JAMRIS 2011 Vol 5 No 4

JOURNAL of AUTOMATION, MOBILE ROBOTICS & INTELLIGENT SYSTEMS

Editor-in-Chief Janusz Kacprzyk (Systems Research Institute, Polish Academy of Sciences; PIAP, Poland)

Associate Editors: Mariusz Andrzejczak (BUMAR, Poland) Katarzyna Rzeplińska-Rykała (PIAP, Poland)

Co-Editors: Oscar Castillo

Statistical Editor: Małgorzata Kaliczyńska (PIAP, Poland)

(Tijuana Institute of Technology, Mexico)

Dimitar Filev

Webmaster: Tomasz Kobyliński tkobylinski@piap.pl

(Research & Advanced Engineering, Ford Motor Company, USA)

Kaoru Hirota Editorial Office: Industrial Research Institute for Automation and Measurements PIAP Al. Jerozolimskie 202, 02-486 Warsaw, POLAND Tel. +48-22-8740109, office@jamris.org

(Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology, Japan)

Witold Pedrycz (ECERF, University of Alberta, Canada)

Roman Szewczyk (PIAP, Warsaw University of Technology, Poland)

Executive Editor: Anna Ładan aladan@piap.pl

Copyright and reprint permissions Executive Editor

Editorial Board: Chairman: Janusz Kacprzyk (Polish Academy of Sciences; PIAP, Poland) Plamen Angelov (Lancaster University, UK) Zenn Bien (Korea Advanced Institute of Science and Technology, Korea) Adam Borkowski (Polish Academy of Sciences, Poland) Wolfgang Borutzky (Fachhochschule Bonn-Rhein-Sieg, Germany) Chin Chen Chang (Feng Chia University, Taiwan) Jorge Manuel Miranda Dias (University of Coimbra, Portugal) Bogdan Gabryś (Bournemouth University, UK) Jan Jabłkowski (PIAP, Poland) Stanisław Kaczanowski (PIAP, Poland) Tadeusz Kaczorek (Warsaw University of Technology, Poland) Marian P. Kaźmierkowski (Warsaw University of Technology, Poland) Józef Korbicz (University of Zielona Góra, Poland) Krzysztof Kozłowski (Poznań University of Technology, Poland) Eckart Kramer (Fachhochschule Eberswalde, Germany) Piotr Kulczycki (Cracow University of Technology, Poland) Andrew Kusiak (University of Iowa, USA) Mark Last (Ben–Gurion University of the Negev, Israel) Anthony Maciejewski (Colorado State University, USA) Krzysztof Malinowski (Warsaw University of Technology, Poland) Andrzej Masłowski (Warsaw University of Technology, Poland)

Patricia Melin (Tijuana Institute of Technology, Mexico) Tadeusz Missala (PIAP, Poland) Fazel Naghdy (University of Wollongong, Australia) Zbigniew Nahorski (Polish Academy of Science, Poland) Antoni Niederliński (Silesian University of Technology, Poland) Witold Pedrycz (University of Alberta, Canada) Duc Truong Pham (Cardiff University, UK) Lech Polkowski (Polish-Japanese Institute of Information Technology, Poland) Alain Pruski (University of Metz, France) Leszek Rutkowski (Częstochowa University of Technology, Poland) Klaus Schilling (Julius-Maximilians-University Würzburg, Germany) Ryszard Tadeusiewicz (AGH University of Science and Technology in Cracow, Poland)

Stanisław Tarasiewicz (University of Laval, Canada) Piotr Tatjewski (Warsaw University of Technology, Poland) Władysław Torbicz (Polish Academy of Sciences, Poland) Leszek Trybus (Rzeszów University of Technology, Poland) René Wamkeue (University of Québec, Canada) Janusz Zalewski (Florida Gulf Coast University, USA) Marek Zaremba (University of Québec, Canada) Teresa Zielińska (Warsaw University of Technology, Poland)

Publisher: Industrial Research Institute for Automation and Measurements PIAP

If in doubt about the proper edition of contributions, please contact the Executive Editor. Articles are reviewed, excluding advertisements and descriptions of products.

The Editor does not take the responsibility for contents of advertisements, inserts etc. The Editor reserves the right to make relevant revisions, abbreviations and adjustments to the articles.

All rights reserved ©

1

JOURNAL of AUTOMATION, MOBILE ROBOTICS & INTELLIGENT SYSTEMS VOLUME 5, N° 4, 2011

CONTENTS 3

35

Simulation of semiautonomy mode for ibis mobile robot with analysis of sensor failure tolerance M. Trojnacki, P. Bigaj, J. Bartoszek

Modeling and identification of actuator for flap deflection M. Ulinowicz, J. Narkiewicz 41

11

Improvisation process for banks capacitor compensation applied fuzzy logic knowledge for nodal detection on electrical network B. Gasbaoui, M. Oudda, A. Nasri

DiaSter - Intelligent system for diagnostics and automatic control support of industrial processes M. Syfert, P. Wnuk, J.M. Kościelny 47

Eye trackers in quality evaluation of compressed video A. Ostaszewska-Liżewska, R. Kłoda, S. Żebrowska-Łucyk, M. Liżewski

17

Application of dependence graphs and game trees for decision decomposition for machine systems A. Partyka, A. Deptuła 27

Hardware-software platform for integrated circuit technology learning and design via Internet V.V. Nelayev, M. Najbuk, T. Breczko 30

Adaptive control of frictional contact models for nonholonomic wheeled mobile robot R. Vivekananthan, L. Karunamoorthy

2

51

Specific issues in management of large international research projects in the field of security and defence A. Bukała, M. Andrzejczak, A. Wołoszczuk

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N° 4

2011

Simulation of Semiautonomy Mode for IBIS Mobile Robot with Analysis of Sensor Failure Tolerance Submitted 12th April 2011; accepted 6th June 2011

Jakub Bartoszek, Maciej Trojnacki, Piotr Bigaj

Abstract:

This work is concerned on sensitivity analysis of semiautonomy algorithm of mobile combat robot to environmental sensors’ damage. The construction of the robot, semiautonomy algorithm and used sensors have been described. This algorithm takes into account environmental sensors’ damage. Simulation research results of semiautonomy algorithm using Matlab/Simulink package was presented. This research was performed for normal environmental sensors’ operation and for selected sensors’ damage. On that basis, sensitivity of semiautonomy algorithm to selected environmental sensors damage was tested.

Keywords: mobile robot, semiautonomy, sensor failure.

1. Introduction

Extraordinarily large funds and and the work of many research centres around the globe are focused on developing increasingly efficient control algorithms for mobile robots. The main objective behind most this work is to achieve a level of development mobile robotics described as full autonomy. The moment in which this goal is achieved will have a profound effect on the direction and speed of the development of our civilisation. Achieving full autonomy will allow for a great increase in functionality and the number of possible uses for autonomous vehicles. The most important asset of these vehicles is going to be their lack of necessity for them to be controlled by a human operator and their need for only minimal supervision.

One of the most discussed issues in the case of research on mobile robot autonomy is the problem of them moving in a dynamic, highly diverse and unknown environment [1]. To put it simply, the issue can brought down to plotting a movement trajectory for the robot, from its current location to a designated target in a way that allows the robot to reach its destination in the shortest time possible while at the same time avoiding collisions [2]. Algorithms enabling the selection of an optimal path based on various quality indicators are being developed in research centres all over the world. Each algorithm makes use of at least 2 types of sensors – state and environment.

The state sensors include all of the sensors that allow for the determination of the robot’s current status. In most cases they are only able to determine the robot’s position.

The methods used to determine position based on available data can be divided into several separate groups.

The first of these includes the use odometry and often involves an error rapidly growing as the driver program loops. The second method relies on knowledge of a statistical map of the terrain and determining position based on markers set up on the robot’s route or based on landmarks identified with the use of visual sensors, for example. The third group is based on GPS and similar systems.

Each of these methods is not fully reliable and must involve safeguards to protect the program from incorrect data.

The data may be a result of a sensor malfunction or, in the case of systems similar to GPS, lack of satellite contact.

Environmental sensors are a group that enables the modelling of a virtual environment, based on the actual environment in which the robot is currently operating.

Depending on the number and type of the sensors used, the model may be 3 or 2 dimensional. Based on the power of the robot’s CPU this model make take various forms.

An approach in which a representation of the model of the environment in which the robot was operating has been used in publications [3], [4], it has been omitted, and control was outlined based only on sensor signals in, for example, publication [5]. The main characteristic of this sensor group is their limited range. In the case of mobile robots designed to work in outdoor environments, these sensors share one other common characteristic – they are vulnerable to various malfunctions. These may be caused by external factors such as collisions with potential obstacles or vibrations caused by the robot’s movement they can also be caused by internal factors such as overheating electronic components.

In most cases, sensor malfunctions are extremely difficult to predict. Their nature is often possible to determine only after they have occurred and need to be properly diagnosed. This is why the methods of diagnosing and reacting to sensor malfunctions needs to be included in the first stage of designing the entire system. The literature mentions 2 approaches to this problem [6], [8].

One of them is based on a model enabling the detection of sensor malfunctions [9]-[12]. Some of them require interference into the structure of the system because of the detected malfunction [13].

Another, completely different approach to the problem of detecting and reacting to a sensor malfunctions has been presented by Martin Soika in [6]. The goal of his work was the development of a method that will enable the detection of a previously unnoticed sensor malfunction and an adequate reaction to it. His method is based on determining the level of credibility of the data sent by the sensors. Healey [14] created a system based on a model of sensor malfunctions, however he utilised an artificial neuron net to detect the malfunctions Articles

3

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

themselves. This solution enables the implementation of the method into the robot’s driver in a way not limiting its autonomy, unfortunately this also increases the complexity of the algorithm.

An important factor enabling automated detection and diagnosis of sensor damage is their redundancy. In this case, the environmental model is being created based on data from a larger number of sensors, which allows for relatively easy detection of a potential malfunction in one of them. A larger number of sensors, often using different measurement methods, requires the use of more elaborate and better optimized algorithms designed for constructing a model of the robot’s surrounding environment. This in turn necessitates the use of more efficient CPUs in the robot’s driver. Because of this, it is necessary to give up on sensor redundancy in some cases.

This article presents the methods used in reacting to sensor malfunctions used in the control algorithm for the IBIS robot. This algorithm has been described in greater detail in [15].

2. Mobile combat robot

The commercial version of the IBIS (Fig. 1a) is a pyrotechnical and combat robot. It has been adapted to operate on difficult and diverse terrain, such as snow, sand or rocky terrain. The robot’s high speed enables it to perform its tasks dynamically. The manipulator gives it a long range of operation, while the design of its drive gives it a great deal of movement flexibility as far as speed is concerned. a)

b)

Fig. 1. IBIS robot: a – commercial version with manipulator, b –autonomy research variant.

4

Articles

N° 4

2011

The basic technical specifications of the robot are as follows: mass: 295 kg, dimensions (length x width x height): 1,3 m x 0,85 m x 0,95 m, maximum speed: 8,5 km/h, manipulator lifting capacity: 30 kg, manipulator maximum range: 3,15 m [16], [17].

A version of the IBIS without the manipulator arm has been developed for research purposes (Fig. 1b).

This version of the robot has been also equipped with an additional installable frame. This contains the module of the robot responsible for performing semiautonomy.

The module contains microprocessors as well as a set of four cameras and sensors designed to detect and locate obstacles. The robot can operate in two modes: teleportation – when controlled by an operator, semiautonomy – when it performs its tasks by itself under the supervision of an operator.

3. Robot sensors

The mobile base of the robot has been extended with a specially designed frame enabling precise positioning of sensors as well as navigational and positioning controllers. The physical architecture of the system is comprised of 4 blocks separated physically and functionally: sensors, positional controller (determining the robot’s current position), navigational controller (processing outgoing data and facilitating control) and the engine drivers. Movement controls in teleportation mode is transmitted to the robot via an ISM modem, while movement in semiautonomy mode is calculated entirely by the navigational controller.

The localization sensors are mounted on the positional controller and are used to determine the robot’s position and orientation. Position is determined in 3 dimensions: longitude, latitude and altitude according to NMEA specifications, thanks to this it can be presented in any form of GIS programming. A monophase GPS receiver supplemented by INS is being used to pinpoint the current position of the robot. Pinpointing of the robot’s coordinates in the WGS-84 system is achieved with the use of Kalaman filtration, GPS positioning and inertial navigation. The robot’s azimuth is determined by use of a digital compass with inclination compensation. Inclination sensors (inclinometers and accelerometers) are being used to determine the robot’s inclination and declination.

Four types of obstacle detection sensors have been mounted on the frame: a 2 dimensional laser scanner, ladars, true-presence radar sensors and tactile sensors (bumpers). These sensors are placed in a way that (Fig. 2), they are able to cover the entire area around the robot and so that most of the gathered information concerns the front.

The purpose of the 2D laser scanner is the detection of obstacles in front of the robot. Its angle scope has been set to 100º. Its maximum range of 80 m is the result of technical limitations of the sensor itself, however the present LOS depends on the angle, at which the beam is directed.

The scanner is tilted downward at a small angle, which enables it to detect obstacles smaller than the height on which the scanner is mounted.

Ladars are used to detect obstacles in the immediate vicinity of the robot. They are mounted at various angles facing downwards, which allows them to detect both concave and convex obstacles. One of them is tilted

Journal of Automation, Mobile Robotics & Intelligent Systems

Fig. 2. Sensor placement: a – left side view, b – top side view. upwards in order to detect obstacles that are too low for the robot. The ladars used enable measuring the distance to the nearest obstacle within a range of 0,5–10 m.

Radar sensors are designed to detect obstacles at long range. The only information they provide is, whether or not there is an object at the specified distance. They work at ranges of 2–15 m. They allow for early warnings about obstacles within the robot’s vicinity.

The tactile sensors are tasked with detecting obstacles that have not been detected by the other sensors i and triggering an emergency stop in case the robot collides with something.

4. Semiautonomy algorithm

None of the algorithms designed so far have allowed mobile robots to perform their tasks without error. This is due to the fact that the creators of algorithms are unable to predict all possible situations, as well as sensors being unable to detect and classify all obstacles. Because of this, the following article uses the term, semiautomy.

This means that the robot will perform its task by itself, but under the supervision of an operator, who can halt the robot at any time.

The goal of the research described in this paper was to analyse the effect of sensor malfunctions on the algorithm enabling the IBIS to perform its objective, while avoiding detected obstacles.

The control algorithm is comprised of 4 subroutines.

The first one of these is responsible for turning the robot towards the target. The second of these is responsible for setting the value of velocity depending on the distance between the robot and the nearest obstacle. If the robot is far away from the target, and the nearest obstacle is located beyond sensor range, the robot will move at maximum speed. The third behaviour is connected to the laser scanner, which is treated as 101 single beams. Modified VHF

VOLUME 5,

N° 4

2011

algorithms [18], [19] are used in order to process the environmental data received from the scanner.

The fourth subroutine is patterned on the Braitenberg algorithm, which is based on directly connecting the sensors with actuators, with every connection having its own weighting factor. Depending on the sensor readings and weighting factor the robots able to perform various tasks. The semiauthonomy algorithm as well as its division into separate subroutines has been described in greater detail in [15].

The weighting factor for each of the subroutines is determined based on the active routine.

If the current routine is ineffective the virtual tank (WaterTank) overflows and a different routine is activated. The new routine is decided based on an assessment of the locations of the target and obstacles in relation to the robot.

Eg., if there are obstacles in front of the robot and the robot must go forward and turn right to reach it, at the moment when the „move to target and avoid obstacles” routine is interrupted, the robot may switch to the „observe obstacles on the left side” routine.

The WaterTank is first emptied before switching to a new routine.

One of the most important characteristics of modern control algorithms is their resistance to sensor malfunctions. Malfunctioning sensors may cause significant changes to the robot’s behaviour which may cause damage to it or pose a threat to people in the vicinity. In order to avoid such situations the presented algorithm has been modified in order to detect improper data transmitted by damaged sensors.

The location sensor in the form of a GPS receiver will trigger a switch from semiautonomy toteleoperation mode if the signal from the transmitters is lost or there are high discrepancies in the robot’s perceived location (above 10 m). The main factors determining the proper functioning of the localisation sensors are discrepancies in the robot’s location and its distance to the target in each iteration of the driver.

The environmental sensors on the robot are seven ladars and a 2D scanner. A malfunction in these sensors may manifest itself in different ways, depending on its type.

The first type of malfunction appears when it is impossible to establish contact with the sensor or if the sensor itself is sending an error message after it has been initialised. In such a case, the microprocessor responsible for gathering information about the sensors and communication with them sends a malfunction message to the main microprocessor. This information is sent in the same way as the proper value sent back by the sensors but the value is 0.01 m. This is equivalent to a situation in which an obstacle is at that distance from the sensor, which is impossible under normal working conditions.

Articles

5

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

The second type of malfunction is harder to detect. It appears when the sensor is returning false data. A malfunction of this type may be caused by multiple factors such as for example large vibrations of the mobile platform which may occur during fast movement on rough terrain or the deflection of a laser beam of a flat surface tilted at a low angle. Detecting and filtering false data in such cases is extremely difficoult. In steering algorithm only values going beyond measurement range which arising from the construction of a mobile robot and installing manner of sensors are filter off.

Detecting and diagnosing a malfunction causes a change in the weighting factor for the proper subroutine.

In the case of detecting a sensor malfunction the weighting for subroutine 3 is set to zero, which means it no longer has any effect on the functioning of the algorithm and the robot itself. In the case of one or more ladars malfunctioning the weighting for subroutine four is being changed. The weighting is changed based on reliability, which is determined using the method shown below.

The controls for the robot’s wheels are determined depending on: 4

ui = ∑ c j uij ,

(1)

j =1

where: i = {L, R}, L and R – respectfully the left and right wheels of the robot, j – the number of the subroutine, cj – the weighting factor of subroutine j defining the influence this algorithm has over general control, uij – controls for wheel i and subroutine j.

The weighting for particular subroutines is as follows: c1 = 1 (malfunctions of the sensor related to the determination of the robot’s orientation are not being analysed), c2 = 1 (malfunctions of the sensor related to the robot’s speed are not analysed), c3 = 0 in case of a laser scanner * malfunction, c3 = c3 = 0,6 if the scanner is functional, while the weighting for the last subroutine is calculated with:

c4 = c4* ∆

(2)

c3 + c 4 < 1 .

(3)

An approach has been adopted in which the robot’s maximum speed is limited in the case of an environmental sensor malfunction, in order to ensure safety for the robot. In a different approach the weighting could be increased to the subroutine, for which the sensors are functional, so that c3 + c4 = 1.

The reliability factor is determined in relation:

1 ∆= − 0,13 n + 0,88

The changing of value for this factor in relation to the number of malfunctioning ladars has been illustrated in Fig. 3.

5. Simulation reasearch

In order to perform simulation research on autonomy methods, a simulated environment has been prepared based on the Matlab/Simulink packet. The programming responsible for environmental simulations and determining the data displayed by virtual sensors has been separated for the programming performing the robot’s control algorithm in teloperation and semiautonomy modes.

This second programming has been prepared so that, to implement it into the robot’s microcontroller quickly and easily. The robot’s simulated environment has been prepared with the V-Realm Builder program objects and scenes have been approximated with the use of cuboids and cylinders.

This work presents the research data in the case where 4 obstacles were placed on the robot’s route to the target.

The robot preformed the complex „move towards the target and avoid obstacles” routine. As part of the simulation research the mobile robot’s semiautonomy algorithm has been evaluated in terms of vulnerability to the malfunction of select environmental sensors.

In order to receive a conclusive assessment of the results, the following quality indicators have been input: a) The square sum of the robot’s distance to the target n

E = ∑ ei2 ∆t ,

where: n means the number of malfunctioning ladars.

6

Articles

(5)

i =1

where ei – the robot’s distance to the target in the i iteration, n – the number of iterations before the robot reaches its target or the simulation is ended before the target is reached, b) Standard variance in the robot’s speed n

S=

∑ (v i =1

i

− v )2

n −1

,

(6)

where: vi – robot’s speed in the i iteration, v – average movement speed, c) Length of the route from the starting position to the target n

(4)

2011

Fig. 3. Change in the reliability factor in relation to the number of malfunctioning ladars.

*

where: c 4 = 0,4 means the weighting of the subroutine related to the ladars if all of them are functional, ∆ – the factor of reliability of the information coming from the ladars depending on the damage sustained by them.

