vii TABLE OF CONTENTS CHAPTER 1 TITLE PAGE NO. TITLE i DECLARATION ii DEDICATION iii ACKNOWLEDGEMENT iv ABSTRACT v ABSTRAK vi TABLE OF CONTENTS vii LIST OF TABLES xii LIST OF FIGURES xiv LIST OF SYMBOLS xvii LIST OF ABBREVIATIONS xviii INTRODUCTION 1.1 Introduction 1 1.2 Problem Statement 3 1.3 Objective 4 1.4 Scope of Study 4 viii 2 LITERATURE REVIEW 2.1 Introduction 6 2.2 System Identification 8 2.3 System Identification Procedures 11 2.4 Model Structure 11 2.4.1 ARMAX Model 12 2.4.2 NARMAX Model 12 2.5 Parametric Identification 13 2.5.1 Recursive Least Square 13 2.5.2 Genetic Algorithm 14 2.5.2.1 Differences between GA and Traditional Methods 2.5.2.2 2.6 16 Population Representation and Initialization 17 2.5.2.3 Fitness Functions 18 2.5.2.4 Selection 18 2.5.2.5 Crossover 19 2.5.2.6 Mutation 20 2.5.2.7 Termination Strategy 20 Nonparametric Identification 21 2.6.1 21 Artificial Neural Networks 2.6.1.1 Definition of Artificial Neural Networks 2.6.1.2 The Structure of Neural Networks (NNs) 2.6.1.3 2.6.1.4 2.6.3 22 23 Multi-layer Perceptron Neural Networks 24 Elman Neural Network 26 Adaptive Neuro-Fuzzy Inference System 2.6.3.1 28 ANFIS Architecture 29 ix 3 SYSTEM IDENTIFICATION APPROACHES 3.1 Introduction 34 3.2 Modeling 34 3.3 Model Structure 35 3.3.1 Parametric Model Structure 36 3.3.2 Nonparametric Model Structure 37 3.4 Parametric Model Estimation 38 3.4.1 Modeling Using Recursive Least Square 3.5 38 3.4.2 Modeling Using Genetic algorithm 40 3.4.2.1 Control parameters of GA 42 Nonparametric model estimation 42 3.5.1 Modeling Using Multi-layer Perceptron 3.5.2 Neural Networks 43 Modeling Using Elman Neural Networks 45 3.5.3 Modeling Using Adaptive Neuro-Fuzzy Inference System 3.6 4 46 Model Validation 47 3.6.1 One Step-Ahead Prediction 47 3.6.2 Mean Squared Error 48 3.6.3 Correlation Test 48 EXPERIMENTAL STUDY 4.1 Introduction 50 4.2 Data Acquisition System (DAQ) 50 4.2.1 Flow of Information in DAQ 52 Test Equipment 54 4.3.1 Rectangular Plate 54 4.3 x 4.3.2 Function Generator Type (TG1010A) 54 4.3.3 Amplifier Type (2706) 55 4.3.4 Electromagnetic Shaker 57 4.3.5 Piezo-beam Type Accelerometer (Kistler-8636C5) 4.3.6 NI Compact-data Acquisition Unit 4.4 5 58 59 4.3.6.1 NI-9234 Module 60 4.3.7 Processer with LabVIEW 61 Experimental Setup 62 4.4.1 63 Experiment Procedure IMPLEMENTATION AND RESULTS 5. 5.1 Experimental Results 64 5. 5.2 Modeling Using Recursive Least Square (RLS) 70 5 5.3 Modeling Using Genetic algorithm (GA) 73 5.4 Modeling Using Multilayer Perceptron Neural Network (MLP-NN) 77 5.5 Modeling Using Elman Neural Network (ENN) 80 5.6 Modeling Using Adaptive Neuro-Fuzzy 5.7 Inference System (ANFIS) 83 Comparative Assessment and Discussion 87 5.7.1 RLS Performance 87 5.7.2 GA Performance 88 5.7.3 MLP-NN Performance 90 5.7.4 ENN Performance 91 5.7.5 ANFIS Performance 93 5.7.6 Overall comparison 94 xi 6 REFERENCES CONLUSION AND FURTHER WORK 6.1 Conclusion 97 6.2 Further Work 98 99 xii LIST OF TABLES TABLE NO TITLE 5.1 Plate specifications 5.2 First five natural frequencies of the plate PAGE NO. 64 with C-C-C-C- boundary condition detected at observation point 5.3 Performance of RLS with different numbers of model order 5.4 92 Performance of ENN with different model structures 5.11 91 Performance of ENN with different numbers of model order 5.10 90 Performance of MLP-NN with different model structure 5.9 89 Performance of MLP-NN with different numbers of model order 5.8 88 Performance of GA with different numbers of population size 5.7 88 Performance of GA with different numbers of generation 5.6 87 Performance of RLS with different numbers of forgetting factor 5.5 67 92 Performance of ANFIS with different number of membership function for model order 2 93 xiii 5.12 Performance of ANFIS with different number of membership function for model order 4 5.13 94 summary of the best Performance achieved in parametric and nonparametric modelling 96 xiv LIST OF FIGURES FIGURE NO TITLE PAGE NO. 1.1 The flow chart of the project 5 2.1 Schematic of ARMAX model 12 2.2 Working principle of GA 15 2.3 Schematic of how artificial neural networks work 22 2.4 Connectives