mining and exploration damage equipment. To complicate matters, the occurrence, geometry, and area of influence of intrusive bodies are difficult to predict ahead of mining, and unexpected intrusive bodies may negatively impact reserve and target estimates.

Structural and in-seam continuity disruptions (faults, lenses) Geological faults disrupt and displace the seam horizon, which impacts adversely on production and mining due to the need to adapt or relocate workings. Regional-scale faults are often known in advance, but small mine- or metre-scale faults are often encountered only during active mining.

Localized in-seam inclusions such as smaller dolerite intrusions and sandstone lenses also present a great challenge to early detection efforts because of their geometry.

Such features may only be a few metres in diameter and they also typically do not have a linear trend or extend up towards the surface. The size-to-depth ratio of these features is relatively small, making them difficult to detect using conventional surface geophysical techniques, and in-mine or borehole-based geophysical methods may have to be used.

Dolomitic pinnacles/uneven basement This problem occurs only where opencast mining activities are conducted in an area with a dolomitic basement that does not constitute a predictable, flat-lying horizon. Finger-like structures (pinnacles) that protrude upward and disrupt the lateral continuity of the overlying coal seams and slump structures (potholes) present extreme difficulties for mining due to the varying floor topography and the relative hardness of the dolomite (Lanham, 2004).

Geophysical applicability considerations There are a number of factors that determine the applicability of a geophysical method to a given problem. These factors are: ➤ Physical property contrast (each geophysical method targets a different physical property) ➤ Range ➤ Required resolution (mapping accuracy) ➤ Geometry and scale of problem.

Property contrast The fundamental requirement that governs the ability of any geophysical method to provide a solution is that there must be a detectable contrast in some physical property between the target and its surroundings (host rock) that can be exploited by geophysical measurements. For example, if the target is a subsurface void located in sedimentary rocks, the void will have a lower density and higher resistivity than the host rock. If, however, the void is filled with weathered material it may have a negligible density contrast with its surroundings, but then it may be relatively conductive compared to the host material.

Range Every geophysical system will have a characteristic minimum detectable signal – below this level the system will essentially detect only noise. Furthermore, there is typically an inverse

▲

876

OCTOBER 2014

VOLUME 114

relationship between the recorded signal strength and the distance between the system and the target. The maximum possible distance between the system and the target of interest is known as the range. The range is determined by a variety of parameters such as the physical properties of the intervening rock mass and the type and properties of the source of the geophysical signal. As an illustration, at one extreme a high-frequency ground penetrating radar (GPR) system might provide a range of only a couple of metres, while at the other extreme, a reflection seismic system could achieve several hundred metres.

Required resolution (mapping accuracy) Resolution technically refers to the smallest distance between two target objects for which the geophysical method can still discriminate between the two distinct objects. As with range, the resolution is determined by a variety of parameters, but primarily the characteristics of the sourced signal is important. As an example, wave-based methods (e.g. GPR) have a trade-off between range and resolution, which can be controlled to some extent through the operating frequency.

Using a lower frequency increases the wavelength and lowers the signal attenuation. The result is a longer range, but at the cost of a decrease in resolution.

Geometry and scale of problem Access to the survey area and proximity to the target play an important role in selecting possible geophysical solutions.

Most traditional geophysical applications involve the taking of measurements on the surface. This approach is, however, not always good enough to achieve the desired depth of investigation and mapping accuracy. For this reason it is often necessary to exploit boreholes and underground developments to get closer to the geophysical target. Some geophysical methods also demand very specific or nonstandard survey geometries; for example cross-hole tomographic imaging technologies will require a pair of coplanar boreholes (or mining tunnels) that straddle the target zone. Borehole radar reflection surveys will require a single borehole drilled sub-parallel to a planar target horizon.

Based on the above considerations and requirements, and the coal problems described earlier, Table I shows the basic applicability guidelines for geophysical techniques.

Selected case studies In this section, selected case studies, illustrating the applicability of geophysics to the previously described problem scenarios, are presented.