It is worth noting that in case of a malfunction of any of the sensor there is a relation:

N° 4

s = ∑ vi ∆t ,

(7)

i =1

d) The time it takes for the robot to reach the target T, assuming the target is achieved for e ≤ 0,5 [m],

Journal of Automation, Mobile Robotics & Intelligent Systems

e) Robot’s maximum speed vmax (within a timeframe from t0 to T, where t0 is the moment when the sensor malfunction occurred), f) Robot’s average speed v (within a timeframe from 0 to T).

It should be noted that for designators a – d one should move towards minimallisation, while for designators e and f their maximalisation.

In the case in which the robot cannot reach the target within the assumed time Tmax, it is assumed that quality designators s and T reach the value of +∞, while remaining designators reach the values calculated for Tmax.

This work assumes that Tmax = 100 [s].

VOLUME 5,

N° 4

2011

Simulation 1 – all environmental sensors are functional

The first presented simulation has been preformed with the assumption that all o the robot’s environmental sensors are functional. The results of this simulation presented in Fig. 4 and in Tab. 1, are used as reference for the remaining simulations, in which the malfunctions of select sensors are assumed.

The ladar indicators shown in Fig. 4e are consistent with Fig. 2. While marking the laser scanner beams on pic. 4f they have been designated Lα, where α is the scanner’s angle (in degrees). A positive angle means a tilt towards the left while a negative one means a tilt towards the right.

The readings from every tenth scanner beam have been shown in Fig. 4f. The control values for the left and right

Quality indicators

E [m2s]

S [m/s]

s [m]

T [s]

vmax [m/s]

_ v [m/s]

Value

8 533

0,284

28,2

38,8

1,34

0,726

Tab. 1. Received values for quality indicators in simulation 1.

a

b

c

d

e

f

g

h

i

Fig. 4. Simulation 1 results: a – robot’s movement trajectory, with the location of the target and obstacles being marked, b – linear speeds for the robot’s left and right wheels, c – robot’s movement speed and changes in the WaterTank parameter, d – robot’s distance and angle to the target, e – ladar readings, f – scanner readings for select 11 beams, g-i – control values and weighting for specific routines.

Articles

7

Journal of Automation, Mobile Robotics & Intelligent Systems

wheels of the robot (uLi and uRi, i = 1 ... 4) for the previously described subroutines as well as general control (uL and uR) have been presented in Figs. 4g and 4h. The weighting for each of the robot’s subroutines remains the same (Fig. 4i) during the robot’s movement. The weighting is changed only if an environmental sensor malfunction is detected.

The simulations indicate that the developed method enables the robot to avoid obstacles and reach its designated target. The robot moves at a reduced speed, which is related to the fact that it moves slower the closer it is to an obstacle. The distance between the target and the robot decreases constantly during its movement. While evading obstacles, the work of the ladars and laser scanners becomes visible; their readings are subject to noticeable change. The effectiveness of the routine is deter-

1)

VOLUME 5,

N° 4

2011

mined with the WaterTank parameter, which has the largest values when the robot is between obstacles, which is visible in Fig. 4c.

Simulation 2 – the ladar marked as ML malfunc-

tions after the time of 1 s The second simulation has been performed in the case of a malfunctioning the ladar marked in Fig. 2 as ML. The malfunction occurred after 1 s of the simulation. The results of this simulation have been illustrated in Fig. 5 and in Tab. 2. After the ladar malfunction has been detected by the semiautonomy algorithm the weighting has been modified for the first subroutine (Fig. 5i) related to the modified Braitenberg algorithm based on readings from the ladars.

Quality indicator

E [m2s]

S [m/s]

s [m]

T [s]

vmax [m/s]

_ v [m/s]

Value

10 459

0,282

29,1

57,0

1,36 1)

0,511

Maximum values at the begining have been omited, as they are the effect of the algorithm’s functions before the malfunction.

Tab. 2. Received values for quality indicators in simulation 2. a

b

c

d

e

f

g

h

i

Fig. 5. Simulation 2 results: a – robot’s movement trajectory, with the location of the target and obstacles being marked, b – linear speeds for the robot’s left and right wheels, c – robot’s movement speed and changes in the WaterTank parameter, d – robot’s distance and angle to the target, e – ladar readings, f – scanner readings for select 11 beams, g-i – control values and weighting for specific routines.

8

Articles

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

2011

at a lower and more stable speed. When compared with Simulation 2 the robot achieved its objective faster with a higher average speed which is related to the fact that the robot had less information about obstacles so it moved more freely.

As part of this work simulation research has been conducted assuming the malfunction o remaining sensors in the case of a single sensor malfunctioning.

Based on all simulations preformed it can be concluded that, in most cases, the developed semiautonomy algorithm is capable of dealing well with the malfunction of a single sensor.

The robot animations for the described simulations have been done with the Simulink 3D Animation tool and can be found under address [17].

Be comparing the results of simulations 1 and 2 it can be concluded that in the case of a malfunction in the ML ladar, the robot achieved its target after a longer time.

Lower values for quality indicators E, s, T and v have been noted. While indicators S and vmax achieved marginally higher values.

Simulation 3 – laser scanner malfunctions – after a time of 1 s The last simulation has been performed under the assumption that the laser scanner has malfunctioned after 1 s. The results of the simulation are displayed in Fig. 6 and in Tab. 3.

In this case lower values for all indicators except indicator S have been noted when compared with Simulation 1 which is related to the fact that the robot was moving Quality indicator

E [m2s]

S [m/s]

s [m]

T [s]

vmax [m/s]

v [m/s]

Value

10 564

0,177

28,9

47,2

0,85

0,6133

2)

N° 4

2

Maximum values at the begining have been omited, as they are the effect of the algorithm’s functions before the malfunction.

Tab. 3. Received values for quality indicators in simulation 3. a

b

c

d

e

f

g

h

i

Fig. 6. Simulation 3 results: a – robot’s movement trajectory, with the location of the target and obstacles being marked, b – linear speeds for the robot’s left and right wheels, c – robot’s movement speed and changes in the WaterTank parameter, d – robot’s distance and angle to the target, e – ladar readings, f – scanner readings for select 11 beams, g-i – control values and weighting for specific routines.

Articles

9

Journal of Automation, Mobile Robotics & Intelligent Systems

6. Summary and conclusions for further research

The results of research into the vulnerability of mobile robot semiautonomy algorithms to environmental sensor malfunctions have been described in this paper. The developed semiautonomy algorithm included the possibility of a sensor malfunction, appropriately modifying the robot’s subroutines. The results of simulation research carried out with the Matlab/Simulink packet for cases of full functionality and single sensor malfunctions, has been presented. A comparison of the results using 6 quality indicators has been carried out. The data gathered indicates that the developed semiautonomy algorithm exhibits high resistance to sensor malfunctions. What’s more, in the case of malfunctions in some sensors better values in the quality indicators are achieved. This behaviour can be explained by the robot passing closer to some obstacles because of the malfunction and as a result shortening its route. However, this does not mean that the algorithm functions better under these conditions or that the malfunctioning sensor should be.

Further research will focus on simulating the malfunctions of a larger number of sensors as well as the implementation of a modified semiautonomy algorithm in the mobile robot and conducting experimental research related to analysing the vulnerability of this algorithm to environmental sensor malfunctions.

ACKNOWLEDGMENTS

The paper is a result of PROTEUS Project (POIG.01.01.0200-014/08) and is co-financed by The European Union from The European Regional Development Fund under the Operation Programme Innovative Economy, 2007-2013.

AUTHORS

Jakub Bartoszek*, Maciej Trojnacki, Piotr Bigaj – Industrial Research Institute for Automation & Measurements PIAP, Al. Jerozolimskie 202, PL-02-486 Warsaw, Poland *Corresponding author. E-mail: jbartoszek@piap.pl

References

[1] A. Wołoszczuk, M. Andrzejczak, P. Szynkarczyk, “Architecture of mobile robotics platform planned for intelligent robotic porter system – IRPS project”, Journal of Automation, Mobile Robotics & Intelligent Systems, vol. 1, 2007, pp. 59-63.

[2] M. Trojnacki, P. Szynkarczyk, “Tendencje rozwoju mobilnych robotów lądowych (3). Autonomia robotów mobilnych – stan obecny i perspektywy rozwoju”, Pomiary Automatyka Robotyka, no. 9, 2008, pp. 5-9. (in Polish) [3] A. Ghorbani, S. Shiry and A. Nodehi, “Using Genetic Algorithm for a Mobile Robot Path Planning”. In: Proc. of International Conference on Future Computer and Communication, 2009, Kuala Lumpur, Malaysia, pp. 164-166.

[4] M. Wang, J. N. K. Liu, “Fuzzy logic-based real-time robot navigation in unknown environment with dead ends”, Robotics and Autonomous Systems, vol. 56, 2008, pp. 625-643.

10

Articles

VOLUME 5,

N° 4

2011

[5] X.-J. Jing, “Behaviour dynamics based motion planning of mobile robots in uncertain dynamic environments”, Robotics and Autonomous Systems, vol. 53, 2005, pp. 99-123.

[6] M. Soika, “A sensor failure detection framework for autonomous mobile robots”. In: Proceedings of the 1997 IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS ‘97, 1997, vol. 3, pp. 1735-1740.

[7] N. Ranganathan, M. I. Patel, R. Sathyamurthy, “An intelligent system for failure detection and control in an autonomous underwater vehicle”, IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans, vol. 31, 2001, pp. 762-767.

[8] K. Bouibed, A. Aitouche, M. Bayart, “Sensor fault detection by sliding mode observer applied to an autonomous vehicle”. In: Proc. of International Conference on Advances in Computational Tools for Engineering Applications, ACTEA ‘09, 2009, pp. 621-626.

[9] I. J. Rudas, I. Ori, A. Toth, “Design methodology and environment for robot diagnosis”. In: Proceedings of IEEE International Symposium on Industrial Electronics, ISIE’93, Budapest, 1993, pp. 367-372.

[10] K. Kroschel, A. Wernz, “Sensor Fault Detection and Localisation Using Decorrelation Methods”, A Sesnsors and Actuators, vol. 25-27, 1991, pp. 43-50.

[11] G. J. S. Ra, S. E. Dunn, “On-Line detection for AUV”. In: IEEE Symp. Autonomous Underwater Vehicle Technology, 1994, pp. 383-392.

[12] T. J. Farrel, B. Appleby, “Using leraning techniques to accommodate unanticipated faults”. In: IEEE Trans. Control Syst. Technol., 1993, pp. 40-49.

[13] M. L. Visinsky, J. R. Cavallaro, J. D. Walker, “Expert System Framework for Fault Detection and Fault Tolerance in Robotics”, Computers & Electrical Engineering, vol. 20, no. 5, 1994, pp. 421-435.

[14] A. Healey, “Toward an automatic health monitor for autonomous underwater vehicles using parameter identification”. In: Amer. Control Conf., 1993, pp. 585-589.

[15] P. Bigaj, M. Trojnacki, J. Bartoszek, “Robot IBIS – realizacja ruchu w trybie teleoperacji i semiautonomii”, Prace naukowe Politechniki Warszawskiej, series: Elektronika, Zeszyt 175, vol. 1, 2010, pp. 135-148. (in Polish) [16] IBIS Robot designed by PIAP – an offer (24/11/2010). Available at: http://www.antyterroryzm.com/robot_bojowy.php (in Polish) [17] http://www.youtube.com/user/osmpiap [18] J. Borenstein, Y. Koren, “High-speed obstacle avoidance for mobile robots”. In: Proceedings., IEEE International Symposium on Intelligent Control, 1998, pp. 382-384.

[19] I. Ulrich, J. Borenstein, „VFH+: reliable obstacle avoidance for fast mobile robots,” in Robotics and Automation, 1998. Proceedings. 1998 IEEE International Conference on, 1572-1577, 1998.

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N째 4

2011

Improvisation Process for Banks Capacitor Compensation Applied Fuzzy Logic Knowledge for Nodal Detection on Electrical Network Submitted 26th March 2011; accepted 19th July 2011

Brahim Gasbaoui, Meriem Oudda, Abdelfatah Nasri

Abstract:

The main problem of electrical distribution systems is the reactive power flow. It causes reduction of active power transmission, diminishes power losses, and augments the drop voltage. In this research we described an efficiency approach FLC-HSO to solve the optimal power flow (OPF) combinatorial problem. The proposed approach employ tow algorithms, Fuzzy logic controller (FLC) algorithm for nodal detection and harmony search optimization (HSO) algorithm for optimal seizing capacitor of OPF combinatorial problem control variables. HSO method is more proficient in improving combinatory problem. The proposed approach has been examined and tested on the standard IEEE 57-bus test system with different objectives that reflect cost function minimization, voltage profile improvement, and voltage stability enhancement. The proposed approach results have been compared to those that reported in the literature recently. The results are promising and show the effectiveness and robustness of the proposed approach.

Keywords: capacitor placement, fuzzy logic, Harmony Search Optimization (HSO), capacitor seizing, power flow.

1. Introduction

Power distribution systems from electric power plants to ultimate consumers are accomplished via the transmission system, and distribution lines. Studies have indicated that as much as 13% of total power generated is consumed as R2 losses at the distribution level. The R2 losses can be separated to active and reactive component of branch current, where the losses produced by reactive current can be reduced by the installation of shunt capacitors. Capacitors are widely used in distribution systems to reduce energy and peak demand losses, release the KVA capacities of distribution apparatus, and to maintain a voltage profile within permissible limits. The objective of optimal capacitor placement problem is to determine the size, type, and location of capacitor banks to be installed on radial distribution feeders to achieve positive economic response. The economic benefits obtained from the loss reduction weighted against capacitors costs while keeping the operational and power quality constraints within required limits.

Fuzzy theory was first proposed and investigated by Prof. Zadeh in 1965.The Mamdani fuzzy inference system was presented to control a steam engine and boi-

ler combination by linguistic rules [3, 4]. Fuzzy logic is expressed by means of if-then rules with the human language. In the design of a fuzzy logic controller, the mathematical model is not necessary. Therefore the fuzzy logic controller is of good robustness. Owing to its easy application, it has been widely used in industry. However, the rules and the membership functions of a fuzzy logic controller are based on expert experience or knowledge database.

The harmony search (HS) algorithm is a recently developed meta-heuristic algorithm, and has been very successful in a wide variety of optimization problems.

HS was conceptualized using an analogy with music improvisation process where music players improvise the pitches of their instruments to obtain better harmony. The HS algorithm does not require initial values and uses a random search instead of a gradient search, so derivative information is unnecessary. Furthermore, the HS algorithm is simple in concept, few in parameters, easy in implementation, imposes fewer mathematical requirements, and does not require initial value settings of the decision variables. Many of the previous strategies for capacitor allocation in the literature are also limited for the application to planning, expansion or operation of distribution systems. Very few of these capacitor allocation techniques have the flexibility of being applicable to more than one of the above problems. Hence, this paper presents a FLC_HSO approach to determine suitable locations for capacitor placement and the seizing of the capacitor. This approach has the versatility of being applied to the planning, expansion, and operation studies of distribution systems. The proposed method was tested on electrical distribution systems consisting of standard IEEE 57-bus test System.

The fuzzy logic controller is employed to detection the critical nodal. The HSO methods have been employed successfully to solve complex optimization problems. It used to seizing the optimal capacitor banks. Simulation results are given to show the effectiveness of FLC_HSO approach.

The structure of the work presented in this paper is organized in the following sequence: Mathematical formulation is set in section 2. Section 3 shows the Fuzzy Logic Controller. Section 4 shows the Harmony Search Optimization (HSO). Section 5 shows the development space vector modulation technique based DTC for Electric vehicle motorization. The proposed structure of the studied propulsion system is given in the section 6. The simulation results of the different studied cases are presented in Section 6.

Articles

11

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

2. Mathematical formulation Load Flow

Fuzzy Controller

2011

m

The Principe of method is presented in Figure 1.

Power System

N° 4

u (t ) =

Harmony search optimization

∑m i =1 m

B

∑m i =1

(ui (t )).ui B

(4)

(ui (t ))

where i : is the output rule after inferring.

Optimal capacitor

Fig. 1. Bloc of intelligent fuzzy-ant approach.

The objective function of placement to reduce the power loss and keep bus voltage within prescribed limits with minimum cost. The constraint are voltage limits.

Following the above notation, the total annual cost function due to capacitor placement and power losses written as [10]: Minimize

N    F = K PL PL + ∑ K Cj BJ  j =1  

(1)

Constraint of voltage Vi min ≤ Vi min ≤ Vi max

i = 2,3,....N

(2)

3.1. Fuzzy based capacitor location Node voltages and power loss indices are the inputs to fuzzy controller to determine the suitability of a node in the capacitor placement problem. The suitability of a node is chosen from the capacitor suitability index (CSI) at each node. The higher values of CSI are chosen as best locations for capacitor placement [1], [2], [3], [4], [5].

The power loss indices are calculated as: PLI (i ) =

Power loss indices

Inference

-1

F

Voltage

Fig. 2. Structure of fuzzy controller.

m  m j (e(t )), m j ( de(t )),  A A2 mB (u (t )) = max  1  j =1  mB j (u (t )) 

(3)

where m A ( PLI ) is the membership function of PLI, m A (V ) is the membership function of V, mB (CSI ) is the membership function of SCI, j is an index of every membership function of fuzzy set, m is the number of rules and is the inference result. Fuzzy output CSI can be calculated by the centre of gravity defuzzification as: j 1

j

Degree of membership

where: F : Fuzziffication, F −1 : Defuzzification 1 L

LM

M

HM

0.5

1 PLI

1.5

H

0.8 0.6 0.4 0.2 0 0

2

Fig. 3. Power loss indices membership.

Degree of membership

Fuzzy logic is expressed by means of the human language. Based on fuzzy logic, a fuzzy controller converts a linguistic control strategy into an automatic control strategy, and fuzzy rules are constructed by expert experience or knowledge database. First, set the power loss index PLI and the voltage V to be the input variables of the fuzzy logic controller. The Capacitor suitable index CSI is the output variable of the fuzzy logic controller. The linguistic variables are defined as {L, LM, M, HM, H}, where L means low, LM means low medium, M means medium HM means height medium and H means height. The membership functions of the fuzzy logic controller are shown in Fig. 3. The fuzzy rules are summarized in Table 1. The type of fuzzy inference engine is Mamdani. The fuzzy inference mechanism in this study follows as:

Articles

(5)

Capacitor suitability index

Base of rule

F

3. Fuzzy Logic Controller

12

i = 2,3,..N

where: LR : Loss reduction, LMIN : Minimum reduction, LMAX : Maximum reduction, N : Number of bus

where: F : Total annual cost function ($), KPL : Annual cost per unit of power losses ($/KW), PL :Total active power loss (KW), KjC :Total active power loss (KW), : Shunt capacitor size placed at bus j (KVAR), Bj

N : Number of buses, Vmin : Minimum permissible voltage,

Vmax : Maximum permissible voltage.

j 2

( LR − LMAX ) ( LMIN − LMAXN )

1 L

LM

M

HM

1 Voltage

1.05

H

0.8 0.6 0.4 0.2 0 0.9

0.95

Fig. 4. Voltage membership functions.

1.1

Degree of membership

Journal of Automation, Mobile Robotics & Intelligent Systems

1L

LM

M

HM

VOLUME 5,

H

2011

the highest value of CSI are the most suitable for capacitor placement.

Step 8: Stop.

0.8 0.6 0.4

4. Harmony Search Optimization algorithm

0.2

Recently, Geem et al. [8] developed a new harmony search (HS) meta-heuristic algorithm that was conceptualized using the musical process of searching for a perfect state of harmony. The harmony in music is analogous to the optimization solution vector, and the musicians’ improvisations are analogous to local and global search schemes in optimization techniques. The HS algorithm does not require initial values for the decision variables.

Furthermore, instead of a gradient search, the HS algorithm uses a stochastic random search that is based on the harmony memory considering rate and the pitch adjusting rate (defined in harmony search meta-heuristic algorithm section), so that derivative information is unnecessary. Compared to earlier meta-heuristic optimization algorithms, the HS algorithm imposes fewer mathematical requirements and can be easily adopted for various types of engineering optimization problems [15], [16], [17], [18], [19], [20], [21], [22].

The optimization procedure of the HS algorithm consists of steps 1–5, as follows: Step 1: Initialize the optimization problem and algorithm parameters.

Step 2: Initialize the harmony memory (HM).

Step 3: Improvise a new harmony from the HM. Step 4: Update the HM. Step 5: Repeat Steps 3 and 4 until the termination criterion is satisfied.

