22i16 ijaet0916945 v6 iss4 1639to1646

Page 1

International Journal of Advances in Engineering & Technology, Sept. 2013. ©IJAET ISSN: 22311963

ADAPTIVE GENETIC ALGORITHM WITH NEURAL NETWORK FOR MACHINERY FAULT DETECTION B.Kishore1, M.R.S.Satyanarayana2 and K.Sujatha3 1Assistant Professor, Dept. of Mechanical Engg., GITAM University, Hyderabad, India 2Professor, Dept. of Mechanical Engg., GITAM University, Visakhapatnam, India 3Associate Professor, Dept. of Computer Sc. & Engg., MIRACLE, Visakhapatnam, India

ABSTRACT Machine malfunctions are pestilence to all production lines. One fault or malfunction leads to another in a multistage system and faults may also be developed over a long period of time or even suddenly. However manual fault detection techniques are error prone. This paper gives solutions for fault detection by applying Adaptive Genetic Algorithm to train a neural network. Adaptive Genetic Algorithms (AGAs) have been built that dynamically adjust selected control parameters during the course of evolving a problem solution. The results from the standard Genetic Algorithm (SGA) and Adaptive Genetic Algorithm are compared for analysis. This comparison stated that AGA reached the local optimum remarkably faster than SGA.

KEYWORDS:

Adaptive genetic algorithm (AGA), Artificial neural network (ANN, Condition monitoring, Fault detection, Standard genetic algorithm (SGA),

I.

INTRODUCTION

Rotating machine faults diagnosis is becoming a challenging role for the researchers. McCormick et al classified the condition of rotating machines [1,2].Genetic algorithms are a class of probabilistic optimization algorithms pioneered by John Holland in the 1970’s and got popular in the late 1980’s. They are based on ideas from Darwinian Evolution inspired by the biological evolution process as shown in the conceptual genetic algorithm in figure 1. They can be used to solve a variety of problems that are not easy to solve using other techniques. A Genetic Algorithm (GA) is a search technique used in computing to find true or approximate solutions to optimization and search problems. They are widely-used in business, Science and Engineering fields in solving complex tasks [3,4]. Genetic algorithms are implemented as a computer simulation in which a population of abstract representations (called chromosomes or the genotype or the genome) of candidate solutions (called individuals, creatures, or phenotypes) to an optimization problem evolves toward better solutions. A matrix formulation of an adaptive [5] genetic algorithm is developed using mutation matrix and crossover matrix. Selection, mutation, and crossover are all parameters free in the sense that the problem at a particular stage of evolution will choose the parameters automatically. The notion of mutation matrix is used in the development of an adaptive genetic algorithm so that the selection process does not require any external input parameter. A lot of research has been done on clustering algorithms based on evolutionary algorithms such as genetic algorithms, greedy randomized adaptive search procedure, honey bees mating optimization algorithm, artificial immune system etc. however because of the limitation of the other optimized systems in the clustering domain, specified Adaptive Genetic Algorithm can be used in training Neural Networks. Neural networks take a different approach to problem solving than that of conventional computers. Conventional computers use an algorithmic approach i.e. the computer follows a set of instructions in order to solve a problem. Unless the specific steps that the computer needs to follow are known the computer cannot solve the problem. That restricts the problem solving capability of conventional computers to problems that we already understand and know how to solve.

1639

Vol. 6, Issue 4, pp. 1639-1646


International Journal of Advances in Engineering & Technology, Sept. 2013. ŠIJAET ISSN: 22311963 Initial Population