within a node 23 2.5 Multiple layers of feedfoward neural network 25 2.6 Structure of the ENN Model 27 2.7 ANFIS process 29 2.8 Type-3 fuzzy reasoning 30 2.9 Structure of type-3 ANFIS 30 3.1 System Identification procedure 35 3.2 Schematic of ARX model 36 3.3 Diagrammatic representation of the RLS algorithm 40 3.4 Diagrammatic representation of the GA algorithm 41 3.5 Diagrammatic representation of the MLP-NN algorithm 43 3.6 Diagrammatic representation of the ENN algorithm 45 3.7 Diagrammatic representation of the ANFIS algorithm 46 4.1 Block diagram of data acquisition system 51 4.2 Experimental setup layout 53 xv 4.3 Function generator of type (TG1010A) 55 4.4 Amplifier type (2706) 56 4.5 Electromagnetic shaker 57 4.6 Piezo-beam type accelerometer (Kistler-8636C5) 59 4.7 NI compact-data acquisition unit 60 4.8 NI-9234 module 61 4.9 Processer with LabVIEW 62 4.10 Experimental setup 62 5.1 Schematic of the plate with all edges clamped boundary conditions 65 5.2 Input force applied at point x=0.25𝑎 m and y=0.25𝑏 m 66 5.3 Lateral deflection detected at x=0.75a, and y=0.75b in time domain 5.4 Lateral deflection detected at x=0.75a, and y=0.75b in frequency domain 5.5 68 Lateral deflection detected at x=0.75a, and y=0.75b point ( Y ) 5.7 67 Schematic of the plate with all edges clamped boundary conditions for the development of AVC 5.6 66 69 Lateral deflection detected at x=0.75a, and y =0.25b point ( Z ) 69 5.8 Actual and predicted RLS output 71 5.9 Error between actual and predicted RLS output 71 5.10 Correlation tests of RLS 72 5.11 Actual and predicted GA output 74 5.12 Error between actual and predicted GA output 75 5.13 The best and mean fitness values in each generation 75 5.14 Correlation tests of GA 76 5.15 Actual and predicted MLP-NN output 77 5.16 Error between actual and predicted NLP-NN output 78 5.17 Mean-squared error vs. number of training passes 78 5.18 Correlation tests of MLP-NN 79 5.19 Actual and predicted ENN output 80 xvi 5.20 Error between actual and predicted ENN output 81 5.21 Mean-squared error vs. number of training passes 81 5.22 Correlation tests of ENN 82 5.23 Initial membership functions of the input variable 83 5.24 Final membership functions of the input variable 84 5.25 Actual and predicted ANFIS output 84 5.26 Error between actual and predicted ANFIS output 85 5.27 Correlation tests of ANFIS 86 xvii LIST OF SYMBOLS A z −1 Polynomials parameters of autoregressive B z −1 Polynomials parameters of exogenous C z −1 Polynomials parameters of moving average 𝑦(𝑘) System output at time sample k 𝑢 𝑘 System input at time sample k 𝜉 𝑘 White noise at time sample k 𝑦 𝑛𝑦 , 𝑛𝑢 ,𝑛𝑒 Pc Estimation output at time t Model orders Crossover probability 𝑃1 , 𝑃2 GA parents strings 𝑂1 , 𝑂2 GA offspring strings 𝐴𝑖 , 𝐵𝑖 Linguistic label of ANFIS structure 𝑧 −1 𝜆 𝜙𝑢𝑒 𝜏 𝑂𝑘 , 𝑂𝑗 , 𝑂𝑖 𝑤𝑘𝑗 𝑤𝑗𝑖 Back-shift operator Forgetting factor Cross-correlation function between u(t ) and e (t) Output values at the output, hidden and input layers Connection weight from unit j at the hidden layer to unit k at the output layer Connection weight from unit i at the input layer to unit j at the hidden layer 𝛼 Momentum factor 𝜂 Learning rate δk The error signal from the NN output to the hidden layer δj The error signal from the hidden layer to the input layer xviii LIST OF ABBREVIATIONS RLS Recursive Least Square GA Genetic Algorithm MLP-NN Multi-layer Perceptron Neural Networks ENN Elman Neural Networks ANFIS Adaptive Neuro-Fuzzy Inference System NI National Instrumentation OSA One step ahead MSE Mean squared error NNs Neural networks TRMS Twin rotor multi-input multi-output system FD Finite Difference ARMAX Autoregressive moving average model with exogenous inputs NARMAX Non-linear autoregressive moving average model with exogenous inputs ARX Auto Regressive with exogenous inputs NARX Non-linear Auto Regressive with exogenous inputs FL Fuzzy Logic RBF Radial-Basis Function BP Backpropagation RNNs Recurrent Neural Networks FIS Fuzzy Inference System SISO Single-Input Single-Output DAQ Data Acquisition System xix A/D An analog to digital converter D/A Digital to an analog converter