0 0

0.2

0.4 0.6 Convenance

0.8

1

Fig. 5. Capacitor suitability index membership function.

To determine the critical busses the voltage and power loss index at each node shall be calculated and are represented in fuzzy membership function. By using these voltages and PLI, rules are framed and are summarized in the fuzzy decision matrix as given in Table 1.

0.6 0.5 Convenance

N° 4

0.4 0.3 0.2 0.1 1.1 1.05 Voltage

1 0.95

0.5

0

1.5

1

2

PLI

Fig. 6. View plot surface of fuzzy controller.

3.2. Algorithm (FLC) for identification of busses identification Following algorithm explain the methodologies to identify critical busses, witches are more suitable for capacitor placement [6], [10].

Step 1: Read line and load data of power system.

Step 2: Calculate power flow by Newton Raphson methods.

Step 3: Determine total active power loss.

Step 4: By compensation the self-reactive power at each node and conduct the load flow to determinate the total active power losses in each case.

Step 5: Calculate the power loss reduction and power flow loss indices.

Step 6: The PLI and the per-unit voltage are the inputs to the fuzzy controller.

Step 7: The outputs of Fuzzy controller are defuzzified.

This gives the ranking of CSI. The nodes having

5. Results

The FLC-HSO is coded in MATLAB environment version 7.6 (R2008a), and run using an Intel Pentium 4, core duo 1.87 GHz PC with 2 Go DDRAM-II and 2 Mo cache memory. All computations use real float point precision without rounding or truncating values. More than 6 small-sized test cases were used to demonstrate the performance of the proposed algorithm. Consistently acceptable results were observed.

The FLC_HSO method has been applied on the network test IEEE 57 buses that represent a portion of the American electric power system (the Midwestern, USA) for December 1961. This electric network is constituted of 57 buses and 7 generators (at the buses Nº: 1, 2, 3, 6,

Table. 1. Decision matrix for determining suitable capacitor.

V

CSI

PLI

L

LM

M

HM

H

L

L

L

L

LM

LM

LM

L

L

LM

LM

M

M

L

L

LM

M

HM

HM

L

LM

M

HM

HM

H

LM

LM

M

HM

H Articles

13

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

8, 9 and 12) injecting their powers for a system nourishing 42 loads through 80 lines of transportation (Shown in Fig 1). The base voltage for every bus is of 135 kV. The proposed method is illustrated with a system, consisting of standard IEEE 57-bus test System. The location for placement of capacitors is determined by fuzzy controller and the capacitor sizes are evaluated using harmony search optimization.

FLC-HSO approach is applied for IEEE 57- approach given above is shows in Table 1.

In primary case we applied the first algorithm (FLC) based fuzzy logic controller logic which gives the critical busses shown in Table 1, follows in makes call the particle swarm optimization conceived for difficult combinative optimization to optimize the objective function (1) all respect limit constraints voltage (2) Finally we obtained the optimal cost function and the value of optimal capacitor for each critical buses all that are illustrated in Table 1.

1.04

Voltage [PU]

1.02

1

0.98

0.96

0.94

0.92

0

10

20

1 17

15 45

16

14

44

5

13

46

12

18 19

20

49

38

48

21

26

6

37

47 39

27 22

28

57

23

29

24

56 36

II

40

42

35

25

7

50

34

30

52

31

33 32

10 51

11 53

43 54

41

55

8

Fig. 10. Topology of the IEEE 57-bus.

14

Articles

30 N째 of Bus

40

50

60

Fig. 9. The levels of voltage (Per Unit) for the IEEE 57bus Electrical Network Before placement of optimal capacitor.

3

II

2011

1.06

2

4

N째 4

9

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

Table 1. The initialized coefficients of the HSO. Coefficients

value

Size of the harmony memory matrix (HMS)

10

Harmony memory considering rate (HMCR)

0.85

Pitch adjusting rate (PAR)

0.45

Table 2. Results of FLC-HSO. Fuzzy logic controller (FLC)

N° 4

2011

After application the new approach FLC-HSO the results are improved, the loss are decreased by 15.75 %, as well as the reactive power injected into the electrical distribution system are diminish by 13.06% and the nodal voltage are improved.

In our application we have compared the FLC-HSO by another approach explained in the Tables 2 and 3: Genetic algorithm based optimal power GA-OPF [8], Ant Colony Optimization (ACO) algorithm for optimal flow ACO-OPF [14], Quasi Newton based optimal power flow QN-OPF and MATPOWER. Our approach FLC-HSO was proved the satisfactory results illustrated in Table 3.

The constraints of security are also verified for the angles and the amplitudes of voltages, the levels of voltage (Per Unit) for the IEEE 57-bus Electrical Network are drawn in the Fig. 5 and Fig. 6.

Number of critical buses

Value of capacitor [MVAR]

10

4.22

1.05

19

2.12

1.04

21

3.11

1.03

28

3.95

32

5.91

33

7.35

52

1.40

Voltage [PU]

1.02 1.01 1 0.99 0.98

Harmony search optimization (HSO)

Power Losses [MW]

0.97

Before placement of optimal capacitor

After placement of optimal capacitor

18.50

15.29

Minimal Voltage [Per Unit]

0.935

0.981

Reactive Power [MVAR]

275.23

239.27

Table. 3. Comparison of the results gotten by ACO-OPF, QN-OPF, MATPOWER and proposed method FLC-HSO on the IEEE 57-bus Electrical Network.

Results

Power Loss [MW]

Reactive Power [MVAR]

GA-OPF [29]

18.60

–

QN-OPF

17.16

–

ACO-OPF

17.96

–

MATPOWER

16.512

270.56

FLC-HSO

15.29

239.27

0

10

20

30 N° of Bus

40

50

60

Fig. 10. The levels of voltage (Per Unit) for the IEEE 57-bus Electrical Network After placement of optimal capacitor.

6. Conclusions

In this paper, a novel approach FLC-HSO based harmony search optimization and fuzzy logic controller to OPF problem has been presented. The proposed approach FLCHSO utilizes the fuzzy logic controller for identification the critical bus and the improvisation Process of HSO to search the optimal seizing capacitor banks. Different objective functions have been considered to minimize losses and, to improve the voltage profile, and to enhance voltage stability. The proposed approach has been tested and examined with different objectives to demonstrate its effectiveness and robustness. The results using the proposed approach were compared to those reported in the literature.

The results confirm the potential of the proposed approach and show its effectiveness and superiority over the classical techniques and genetic algorithms.

AUTHORS

Brahim Gasbaoui*, Meriem Oudda and Abdelfatah Nasri – Bechar University, Faculty of Sciences and Technology, Department of Electrical Engineering, B.P

417 BECHAR (08000) Algeria.

*Corresponding author.

E-mail: gasbaoui_2009@yahoo.com.

Articles

15

Journal of Automation, Mobile Robotics & Intelligent Systems

References

[1] Baran M.E., Wu F.F., “Optimal capacitor placement on radial distribution systems”, IEEE Trans. Power Delivery, vol. 4, , Jan. 1989, pp.725–734.

[2] Baran M.E., Wu F.F., “Optimal seizing of capacitors placed on radial distribution systems”, IEEE Trans.

Power Delivery, vol. 4, Jan. 1989, pp.735–743.

[3] Ponnavaiko M., Prakasa Rao K.S., “Optimal choice of fixed and switched capacitors on radial distribution feeders by the method of local variations”, IEEE Trans. Power Apparatus and Systems, vol.

102, Jun. 1983, pp.1607–1615.

[4] Grainger J.J., Lee S.H., “Optimum size and location of shunt capacitors for reduction of losses on distribution feeders“, IEEE Trans. Power Apparatus and Systems, vol. 100, March 1981, pp. 1105–1118.

[5] Das D., “Novel method for solving radial distribution networks”, IEE Proc.-C, vol.141, Jul. 1994, pp. 291–298.

[6] Bouri S., Zeblah A., Ghoraf A., Hadjeri S., H. Hamdaouil, “Ant colony optimization to shunt capacitor allocation in radial distribution systems“, Acta Electrotechnica et Informatica, no. 4, vol. 5, 2005.

[7] Prasad P.V., Sivana S., Sreenivasulu N., “A fuzzygenetic algorithm for optimal capacitor in radial distribution systems”, ARPN Journal of Engineering and Applied Sciences, 2007.

[8] Younes M., Rahli M., Koridak L.A., “Optimal Power Flow Based on Hybrid Genetic Algorithm”, Journal of Information Science and Engineering, vol. 23, 2007, pp. 1801-1816.

[9] Omran M.G.H., Mahdavi M., “Global-best harmony search”, Appl. Math. Comput. In press. DOI:10.1016/j. amc.2007.09.004.

[10] Mahdavi M., Fesanghary M., Damangir E., “An improved harmony search algorithm for solving optimization problems”, Appl. Math. Comput., no. 188(2), 2007, pp. 1567–79.

[11] Alloua B., Laouifi A., Gasbaoui B., et al., “Intelligent Controller Design for DC Motor Speed Control based on Fuzzy Logic-Genetic Algorithms Optimization”, Leonardo Journal of Sciences, Issue 13, July-December 2008, pp. 90–102.

16

Articles

VOLUME 5,

N° 4

2011

[12] Alloua B., Laouifi A., Gasbaoui B., et al., “The Efficiency of Particle Swarm Optimization Applied on Fuzzy Logic DC Motor Speed Control”, Serbian Journal of Electrical Engineering, vol. 5, no. 2, November 2008, pp. 247-262.

[13] Yen J., Langari R., Fuzzy Logic: Intelligence, Control, and Information, Prentice-Hall, New York, 1999.

[14] Gasbaoui B.,Alloua B., “Ant Colony Optimization Applied on Combinatorial Problem for Optimal Power Flow Solution”, Leonardo Journal of Sciences, Issue 14, January-June 2009, pp. 1–16.

[15] Geem Z.W., Tseng C., Park Y., “Harmony search for generalized orienteering problem: best touring in China”, Springer, Lecture Notes in Computer. Science, 2005, vol. 3412, pp. 741–50. DOI: 10.1007/11539902_91 [16] Kim J.H., Geem Z.W., Kim E.S., “Parameter estimation of the nonlinear Muskingum model using harmony search”, Journal American Water Resources Association, vol. 37, no. 5, 2001, pp.1131–8.

[17] Lee K.S., Geem Z.W., “A new structural optimization method based on the harmony search algorithm”, Comput. Struct.,;82(9–102004), pp. 781–98.

[18] Lee K.S., Geem Z.W.,. Lee S.H, Bae K. W., “The Harmony Search Heuristic Algorithm for Discrete Structural Optimization”, Engineering Optimization, vol. 37, no. 7, 2005 , pp. 663–684.

[19] Geem Z.W., Tseng C.L., Williams J.C., “Harmony Search Algorithm for Water and Environmental Systems”. In: Geem Z.W.,Music-Inspired Harmony Search Algorithm, Springer, SCI191, 2009, pp. 113–127.

[20] Geem Z.W., Improved Harmony Search from Ensemble of Music Players, Springer-Verlag, KES 2006, Part I, LNAI 4251, 2006, pp. 86-93.

[21] Ayvaz M.T., “Application of Harmony Search algorithm to the solution of groundwater management models”, Advances in Water Resources, vol. 32, 2009, pp. 916–924.

[22] Wang C.M., Huang Y.F., “Self-adaptive harmony search algorithm for optimization”, Expert Systems with Applications, vol. 37, 2010, pp. 2826–2837.

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

N° 4

2011

APPLICATION OF DEPENDENCE GRAPHS AND GAME TREES FOR DECISION DECOMPOSITION FOR MACHINE SYSTEMS Submitted 6th January 2011; accepted 15th February 2011

Adam Deptuła, Marian A. Partyka

Abstract: The paper concerns application of dependence graphs and parametric game trees for analysis and synthesis of dynamic properties of machine systems. Different graph solutions mean connections of input and output quantities as well as constructional parameters. The method of dependence graph distribution into the game tree structure was described. Multiple vertex numeration was introduced in order to determine subordination of elements in the system, and the additional time vertex allowed to obtain the structure with the closed loop of the feedback.

The obtained graph solutions were evaluated taking into account decision decomposition into single constructional and/or service parameters.

Keywords: dependence graph, vertex complexity, system structure, game tree, decision decomposition, multiple vertex numeration.

1. Introduction Engineering practice requires correct evaluation of the mathematical model describing a given system with some variables. A proper mathematical model contains a group of functions joining different variables and describing connections between quantities in the system. Decision tables [1] and logical functions [8], [18], [20], [22] can be applied in simulation of machine systems, for example described by ordinary or partial differential equations. It results from the fact that the occurring nonlinear elements can be divided into the finite number of linear elements (parts), and in a consequence we obtain some linear systems as for simulation from the primary single nonlinear system [13], [18]. The occurring constructional parameters influence courses of unknown functions dependent on time. Traditional analysis such as Input ® Output and Output ® Input of the given system with the dependence graph method allows to obtain the vertex groups of the following properties: [2], [3]: elements inside the given group have many information connections, particular groups have a little reciprocal information connections.

In this way we obtain constructional recommendations. Let us see that it is possible to obtain multiple solutions from the dependence graph, so selection of suitable subsolutions can be done with use of tree classifiers from the neuron networks [7], [14], [19].

Let us consider a different approach, being translation of the dependence digraph into the tree game structure

[9], [10]. Constructional and service parameters can be joined in different mathematical ways, it is necessary to determine precise decomposition of the dependence graph in order to define decision variables. Such procedure allows to determine the area of feasible solutions and to select a proper optimization procedure.

2. Application of the dependence graph for analysis of dynamic properties in the case of machine systems The equations of dynamics can be used for determination of mutual connections of all the functions dependent on time. As a result of notation and decomposition of the dependence graph of those functions we obtain the groups of distribution which describe properties of successive subsystems of the considered machine system and a set of suitable constructional and service parameters.

Example Fig. 1 [12], [18] presents a simplified hydraulic system including a gear pump, an overflow valve, a divider and a motor loaded by a big mass moment of inertia. The paper does not include numerical values of constructional and service parameters because diagrams will not determined while simulation.

Fig. 1. Hydraulic scheme of the system.

The mathematical model of the above system has the following form: 1. equation of intensity of the flow from the pump , where

(1)

Articles

17

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

and: Pp - pressure in the forcing line of the pump, Qp - theoretical capacity of the pump, Qzp - flow intensity through the overflow valve, Qs - flow intensity given to the receiving part of the system;

N° 4

2011

1. The signals from which the given signal formed:

2. The signals forming the given signals:

2. equation of the overflow valve Qzp = 0 for Pp ÂŁ P0 ;

dQzp dt

=

1 K Pp - Qzp for Pp > P0 (2) T T

3. equation of losses of pressure Pp = Rl Qs + Ps

In this case the obtained graph solution (for the initial vertex Qp) is polysemantic from the point of view of successive obtaining the subgroups:

(3)

where Ps - pressure drop between working chambers of the motor; 4. flow equation of the motor (4) where w - angular velocity of the motor shaft; 5. equation of the motor moments , i.e.

(5)

The unknown functions Pp, Ps, Qs, Qzp, w are calculated on the basis of the given input of the system Qp, so there is a system structure (Fig. 2). Thus, the following notations of the dependence graph are resulting:

3. The dependence graph for tree game structures The considered hydraulic system containing the elements joined in a way allowing for a flow of signals can be written with the dependence graph for tree game structures (Fig. 3). The graph distribution from any vertex in the first stage leads to a tree structure with cycles, and next to a general tree game structure. Each structure has a proper analytic notation ( and , where i is a vertex, from which the graph decomposition started) determining a way of transition from the dependence graph to the tree structure.

1. The signals from which the given signal formed:

2. The signals forming the given signals:

Fig. 3. Dependence digraph of the signal flow.

Fig. 2. System structure of the hydraulic system.

Generally, we can obtain a graph solution in the tree approach for the hydraulic system shown in Fig.1 taking into account constructional parameters and the following notations:

18

Articles

Let us present decomposition of the dependence graph of the signal flow from the defined initial vertex Pp [16]. First of all, we obtain a tree structure with cycles (Fig. 4), described by Eq. (6). Since it is possible to come back from a given vertex to the former vertex or even the initial vertex, we obtain the analytic notation (7) determining the obtained tree structures with cycles. At the next stage we obtain a tree game structure shown in Fig. 5.

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N째 4

2011

(6)

Fig. 4. Tree structure with cycles and the initial vertex Pp.

(7)

Articles

19

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N째 4

2011

Fig. 5. Tree game structure from the initial vertex Pp.

According to the proper optimization procedure, the mathematical model describing the considered system should provide the maximum possible range of information. Thus, the dependence graph decomposition is performed from the given vertex, taking into account all

the constructional and/or service parameters separately written (not in the interaction method). Such decision decomposition leads to the tree structure with cycles and the tree game structure (Figs. 6 and 7) described by relationships (8) and (9).

(8)

20

Articles

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

N째 4

2011

(9)

Fig. 6. Tree structure with cycles and the initial vertex Pp including decision decomposition.

Occurrence of numerous straight, oblique and complex feedbacks is one of dynamic properties of machine systems, resulting from their nature. The feedback provides influence of the results from the final block of the optimization process (in form of time characteristics of timedependent quantities) on selection of and possibility of application of essential information flowing into the initial block. The given block can be considered as the closed system when there is a feedback causing that the input signal does not depend only on the internal state of the object, but on the present value of the output signal as well.

Thus, it is necessary to introduce an additional initial vertex Qp to the dependence graph notation. This additio-

nal vertex should be joined with the final vertex w by the return decision of transition k. Finally, the mathematical model of the considered hydraulic system can be written with the dependence digraph of the signal flow shown in Fig. 8 [4], [5], [6].

Decomposition of the dependence graph from Fig. 8 from the initial vertex Qs including decision decomposition leads to the tree structure with cycles (Fig. 9) and the game structure (Fig. 10), described by the relationship (10). The additional feedback on the tree structures is determined by suitable return arcs and decision allowing to change the input signal on the basis of the output signal [5], [6].

Articles

21

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N째 4

2011

Fig. 7. Tree game structure from the initial vertex Pp including decision decomposition.

Fig. 8. Dependence digraph of the signal flow with the additional time vertex Qp.

(10)

22

Articles

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

N째 4

2011

Fig. 9. Tree structure with cycles and the initial vertex Qs with the additional time vertex Qp.

Fig. 10. Tree game structure from the initial vertex Qs with the additional time vertex Qp.

Considering tree structures we must determine element subordination in the system. Each structure has its proper analytic notation ( and , where i is the vertex from which the graph decomposition was started). Each element qr has always subordinated elements qi. Both qr and qi elements can occur many times in the expression in

the brackets (k ...)k with different values of k, i.e. at various stages of the tree structure. Thus, it is necessary to introduce multiple numeration of vertices [4], [15]. Fig. 11 shows a tree structure from Fig. 10 including multiple vertex numeration described by relationship (11).

Articles

23

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N째 4

2011

Fig. 11. Tree game structure from the initial vertex Qs with the additional time vertex Qp and multiple vertex numeration.

(11)

Fig. 12. Tree game structure from the initial vertex Pp including the time vertex Qp and multiple vertex numeration.

24

Articles

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

N° 4

2011

(12)

Decomposition of the dependence digraph from Fig. 8 from the initial vertex including decision decomposition and multiple vertex numeration leads to the game structure (Fig. 12) expressed by relationship (12) [4].

4. Conclusions Application of dependence graphs and game tree structures allows to present a sequence of changes of arithmetic values of constructional and service parameters in order to obtain the required behaviour of the system (for example machine system). Unlike traditional dependence graphs and tree classifiers, the dependence graph with game tree structures includes connection of importance rank of vertices (states) and height of the tree structure.

Application of decision decomposition for game graphs and trees does not change types and graphical shapes of such structures. They are more complex but they keep the given structural properties resulting from the initial dependence graph. Thus, a local role of decomposition can be distinguished. The introduced decision decomposition eliminates interaction of constructional and service parameters because a designer can make a decision about only single changes and observation at the successive stages.

The introduced additional time vertex resulting from the physical model well describes the feedback loop on the dependence graph. It appears from the following statements: an additional decision is a guideline for a designing engineer when he/she is going to change the output signal on the basis of the input signal, the tree structure with cycles and the tree game structure preserve identical shapes like before introduction of the additional time vertex, but they have complementary branches at suitable floors.

The introduced multiple vertex numeration allows to consider and distinguish the same elements occurring on different floors of the tree structure. That also allows to

preserve a general shape of the graphical structure existing even before decomposition.