Fitness Evaluation

Yes Fitness Value Met? No Individuals Selection

Survivors Selection

Display results Figure 1: Conceptual genetic algorithm

But computers would be so much more useful if they could do things that we don't exactly know how to do [6]. Neural networks process information in a similar way the human brain does. The network is composed of a large number of highly interconnected processing elements (neurons) working in parallel to solve a specific problem. Neural networks learn by example. They cannot be programmed to perform a specific task. The examples must be selected carefully otherwise useful time is wasted or even worse the network might be functioning incorrectly. The disadvantage is that because the network finds out how to solve the problem by itself, its operation can be unpredictable. Conventional computers use a cognitive approach to problem solving where the way the problem is to solved must be known and stated in small unambiguous instructions. These instructions are then converted to a high level language program and then into machine code that the computer can understand. These machines are totally predictable; if anything goes wrong is due to a software or hardware fault. Neural networks and conventional algorithmic computers are not in competition but complement each other. There are tasks are more suited to an algorithmic approach like arithmetic operations and tasks that are more suited to neural networks. Even more, a large number of tasks, require systems that use a combination of the two approaches (normally a conventional computer is used to supervise the neural network) in order to perform at maximum efficiency. This paper is organized as follows. Vibration data extractions covered in Section2. Overview of a standard genetic algorithm and adaptive genetic algorithms covered In Section 3 and 4 respectively. In section 5 artificial neural networks concept is presented. Section 6 describes the learning algorithm of AGA for neural network design. In Section 7, design of ANN by AGA are simulated and applied to intelligent fault diagnosis and the results were presented. Finally, in Section 8, conclusions are presented.

II.

VIBRATION DATA AND FEATURE EXTRACTION

The main air blower system setup illustrated in figure 2 in the sulphuric acid plant. It intakes air from filters through the drying tower in which the air mixes with the sulphuric acid in the counter current direction reducing its temperature. The MAB is turbine driven and runs at a rated speed of 5000rpm. The turbine and blower are connected by flexible metallic coupling i.e., a spring plate coupling. The blower and the turbine are lifted by a total of five journal bearings as shown in the figure. They are located at various locations like Turbine non drive end, Turbine drive end, Blower thrust end, Blower drive end and Blower non drive end. All the bearings are a five piece babbitt bearings with four portals for lubrication inlet points. The steam enters the turbine at a temperature of 3500C. The turbine is driven by a two step starting

1640

Vol. 6, Issue 4, pp. 1639-1646


International Journal of Advances in Engineering & Technology, Sept. 2013. ŠIJAET ISSN: 22311963 process. First an auxiliary oil pump drives the main oil pump which then starts the turbine. Once the main oil pump comes on line, turbine starts pumping the steam thus rotating the blower. Condition monitoring is done on the main air blower via vibration measurement. It is done through online and manual observation. Manual vibration analysis is done using software PRISM 4. CMVA 65 MICROLOG DATA COLLECTOR / ANALYSER is used to collect and analyze the vibrations of the rotating equipment. Its frequency range is up to 40 KHz. A magnetic sensor receives the vibrations when placed on the bearing housing. The sensor is placed in horizontal (H), vertical (V) and axial (A) locations on the housing to observe the vibrations and analyze the condition properly. The readings are uploaded in the software and it displays the spectrum of the vibrations. By studying the spectrum and observing physically the condition of the equipment, its condition and abnormality can be detected. The maximum limit of vibrations to this blower is 7mm/sec. The various problems that this blower undergoes are Misalignment, Imbalance (system induced), Unbalance (by default), Surging, Bearing wear, Coupling failure etc.

III.

STANDARD GENETIC ALGORITHM

Genetic algorithms are categorized as global search heuristics. Genetic Algorithms are a particular class of evolutionary algorithms that use techniques inspired by evolutionary biology such as inheritance, mutation, selection, and crossover (also called recombination). Particularly well suited for hard problems where little is known about the underlying search space. A genetic algorithm maintains a population of candidate solutions for the problem at hand, and makes it evolve by iteratively applying a set of stochastic operators. The standard genetic algorithms have the steps like to choose initial population, assign a fitness [7] function, perform elitism, perform selection, perform crossover and Perform mutation. The basic algorithm for Standard Genetic Algorithm (SGA) is stated as below: Algorithm: SGA Step 1 : Initialize population Step 2 : Calculate fitness function Step 3 : While(fitness value != termination criteria) Step 4 : Perform Selection Step 5 : Perform Crossover Step 6: Mutation Step 7 : Calculate fitness function; {End of Step 3 while loop} Step 8 : End The GA searches a problem space with a population of chromosomes each of which represents an encoded solution. A fitness value is assigned to each chromosome according to its performance in which the more desirable the chromosome, the higher the fitness value becomes. The population evolves by a set of operators until some stopping criterion is met. A typical iteration of a GA, a generation, proceeds as follows. The best chromosomes of the current population are copied directly to the next generation (reproduction). A selection mechanism chooses chromosomes of the current population in such a way that the chromosome with the higher fitness value has a greater probability of being selected (roulette wheel) [8]. The selected chromosomes mate and generate new offspring (crossover). After the mating process, each offspring might mutate by another mechanism called mutation. The new population is then evaluated again, and the whole process is repeated.