Delineation of old workings Old workings are often filled with mine water, which results in a relatively high bulk electrical conductivity compared to the virgin coal. This contrast in conductivity can be exploited by the TDEM method as illustrated in Figure 1a. This trial survey was done over a known old workings boundary, and the TDEM depth slice clearly shows the cross-over from relatively resistive virgin coal (cold colours) to conductive water-filled workings (orange-red), located at a depth of just over 30 m.

The Journal of The Southern African Institute of Mining and Metallurgy

The application of geophysics in South African coal mining and exploration Table I

Geophysics applicability matrix for pertinent coal mining and exploration problems Exploitable property contrast

Typical required range

Typical required resolution

Most applicable geophysical methods

Workable survey geometry

Air-filled: density, resistivity, conductivity Water-filled: conductivity

0–50 m

Accuracy < 5 m

TDEM Micro-gravity Resistivity/IP

Grid surveys on surface

Voids: density, resistivity, conductivity, dielectric, thermal Filled: conductivity, resistivity, dielectric

0–20 m

1–2 m

Thermal imaging GPR FDEM / TDEM Resistivity/IP

Grid surveys – ideally on low-altitude airborne platform

Detection of intrusive dykes and sills

Magnetic susceptibility OR electromagnetic properties

0–200 m

3–5 m

Magnetics FDEM/TDEM

2D grid surveys (usually airborne)

Structural and in-seam continuity disruptions

Magnetic, electromagnetic, resistivity, conductivity

0–50 m

2–4 m

Radio imaging Borehole rada Seismic tomography developments

Tomographic imaging or reflection surveys using in-seam boreholes /

Dolomitic pinnacles / uneven basement

Resistivity, electromagnetic

50–80+ m

~5m

FDEM/TDEM Resistivity imaging

Grid surveys on surface

Delineation of old workings

Near surface cavity detection

TDEM: Time-domain electromagnetic method FDEM: Frequency-domain electromagnetic method

IP: Induced polarization method GPR: Ground penetrating radar

(a)

(b)

Figure 1—TDEM depth slice (left) and microgravity contour map (right) revealing the presence of previously mined zones (Styles, 2005)

Near-surface cavity detection Figure 2a shows the result of a ground FDEM survey conducted over an area where a known abandoned mine shaft was buried under spoils. The EM-31 grid survey clearly imaged the highly resistive (low-conductivity) anomaly associated with the buried void. In cases where the suspected cavities are shallow and the overburden is not too conductive, GPR would arguably be the solution of choice because the method lends itself to fast data acquisition rates.

The Journal of The Southern African Institute of Mining and Metallurgy

Figure 2b shows an example ffrom a Chinese coalfield f where GPR was used to successfully detect previously mined-out zones as well as ’rat-holing’ caused by such mining cavities.

It should be noted that at many local sites affected by nearsurface cavities, safety considerations may dictate that any geophysical surveying should be done from aerial platforms rather than by ground surveys,. However, it is technically fairly challenging to conduct high-resolution surveys at fine enough line and station spacings and at low enough altitudes in order to achieve the required metre-scale resolution.

Detection of intrusive dykes and sills The value of airborne magnetic and EM methods is well known for the mapping of regional-scale structures. The example presented in Figure 3 is the result of a Coaltech Research Organisation project that was aimed at merging all available aeromagnetic data-sets for the Witbank Coalfield.

VOLUME 114

OCTOBER 2014

877

▲

Figure 1b illustrates the application off the microgravity method to the delineation of old workings. This example is from a coalfield in the UK; the blue polygons indicate zones of contrasting (low) gravity. Subsequent drilling confirmed the correlation between these low gravity anomalies and airfilled old workings occurring at a depth of approximately 12–14 m.