The algorithmic method of formation of graphical structures from the mathematical model of the system describes the optimization method of systematic search.

The game structure describes a space of possible solutions in order to find optimum objective functions. There is connection with other graphical structures which can be graphs in another sense, or even decision trees with node and/or branch coding. Such interpretation can lead to different types of logical trees [2], [11], [13], [17], [21], [23], [24].

ACKNOWLEDGMENTS Work co-financed by European Social Fund.

AUTHORS Marian A. Partyka, Adam Deptu³a* - Opole University of Technology, Faculty of Production Engineering and Logistic, Department of Knowledge Engineering, 75 Ozimska Street, 45-370 Opole, Poland. Telephone/fax: (77) 423-40-44; (77) 423-40-31. Telephone: (77) 453-8448 in.158. E-mail: a.deptula@po.opole.pl. * Corresponding author

References [1]

[2]

Cholewa W., KaŸmierczak J., Diagnostyka techniczna maszyn. Przetwarzanie cech sygna³ów, Course book no.

1904, Silesian University of Technology, Gliwice, Poland, 1995. (in Polish) Coner L., Koziarska A., Partyka M.A., „Zastosowanie klasyfikatorów drzewiastych i grafów zale¿noœci o ró¿nym stopniu szczegó³owoœci w CAD procesów decyzyjnych na przyk³adzie uk³adów maszynowych”. In: Conference. Cylinder 1999. Badanie-Konstrukcja-Wytwarzanie-Eksploatacja Uk³adów Hydraulicznych, Zakopane, Poland, 1999. (in Polish) Articles

25

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

[3]

[4]

[5]

[6]

[7] [8]

[9] [10] [11]

[12] [13]

[14]

[15]

[16]

[17]

26

Coner L., Partyka M. A., “Application of dendritic classifiers and dependence graphs in CAD of decision processes with use of the machine systems as an example”. In: 4th Confer. Neural Netw. and Their Applic., Zakopane, Poland, 1999, Publ. by Depart. of Comput.

Engin., Techn. Univ. of Czêstochowa, Czêstochowa, Poland, 1999.

Deptu³a A., „Analiza w³aœciwoœci dynamicznych uk³adu hydraulicznego za pomoc¹ grafów rozgrywaj¹cych parametrycznie”. In: Conference. Cylinder 1999 Badanie-Konstrukcja-Wytwarzanie-Eksploatacja Uk³adów Hydraulicznych, Instytut Techniki Górniczej KOMAG, Gliwice 2010, pp. 325-340. (in Polish) Deptu³a A., Partyka M.A, “Application of game graphs in optimization of dynamic system Structures”, International Journal of Applied Mechanics and Engineering, vol. 15, no.3, 2010, pp. 647-656.

Deptu³a A., Partyka M.A, „Zastosowanie grafów rozgrywaj¹cych parametrycznie i dekompozycji w procesie optymalizacji dynamicznych struktur systemowych”, Górnictwo Odkrywkowe, no. 3, pp. 2010. (in Polish) Devroye L., Gyori L., Lugosi G., Probabilistic Theory of Pattern Recognition, Berlin, Springer Verlag, 1995.

Kaczanowski S., Olszewski M., Wañski Z., P³ynowe elementy i uk ³ ady logiczne , publ. by WK& £ , Warsaw 1971. (in Polish) Kazimierczak J., System cybernetyczny, Wiedza Powszechna-Omega, Warsaw 1978. (in Polish) Kazimierczak J., Teoria gier w cybernetyce, Wiedza Powszechna-Omega, Warsaw 1973. (in Polish) Koziarska A., Partyka M.A.,” Similarities and differences between logical trees and dendritic classifiers in CAD of decision processes:. In: 4 th Conf. Neur. Netw. and Their Applic., Zakopane 1999, Depart. of Comput.

Engin., Techn. Univ. of Czêstochowa, Czêstochowa 1999.

Lipski J., Napêdy i sterowania hydrauliczne, WK&£, Warsaw 1981. (in Polish) Palczak E., „Modelowanie uk³adów hydraulicznych za pomoc¹ grafów Coatesa”, Sterowanie i Napêd Hydrauliczny, Zeszyt 3/93 (in Polish) Partyka M.A., “Application of the tree classifiers for analysis of decision and minimal multiple- valued logical functions- for example in machine systems”, 3rd Inter. Confer. Neur. Netw. And Applic., Kule 1997, Kated. In¿yn. Komput. Polit. Czêst., Czêstochowa 1997.

Partyka M.A, Deptu³a A., „Badanie w³asnoœci dynamicznych uk³adów maszynowych z uwzglêdnieniem wielokrotnej numeracji wierzcho³kowej dla drzew rozgrywaj¹cych parametrycznie”, Napêdy i Sterowanie, no.

3, 2010. (in Polish) Partyka M.A., Deptu³a A., „Zastosowanie grafów zale¿noœci i drzew rozgrywaj¹cych parametrycznie w procesie innowacji na przyk³adzie uk³adów maszynowych”. In: XIII Konferencja Komputerowo Zintegrowane Zarz¹dzanie, Zakopane 2010, Pol. Towarz. Zarz.

Prod. PTZP, Opole 2010 (in Polish) Partyka M.A., Krzy¿ak A., „Podobieñstwa i ró¿nice miêdzy logicznymi drzewami decyzyjnymi a drzewami binarnymi z klasyfikatorów drzewiastych i rozpoznawania obrazów”. In: XXV Konfer. Zast. Matem. Zakopane 1996, Warsaw 1996. (in Polish)

Articles

[18]

[19]

[20]

[21]

[22]

[23] [24]

N° 4

2011

Partyka M.A., Logika systemów projektowania na przyk³adzie CAD uk³adów maszynowych, St. i Monogr. nr 105, Ofic. Wydaw. Polit. Opol., Opole 1999. (in Polish) Partyka M.A., “Similarities and differences between neuron networks and dendritic structures for CAD of decisive processes”. In: 2nd Conf. Neur. Netw., Szczyrk 1996., Czêstochowa 1996.

Partyka M.A., “Some remarks on the Quine Mc Cluskey minimization algorithm of multiple-valued partial functions for design structures”. In: 7th Inter. Cong. Log. Method. Phil. Sc., Salzburg 1983,Austria.

Partyka M.A., “The application of structural multiple valued logical decisions in Knowledge engineering based on the example of mechanical systems”. In: XXXV Symp. Model. in Mech., Wis³a 1996, Zesz. Nauk. Kat. Mech. Tech. Polit. Œl¹s. no.1, Gliwice 1996.

Partyka M.A., “The Quine- Mc Cluskey minimization algorithm of individual multiple-Valued partial functions for digital control systems”/. In: 3rd Inter. Confer.

Syst. Engin., Wright State University, Dayton 1984, USA. Reingold E.M., Nievergelt J., Deo N., Algorytmy kombinatoryczne, Warsaw, PWN, 1985. (Polish edition) Tadeusiewicz R., Sieci neuronowe, Akad. Ofic. Wydaw.

RM, Warsaw 1993. (in Polish)

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N° 4

2011

Hardware-software platform for integrated circuit technology learning and design via Internet Submitted 2nd January 2011; accepted 15th April 2011

Vladislav V. Nelayev, Mirosław Najbuk, Teodor Breczko

Abstract:

The module GUI (Graphical User Interface)-SUPREM III for design and training of microelectronic technology via Internet is described. The module is the part of the software-hardware suit intended both for studying principles of design in computer integrated circuit technology, and for simulation/design of a technological route of integrated circuit manufacturing. Program package SUPREM III is the base platform for physical simulation of processes in microelectronics. Modern information technologies (the server Apache, programming languages PHP and PERL, standard GnuPlot program) are utilized for realisation of the described platform. The module is used at Belarusian universities and abroad during lectures and computer training classes as part of disciplines dedicated to design in microelectronics.

Keywords: e-learning, internet, CVS, design, hardware, software.

1. Introduction

Due to progress in the field of information technologies and constant increase of accessibility to the global Internet network, on-line education in form of a variety of so-called distance learning (e-learning, е-education) is becoming more and more popular. Modern information technologies provide new possibilities for increased learning and training efficiency in the such scientific-capacious sphere of human activity as microelectronics, including integrated circuit (IC) and technology design. Use of Internet-learning reduces the financial expenses, as well as allows getting knowledge in the suitable time and at a rate suitable for each person.

Nowadays, e-learning methodologies and tools are well-perfected, but simulation and design via I Internet have specific particularities and unsolved problems. This new service is intended to help designers overcome the expense and time-consuming effort of establishing and maintaining a state-of-the-art, comprehensive IC design environment. Using this service allows to focus on own unique intellectual property (IP) while avoiding expensive infrastructure development.

Such Internet sites provide customers (in particularly small and medium-size enterprises, SMEs) with the tools and support needed to quickly and easily take advantage of the cost and performance benefits offered by modern (but expensive!) design tools. For example, companies Synopsys and Avant! have joined together to provide Internet-based design via DesignSphere Access [1]. The alliance combines electronic design automation tools with extensive computing resources, network capability and security services acces-

sible via Internet. DesignSphere Access ensures data safety and integrity with a strong system of firewalls, physical access security, customer-dedicated hardware configurations, and log-in and password restrictions.

NEC Electronics launched in 2001 Internet-based collaborative design environment for gate arrays as part of the NEC Gate Array Design Center Internet site [2]. The easy-to-use site includes real-time pricing, turnaround time and packaging in-formation to speed the evaluation process and also facilitates quick product development through online training, design tutorials and technical and sales support. There are a lot of other references devoted to use of Internet as a power tool for remote design and experiments (see for example [3]).

One more aspect of the use of Internet network for design is the possibility of collaboration in the frame of one project of many specialists representing various kinds of skills and levels of experience and working in different places.

The CVS technology, widely used in the SourceForge, is the example of such a system for group development of applications when large projects are carried out.

Essential problem, particularly in the field of microelectronics, is high cost and, as a result, inaccessibility of modern software for IC design. The overall value of the software for IC design on the world market was well above billion dollars in 2000, and it constantly increases with the rate near 8% per year. The design software distributed free of charge (with Open Source-type licenses) has limited functionality, contains many errors and often does not follow industrial standards. Lack of possibility to study and use modern and fully functional tools for IC design adversely affects, in particular, the level of skills and qualifications of IC designers.

Obviously, use of the Internet network for learning of software inaccessible directly for the designer, as well as for design in the client-server’ mode, by means of appropriate software running on a central server, can be very efficient.

Here it is assumed that end user license does not restrict Internet access to the design software.

At the Department of Micro- and Nanoelectronics of the Belarusian State University of Informatics and Radioelectronics [4]-[6] investigations are undertaken for the development of methods and software both for learning, and for design of microelectronic technology via Internet/ Intranet network.

Here we describe approaches and the realized hard-waresoftware platform for a distance learning of integrated circuit (IC) manufacturing, physical modeling of technology processes and simulation/designing of technology route with the use of modern program package maintained at the central server.

Articles

27

Journal of Automation, Mobile Robotics & Intelligent Systems

Presented module GUI (Graphical User Interface) -SUPREM III with the created by authors software-hardware package allows to realize learning and design of integrated circuit technology both at local (Intranet), and global network Internet.

2. Module GUI-SUPREM III

The module GUI (Graphical User Interface)-SUPREM III is based on the free distributive of Linux Slax [11] and free package SUPREM III for IC technology simulation and design [12]-[13]. SUPREM III is an analogue of the commercial program Suprem3, which costs more than a hundred of thousands dollars. Suprem3 is the part of module ATHENA of the Silvaco package [14]. Silvaco is the leader company amongst developer of modern software programs intended for design in microelectronics. SUPREM III is used for 1-dimensional physical simulation of separate processes and technological flow of IC manufacturing.

The Program SUPREM III allows realizing physical modeling of basic technological steps, including diffusion and implantation doping, diffusion redistribution of decants, oxidation, epitaxy, etching, etc. Output file SUPREM III contents information about concentration profiles and geometry of the modeled device structure.

VOLUME 5,

N° 4

2011

3. Methodology of the design and learning via internet

The realized approach to the design/learning of IC technology via Internet network is shown in the Figure 2.

The main part on this diagram belongs to a Web server on which the design software is placed. Web server provides user access to the site, as well as processes all requests received from users who use their Web browsers for communication with the server. Another function of the Web server is to provide information about the current status of the project (i.e. the status of actions requested by the user).

Fig. 2. The block-diagram illustrating the softwarehardware platform for Internet learning/design with use GUI-SUPREM III.

Fig. 1. Module GUI-SUPREM III. Screen-shot of the main page.

Format of the output file of module GUI-SUPREM III is compatible with the format of the input file of the package PISCES [15] intended for modeling of device electric features. This important particularity allows to realize the closed cycle of simulation including modeling and design both technology and device. The used distributive Linux Slax LiveCD does not require installation on hard disk. It is loaded from removable disk such as USB disk or compact disk. Distributive Slax has a module structure. Such preference property allows easy to modify the distribution programs under particular requirements of the user. Standard start of the module GUI-SUPREM III is realized by use of self-loading compact disk Linux Slax LiveCD. As a result the main page of the module GUISUPREM III is opened (Figure 1). The main page of the module GUI-SUPREM III consists of a standard menu, instrumental panels, area for entering input parameters in file with the task for modeling and area for image of simulation result in tabular form.

28

Articles

Thereby, the Web server is a connecting link between the hardware of the Internet network and the user’s browser.

The Web server Apache is most suitable and the most efficient tool in the respect of specified requirements. Apache is easily configurable, allowing adjusting the Web services in accordance with needs of individual and corporate users.

Configuration of Apache is achieved by means of directives kept in configuration files. Apache allows to create virtual Web nodes, as well as to execute the functions of a proxy server.

The allocated computer complexes can be built on the base of protocol HTTP (Hyper Text Transfer Protocol), which supports the interface with CGI (Common Gateway Interface) applications on Web-servers and allows ensuring an access to server computing resources from browsers. The Java language gives an opportunity to create the computing system even more allocated, transferring a part of computing loading, which carries out Web-server, to browser. In result there is a possibility to construct more flexible and productive computing systems.

The next problem is to make dynamic template pages, written in HTML language. Use of the PERL language (Practical Extraction and Report Language) is the most efficient for the development of applications realizing interactions with databases and with templates.

Another function implemented in PERL is the organization of the file with the input data and with definition of the design or simulation task for the software placed on the server (for instance, program SUPREM for simulation of IC technology).

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N° 4

2011

Processing of the forms, which contain the requests for design or simulations, as well as results of calculations, is also performed by means of PERL, as a script language well tailored to OS UNIX. Integration of modules of such programming and scripting languages as PERL and PHP with Apache is often used to enhance capabilities of HTML in processing and creating of dynamic Web pages.

It is suitable to use standard, free program gnuplot [16] for saving simulation results in graphical format (for instance, in png) since it is a flexible format that can be easily adjusted and is compatible with widely used operation system UNIX. Presented Internet software-hardware suite, realized as depicted in Figure 2, is placed at the server Apache and is available at Web-address: http://kim.uwb.edu.pl

Plot and are sent to user via e-mail. Typical presentation of obtained results of IC technology simulation/design is shown in Figure 4. Here, concentration profiles of impurities (boron, arsenic and antimony) in the structure of n-p-n bipolar transistor are shown.

Fig. 3. Screen-shot of the dynamic shell for entering commands and parameters in the file for simulation task in the GUI-SUPREM III module.

AUTHORS

Dynamic shell (Figure 3) was developed for entering commands and parameters in the file necessary for a simulation task performed by GUI-SUPREM III module.

This procedure may be realized directly at the Web-site.

Dynamic shell is written with use the interpreted language JavaScript.

Processed data (the results of calculations) may be sent to client via e-mail if hardware resources are insufficient for real time communication. The server POSTFIX is used for sending the e-mail as an efficient way for automation of the sending of the postal messages. The server POSTFIX provides reliable protection from malfunctions and warranty of the message reception by the client.

Results of simulation are transformed in to graphical attachments (files in the png format) using program Gnu-

Fig. 4. Results of simulation via Internet the concentration profiles of impurities (boron, arsenic and antimony) in the structure of n-p-n bipolar transistor.

4. Conclusions

The module GUI (Graphical User Interface)-SUPREM III for organization of design and training of microelectronic technology via Internet is developed. The module is the part of the created software-hardware suite intended both for studying of principles of computer integrated circuit technology design, and for simulation/design of a technological flow of integrated circuit manufacturing via Internet. Modern means of information technology are used in the complex including the server Apache, programming languages PHP and PERL, database MySQL, standard GnuPlot program. It is obvious that presented methodology and principles for microelectronic technology may be successfully used in any applications for organization of calculations in the client-server mode via Internet. Website (http://kim.uwb.edu.pl) with described instruments is used successfully in Belarusian universities as part of “Computer Aided Design in Microelectronics” course.

Vladislav Nelayev – Belarusian State University of Informatics and Radioelectronics, Minsk, Belarus, E-mail: nvv@bsuir.by.

Teodor Breczko – University of Bialystok, Bialystok, Poland, E-mail: tbreczko@uwb.edu.pl Mirosław Najbuk* – University of Bialystok, Bialystok, Poland, E-mail: najbuk@gmail.com * Corresponding author

References:

[1] Fjeldy T.A., Shur M.S., Lab on the Web: Running Real Electronics Experiments via the Internet, Wiley, 2003.

[2] Nelayev V.V., Kolosnicin B.S., “Computer training programs for a microelectronics technology curriculum”. In: Proc. 3nd East-West Congress on Engng.

Educ., Gdynia, Poland, 1996, pp. 123–124.

[3] Kazitov M.V., Nelayev V.V., “Active virtual laboratory at Internet as an effective tool for learning”. In: Proc.

1st Global Congress on Engineering Educ., Kraków, Poland, 1988, pp. 269–272.

[4] Kudrjavtsev P.A., Maximenya A.I., Nelayev V.V., “Virtual laboratory at the Internet for a distance learning in integrated circuit technology”. In: Proc. 2nd Global Congress on Engng. Educ., Wismar, Germany, 2000, pp. 144–147.

[5] Kuzmicz W., Nelayev V., Stempitsky V., Kudin K., “Design and simulation via Internеt”. In: Proc. 9th Advanced Training Course on Mixed Design of Integrated Circuits and Systems. Education of Computer Aided Design of Modern ICs and Systems, Łódź, Poland, 665–668 (2003).

[6] Nelayev V.V., “The experience of distance design and learning via Internet”. In: Proc. 7th World Multiconference on Systemics, Cybernetics and Informatics, Orlando, USA, 12003, pp. 97–101.

Articles

29

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N° 4

2011

Adaptive Control of Frictional Contact Models for Nonholonomic Wheeled Mobile Robot Submitted 14th March 2011; accepted 25th August 2011

R.Vivekananthan, L. Karunamoorthy

Abstract:

Mobility of the robot depends on the vehicle dimensions, locomotion principles and wheel characteristics.

The function of the wheel is to carry the load and to produce the traction force. The main factors of wheel terrain interaction are bearing capacity of ground, traction performance of the wheel and geometry of terrain profile. In this paper the system and control concepts of the wheeled robot is discussed in more detail, within the framework provided by the wheel terrain contact model. The dynamic model of the wheeled robot is presented by considering contact forces of the wheel due to their relative motion of the wheel and contact plane. Finally, a dynamic relation is introduced and results are presented in terms of forces, torques and displacements related to wheel terrain interaction. To estimate the forces in the system arising from the interaction between a deformable wheel and rigid terrain using the software package Ansys 10.0. Simulations were performed using MatlabSimulink program and the results are shown that the proposed controller can overcome the influences the effect of contact forces in order to achieve the desired trajectory.

Keywords: wheeled robot, dynamic model, wheel terrain interaction, Ansys analysis, Matlab-Simulink.

1. Introduction

The required condition of wheeled mobile robot in an environment is stable and fast navigation to reach the target. But the change of characteristics of the robot movement will cause unstable drive according to the relationship between the driving wheel and ground.

Most of the control algorithms do not consider the physical dimensions and capabilities of the mobile robot within its environment. When the wheel torque generates a turning momentum along the wheel rim, it develops resistive forces on the motion. The integration of longitudinal shear stresses over the entire contact path represents the tractive force. The tractive force can be used to overcome the rolling resistance and to generate pulling force. The actual wheel ground interaction needs to be considered in order to improve the robot motion control. Wheeled robots are almost always designed so that all wheels are in ground contact at all times. Thus, three wheels are sufficient to guarantee stable balance. Instead of worrying about balance, wheeled robot research tends to focus on the problems of traction and stability, maneuverability, and control which can provide sufficient traction and stability for the robot to cover all of the desired.