IV.

ADAPTIVE GENETIC ALGORITHM

Adaptation of strategy parameters and genetic operators has become an important and promising area of research on GAs. Many researchers are focusing on solving real optimization problems [9, 10] by using adaptive techniques, like probability matching, adaptive pursuit method, numerical optimization, and graph colouring algorithms [11]. The value of parameters and genetic operators are adjusted in Gas. Parameter setting and adaptation in mutation was first introduced in evolutionary strategies. Basically, there are two main types of parameter settings: parameter tuning and parameter control. Parameter tuning means to set the suitable parameters before the run of algorithms and the

1641

Vol. 6, Issue 4, pp. 1639-1646


International Journal of Advances in Engineering & Technology, Sept. 2013. ŠIJAET ISSN: 22311963 parameters remain constant during the execution of algorithms. Parameter control means to assign initial values to the parameters and then these values adaptively change during the execution of algorithms. According to A. E. Eiben et al [12], parameters are adapted according to one of three methods. The first one is deterministic adaptation adjusts the values of parameters according to some deterministic rule without using any feedback information from the search space; the second one is adaptive adaptation modifies the parameters using the feedback information from the search space and the third is self-adaptive adaptation adapts the parameters by the GA itself. Adaptive Genetic Algorithm has many advantages compared to Conceptual Genetic Algorithm. This requires less computation and computing time and is more accurate and reliable.

V.

NEURAL NETWORKS

Neural networks, with their remarkable ability to derive meaning from complicated or imprecise data, can be used to extract patterns and detect trends that are too complex to be noticed by either humans or other computer techniques. A trained neural network can be thought of as an "expert" in the category of information it has been given to analyze. This expert can then be used to provide projections given new situations of interest and answer "what if" questions. Initial Population

Training Data

Train ANN

Testing Data

Find Fitness

Input data

Selection

Cross over

Adaptive Mutation

Termination Condition

No

Yes Predict fault type with best weights Figure 2 : Neural Network with Adaptive GA Implementation (NNAGA) Process

The computing world has a lot to gain from neural networks. Their ability to learn by example makes them very flexible and powerful. Furthermore there is no need to devise an algorithm in order to perform a specific task; i.e. there is no need to understand the internal mechanisms of that task. They are also very well suited for real time systems because of their fast response and computational times which are due to their parallel architecture [13].

1642

Vol. 6, Issue 4, pp. 1639-1646


International Journal of Advances in Engineering & Technology, Sept. 2013. ŠIJAET ISSN: 22311963

VI.

NEURAL NETWORK

WITH ADAPTIVE GA IMPLEMENTATION (NNAGA)

In the proposed method the neural network learns using adaptive genetic algorithms. This means that the weights [14, 15] and biases of all the neurons are joined to create a single vector. A certain set of vectors is a correct solution to the classification problem at hand. One of these vectors is found to be the best using an Adaptive Genetic Algorithm. The flowchart for the Neural Network with Adaptive GA (NNAGA) is as shown in figure 2.

VII.