The application of geophysics in South African coal mining and exploration

(a)

res

(Donnelly and McCann, 2000)

(Hu et al., 2012) (b)

Figure 2—FDEM contour map (top) and GPR section revealing the presence of a buried mine shaft and near-surface cavities related to coal mining

developments or boreholes, it is possible to apply a tomographic imaging approach to search for any localized seam disruptions. The example shown in Figure 5 is from a US coal mine where the radio imaging method (RIM) was successfully applied between developments to map the continuity of coal seams within longwall panels. The extent of sandstone palaeochannels in the coal seam ahead of mining could be inferred from the RIM survey results. Due to

Figure 3—Total field aeromagnetic image for the Witbank Coalfield (Du Plessis, 2006)

Over 40 individual data-sets were merged and the output was used to perform lineament (dyke) and fault interpretations.

While airborne magnetics is generally well suited to detecting intrusions, the occurrence of non-magnetic dykes in South African coalfields is well documented. In such cases one needs to resort to the airborne EM method. Figure 4 shows the result of a helicopter-based Dighem survey. The EM method succeeded in detecting most of the previously mapped magnetic dykes, as well as other dykes that were not evident on the corresponding magnetic image.

Structural and in-seam continuity disruptions If it is possible to straddle a to-be-mined block with two lines of co-planar access; for example, two adjacent in-seam

▲

878

OCTOBER 2014

VOLUME 114

Figure 4—Dighem aeromagnetic image (top) and schematic of survey area (bottom) showing the location of magnetic and non-magnetic dykes (Du Plessis and Saunderson, 2000) The Journal of The Southern African Institute of Mining and Metallurgy

The application of geophysics in South African coal mining and exploration

Figure 5—Example of a successful in-seam RIM survey from Pittsburgh, USA (Stolarczyk et al., 2003)

well-planned geophysical surveys can provide advance warning of any deviations from the anticipated coal seam continuity and of any geological or secondary features that may constitute a production or safety hazard. However, to extracting maximum useful information from geophysical surveys it is essential to select the most appropriate method to apply to a given problem, as each method has its own strengths, weaknesses, and niche applications.

Finally, it should be noted that significant advances are continually made in various geophysical technologies. For example, enhancements in electronics, computer hardware, and software algorithms have enabled the acquisition and processing of significantly larger data-sets at much higher productivity rates and with better accuracies than was possible a decade or two ago. Consequently, advanced techniques such as 3D data acquisition and 3D forward/inverse modelling have become reality. It is therefore strongly advocated that coal industry practitioners remain up-to-date with the latest developments in geophysical research and development; this knowledge will aid them in the optimal selection and application of geophysical methods.

References DONNELLY, L.J. and MCCANN, D.M. 2000. The location of abandoned mine workings using thermal techniques. Engineering Geology, vol. 57. pp. 39–52.

Figure 6—2D resistivity image showing the topography of a dolomitic basement (Zhou et al., 2000)

DU PLESSIS, G.P. 2008. The relationship between geological structures and dolerite intrusions in the Witbank Highveld coalfield, South Africa. MSc thesis, University of the Free State, Bloemfontein.

DU PLESSIS, S.J. 2006. The merging of the aeromagnetic data of the Witbank

their relatively high conductivity the sandstone channels are associated with an increase in radio wave attenuation, and this is depicted by the warmer colours in the image.

Coalfield. Coaltech Research Association, Johannesburg.

DU PLESSIS, S.J. and SAUNDERSON, R.D. 2000. The successful prediction of nonmagnetic dykes using a Dighem survey. Coal Indaba, Johannesburg, 15–16 November 2000. Fossil Fuel Foundation, Johannesburg.

Dolomitic pinnacles/uneven basement The final case study example relates to the mapping of uneven floor conditions – typically associated with dolomitic basement structures. Unweathered dolomite usually has a relatively high electrical resistivity compared to overlying shale, coal, and sandstone layers. Variations in the floor topography such as pinnacles and depressions can be imaged using either the EM or resistivity method. Figure 6 shows an example of the application of the 2D electrical resistance tomography (ERT) method to this type of problem. It should be noted that the mapping accuracy of the surface through the ERT method decreases with increasing depth, and for deeper basements it may be better to resort to the TDEM method.