30

Articles

Mobile robots have actuated wheels whose slip rate, rolling, inertia moments, and mass distribution contribute to the forces exerted on the structure of the vehicle thus affecting the accuracy and maneuverability of the robot [1]-[3]. In the model [4], the steady state wheel forces and torques are generated as functions of the longitudinal and lateral slip. Depending on the slip definition the dynamics of a wheel depends on the vehicle velocity or the angular velocity of the wheel. The Hertzian pressure distribution was assumed for the normal surface contact load over a contact area. The tangential forces in both the rolling and lateral directions were considered and were assumed to be proportional to the Hertzian pressure. Theory of vehicle dynamics [3], a well established discipline in automobiles dealing with dynamic properties of rolling motion, has revealed that different nonlinear dynamic effects and disturbances will be generated in different wheel ground interaction conditions.

Motion planner minimizes the distance between the present robot location and the desired end location. A local level motion planner attempted to attain the goals set by the higher level [5]. This was done by computing wheel accelerations, contact forces, equations of motion and the new state of the deformable regions in the terrain [6]-[8].

This algorithm also incorporated the wheel ground interaction and a bounded control torque constraint.

Nilanjan Chakraborty and Ashitava Ghosal [9] developed a hybrid parallel mechanism with the wheel ground contact described by differential equations which take into account the geometry of the wheel, the ground and the nonholonomic constraints of no slip. The workspace for the WMR is not always ideal and usually packed with various forms of disturbances including frictions, irregular terrains, obstacles in robot’s path, parametric changes and uncertainties within and outside the system, making it almost impossible to model all these disturbances and incorporate them into the dynamics of the WMR [10], [11]. Recently adaptive methods are used to compensate the effect of uncertainties in dynamic model and to configure the vehicle to adapt to terrain variations and allow rolling of wheels. Thus, in order to ensure a more robust and accurate operation of the mobile robot, a disturbance compensation scheme should be incorporated into the operation of the WMR.

2. Wheeled Mobile Robot Model

This paper analyzes the vehicle dynamics of wheeled mobile robots with contact forces. Condition that describes the limits of contact stability in terms of contact forces, it is derived from the interaction between

Journal of Automation, Mobile Robotics & Intelligent Systems

a deformable wheel and rigid terrain. The best model for the continuous nature of the deformation and contact area is non linear finite element method. Implicit finite element methods (FEM) have traditionally been used to determine contact parameters during static and quasi-static loading conditions. The first one is based on a formulation using displacements and the second one is based on a mixed formulation using displacements and contact forces. The normal force is generated by the supporting normal load.

The tractive force is generated by the forward friction force and the lateral force, which is existed.

y- Direction

2.1. Geometric model of the robot The actual wheel-ground interaction needs to be considered in order to improve the robot motion control.

Here the terrain assumed to be rigid and the wheel deformable. Consider a wheel that rolls on a plane while keeping its body vertical as shown in Fig.1. Configuration of the robot can be described by a vector q = (x, y, q, j) of generalized coordinates, where x, y are the Cartesian coordinates of center of the rear axle, q measures the orientation of the robot body with respect to the X axis, and j is the rolling angle of the wheel. At the wheel ground contact point, the holonomic constraint is uzc = 0, which ensures wheel ground contact is always maintained.

Moreover, at each instant, nonholonomic constraints which prevents instantaneous sliding and these are uxc = 0 and uyc = 0.

VOLUME 5,

N° 4

2011

sistances, the dynamics of a single wheel as shown in Fig. 2 is written as, w = t r − M y − Fx re (2) Irj

I s d = t s − M z

(3) where Ir moment of inertia of the wheel about rolling, Is moment of inertia of the . wheel about turning, jw rolling velocity of the wheel, d turning velocity of the wheel, tr rolling torque, ts steering torque, My moment of rolling resistance, Mz moment of turning resistance and re effective radius of the wheel.

Mz

υ

Ry

My

Rx

Fig. 2. Wheel ground interactions.

The rolling motion generates a horizontal reactive force Rx and a lateral reactive force Ry while the twisting motion generates a pure reactive turning moment Mz in the vertical direction. Assuming that the ground is flat and does not deform, the above three quantities are defined in vehicle dynamics as, (4) (5)

l

(6) (7)

x- Direction

Fig.1. Three wheeled differentially driven mobile robot.

For simple dynamic model of the wheel is a thick cylinder that represents the middle cross section of the wheel and the linear velocity of the wheel center lies in the body plane of the wheel. The general dynamic equation of the wheel robot is given below, M ( q) q + C ( q, q ) + G ( q) + t d = B ( q) t + AT ( q) l

(1)

. where M(q) is the inertia matrix, C(q, q ) is a matrix containing the centrifugal and coriolis terms, G(q) is the gravity force matrix matrix, B(q) is the input transformation matrix, t is the input torque, AT(q) is the Jacobian matrix associated with the constraints, l is the constraint force vector and q is the state vector representing the generalized coordinates. td denotes the bounded unknown external disturbance. For the continuous nature of the deformation and contact, the non-linear finite element method is selected for the best model and the contact force is measured from the built in geometric model of a wheel and a terrain. When considering the motion re-

Where b – width of the wheel, m – longitudinal friction coefficient, mt – lateral friction coefficient. Based on the force and moment analysis for wheel in Fig. 2, the total resistive force acting on each wheel Req can be derived as:  Rx    Req ( q ) =  Ry   M eq   

The dynamic model is obtained from dynamic properties of mass, inertia moments, friction force, gravitation and wheel ground interaction. The orthogonal force components are vertical, longitudinal and lateral.

The lateral frictional forces also prevent the vehicle from sliding to unwanted directions. Several parameters of the terrain are used to estimate normal, lateral and longitudinal forces at the wheel contact patch. If the frictional force is less than the maximum value, the wheel position is not changed, if it is greater than or equal to maximum value, wheel is pulled in direction opposite to the friction force from the wheel position. The total resistive quantities are defined in vehicle dynamics as: Articles

31

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

(8) (9)

(10)

2.2. Contact model of the robot The resulting frictional forces can be defined by integration of all forces acting on the contact surface. The pressure distribution resulting from the normal contact can be calculated in the local reference. As a consequence, the tangential and the normal forces in the global reference can be calculated by integrating the contact pressures on contact of the X and Y axes for the tangential forces and for the normal force on the Z axis.

Fx = ∫∫ px dxdy

(11)

Fy = ∫∫ p y dxdy

(12)

M z = ∫∫ ( xp y − ypx )dxdy

(13)

At a point of the contact surface the projected force Fy on the Y axis is zero due to the symmetry of the vehicle structure. As a result, contact friction leads not only to a resultant force applied to the center of the area but also to a non-vanishing moment about the normal axis through the center of that area. This moment, Mz is a function of the size of the contact area A, wheel material, type of wheel ground contact, weight of the vehicle, etc. Since Mz opposes the steering motion, it should be added to Eq. (18) using a sign function. At the contact point, the contact force can be decomposed into normal and tangential components. Let Fx be the horizontal component of contact force and Fz be the normal component of contact force. Assume that the coordinate frame and centre of gravity are lying in symmetry axis of the wheels. So that the contact force Fy = 0 and Fz is expressed as the function of contact pressure.

This resultant frictional force is still acting, but the new distribution of the normal forces creates a net torque opposing the rotational contribution of the friction and causing an overall deceleration of the wheel’s forward velocity. The lateral wheel friction is a coloum friction, whose force takes two sign opposite values depending on the direction of turning of the vehicle. Therefore Mz can be rewritten as, (14) = ( Fxl − Fxr ) l / 2 − M z Iw

(15) (16)

(17) 32

Articles

N° 4

2011

Considering the motion resistances, the dynamic model of the robot Eq.1 is rewritten as: M ( q) q + C ( q, q ) + G ( q) = B ( q) t + t c

(18)

where is the torques generated by the contact forces.

The contact torques can be written under the following equation: t c = J ( q)T Req ( q, q )

Where J(q) is the Jacobian matrix of the constraint on the position of the points on which these contact forces are applied. In real situations, motion resistance generated by the wheel ground interaction always exists, so the actual governing dynamic equations of motion of the robot are given by equation (19) rather than equation (1). where

M ( q) q + R ( u, w) = B ( q) t

m 0 M ( q) =   0 I

 RX R ( u, w) =  0

(19)

0   sgn ( u)  M z   sgn ( w) 

1  1 1 B( q) =  r l / 2 − l / 2 

In the case of trajectory tracking, control algorithms that consider wheel ground interaction are expected to demonstrate better tracking performance than those that do not consider the wheel ground interaction.

3. Simulation Results and Discussion

In this section, motion control scheme is considered with wheel ground interaction. Based on the dynamics of the mobile robot represented in equation (19), a control algorithm is proposed and modeled using Matlab-Simulink.

Simulink model of the robot motion control is shown in Fig. 4. The model parameters taken for this simulation are M = 100 kg, r = 0.1 m, b = 0.05 m, l = 0.5 m, I = 20 kg/m2, u = 3 m/s, w = 0.5 rad/s.

In this simulation the controller provides the desired trajectory, desired velocity and desired accelerations to the robot body. The 3D model of mobile robot wheel and terrain is created and analyzed using ANSYS10.0. All the external loads are applied at the wheel center. The contact was created by using Ansys software; here, wheel is contact element and terrain is target element. For contact CONTA174 and TARGET170 elements are used for 3D model. Friction effect is included into the material properties of the contact element. The material properties are listed in the Table 1. Then the contact region is finely meshed using a sub model approach. Next, quasi-static analysis is performed for the full model and the contact pressure results is plotted as shown in Fig. 3, and the value of contact force of wheel is calculated from the simulated results.

The developed fuzzy controller for this simulation is: angle and distance errors as inputs. During the robot movement, it moves whether in a straight line or circular arc, and creates the position and orientation errors which depend on the path. Designed FLC has three inputs and two outputs. Inputs are: linear distance errors and orientation angle error. Outputs are: the linear and angular

Journal of Automation, Mobile Robotics & Intelligent Systems

Table 1. Materials properties of the wheel.

Material

Youngâ&#x20AC;&#x2122;s modulus, N/mm2

Poisson ratio

Density, kg/mm3

Concrete

48x103

0.2

2.5x10-6

0.025x103

0.499

1.2x10-6

Polyurethane

VOLUME 5,

N° 4

2011

(a)

velocities uc and wc. The adaptive control of WMR with the dynamic model is to implement an adaptive control with the set of frictional contact force parameters in order to achieve the desired trajectory. To set the desired . . accelerations ud and wd by specifying required forces and torques by the equations 16 and 17.

(b)

Fig. 3. Contact stress distribution on the robot wheel.

Simulated calculations: Contact force at node 1281 = 277.83 N Contact force at node 2148 = 328.06 N Contact force at node 1669 = 274.46 N Total contact force = 277.83 +328.06 + 274.46 Fc = 880.35 N It can be observed from Fig. 5 that the tracking errors of dynamic controller with wheel ground interaction are much less than that of simple dynamic controller. This result shows that both position and orientation tracking errors can be reduced substantially when the wheel

Fig. 5. The convergence of the state errors in trajectory tracking a) without contact force b) with contact force.

Fig.4. Simulink model of integrated kinematic and dynamic with Fuzzy controller.

Articles

33

Journal of Automation, Mobile Robotics & Intelligent Systems

(a)

VOLUME 5,

N° 4

2011

(b)

Fig. 6. The developed torques in left and right wheels a) without contact force b) with contact force.

ground interaction is considered in resistive forces. The controller is able to recover from the error and stabilizes the robot to the desired trajectory, even if wheel ground interaction parameters are variable during motion.

When implementing the computation of Fc will add the contributions to the dynamic model. While position and velocity errors are rapidly compensated for very small changes in wheel ground interaction parameters and Fig. 6 shows the results of the controller that represents the torques and to the wheels of autonomous robot.

4. Conclusion

The dynamic model of the wheeled mobile robot was constructed with wheel ground interaction and the robot parameters were computed to provide robot motion. This paper analyzes the vehicle dynamics of wheeled mobile robots with resistive moment of the contact forces.

Adaptive contact force distribution scheme is proposed to satisfy the stable contact condition. The genetic fuzzy controller is proposed is used to estimate the influences the effect of contact forces and its effectiveness is demonstrated by simulation. Future works may integrate the resistive contact forces into dynamic model to implement the proposed methods on the real robot.

AUTHORS

R. Vivekananthan* – 1 Assistant Professor, Government College of Engineering, Salem, India.

E-mail: rvivek704@rediffmail.com, L. Karunamoorthy – Professor, College of Engineering, Guindy, Anna University, Chennai, India.

E-mail: karun@annauniv.edu

References

[1] Tanner H.G., “ISS properties of nonholonomic vehicles”, Systems & Control Letters, no. 53, 2004, pp. 229 – 235.

[2] D’Andrea Novel B., Bastin G., Campion G., “Control of nonholonomic wheeled mobile robots by statefeedback linearization”, Int. Journal of Robotics Research, vol. 14, no. 6, 1995, pp. 543–559.

34

Articles

[3] Danwei Wang, Guangyan Xu, “Full-State Tracking and Internal Dynamics of Nonholonomic Wheeled Mobile Robots”, IEEE/ASME Transactions on Mechatronics, vol. 8, no. 2, 2003, pp. 203-214.

[4] Van Wyk D. J, Spoelstra J., de Klerk J. H., “Mathematical modelling of the interaction between a tracked vehicle and the terrain”, Appl. Math. Modelling, vol. 20, November 1996.

[5] Rill G., “Wheel Dynamics”. In: Proceedings of the XII International Symposium on Dynamic Problems of Mechanics DINAME 2007, ABCM, Ilhabela, SP, Brazil, 26th February – 2nd March 2007.

[6] Sun S., “Designing approach on trajectory-tracking control of mobile robot”, Robotics and Computer-Integrated Manufacturing, vol. 21, no. 1, 2005, pp. 81–85.

[7] Siegwart R., Lamon P., Estier T., Michel Lauria, Ralph Piguet, “Innovative design for wheeled locomotion in rough terrain”, Robotics and Autonomous Systems, vol. 40, 2002, pp. 151–162 [8] Chang Y.-Ch., Yamamoto Y., “Path planning of wheeled mobile robot with simultaneous free space locating capability”, Intel Serv Robotics, no. 2, 2009, pp. 9–22.

[9] Ray L.R., Brande D.C., Lever J.H., “Estimation of net traction for differential-steered wheeled robots”, Journal of Terramechanics, vol. 46, 2009, 75–87.

[10] Chakraborty N., Ghosal A., “Kinematics of wheeled mobile robots on uneven terrain”, Mechanism and Machine Theory, no. 39, 2004, pp. 1273–1287.

[11] Kozlowski K., Pazderski D., “Modelling and control of a 4-Wheel skid steering mobile robot”, International Journal of Applied Mathematics and Computer Science, vol. 14, no.4, 2004, pp. 477–496.

[12] Gonzalez A, Ottaviano E., Ceccarelli M., “On the kinematic functionality of a four-bar based mechanism for guiding wheels in climbing steps and obstacles”, Mechanism and Machine Theory, vol. 44, 2009, pp. 1507–1523.

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N° 4

2011

Modeling and Identification of Actuator for Flap Deflection Submitted 26th April 2011; accepted 31st August 2011

Martyna Ulinowicz, Janusz Narkiewicz

Abstract:

The electromechanical actuator (EMA) model is presented with the methods for identifying its design parameters. The actuator is a part of the system for flap deployment on the commercial transport airplane. The differential equations with the feedback control describe behaviour of the actuator deflection. There are two concepts of drive system simultaneously considered: a high torque/low speed (HT/LS) and a geared low torque/high speed (LT/HS). For parameter identification in both cases the Maximum Likelihood (ML) method with two minimisation algorithms: linearized Gauss-Newton and Levenberg-Marquardt is applied. Both approaches for each of two design solutions were effective, while tested on hypothetical data.

Keywords: electromechanical actuator, system identification, maximum likelihood method.

1. Introduction

The concept of More Electrical Aircraft MEA, which has been investigated for some time, recently attracts more interest due to trend for “greening” aviation operations. In aircraft control, the electric system will replace the hydraulic one used nowadays, resulting in saving weight and operational costs [1].

In the project NEFS - New Track integrated Electrical Single Flap Drive System, funded by EC under 6 FP the concepts are investigated for replacing conventional hydraulic drive system used for deploying wing flaps of large transport aircraft by individual, distributed electrical drive actuators integrated into each flap track beams [5].

The system performance is tested at the laboratory rig, but the system failures and their influence on aircraft is investigated by simulations. To make the integrated simulations of system and aircraft reliable, the actuator simulation model should reflect behaviour of a real device in a correct way. The required accuracy level of simulation model may be achieved by identification of the system model parameters using data from laboratory tests.

In this paper the actuator model and identification methods are described with some preliminary sample results.

System identification is a process of determining the parameters of mathematical model of the system using data from experiments, sometimes specially planned for the model used. Developed model should describe correctly system behavior to be reliable for further investigation of system dynamics. For assumed structure of the model, the estimation method is selected. The choice of identification algorithm depends on the knowledge about

object being identified. Usually computation efficiency and available experimental data are major determinant.

The final validation of the identified parameters is based on the experimental data not used for estimation of model parameters [1].

In the paper parameters of electromechanical actuator system nonlinear model are identified using ML method.

2. Actuator model

In the project two concepts of drive systems are considered: a High Torque/Low Speed (HT/LS) and a geared Low Torque/ High Speed (LT/HS).

2.1. The High Torque/ Low Speed concept In case of HT/LS solution the EMA system contains two drive stations 1 and 2 connected to aircraft flap. Each drive station contains a ball-screw driving the flap connected to common DC motor by U-joint and a gearbox.

The motor is controlled by performing assumed function of flap deflection.

The actuator is described by the mechanical equation: J

d W(t ) = M n − ( M B1 + M TPER1 + M TPER 2 ) , dt

(1)

where: Ω – motor shaft angular velocity, J – inertia moment of rotating parts reduced to the shaft [3].

The torque generated by the motor is calculated as: M n = kn I A ,

(2)

where: kn – motor constant, I A – armature current.

The M TPER1 and M TPER 2 are the external torques from station 1 and 2.

Each station contains bevel gear, U-joint and ballscrew.

The M B is the primary brake torque applied to the motor modelled as:  W   − KB    W0,5 Tmax   , (3) M B = M B max − 0,5M B max 1 − e     where: K B – electromagnetic brake constant coefficient, M B max – maximal torque value (for Ω = 0 ), Ω0,5 M Bmax – angular velocity value for half a maximal torque M B max .

In bevel gear the motor angular velocity Ω is reduced to ΩTP with reduction ratio iTP : WTP (t ) =

W iTP

(4)

External torque acting on the motor shaft is calculated as the sum of external torque from the U-joint M UER , torque resulting from the losses in bevel-gear M TP and Articles

35

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

secondary brake torque M B 2 ,which is equal to zero when the brake is not released: M TPER

M = UER + M TP + M B 2 iTP

(5)

The bevel gear moment due to mechanical energy losses is composed of viscous damping represented by BTP coefficient and Coulomb friction CTP [3]: M TP = BTP ΩTP (t ) + CTP sign(WTP (t ))

(6)

The screw is connected to the gear-box by U-joint where the angular rotation is changed: WU (t ) = gU WTP

where:

  cos b gU =  , 2 2  1 sin cos − b q  TP  qTP

= q i , TP

(7) (8) (9)

β – angle between the input and output shaft axis, θ – motor shaft angle of rotation.

The additional moment M U due energy losses in U-joint is assumed in the same form as for the gear box: M U = BU ΩU (t ) + CU sign(WU (t )) ,

(10)

so the U-joint output shaft moment takes the form: M UER =

M TSER + MU , cos b

(11)

In the ball screw the shaft rotation is transferred into nut translation: dy (t ) = r ⋅ tg g ⋅ WU (t ) dt

(12)

where: r – screw rolling radius, γ – nominal lead angle.

External loads FER from the flap act on the ball-screw.

The moment on the screw resulting from external loads is calculated as: M ER = − r ⋅ FER1 sin g cos g ,

(13)

where r denotes screw rolling radius and γ stands for nominal lead angle.

The output moment from the ball screw is composed from moment resulting from external loads M ER and additional torque losses in the ball screw M TS modelled analogically to the damping in bevel-gear and U-joint: where:

M TSER = M TS + M ER , M TS = BTS WU (t ) + CTS sign (WU (t ) ) .