RESULTS

We implemented a NNAGA with three inputs and one output with a sample database whose values are gathered from industry. The input for the NNAGA is Velocity, Displacement and Speed of an air blower. The first network is trained by using training samples. While the learning takes place, a textual indication of the learning process is presented on the standard output. Weight NO 1 2 3 4 5 6 7 8 9 10

GA 0.14372 0.98567 0.14327 0.96698 0.19430 -0.79612 -0.64219 0.80994 0.00314 -0.49664

Table 1 : Optimized weights with GA and AGA Weight Weight AGA NO GA AGA NO -0.34434 11 -0.61767 -0.72833 21 -0.47389 12 0.873507 -0.60901 22 0.73715 13 0.934699 0.088958 23 -0.19517 14 -0.98628 0.840283 24 0.06645 15 0.71278 0.26224 25 0.18221 16 -0.58686 0.819196 26 0.93238 17 0.558872 0.892318 27 -0.85577 18 -0.9787 -0.01603 28 0.40100 19 -0.79584 0.238023 29 -0.35399 20 0.846129 0.071908 30

GA 0.738447 0.380673 0.591644 0.651637 -0.49028 -0.7035 0.711918 0.163061 -0.26214 -0.45242

AGA 0.625848 0.811345 -0.56355 0.237752 -0.60007 -0.18514 -0.86337 0.536593 0.784929 -0.42388

Figure 3: AGA optimized Weights in comparison with GA optimized weights

This includes the fitness of the best individual in each population on each generation, and a schematic textual division of the plane once every 50 generations, to allow the user to inspect the progress. Then testing is performed and this network is found to be more accurate and efficient compared to Neural Network with Conceptual GA. Obtained optimized values are tabulated in the table1 for fist 30 values out of 100 values and optimized weights with GA and AGA is given in figure 3.

1643

Vol. 6, Issue 4, pp. 1639-1646


International Journal of Advances in Engineering & Technology, Sept. 2013. ŠIJAET ISSN: 22311963

Figure 4: GA Best Weight analysis using the WV tool box

Figure 5: AGA Best Weight analysis using the WV tool box

We obtained the values of Sensitivity, Specificity, Accuracy and Time for GA: 0.36111, 0.68056, 0.57407 and 102.0689Seconds respectively and for AGA: 0.41667, 0.70833, 0.61111 and 99.091Seconds respectively. The best weights through GA and AGA compared by using signal processing Window Visualization (WV tool box) Technique and the results are plotted in figure 4 and figure 5 respectively.

VIII.

CONCLUSION

In this paper application of an Adaptive Genetic Algorithm with Neural Network is proposed for fault detection in machinery. The reported method is a useful alternative to the conventional synthesis procedures. This optimization technique has been used to compute the network parameters satisfying all the requirements in terms of storing and correct recall in a predefined search space. Moreover, the performances may be considered improved if they are compared to the deterministic method used as reference. The variants of the proposed approach and results confirm that the designed networks does fault diagnosis correctly and guarantee good performances in terms of sensitive data obtained.

1644

Vol. 6, Issue 4, pp. 1639-1646


International Journal of Advances in Engineering & Technology, Sept. 2013. ©IJAET ISSN: 22311963

IX.

FUTURE SCOPE

Adaptive Genetic Algorithm with Neural Network has to be tested on wide range of applications that involve multiple iterative computations and heavy mathematical equations in order to save time and computational complexities.

ACKNOWLEDGEMENTS The authors would like to thank everyone, just everyone!