HU, M-S., PAN, D-M., DONG, S-H., and LI, J-J. 2012. GPR response characteristics of shallow loose coal seam. Geophysical and Geochemical Exploration, vol. 36, no. 4. pp. 599–606.

LANHAM, A. 2004. New Vaal gears up to meet power demand. Mining Weekly, 3 September 2004. http://www.miningweekly.com/print-version/new-vaalgears-up-to-meet-power-demand-2004-09-03 STOLARCZYK, L.G., PENG, S., and LUO, Y. 2003. Imaging ahead of mining with RIM-IV instrumentation and 3-D tomography software. 22nd International Conference on Ground Control in Mining, g Morgantown, WV, 5–7 August 2003.

STYLES, P. 2005. High resolution microgravity investigations for the detection and characterisation of subsidence associated with abandoned coal, chalk

Geophysics can play a significant role in addressing a wide range of coal mining and exploration problems. The primary advantage of using geophysics is that it often provides a noninvasive way of obtaining quantitative information about the subsurface geological structure and of potentially hazardous ground conditions. The application of geophysics can thus contribute to optimizing extraction and to mining safety: The Journal of The Southern African Institute of Mining and Metallurgy

and salt mines. Post-Mining 2005, Nancy, France.

VAN SCHOOR, M. and FOURIE, C.J.S. 2014 (eds.) (in press). A guide for applying geophysics to coal mining problems in South Africa. Randomstruik. Cape Town.

ZHOU, W., BECK, B.F., and STEPHENSON, J.B. 2000. Reliability of dipole-dipole electrical resistivity tomography for defining depth to bedrock in covered karst terranes. Environmental Geology, vol. 39, no. 7. pp. 760–766.

VOLUME 114

OCTOBER 2014

879

◆ ▲

Conclusions and recommendations

PRE-CONFERENCE WORKSHOPS Pre-conference workshops will be organised in conjunction with World Gold 2015: Geometallurgy Workshop Gold Processing Workshop

2015 Building a Resilient Gold Mining Industry

For further details contact: SAIMM, Head of Conferencing, Raymond van der Berg Tel: 27 (11) 834-1273/7 Facsimile 27 (11) 838-5923 E-mail: raymond@saimm.co.za · Website: www.saimm.co.za

28 September 2015 · Workshop 29 September–1 October 2015 · Conference 2 October 2015 · Technical Visits Misty Hills, Gauteng, South Africa Incorporating geology, metallurgy and mining

TOPICS World Gold 2015 will reflect on these key issues with greater focus on improved efficiency and the latest technology in: ➺ geological mapping ➺ gold mining ➺ mineral processing, and ➺ extraction and refining ➺ human resources ➺ financial resources ➺ computer assisted exploration targeting in gold exploration ➺ application of partial extraction methods in gold exploration ➺ geochemical and/or mineralogical haloes as indicator for gold targets ➺ Black Smokers as exploration targets for gold ➺ are there still mega gold deposits waiting to be discovered? ➺ Brownfields gold exploration, as success story.

The Southern African Institute of Mining and Metallurgy (SAIMM), the Canadian Institute of Mining, Metallurgy and Petroleum (CIM) and the Australasian Institute of Mining and Metallurgy (AusIMM) will jointly convene a World Gold Conference every two years. In 2015 it will be held in Johannesburg, South Africa and hosted under the auspices of the SAIMM. Some important aspects of the current mining environment will provide opportunities and threats for the industry in the foreseeable future, which include: ➺ Gold price volatility ➺ Minimal exploration success for the last 10 years and little immediate prospect for revolutionary success is leading to revisiting known old (sub-marginal) deposits ➺ Lower precious metal content ➺ Increasing refractoriness ➺ More energy efficient mining and processing ➺ Decreasing equity risk for juniors and mid-tiers ➺ Maximising long-term optionality

Thus, we are focusing on more environmentally friendly, resource efficient and energy efficient mining and recovery methods.

Conference Chair: Andries Swart AngloGold Ashanti

The Canadian Institute of Mining, Metallurgy & Petroleum

Sponsor:

SPONSORSHIP Sponsorship opportunities are available.