(14) (15)

2.2. The Low Torque/ High Speed concept The LT/HS concept is asymmetric in contrary to the previously described. The motor is controlled by the voltage and form electrical side is modelled in the following form: 36

Articles

k R U I A = − A I A − s W + A LA LA LA ,

N° 4

2011

(16)

where: I A – armature current, U A – control voltage, k s – motor magnetic flux coefficient, LA – inductance of the armature, RA – resistance of the armature.

The torque generated by the motor is calculated analogically to these in HT/LS solution: M n = kn I A

(17)

The whole EMA is described by the inertia moment

J of all rotating parts reduced to the motor shaft axis

multiplied by angular acceleration of the motor shaft Ω equate to the difference between motor torque and external moment M MGER from all connected devices from station 1 and 2. = M −M JΩ n MGER

(18)

Output moment from the motor is transferred by the shaft which changes it due to the stiffness: M ISER = M MGER + DIS ⋅èMG

(19)

,

where: θ MG – magnetic gearbox output shaft angle of rotation, M TPER – external torque from the bevel gear, DIS – input shaft stiffness coefficient, which is assumed to be zero in the model.

The magnetic gearbox placed just after the motor changes output angular velocity Ω from the motor into Ω MG with reduction ratio iMG .The external torque on the motor shaft is calculated as: M MGER =

M ISER + M MG (WMG ) iMG

(20)

where: iMG – bevel gear reduction ratio, M ISER – torque on the output shaft, M MG – moment due to losses in the gearbox analogically calculated as in eqs. 6 and 10 : M MG = BMG WMG + CMG sign (WMG ) .

(21)

Magnetic gearbox is connected with the first bevel gear by the input shaft which modifies output moment in a following way: M ISER = M TPER 2 + DIS ⋅ q MG

,

(22)

M TPER 2 – external torque from the shaft of station 2, DS – shaft stiffness coefficient, which is assumed as zero in this research.

There are two shafts connected to the output of the first bevel gear. One of them leads to the first ball-screw by the first U-joint and the second shaft which, after one more bevel-gear and U-joint, connects second ball-screw to the system. The first bevel gear reduces angular velocity Ω MG to ΩTP1 with ratio iBG1 in case of the first station and at the same time but with different ratio iBG 2 it reduces angular velocity to Ω MG 2 for the second station: WTP1 =

WMG , iBG1

(23)

Journal of Automation, Mobile Robotics & Intelligent Systems

WMG 2 =

WMG iBG 2

VOLUME 5,

(24)

External torque on the input shaft of the first bevel gear is calculated as the sum of moment coming from the first (MUER1) and second (MMG2) station divided by appropriate reduction ratio, moment of losses in the gearbox M TP1 and moment from the brake M B 2 , which is equal to zero when the brake is not released: M TS = BTS WU (t ) + CTS sign (WU (t ) )

(25)

The first station contain U-joint which influence on the angular velocity. These change may be described by coefficient gU defined as in eq.8 : WU (t ) = gU WTP .

(26)

2.3. Control The input signal for the system is the required flap deflection angle ϕ r (Fig.1).

This signal is transformed into required carriage position yr which is compared with the position measured by the sensor ys and changed proportionally into angular velocity of the screw Ω r command in ACE module using proportional regulator with p ACE coefficient: Ω r = p ACE ( yr − ys ) ,

(27)

the U-joint output shaft moment is calculated as: M UER1 =

M TSER1 + M U 1 (WU 1 ) , cos b1

(28)

where: β1 – angle between input/output shaft directions.

The U-joint couple ball-screw with the first bevel-gear.

On the ball screw of drive station 1 there are moments resulting from external loads M ER1 acing on the flap mounted to the carriage, which is connected to the ball nut and torques due to losses in the ball screw M TS 1 : M TSER1 = M TS 1 + M ER1

(29)

The second shaft, which is connected to the output of the first bevel gear, leads to another bevel gear in which angular velocity as well as the moment are changed into, respectively: WTP 2 = M MG 2

WMG 2 iTP 2

M = UER 2 + M TP 2 + M B 22 . iTP 2

(30)

2011

ond bevel gear is coupled by U-joint with the ball-screw, models of each elements are similar to these from the drive station 1.

The output torque from the U-joint M UER 2 is calculated analogically to the M UER1 . The moment due to losses in the gearbox M TP 2 is defined in the same way that those from the first bevel gear and magnetic gearbox (eq. 21), while moment from the second brake M B 22 is equal to zero when the brake is not released.

As the moment due to losses has a form: M U 1 = BU 1WU 1 + CU 1 sign(WU 1 )

N° 4

(32)

where: ys = 0.5 ⋅ ( y1 + y2 ) , (33) y1 , y2 – carriage position of station 1 and 2 respectively.

The control signal for the motor is calculated in PCE module as: – in HT/LS concept IA =

pPCE (Wr − Ws ) , kn

(34)

– in LT/HS concept UA =

pPCE (Wr − Ws ) , kn

where Ω s stands for measured angular velocity and pPCE is a proportional control coefficient. Appropriate value of the signal should be contained in the boundaries follow from DC motor performance: M min ≤ pPCE (Wr − Ws ) ≤ M max

(35)

As a result in HT/LS solution actuator is described by the state equations:

(31)

From this moment drive transmission in drive station 2 is analogical to these from drive station 1. As the sec-

 1 W = J  M n − ( M B1 + M TPER1 + M TPER 2 )  ,  y 1 = r ⋅ tg g ⋅ WU 1  y = r ⋅ tg g ⋅ W U2  2 

(36)

Fig. 1. Electromechanical actuator control system.

Articles

37

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

where the model states are: x = [W

y1

y2 ]

(37)

T

N° 4

2011

3.1. The Low Torque/ High Speed concept The output vector of the LT/HS system is formulated as:

The control equation is: IA =

pPCE ( pACE ( yr − ys ) − Ws ) kn

(38)

The observation parameters

(42)

z are:

While in the LT/HS concept EMA can be described by the fallowing system of equations: ks RA UA  I A = − L I A − L W + L A A A   1 W = ( M n − M MGER ) J   y 1 = r ⋅ tg g ⋅ WU 1   y2 = r ⋅ tg g ⋅ WU 2

(39)

where the model states are: x = [I A

W

y1

y2 ]

T

(40)

The control equation is: UA =

pPCE ( pACE ( yr − ys ) − Ws ) kn

(41)

The actuator model is nonlinear with respect to states and is formulated in time domain.

3. Identification methods The aim of the EMA system identification is to calculate the actual values of mathematical model parameters.

(43) where: I As denotes the current passed to the DC motor, Ω s – angular velocity on the motor output shaft, ΩTP1s , ΩTP 2 s - angular velocities on the bevel-gears, ΩU 1s , ΩU 2 s – angular velocities on the U-joints. The external torques acting on ball-screws of two stations are M ER1s and M ER 2 s , while torques before the ball-screws states as M TSER1s and M TSER 2 s . The torques M UER1s and M UER 2 s are those before the U-joints, M TPER1s , M TPER 2 s are moments before the bevel-gears.

The system is observable. The proportional regulators coefficients p ACE and pPCE are taken from regulator adjustment on the test rig.

The estimated parameters are combined in column vector Θ : Q = [J CTS 2

BTS 1 CTS 1 BU 2

CU 2

BU 1 CU 1

BTP 2

CTP 2

BTP1 CTP1 kn

p ACE

BTS 2 pPCE ] (44) T

Considering that in U-joint the β angle reaches maximally 5 degrees, the values of both gU depending of angle θTS is close to 1, thus it is assumed that gU 1 = gU 2 = 1 .

Parameters β1 , β 2 , γ , iTP1 , iTP 2 , r are known from the system design.

Fig. 2. Comparison of output signals of the HT/LS model with identified parameters using ML_GN algorithm and ML_LM one with outputs considered as measurements.

38

Articles

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

3.2. The Low Torque/ High Speed concept In LT/HS solution system there are following outputs: y = [W WTP 2

y1

y2

IA UA M ER1

WU 1 WU 2

M UER1

M UER 2

M MG 2

WMG

WTP1 WMG 2

M ER 2

M TSER1

M TPER

M MGER

Readouts from sensors creates vector z = [Ws WTP 2 s

y1s WU 1s

M UER1s

y2 s

I As U As

WU 2 s

M ER1s

M UER 2 s

M MG 2 s

WMGs M ER 2 s

M TPERs

M TSER 2

p ACE ]

z as :

WTP1s M TSER1s M MGERs

WMG 2 s M TSER 2 s

p ACEs ]

(46)

The estimated parameters vector Θ is defined: Q = [J

BTS 1 CTS 1

BTS 2

CTS 2

CMG

kn

BU 2 RA

LA

BU 1 CU 1

CU 2

BTP 2

ks

p ACE

BTP1 CTP1 CTP 2

pPCE ]

BMG

J (Q) = det( R) ,

(47)

The assumption from the previous case concerning gU is still valid ( gU 1 = gU 2 = 1 ). System design impose values of such parameters as: β1 , β 2 , γ , iBG1 , iBG 2 , iTP 2 , iMG , r.

3.1. Identification algorithm The Maximum Likelihood (ML) method in time domain with two alternative minimisation methods: linearized Gauss-Newton (ML_GN) and Levenberg-Marquard (ML_LM) was chosen for identification in this research.

In both cases, ML estimates are obtained by minimization of the cost function J ( Θ, R ) , which in case of considered EMA system is assumed as [1], [4]:

2011

(48)

where covariance matrix R is defined in the following form: R=

(45)

N° 4

1 N data

N data

∑  z (t ) − y (t ) ⋅  z (t ) − y (t ) k =1

k

k

k

T

k

(49)

The ML_GN algorithm with relaxation strategy contains the following steps [2], [4]: 1) Assuming initial values of parameters Θ ( t0 ) and states x ( t0 ) .

2) Computation of gradients of cost function with respect to parameters that are identified and gathering them in the matrix G: Gi =

∂J . ∂Q

(50)

3) Information matrix (Hessian) F evaluation .

4) Solving the following equation with respect to ∆Θi: Fi ⋅ DQi = −Gi .

(51)

5) Updating parameters: Qi +1 = Qi + DQi .

(52)

6) Checking convergence or maximum iteration number limit J i − J i −1 < 10−5 Ji

(53)

The same algorithm but with Levenberg-Marquard (ML_LM) method of optimization used for cost function minimization may be applied in the fallowing way [2], [4]:

Fig. 3. Comparison of output signals of the LT/HS model with identified parameters using ML_GN algorithm and ML_LM one with outputs considered as measurements.

Articles

39

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

1) Fallow steps 2-5 from the ML_GN algorithm and compute the cost function J ( Θ ) = det( R) , which will be further considered as J i −1 ( Θ ).

2) Solve the equation:

( F + lI ) ⋅ DQi = −G

(54)

with respect to ∆Θi , for Levenberg-Marquard (LM) parameter λ = λ i −1 and λ = λ

i −1

, where v is a reducv tion factor ( v > 1) and I -the identity matrix.

The LM parameter λ enables to control the update search direction. If λ → ∞ the algorithm is reaching steepest- descent variant but while λ → 0 it becomes closer to the Gauss-Newton.

3) Update parameters for each of above solution

i −1 ∆Θi ( λ i −1 ) and ∆Θi λ .

(

ν

)

4) Compute respective cost functions

(

)

( (

i −1 J i ∆Θ ( λ i −1 ) and J i ∆Θ λ

ν

))

5) Compare two above cost functions:

(

)

( (

i −1 J i ∆Θi ( λ i −1 ) and J i ∆Θi λ

ν

))

with the one from previous iteration and choose those which corresponds to the greatest reduction by reaching the smallest value.

6) Select parameters update corresponding to the cost function chosen in previous step.

7) Update parameters vector (Eq. 31).

8) Check convergence or maximum iteration number limit (Eq. 32).

Both identification procedures were implemented in Matlab software environment and applied for the system parameters estimation.

4. Results

Instead of laboratory test data, there were used perturbed by random signal simulation ones to make sure that the algorithms are working correctly and the calculations derived by them are reliable. The results tests of both algorithms for the HT/LS and LT/HS concepts are presented on the Figure 2 and 3 respectively. On each diagram of these figures, the measured output signal is compared to output signals of the model with identified parameters using ML_GN algorithm and ML_LM one.

The results indicates that implementation of both methods leads to accurate parameters and recreates behaviour of EMA system in each configuration (HT/LS and LT/HS) properly.

Currently, the laboratory test are in progress and EMA system will be identified as soon as the data will be available.

5. Conclusions

Two concepts of the EMA system were considered: a High Torque/Low Speed (HT/LS) and a geared Low 40

Articles

N° 4

2011

Torque/ High Speed (LT/HS). Dynamical models describing EMA systems behaviour in both configurations in form of differential equations is developed and Maximum Likelihood method with two alternative optimization algorithms (linearized Gauss-Newton and Levenberg-Marquardt) is implemented in Matlab software environment. As a result the system model parameters are successfully estimated using test data. After laboratory test completion, the experiment data will be used for the system parameters identification and more reliable model for simulation will be obtained.

Acknowledgments

The paper was prepared under EC funded 6 FP project NEFS – New Track integrated Electrical Single Flap Drive System, Contract No 030789, coordinated by EADS IW.

AUTHORS

Martyna Ulinowicz* – Ph.D. Student, Department of Automation and Aeronautical Systems, Warsaw University of Technology, Warsaw, 00-665, Poland, martyna. ulinowicz@meil.pw.edu Janusz Narkiewicz – Professor, Department of Automation and Aeronautical Systems, Warsaw University of Technology, Warsaw, 00-665, Poland, janusz.narkiewicz@mel.pw.edu.pl *Corresponding author

References [1] Andrzejczak M., Methods applied to aircraft identification, III International Conference “Intelligence, Integration, Reliability” papers, Kijev, 2010.

[2] Bonnas J.F., Numerical optimization: Theoretical and practical aspects, Springer-Verlag: Berlin, 2006.

[3] Esfandiari R.S., Bei L., Modeling and analysis of dynamic systems, CRC Press/Taylor& Francis Group: Boca Raton, 2010.

[4] Jategaonkar, Ravindra V., Flight vehicle system identification, A Time Domain Methodology, American Institute of Aeronautics and Astronautics, Reston, Virginia, 2006.

[5] New Track integrated Electrical Single Flap Drive System (NEFS), 6th European Union Framework Programme research project, http://www.nefs.eu/

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

N° 4

2011

DIASTER - INTELLIGENT SYSTEM FOR DIAGNOSTICS AND AUTOMATIC CONTROL SUPPORT OF INDUSTRIAL PROCESSES Submitted 27th June 2011; accepted 1st September.

Michał Syfert, Paweł Wnuk, Jan Maciej Kościelny

Abstract: The paper presents general description of the DiaSter system implementing advanced methods of modeling, diagnostics and supervisory control for industrial processes. The scope of the tasks realized in the system as well as the system software platform were characterized, in particular: the software structure, central archival and configuration databases, the way of data exchange in the system and the modules of modeling and calculations.

ling, supervisory control, optimization, fault detection and isolation. Thanks to its open architecture, connections to virtually any automation system are possible and easy to implement. The position of DiaSter system in industrial process management tasks is presented in Fig. 1.

Keywords: diagnostic and supervisory systems, automatic control support, software platform.

1. Introduction In recent years there have been significant developments in techniques for modeling [3], [5] advanced control [1], [11] and process diagnostic [2], [4], [6], [10]. Modern computer systems enable the application of complex computational algorithms, developed on the basis of recent re-search in computing, automatic, diagnostics and know-ledge engineering. They use artificial intelligence techniques such as artificial networks, fuzzy logic, rough sets, evolutionary algorithms and methods for knowledge discovery from databases [9].

Development of methods outlined above in conjunction with the rapid progress of computer technology (computing power of new generation processors, memory capacity, speed of data transmission in LAN networks and field bus, the Internet growth) leads to a new generation of control systems. It is characterized by the introduction of advanced software for modeling, control and diagnosis processes. This software is a special software modules, which are part of the automation system, or expert systems integrated with automation systems.

Such software package is DiaSter system. It is developed by a research team composed of specialists from the Warsaw University of Technology, Silesian University of Technology, Rzeszów University of Technology and University of Zielona Gora and supported by polish grant: An intelligent system for diagnosis and control of industrial processes support DiaSter. It is brand new, functionally and software extended version of AMandD system, developed in Institute of Automatic Control and Robotics, Warsaw University of Technology [7], [8]. The system is dedicated for use in the energy industry, chemical, pharmaceutical, metallurgical, food and many other.

The system is world-wide unique solution. It includes implementation of a wide range of the latest algorithms in the field of intelligent computation, used to system mode-

Fig. 1. The position of DiaSter system in respect to the hierarchy of tasks (and system classes) connected with production process.

2. DiaSter system functions DiaSter system allows to realize following functions: Process variables processing. System gives a possibility to freely design processing paths for each variable.

Simulation and modeling. In the system is possible to create models on the basis of real physical description of modeled phenomena, or to identify model on the basis of measured data.

Virtual sensors and analyzers. Virtual sensors based on the analytical, neural-network or fuzzy models can be treated as information redundancy for real measurements.

Process simulators. Developed within DiaSter system process models can be used as simulators for operator training, or to test new strategies of control or to optimize the process set-points.

Fault detection. In the system methods based on analytical, neural-network or fuzzy models as well as heuristic tests utilizing different kinds of relations between process variables are used for fault detection.

Such detection methods give a possibility for early detect of much higher number of faults than classic alarm system.

Articles

41

Journal of Automation, Mobile Robotics & Intelligent Systems

Fault diagnosis. In DiaSter system two independent diagnostic inference mechanisms are implemented.

First one allows fault isolation based on the analysis of the set of actual fuzzy diagnostic signal values and the relation between faults and symptoms stored in the base of knowledge. Second uses belief networks and multi-facet models.

Monitoring of degradation degree of technological equipment. The system allows to early detection and tracking of slowly developing destructive changes. In addition an approximation of time left to critical state can be done.

Support of process operators decisions. On the basis of elaborated diagnosis the DiaSter system can additionally support the process operators decisions in abnormal and emergency states.

Knowledge discovery in databases. There are mechanisms for knowledge discovery in SCADA and DCS databases implemented in system, designed mainly to support diagnostic reasoning.

Advanced control and optimization. The algorithms of superior predictive control (DMC, GPC) using linear models, as well as algorithms based on nonlinear models, in particular fuzzy and neural networks based, are implemented in the DiaSter system.

Superior tuning and adaptation of control loops.

DiaSter is able to carry out both preliminary (pre-tune) and precise (fine) tuning of control loops with step response or frequency methods.

VOLUME 5,

N° 4

2011

There are also available several utility module. There are not responsible for main information processing but are very useful during system configuration and tests.

Good examples are communication scanner monitoring messages exchanged between calculation modules or module reconstructing archival process variables from database (or files) in real-time, possibly with some acceleration factor.

Advanced system functionality is delivered, developed and realized by specialized packages, called user packages. They are realized as independent software modules cooperating with other system components or in the form of plug-ins of software platform modules (modeling, processing and visualization).

4. Central archival and configuration databases In order to make possible the cooperation of all system components the following elements were worked out: common information model, central configuration environment and central archival database.

3. System structure DiaSter system consists of a software platform and several specialized packages cooperating with its use.

Fig. 3. Main objects and relations of the information model.

Fig. 2. General structure of DiaSter system. Black blocks symbolize specialized packages realized as different system components cooperating with the platform.

The software platform is a core of a system. Its main components are: archival data processing and model identification module, module of on-line processing of system variables, visualization module, central configuration and archival databases and communication server. They are described in the consecutive sections.

42

Articles

The information model specifies the definition of the information being exchanged between the system components (Fig. 3). It constitutes the basis of the data exchange in respect to configuration as well as processed system variables. First of all, the information model defines: the elements connected with process/installation description, e.g., division into subsystems and particular components including the specification of logical as well as physical units, specification of process variables and their groups,

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

N° 4

2011

the logical components of processing algorithms connected with realized by the system tasks, e.g., models, residuals, calculated and simulated variables, faults, the relation between the above elements, e.g., the relation “is a part of” and “is connected as input / output” defined between process subsystems and components or the relation “controls the behavior of” between residuals and monitored process components.