REFERENCES [1]. McCormick A.C. And Nandi A.K.,(1997) “Classification of the rotating machines condition using Artificial Neural Networks”, Proceedings of Institution of Mechanical Engineers, Part C 211, pp439 450. [2]. A. Azadeha, M. Saberib, A. Kazemc, V. Ebrahimipoura, A. Nourmohammadzadeha, Z. Saberid,(2013) “A flexible algorithm for fault diagnosis in a centrifugal pump with corrupted data and noise based on ANN and support vector machine with hyper-parameters optimization”, Applied Soft Computing 13 pp1478–1485. [3]. Bäck T, Fogel DB, Michalewicz Z (eds) Evolutionary computation 1: basic algorithms and operators.Institute of physics publishing, Bristol, UK[2000a]. [4]. Bäck T, Fogel DB, Michalewicz Z (eds) Evolutionary computation 2: basic algorithms and operators.Institute of physics publishing, Bristol, UK [2000b]. [5]. Gang Xu(2013), “An adaptive parameter tuning of particle swarm optimization algorithm”, Applied Mathematics and Computation, Vol. 219, pp4560–4569 [6]. Cowgill M, Harvey R, Watson L,(1999) “A genetic algorithm approach to cluster analysis”, Comput Math Appl,37:pp99–108. [7]. R.J. Kuo a,⇑ , Y.J. Syu b, Zhen-Yao Chen c, F.C. Tien d(2012) “Integration of particle swarm optimization and genetic algorithm for dynamic clustering”, Information Sciences, Vol. 195,pp 124– 140. [8]. R.J. Kuo a,⇑, Y.J. Syu b, Zhen-Yao Chen c, F.C. Tien dLiang Zhang, Lindsay B. Jack, Asoke K. Nandi,(2005) “Fault detection using genetic programming”, Mechanical Systems and Signal Processing 19 ,pp271–289. [9]. D.B. Fogel, Evolutionary Computation: Toward a New Philosophy of Machine Intelligence, IEEE Press, Piscataway, NJ, [1995]. [10]. X. Yao (Ed.), Evolutionary Computation: Theory and Applications,World Scientific, Singapore, [1999]. [11]. D. Thierens. Adaptive strategies for operator allocation. In F. Lobo, C. Lima, and Z. Michalewics (Eds.), Parameter Setting in Evolutionary Algorithms, Chapter 4, 77–90, [2007]. [12]. A. E. Eiben, Z.Michalewics, M. Schoenauer,(2007) and J. E. Smith. Parameter Control in Evolutionary Algorithms. In F. Lobo, C. Lima, and Z. Michalewics (Eds), Parameter Setting in Evolutionary Algorithms, Chapter 2,pp19–46. [13]. Shelly Xiaonan Wu, Wolfgang Banzhaf ,(2010) “The use of computational intelligence in intrusion detection systems: A review”, Applied Soft Computing 10,pp1–35. [14]. Tian Xuguang, Song Tong, Liu Yuxin, (2004) “Improving the structure and the parameter of the BP nerve network with Genetic Algorithm, Journal of Dalian University of Technology”, Vol. 2, No .6 pp69-71. [15]. Luo Chengfeng, Liu Zhengjun, Wang Changyao, Liu Zheng, (2006) “Classification of remote Agriculture land cover sensing data based on BP neural network improved by Genetic algorithm”, Journal of Agricultural Engineering, Vol. 22,No. 12 pp133-137.

AUTHORS B. Kishore is currently pursuing Ph.D in Mechanical Engineering from GITAM University, Visakhapatnam, India. He holds an M.Tech degree with CAD/CAM specialization from Andhra University, India. He is currently working as an Assistant Professor in the department of Mechanical Engineering at GITAM School of Technology, Hyderabad, India. He has 9 years experience in teaching. He is a member of ASME. His research interests include Condition Monitoring, Speech Recognition, Pattern Recognition, Artificial Neural Networks, and Genetic Algorithm.

1645

Vol. 6, Issue 4, pp. 1639-1646


International Journal of Advances in Engineering & Technology, Sept. 2013. ŠIJAET ISSN: 22311963 M.R.S. Satyanarayana is working as Professor in the Department of Mechanical Engineering at GITAM University, Visakhapatnam, India. He holds a Ph.D in Mechanical Engineering. He has 20 years’ experience in teaching and research in GITAM. He is a life member of Institute of Engineers (I), Condition Monitoring Society of India, Acoustic Society of India. He has published papers in several national and international journals. His research interests include Artificial Neural Networks, Rotor Dynamics, Vibrations, CAD/CAM. K.Sujatha holds M.Tech in Computer Science and Engineering from JNTU, Kakinada, India and Degree in Computer Science from Institute of Engineers and Master of Business Administration from Pondicherry University. She is presently working as Sr. Assistant Professor at Miracle Engineering College. She has 15 years of experience comprising of both teaching and programming. She is a member of Institute of Engineers (I). Her research interests include Speech Recognition, Network Security, Biometric and Cryptography, Neural Networks.

1646

Vol. 6, Issue 4, pp. 1639-1646


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.