Companies wishing to sponsor or exhibit should contact the Conference Coordinator.

Conference Announcement

INTERNATIONAL ACTIVITIES 20–24 October 2014 — 6th International Platinum Conference Sun City, South Africa

E-mail: raymond@saimm.co.za

E-mail: raymond@saimm.co.za

14–17 June 2015 —

14–17 June 2015 —

3–7 November 2014 — 16–20 June 2015 — E-mail: r.prior@mweb.co.za 12 November 2014 —

6–8 July 2015 — Copper Cobalt Africa Incorporating The 8th Southern African Base Metals Conference Zambezi Sun Hotel, Victoria Falls, Livingstone, Zambia E-mail: raymond@saimm.co.za

E-mail: raymond@saimm.co.za 18–19 November 2014 —

13–15 July 2015 — Production of Clean Steel Emperors Palace, Johannesburg E-mail: yolanda@saimm.co.za 15–17 July 2015 — Virtual Reality (VR) applications in the mining industry

19–20 November 2014 — Emperors Palace, Hotel Casino Convention Resort, Johannesbur

E-mail: raymond@saimm.co.za 11–14 August 2015 —

11–13 March 2015 — E-mail: raymond@saimm.co.za 8–10 April 2015 — 5th Sulphur and Sulphuric Acid 2015 Conference Southern Sun Elangeni Maharani KwaZulu-Natal, South Africa

E-mail: camielah@saimm.co.za 28 September-2 October 2015 — Misty Hills Country Hotel and Conference Centre, Cradle of Humankind, E-mail: raymond@saimm.co.za 12–14 October 2015 —

E-mail: camielah@saimm.co.za 12–13 May 2015 — E-mail: raymond@saimm.co.za 28–30 October 2015 — E-mail: yolanda@saimm.co.za 27–28 May 2015 — E-mail: raymond@saimm.co.za E-mail: aims@bbk1.rwth-aachen.de

8–13 November 2015 —

9–10 June 2015 —

The Journal of The Southern African Institute of Mining and Metallurgy

OCTOBER 2014

vii

▲

Raj Singhal, E-mail: singhal@shaw.ca or E-mail: raymond@saimm.co.za, Website: http://www.saimm.co.za

Company Affiliates The following organizations have been admitted to the Institute as Company Affiliates

â&#x2013;˛

viii

OCTOBER 2014

The Journal of The Southern African Institute of Mining and Metallurgy

Forthcoming SAIMM events...

IP PONSORSH EXHIBITS/S ng to sponsor ishi e Companies w t at any of thes and/or exhibi contact the events should rdinator -o conference co ssible as soon as po

2014 ◆

F

6th International Platinum Conference 20–24 October 2014, Sun City, South Africa

◆ 12 November 2014, ◆ 19–20 November 2014, Emperors Palace, Hotel Casino Convention Resort, Johannesbur

2015 ◆ 11–13 March 2015, ◆

Sulphur and Sulphuric Acid 2015 Conference 8–10 April 2015, Southern Sun Elangeni Maharani KwaZulu-Natal, South Africa

◆ 12–13 May 2015, Johannesburg, South Africa ◆ 9–10 June 2015, ◆

◆

◆

Copper Cobalt Africa Incorporating The 8th Southern African Base Metals Conference 6–8 July 2015, Zambezi Sun Hotel, Victoria Falls, Livingstone, Zambia Production of Clean Steel 13–15 July 2015, Emperors Palace, Johannesburg Virtual Reality (VR) applications in the mining industry 15–17 July 2015

◆

For further information contact: Conferencing, SAIMM P O Box 61127, Marshalltown 2107 Tel: (011) 834-1273/7 Fax: (011) 833-8156 or (011) 838-5923 E-mail: raymond@saimm.co.za

11–14 August 2015, ◆ 28 September-2 October 2015,

Website: http://www.saimm.co.za

Integrated systems of support

Applying Poka Yokes in the mining industry

+27 11 494 6000 www.ncm.co.za