The information model also defines the types of system variables and the types of their values processed by the platform. It constitutes the framework of process and processed variables description. It is designed in such a way to be extendable by the particular user packages, or a group of packages. In this area the platform enables: creation (registering in the platform) user-defined data types, e.g. fuzzy signals, creation (registering in the platform) user-defined system variables types, e.g., efficiency indexes, the division of system variables into user-defined groups, e.g. control (CV), set-point (SP) and controlled (PV) and disturbance (D) signals, creating user-defined types of the process components, e.g. pipelines, valves, the possibility to register user-defined relations between different system elements.

The central configuration environment is a repository of configuration data for all system modules.

It consists of: Central configuration database. It is a relational database that stores the whole system configuration for particular application.

The set of configuration interfaces. They are used by all system modules and their components (including plug-ins) to get standardized access to the configuration data. They delivers the set of necessary and useful functions for manipulating the elements of system configuration.

The interface to user-defined databases. This interface enables to create, manage and access specialized databases assigned to system packages, e.g. the database of cases used by case-based reasoning methods (CBR) or diagnostic messages database. Such databases can be used to store any data (configuration or variables) typical for particular package, or a set of packages. The data stored in those databases do not have to be consistent with information model of the system platform.

Central configuration module. It is used to manage common for all system modules configuration data that is consist with information model.

The central archival database is capable to store any system variables defined in the system configuration. It can store the variables of build-in data types (process variables, residuals, faults, etc.) as well as user-defined data types. Also any type of variable values can be stored. There build-in as well as user-defined value types are handled.

The variable values can be stored in short- or long-term archives. Each of the system component (modules, plugins, build-in and user-defined) has access to specialized interface constituting common access to archival data.

Fig. 4. Example of extending platform information module by adding user-defined variable types and groups.

5. Data exchange DiaSter system is fully distributed software system with possibility to work on many PC-s connected with local area TCP/IP compliant network. Communication subsystem is a native solution, working with specially designed protocol independent on any external software.

Communication library delivers simple API to system modules programmers. Thanks to this library the connection of an external program to DiaSter system is reduced to loading library and setting IP address of communication server. To send a message only one function call is needed.

Base communication method in DiaSter platform is based on message transmission between modules in two modes: direct module to module, and broadcast to topics (publish/subscribe). In both modes in data exchange active role acts MRIaS server included in the platform. Server is only used to transmit information between modules and to provide transmission errors maintenance. It is not responsible for message analysis or storage. Communication is always made through TCP/IP sockets, regardless of whether the modules work on standalone computer or on multiple networked PCs. Mentioned mechanisms are mainly used to transmit the data rapidly and frequently changing, e.g. the values of measurements, the result of current calculations, etc. Any type of value acceptable by the software platform (floating point numbers, binary signals, fuzzy values, arrays, etc.) can be transmitted this way.

Additionally, it is possible to call remote procedures - both in direct communication (module A calls procedure executed by B and receives the results) and broadcasting (e.g. supervisory or configurator module send a “reconfigure” request to all modules within a single RPC call).

More sophisticated communication mechanisms are provided by some modules working on-line. Additional communication interfaces build on the basis of CORBA standard are used to access the properties and methods of each function block embedded in the on-line calculation module (PExSim), and to remotely control the process of simulation / calculations realized by this module. Similar Articles

23

Journal of Automation, Mobile Robotics & Intelligent Systems

mechanism is used in visualization. It allows to transmit complex and infrequently changing information and direct access to system objects.

6. Modeling module For the identification purposes the MITforRD (Model Identification Tool for Diagnosis and Reconstruction) module is designed in DiaSter system. It allows to create models without the knowledge of the analytical form of the relationship between modeled variables. MITforRD module allows to identify both static and dynamic models of different types starting from well-known linear transmittances to neural network or fuzzy logic based models.

The identification is carried out off-line using measurement data from central system archives. The models implemented in MITforRD module belongs to the group of partial parametric models of the process variables (time series). Models are obtained in a semi-automatic identification process. The deep knowledge of processes and their physical characteristics is not necessary – module effectively supports users not familiar with identification techniques during the whole identification process. The main features of the software includes: common interface to many kinds of models, wizards with default parameters for each identification step flexible, self-configurable distributed calculation environment allows the use of free computing power of offices PCs, plug-in based architecture allows to easily extend module functionality by the independent software vendors.

MITforRD supports the user in each identification step, starting at data acquisition, up to final analysis and verification of received model. All the time user can access to program options with simple and intuitive menu.

During process data analysis MITforRD allows to edit process archives inside advanced embedded editor and data visualization. Additional features are available via attached plug-ins: import and export of data, histograms, statistical parameters, frequency-domain analysis, correlations, transformation according to given mathematical expression, filtering, numerical differentiation and integration.An extra functionality is provided to work with process archives filling-in missing samples, validation of signals. Missing features can be added by the user by creating a new plug-in.

One of the main assumptions for DiaSter system was to make it flexible and expandable. Therefore, the main program MITforRD Model Builder does not implement a possibility of model estimation of any type. All model types are implemented as plug-ins. Similar mechanism is provided also in other DiaSter modules. All plugins are based on dynamic link libraries (dll's). In the basic specification of dll libraries only the export of functions is allowed. Thus, in whole DiaSter system additional layer is added, that allows to work with objects. This approach allows to create new plug-in in different programming languages (most commonly used is C++).

MITforRD module may employ a wide range of evolutionary algorithms in order to explore the structure of 44

Articles

VOLUME 5,

N° 4

2011

linear or fuzzy models. Such algorithms, besides many advantages, also have one serious disadvantage – usually are very time consuming. To reduce computation time MITforRD provides distributed computing environment.

It can be used to utilize free computation power of the classic of?ce PCs running on Windows and connected to the local area network. The environment does not require any change in the local PCs configuration and does not disturb its normal operation. The main PC with running MITforRD Model Builder is a control center for the calculations. There is a built-in calculation manager inside the Model Builder capable to distribute the calculations among local host as well as remote PCs available at the moment. The computation environment built-in into MITforRD module do not require configuration phase, and do not add time overhead caused by technologies like COM, CORBA, etc.

7. On-line calculation module The main element of DiaSter platform for on-line use is an calculation module called PExSim. It is dedicated for advanced system variables processing. The processing algorithms are written and stored in the form of configurable function block diagrams. The primary task for PExSim module is to process the information circulating in the system in a way defined by the user. From this point of view PExSim module can be treated as specialized programming language. An algorithm of information processing is defined graphically by creating so-called processing paths. Each path consists of a set of interconnected function blocks, which carries out various tasks on signals (see Fig. 5).

Fig. 5. Graphical programming of desired calculations in a form of processing paths.

Function blocks are provided as PExSim plugins, and are grouped into thematic libraries, e.g.: surces, sinks, statistical operations etc. Each function block has the set of parameters defining the way of signal processing by this block, and stored by the platform. Various block inputs and outputs can transmit data of various types (e.g.: floating numbers, fuzzy values, vectors etc.) including user data types. PExSim can run as stand-alone tool (simulator mode), or as a module working in a distributed system (multi-module mode). Multi-module mode allows to exchange the data with other DiaSter modules thus there is a need for synchronization external data with internal

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

simulation clock. To realize synchronization different modes of simulation triggering are available.

Each main path of the calculation module can be assigned to one of the following groups: Â&#x2014; synchronous paths. In this case the simulation kernel triggers given path with constant, previously defined time period, calling sampling time, or its multiplicity, Â&#x2014; asynchronous paths. These paths are triggered by events generated internally (via specialized block) or coming from outside world (a message, RPC call, etc.).

8. Visualization module The platform delivers also a visualization module (graphic user interface) called PExSim. It can be used to realized advanced operator interfaces. It is organized in a similar way as tools for configuring process mimics in supervisory, control and data acquisition systems (SCADA) or decentralized control systems (DCS). The set of synoptic screens organized in hierarchical structure is prepared during the configuration stage for particular application. Dedicated displays visualizing the values of system variables in particular way are placed on that mimics.

N° 4

2011

The current values of the system variables are automatically delivered by the communication server (including user-defined data types). The module automatically collects the data from central archival database when the history (short- or long-term) is analyzed by the user. Additionally, each display can retrieve the data from the user databases (specialized databases registered by specialized packages started on the platform) with the use of standard SQL interface.

In the case of a need to elaborate very specialized visualization there is a possibility to elaborate stand-alone module. Such module can use the data exchange mechanisms available in the system and cooperate with other platform modules.

Fig. 7. Exemplary process operator interface dedicated to deliver diagnostics information about process state.

9. Summary Fig. 6. Integration of user interface with configuration of particular application.

This module is not designed as competitor of wellgrounded visualization modules available in SCADA systems. Its main advantages result from its full integration with platform information module. For example, dedicated displays are used to automatically present the information that was used to generate the diagnosis the set of useful residuals in respect to particular fault is retrieved automatically from the diagnostics relation configuration stored in diagnostic package private database. It is possible because the visualization pages can be connected with particular assets, while displays are related with defined faults and residuals. The idea of the integration of user interface components with configuration objects is presented in Fig. 6, while its implementation is shown in Fig. 7.

The displays are realized as visualization module plugins. Such approach enables to elaborate specialized displays designed by the users. Such displays can even visualize process variables of user types (unknown by the platform itself), e.g. visualization of fuzzy signals or dedicated diagnostic messages. The only limitation is that the plug-in creator must have the knowledge about the processed signals structure.

Presented system was created as a result of development grant that finished in 2009. Currently test of the system are conducted. Their aim is to remove bags and finally prepare the system to implementation, commercial ones as well as research-development.

The most important part of the system is its software platform and the possibility to easy extend its functionality. The tasks that can be realized by the system can be also fulfilled, however usually in limited scope, by the class of commercial SCADA, DCS or similar systems that are available on the market. However, due to specialized system structure, there is a possibility to implement, apply and test modern and innovative techniques in the field of monitoring, advanced supervisory control, modeling and diagnostics of industrial processes which application in classical control and monitoring systems is difficult or even impossible.

Wide further system development is possible due to implemented plug-ins technology, open communication mechanisms and the possibility to introduce user-defined types of variables and their values (on each stage of system operation). This development can be realized not only by the research centers that took part in system creation but also by other, independent units.

Articles

45

VOLUME 5,

Journal of Automation, Mobile Robotics & Intelligent Systems

The system structure was designed taking into account wide range of its possible fields of application. The use of the software platform and elaborated packages is planned not only in commercial applications but also in researchdevelopment and didactic tasks. The development of new system packages is conducted all the time. In respect to commercial application it is planned to used the system to realized the set of simulators of power generation units of conventional power stations. It is planned to use that simulators in the process of training the operators and other power station technical stuff. In the field of research and development the application of a system as monitoring system of a gas network is currently realized. The didactic use of the system is realized all the time. The system is used during the laboratories and project of courses related with the problems of process modeling (physical as well as based on process data and parametric models), diagnostics, automatic-control and applications of artificial intelligence methods in automatic-control.

ACKNOWLEDGMENTS This work was supported in part by the Polish National Committee of Research under Grant R01 012 02 (DIASTER).

AUTHORS Micha³ Syfert, Pawe³ Wnuk*, Jan Maciej Koœcielny Politechnika Warszawska, Instytut Automatyki i Robotyki, ul. Œw.A. Boboli 8, Warsaw, 02-525, Poland.

E-mails: m.syfert@mchtr.pw.edu.pl, p.wnuk@mchtr.pw.edu.pl, jmk@mchtr.pw.edu.pl. * Corresponding author

References [1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

46

M. A. Brdyœ, P. Tatjewski, Iterative Algorithms for Multilayer Optimizing Control, Imperial College Press, London, 2005.

J. Chen, R. Patton, Robust model based fault diagnosis for dynamic systems, Kluwer Akademic Publishers, Boston, 1999.

W. Cholewa, J. Kiciñski, Technical diagnostics. Inverse diagnostic models. Wydawnictwo Politechniki Œl¹skiej.

Gliwice, 1997. (in Polish) J. Gertler, Fault Detection and Diagnosis in Engineering Systems. Marcel Dekker, Inc. New YorkBasel-Hong Kong, 1998.

K. Janiszowski, Identification of parametric models in examplesc. Akademicka Oficyna Wydawnicza EXIT, Warszawa, 2002. (in Polish) J. Korbicz, J. M. Koœcielny, Z. Kowalczuk, W. Cholewa (ed). Fault Diagnosis. Models, Artificial Intelligence, Application, Springer, 2004.

M. Koœcielny, M. Syfert, P. Wnuk, Advanced monitoring and diagnostic system "AMandD". Problemy Eksploatacji, vol. 61, no. 2, 2006, pp. 169-179 .

J. Koœcielny, M. Syfert, P. Wnuk, Advanced monitoring and diagnostic system "AMandD". In: Safeprocess2006 A Proceedings Volume from the 6th IFAC Symposium on Fault Detection, Supervision and Safety of Technical Processes, Beijing, P.R. China, August 29-September 1,

Articles

[9]

[10] [11]

[12]

N° 4

2011

Elsevier, vol. 1, 2006, pp. 635-640.

W. Moczulski, Technical diagnostics. The methods of knowledge acquiring. Wydawnictwo Politechniki Œl¹skiej, Gliwice, 2002. (in Polish) R. Patton, P. Frank, R. Clark (Eds.), Issues of fault diagnosis for dynamic systems. Springer, 2000.

M. Syfert, P. Rzepiejewski, P. Wnuk, J. M. Koœcielny, Current diagnostics of the evaporation station. In: 16th IFAC World Congress, Praga, 4th-8th July, 2005.

P. Tatjewski, Advanced Control of Industrial Process.

Structures and Algorithms. Springer, 2007.

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N° 4

2011

Eye Trackers in Quality Evaluation of Compressed Video Submitted 20th June 2011; accepted 21st September 2011

. . .

Anna Ostaszewska-Li zewska, Rafał Kłoda, Sabina Zebrowska-Łucyk, Maciej Li zewski

1. Introduction

In case of video, because of the size of the source files, the use of lossy compression is almost a rule. The lossy compression is the process of discarding the part of information, that is the least important for the human perception [3]. Nevertheless the side effect of this process may occur as compression artifacts, which decrease perceived quality of the video. The final effect of compression depends on many factors such as the coding algorithm itself or its parameters used, the character of the scene (i.e. static or dynamic, amount of spatial information [4, 5], etc).

In fact, it is impossible to predict the quality of the output video, thus the only way to control it is measurement of the final results. There are two concepts of compressed video quality evaluation. The first is to use the human audience and to conduct the normalized experiment, where subjects give scores to a test material [6, 7, 8, 9]. The second idea is to automate the whole process with the use of the mathematical model of human perception [10, 11, 12, 13]. The model is created on the base of results of experiment with human audience [14, 15], so it is crucially important to make this kind of measurement as reliable and accurate as possible. The main problems in this area are scores given by unreliable viewers and the lack of a tool that would enable for linking the score with time and space of the test material. Both problems can be solved by the use of an eye tracker. Hence there is a need to obtain results for the same test material both from subjective method and an eye tracker and to develop a dedicated tool that would make such results possible to analyze.

2. Single Stimulus Continuous Quality Evaluation (SSCQE)

2.1. Methodology Single Stimulus Continuous Quality Evaluation (SSCQE) is one of the subjective methods of compressed video quality assessment that were specified in recommendations [1, 2] developed by International Telecommunication Union (ITU). SSCQE method was specially designed to correspond to an actual home viewing situation: a series of video sequences is presented once to a viewer and a reference is not available. Subjects evaluate the instantaneous quality in real time using a slider with a continuous scale. The ratings are sampled with the frequency of 2 Hz. The biggest advantage of this method is the relatively large amount of data – it records each temporal variation of the quality perceived.

2.2. Experiment design and carrying out The SSCQE experiment was carried out with the use of 10-minute test material built of four 15-seconds sequences: ‘bbc3’, ‘cact’, ‘mobl’ and ‘susi’ (Fig. 1) coded in MPEG-2 standard with 10 levels of bitrate, ranging from 1 to 5 Mbps. The length of Group of Pictures (GOP) was 13 and it included two B-frames. Soundtrack was not included. a)

b)

c)

d)

Fig. 1. Source sequences used in the test material: a) ‘bbc3’, b) ‘cact’, c) ‘mobl’, d) ‘susi’.

Test equipment used in subjective assessments consisted of a 20” professional-grade monitor (SONY PVM20M4E) and a professional DVD player (Pioneer DVDV7300D). A slider, which the subjects used to evaluate the sequences, was a stand-alone hardware device with a hundred-level scale attached. SSCQE ratings were entering the PC directly through the NI 6013 card. Incoming data was synchronized with the timecode.

In the experiment 45 typical end-users (mostly university students) participated in the experiment. Each of them was screened for normal visual acuity or corrective glasses and normal color vision (per Ishihara test).

2.3. Results and temporal inconsistency of scores The raw data were processed according to the ITU recommendations [2] to cumulative probability plot (Fig. 2) and used for bitrate optimization.

Articles

47

Journal of Automation, Mobile Robotics & Intelligent Systems

cumulative probablility (%)

100

VOLUME 5,

N° 4

2011

bitrate (2 – 3 Mbps, Fig. 3). For this particular scene, coded with bitrate less than 3 Mbps, compression artifacts were clearly visible to an experienced viewer, but they were appearing only in the 7-th second of the sequence and mainly in the area of hair. The same time the area of the face was almost free of any distortion (Fig. 4).

80 60 40 20 0 0

10 20 30 40 50 60 70 80 90 score 2 Mbps 2,6 Mbps 3 Mbps 4 Mbps

5 Mbps

Fig. 2. Cumulative probability plot, the ‘mobl’ sequence.

40 35 30

Fig. 4. Compression artifacts in 7th second of the ‘susi’ sequence coded with the bitrate of 2 Mbps, blockiness effect appears mainly in the area of hair.

25 20 15 10 5 600

540

480

420

360

300

240

180

120

60

0 0

standard deviation of scores/bitrate (Mb/s)

45

time [s] bbc3

cact

mobl

susi

Fig. 3. Standard deviation of scores given by the audience for consecutive samples.

In general the final results were in accordance to expectations. Hence the analysis of the standard deviation of the scores given by the whole audience for each sample revealed peaks for the ‘susi’ sequence coded with low

a)

b)

Therefore a question was raised: is the high difference between individual scores caused by the fact that subjects observe the whole picture, but some of them don’t perceive compression artifacts or some of viewers prefer to look just in the face of the actress and their scores relate only to that certain part of the picture.

In order to find the answer, the experiment with eye tracker was designed.

3. The experiment with eye tracker

3.1. Tobii T60 test equipment and Tobii Studio™ statistical analysis tool Eye tracker is a device that enables for measuring eye positions and eye movement. For the purpose of the planned experiment, Tobii T60 Eye Tracker was

c)

Fig. 5. Heat maps for the ‘susi’ sequence: a) 2 Mb/s, b) 3,6 Mb/s, c) 5 Mb/s.

a)

b)

c)

Fig. 6. Heat maps for the ‘mobl’ sequence: a) 2 Mb/s, b) 3,6 Mb/s, c) 5 Mb/s.

48

Articles

Journal of Automation, Mobile Robotics & Intelligent Systems

used. It is integrated in a 17-inch TFT monitor and allows for a large degree of head movement, providing a distraction-free test environment that ensures natural behavior, which is especially important in the case of SSCQE method. Eye tracker system was equipped with Tobii Studio™ - a dedicated software, which provides a platform for stimuli presentation, recording, observation, visualization and analysis of eye tracking.

3.2. The experiment design and carrying out The test material was built on the basis of the material used in SSCQE experiment – three levels of bitrate were chosen: 2, 3 and 5 Mbps. Therefore the whole test material was only 3 minutes long.

Each test session started with eye tracker calibration for the individual subject.

16 viewers (the group similar to the group in SSCQE experiment) participated in the experiment.

3.2. Results obtained in Tobii Studio™ The first step of the data analysis was to prepare heat maps and gaze plots with the use of Tobii Studio™, for each of four sequences coded on each level of bitrate.

Heat maps were based on the summary of gaze time data from all recordings, for the whole sequence. In case of ‘susi’ it is clear, that for most time the girl’s face was observed (Fig. 5). Although the green color appears in the area of hair also, but it is necessary to examine, if observers were looking at it when the artifacts occurred.

In case of other sequences, there were also parts of the picture that turned to be more interesting to subjects (Fig. 6.), but at the same time those areas were prone to compression artifacts occurrence. Additionally comparison of heat maps for each level of bitrate proved, that the quality does not have a strong effect on the way that viewers scan the picture.

Gaze plots, which display gaze points, fixations, and scan paths superimposed over the whole of the sequence also seem to be resistant to bitrate changes. Another interesting discovery was the fact that some observers used to have episodes of distraction and they were taking a look at surroundings (Fig. 7, the scan paths which exceed the area of the monitor).

Fig. 7. Gaze plots for the ‘susi’ sequence, 2 Mbps.

VOLUME 5,

N° 4

2011

Presented gaze plots and heat maps are a static summary of the whole data recorded for the time of sequence duration. For the purpose of data analysis in space and time, Tobii Studio™ offers animated versions of graphs.

They can be used for rough assessment of the number of gaze points in a specified area of the picture (area of interest, AOI). However there is still no tool which would enable for exact calculation.

For this reason an independent application was developed, which in assumption would make possible the areas of interest move and change its shape according to the content of the video test sequence.

3.4. Common Sense for Tobii and the results obtained The application called Common Sense for Tobii [16] consists of two modules. The first of them written in C++ as a plugin for VirtualDub video editing software, on the basis of the test sequence and the data from Tobii, generates the video in the original resolution of the Tobii monitor (to get the same view as during the experiment) with the animation of gaze points. There is a possibility of choosing viewers and, if necessary, not choosing any of them to get the view without any gaze points. Second part is written in C# and VideoLab library as Windows application.

Gaze points are animated – their growing diameter reflects the time of fixation. The color is matched with the particular observer. The video generated by the first module and the original data from Tobii are the input to the second module, which enables for defining animated areas of interest. Areas of interest are rectangles set in keyframes. The user controls their attributes: position, size, rotation and visibility (the areas can disappear and reappear). Tweening of areas between keyframes is used.

There is a possibility of selecting the more sophisticated shape of the area of interest just by grouping several rectangles which are still controlled independently. Grouping can be nested as a tree with regions as leaves and grouping nodes.

The final result of using the application is the video with animated gazepoints and the statistics: the total number of all gazepoints and the number of the gazepoints per each area and group of areas of interest (with the information about the percentage).

CS for Tobii enabled for finding the reason for high standard deviation of scores given for ‘susi’ coded with low bitrate. In the first step of analysis, four independent, animated areas of interest were assigned: face, hand, hair and background. Thanks to use of 17 keyframes, the areas were changing shape and dimensions in order to adjust to object movement and perspective. The obtained statistics for three levels of bitrate are very similar (difference does not exceed 5% of the final result). The Fig. 8 presents the average of the percentage of gaze points following to each of defined, animated area of interest.

For the whole duration the face was observed by 65 % of the viewers, and the hair (prone to compression artifacts) only by 11 %. In the second step, analysis concerned the aforementioned 7th second of ‘susi’ coded with bitrate of 2 Mbps. Two areas were defined: the face and the piece of hair that was seriously affected by blockiness effect.

It appeared that even though the half of the picture for one second turned into dynamic, visible blocks, only Articles

49

Journal of Automation, Mobile Robotics & Intelligent Systems

4% of people were puzzled by this sudden change and gazed at it. The same time 44 % were observing the fast moving face and the rest were scanning the background.

This proves that only some observers are able to score the actual quality of the whole material. Most of them fail to take into account even very severe distortions, which is the reason for significant differences between scores given by individual subjects.

face hand hair background

Fig. 8. The number of gaze for the ‘susi’ sequence, 2 Mbps.

4. Conclusion

To sum up, the experimental station for conducting subjective quality evaluation should be integrated with the eye tracker. That would enable for rejecting scores given at the time of the episode of lack of attention, when the test material was not observed. This kind of data filtration would make the final result more reliable.

Besides it would be useful to link up the score with the part of the picture that the observer was looking at, as the quality of the frame hardly ever is uniform. This would provide researchers working on human visual system with interesting data both on human perception and the usability of certain sequence for quality evaluation of compressed video.

AUTHORS

Anna Ostaszewska-Liżewska* – Institute of Metrology and Biomedical Engineering, Warsaw University of Technology, Warsaw, Poland. anna.ostaszewska@gmail. com Rafał Kłoda – Institute of Metrology and Biomedical Engineering, Warsaw University of Technology and Industrial Research Institute for Automation and Measurements PIAP, Warsaw, Poland. E-mail: kloda@mchtr. pw.edu.pl Sabina Żebrowska-Łucyk – Institute of Metrology and Biomedical Engineering, Warsaw University of Technology, Warsaw, Poland. E-mail: szl@mchtr. pw.edu.pl Maciej Liżewski – 3e Internet Software House, CIO, ul. Podbipięty 51, 02-732 Warsaw, Poland. E-mail: maciej.lizewski@3e.pl

50

Articles

VOLUME 5,

References

N° 4

2011

[1] ITU-T Recommendation P.911, Subjective audiovisual quality assessment methods for multimedia applications,1996.

[2] ITU-R BT.500-12 Methodology for the subjective assessment of the quality of television pictures, 2009.

[3] K. Sayood, Introduction to Data Compression, Morgan Kaufmann, 3rd ed., 2005.

[4] A. Ostaszewska, R. Kłoda, “Quantifying the amount of spatial and temporal information in video test sequences”, Recent Advances in Mechatronics, Springer Verlag, Berlin Heidelberg 2007, pp. 11–15.

[5] R. Aldridge, J. Davidoff, M. Ghanbari, D. Hands, D. Pearson, “Measurement of scene-dependent quality variations in digitally coded television pictures”, IEE Proceedings – Visual Image Signal Process, vol. 142, no. 3, June 1995.

[6] Th. Alpert, J.P. Evain, “Subjective quality evaluation – The SSCQE and DSCQE methodo-logies”, EBU Technical Review, Spring 1997.

[7] F. Kozamernik, P. Sunna, E. Wyckens, D.I. Pettersen, “Subjective quality of Internet video codecs”, EBU Technical Review, no. 1, 2005, pp. 1-22.

[8] M. Pinson, S. Wolf, “Comparing subjective video quality testing methodologies”, SPIE Video Communications and Image Processing Conference, Lugano, Switzerland, Jul. 8-11, 2003.

[9] R. Aldridge, D. Hands, D. Pearson, N.K. Lodge, “Continuous quality assessment of digitally-coded television pictures”, IEE Proceedings online, no.

1998, 116-123.

[10] M.P. Eckert, A.P. Bradley, “Perceptual quality metrics applied to still image compression”, Signal Processing, 1998, pp. 177-200.

[11] P.N. Gardiner, M. Ghanbari, D.E. Pearson, K.T. Tan, “Development of a perceptual distortion meter for digital video”, IEE Conference Publication Proceedings of the 1997 International Broadcasting Convention, vol. 1, no. 447, Amsterdam, 1997, pp. 493–497.

[12] D.J. Heeger, P.C. Teo, “A model of perceptual image fidelity”, Proceedings of the International Conference on Image Processing, Washington, DC, 23rd–26th October 1995, pp. 343–345.

[13] M.A. Masry, S.S. Hemami, “CVQE: A metric for continuous video quality evaluation at low bit rates”, Proceedings of the SIGCHI conference on Human factors in computing systems, Vienna, Austria, 2004, pp. 535–542.

[14] D.S. Hands, M.D. Brotherton, A. Bourret, D. Bayart, “Subjective quality assessment for objective quality model development”, Electronic Letters, 2005, vol.41, no. 7, pp. 408-409.

[15] J. Lubin, “Human vision system model for objective picture quality measurements”, IEE Conference Publication Proceedings of the 1997 International Broadcasting Convention, vol. 447, Amsterdam, 1997, pp. 498–503.

[16] http://www.common-sense.com.pl

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N° 4

2011

Specific Issues In Management of Large International Research Projects in The Field of Security and Defence Submitted 20th April 2011; accepted 15th July 2011

Aleksandra Bukała, Mariusz Andrzejczak, Adam Wołoszczuk

Abstract:

This article presents specific characteristics of large international research projects in the area of security and defence. It is based on the experience of the authors gathered while managing international research projects realised within 7th EU Framework Programme, European Defence Agency (EDA) programmes etc. The issues of project size, international and research character are described in the article. Specific elements that characterise projects in security and defence field are also identified and discussed. Finally selected key factors determining the success of such projects are mentioned.

Keywords: management, defence, security project

each other. Therefore the division shown on Figure 1 is only a simplified display to make the presentation of the topic easier.

research

1. Introduction

International research projects are extremely interesting to work on and to manage. Their participants are at the same time co-creators and observers of the birth of new knowledge, new technology and new quality. Companies that normally compete with each other, here decide to cooperate in creation of new knowledge, new technologies and in result, new products and services. One of the reasons laying behind the decision to cooperate is synergy that is created between partners having different knowledge and background. The synergies are created between organisations having competences in different research and technology areas. The other reasons are connected with risks occurring in every research and development activity. Doing R&D together with other companies, research centres and academia leads to sharing and thus reducing the risk of a project. Cooperation in research projects also help in establishing business links with partners operating in other countries and on other markets.

The authors of the article analysed a number of international technological research projects from security and defence area carried out by Polish institutions (research institutes, technical universities, enterprises). The aim of this article is to highlight and discuss the characteristics of such research projects.

2. Specific issues of large international research projects

The four main characteristics of such projects (large, international, research, and security) are described in this section. It is followed by description of the consequences resulting from the project’s size, international nature and scientific character. Above mentioned aspects cannot be described separately as the merge with, and influence

Figure 1. The characteristic of large international security and defence research projects.

2.1. Large In the „Security” priority of the 7th Framework Programme (FP7), projects defined as large are so called Integrating or Demonstration Projects which have received a grant higher than 3,5 million Euro from the European Commission and are being realised in a consortium comprised of at least 3 entities from at least 3 member or associate countries1. In practice however, these projects have budgets of above 9 million Euro and are developed by at least 10 different partners.

2.2. Preparation phase The first characteristic of large projects is the long development process, especially in its first stage, which is the preparation phase. In research projects developing a project concept, analysing current state of the art in key technologies and identifying technological gaps are especially important to minimise the risks with which the research nature of the project is bound. The planning process of such an undertaking is also a challenge. The schedule and budget of the research project must be flexible enough to reflect dynamic character of research yet must enable coordination of efforts of many different project stakeholders.

1 Work Programme 2011, Cooperation, Theme 10, Security, European Commission C(2010)4900 of 19th July 2010.

Articles

51

Journal of Automation, Mobile Robotics & Intelligent Systems

The second element influencing the duration of the preparation phase of the project is composing the consortium. International nature as well as the necessity of the participation of various entities with vastly different competences and research experience forces the creation of large consortiums. The process of creating a consortium begins by creating a “competence map” necessary for the proper development of the research project. Creating such a map is possible after analysing the current state of the art in the technologies related to the project, identifying the key technologies and determining technological gaps. This analysis enables identification of competences that are needed for the successful execution of the project. Partners may come from previous joint research projects and other actions, may be contacted on conferences, fairs, and thematically related workshops or by dedicated services (eg. FP7 Partner Search portal etc.). After choosing the partners and making the initial agreements the next phase is dedicated to further development of the project concept based on the experiences and competences of the members of consortium, as well as negotiations related to the resources committed by each of the partners and creating the project’s budget.

When building a consortium in research projects it is important to clearly define the issue of ownership of the knowledge (Intellectual Property Rights) created during the project as well as access to already existing knowledge owned by the partners. Usually these issues are regulated by the consortium agreement. In security related projects, it is also important to set the framework for information protection, identification of sensitive information, rules for information disclosure etc. Non-disclosure agreement signed at the very beginning of cooperation is usually the first step in this process.

One must take into account that preparation phase that is long and resources consuming is being funded entirely from the future consortium’s own assets. This means that each idea goes through a form of natural selection within the partner’s organisations, which need to determine the importance of the project scope and their future tasks in relation to overall institution strategy. In result of this process partners who decide to enter the project are naturally motivated to ensure its success.

VOLUME 5,

N° 4

2011

2.3. Development phase The development phase is bound with the creation of management plan, which includes planning at least for: •Communication; •Risk management; •Change management; •Quality management.

Each of above mentioned plans should be based on chosen project management methodology, which should be adopted to the particular project’s needs. All management plans must be agreed by the representatives of all the partners in the consortium.

Conducting large projects requires clearly defined goals in the preparation phase and an appropriately detailed division of tasks in the development phase. Clear assignment of responsibilities between the members of the consortium is crucial for successful management of the project. It also help partners to present clear message within their organisations what is their role in the project.

On the other hand sharing responsibility may lead to blurring responsibility. Sharing tasks may lead to doubling the work or causing problems with interfaces. That is why it is very important to appoint technical integrator whose main responsibility is to turn all partners in the same direction. Such person should be well oriented in systems engineering methodologies and his/her most important responsibility at the very beginning of the project is to agree with all partners system engineering approach and tools to be used on the next steps of the project.

Since there are many partners working on many issues at the same time within large projects another important issue is to develop tools for monitoring the state of the project as well as the use of resources in order to be able to early identify risks and issues caused by delays getting behind the schedule or exceeding the budget.

It is also important to identify the key partners early, the ones responsible for supplying the most important components, technologies, deliverables to the project.

A consortium with many partners from several countries is very difficult to coordinate. Inertia of such structure influences significantly time it takes to reach decisions.

This creates a dangerous paradox, in which on one hand we have many independent project executors responsible for different parts, elements of the project and on the other hand efficient coordination and management are crucial to reach successfully expected goals. Through identi-

Figure 2. Stages of the preparation phase in an international research project.

The progress of each of the stages in the preparation phase is illustrated in Figure 2.

52

Articles

Figure 3. Selected phases in the development phase of an international research project.

Journal of Automation, Mobile Robotics & Intelligent Systems

fication of key performers in the consortium it is possible to establish decision making bodies of smaller size where coordination is much easier and consensus decisions can be reached more easily. Identification of roles inside the consortium also helps in managing the project. Project management bodies as well as roles and responsibilities are defined in consortium agreement mentioned earlier.

Above described stages of the development phase of an international research projects are presented in Figure 3.

2.4. International The international nature of the project requires a specific approach in its management. At first, one must take into account the cultural variety of partners coming from different countries and even continents. It is not only about the differences in mentality, different behaviour patterns and way of seeing and understanding the world.

Cultural variety influences project management on many levels – for example different dates of public holidays and religious events, sometimes different weekend days (eg. Friday in Israel) and differences in working hours coming also from time zones influences significantly project schedule. When planning meetings, conferences, management sessions and other events Project Coordinator must take all these factors into account in order to avoid putting the partners into situations having strong impact on their usual private time.

The international nature of the project also means a variety of languages used within the consortium. English is usually used as the “working” language of the project. However, as most of the consortium partners are not English natives, misunderstandings and misinterpretations are likely to occur. It is important to use simple and precise terminology in the project documentation.

Creating common dictionaries and writing down clear project management procedures also help to avoid communication problems. The role of face to face contacts is very important in international research projects, which is understood by the European Commission, which requires regular F2F meetings in EU programmes. This way of communication strongly supports the exchange of views and ideas and significantly increases the level of understanding among the consortium partners. Regular teleconferences usually support meetings, however the entire consortium should not meet less than once in every 4-6 months. Key partners that manage consortium on executive level should have meetings at least once in every 2-3 months.

2.5. Research Main problem of a research project mentioned earlier is that it takes much time and effort to come to practical results. Return on investments in research may come in years or not come at all if the idea is not as promising as it looked at the very beginning. Working on the edge of technology is always a bit risky. There are many ways to reduce risks in research. One of them is cooperation.

While composing the consortium to run international project we must have it in mind. Another very important thing is careful analysis of the current state of the art in relevant technologies, identification of key technologies and technological gaps. Finally created consortium must

VOLUME 5,

N° 4

2011

have appropriate competences, especially in relation to technological gaps.

Another characteristic of research projects run in international consortiums of independent partners is that usually that partners are motivated to join if the project itself and their role in it is in-line with their internal development strategies. This has two important consequences.

First is connected with protection of sensitive information. Companies usually are reluctant to disclose to wider audience (and especially their competitors) the technologies that they are developing. That is why trust and confidence is very important in international projects. Additional issue is that partners are also reluctant to share some of their discoveries even within the consortium.

Access rights to knowledge developed during the project are usually addressed in detail in grant agreements and consortium agreements. The second consequence occurs when development strategy of the partners change as a result of internal factors like changes in management or external as for example crisis. In such cases motivation for working in the project may drop significantly leading even to leaving the consortium.

2.6. Characteristics of security projects Security research projects, apart from the characteristics mentioned above, also have additional aspects which must be taken into account in the management process.

First and foremost security is a very sensitive field.

That is why a greater degree of trust amongst the partners as well as between the consortium and the users of the end product is required. Such trust cannot be built in a day. That is why it is very difficult for new players in this sector to enter into closer cooperation. Also people involved in such projects should be reliable, trustworthy and discrete. As a rule, all personnel involved with the project must have appropriate certifications.

While working on projects in this field it is also important to develop procedures for identifying and handling sensitive information. Authors experience show that cooperation with end users on that field is crucial. These are people who can best identify areas where sensitive information may appear.

In security research projects, especially the ones where hardware results are foreseen, there is a problem of dual use and military equipment purchase and production.

While purchasing of such equipment is put under specific national and international regulations, the management procedures of the project must address the issue of producing such equipment within the project. Appropriate procedures must be created to ensure identification and handling of dual use and military equipment developed during the project lifetime.

Next aspect of security related projects is ethical issues. Technologies developed within research projects may affect ethical standards and their use can be subject to legal restrictions. Authors experience show that it is reasonable to gather a group of specialists in the field of ethics and human rights acting as advisory body, especially to address legal aspects of technologies and systems developed in research projects.

Figure 4 presents dependencies between the main characteristics of the mentioned projects and their attributes, which influence each other.

Articles

53

Journal of Automation, Mobile Robotics & Intelligent Systems

VOLUME 5,

N° 4

2011

Fig. 4. A schematic of the dependencies between the characteristics of a research project.

2.7. Key factors for success in international security research projects Of course it is impossible to create a universal recipe for success of a research project in any field, including security. In this section authors would like to highlight some factors, without which success is very difficult, if not impossible, to achieve.

The most important factor is to have an interesting idea. The successful research projects are usually built on innovative connection of different technologies offering new or improved functionalities for end users or clients.

This needs both deep competences in technological areas developed within the projects and constant, close and good relations with customers.

This leads us to the second key factor for successful project, which is the technological competences of the partners involved with the project. This seems obvious for any research activity but here it means that it is very important to carefully build the consortium based on competence map or matrix to ensure that the necessary knowledge is on board of the project when needed.

The last factor influencing success of the project, worth to mention in this short article, is the project team and its spirit. It is a challenge in projects developed by independent entities from different countries and cultures, when people have chance to meet only for couple of days two, three times a year.

3. Summary

The article presents the specific aspects of preparing and running large international research projects. It describes specific elements influencing the management process related to the international and research nature of the project and its scope. It describes additional issues related to field of security and highlights selected key success factors of such projects.

54

Articles

AUTHORS

Aleksandra Bukała*, Adam Wołoszczuk – Industrial Research Institute for Automation & Measurements PIAP, Al. Jerozolimskie 202, PL-02-486 Warsaw, Poland.

Mariusz Andrzejczak – Bumar Sp. z o.o., Al. Jana Pawła II 11, PL-00-828 Warsaw, Poland. *Corresponding author

References

[1] Christensen C. M., The innovator’s solution: creating and sustaining successful growth, Harvard Business Press, 2003.

[2] Christensen C. M., Anthony S. D., Roth E. A., Seeing What’s Next, Harvard Business School Press, 2004.

[3] Dalkir K., Knowledge Management in Theory and Practice, Elsevier, Oxford, 2005.

[4] Dierkes M., Antal A. B., Child J., Nonaka I., Handbook of Organizational Learning and Knowledge, Oxford University Press, Oxford, 2001.

[5] Drucker P., Innovation and Entrepreneurship, Butterworth-Heinemann, Amsterdam 2007.

[6] Drucker P., The Practice of Management, ButterworthHeinemann, Amsterdam 2007.

[7] Grucza B., Ogonek K., Trocki M., Zarządzanie projektami, Polskie Wydawnictwo Ekonomiczne, Warsaw, 2002. (in Polish) [8] Leifer R., Radical Innovation: How Mature Companies Can Outsmart Upstarts, Harvard Business School Press, Boston, 2000.

[9] Meyer M. H., The Fast Path to Corporate Growth: Leveraging Knowledge and Technologies to New Market Applications, Oxford University Press, Oxford/New York, 2007.

[10] Porter M. E., On competition. Updated and expanded edition, Harvard Business Press, 2008.

[11] Pritchard C. L., Risk Management: Concepts and Guidance, ESI INTL, 2010.