## Artificial Neural Networks for the Modelling and Fault Diagnosis of Technical ProcessesAn unappealing characteristic of all real-world systems is the fact that they are vulnerable to faults, malfunctions and, more generally, unexpected modes of - haviour. This explains why there is a continuous need for reliable and universal monitoring systems based on suitable and e?ective fault diagnosis strategies. This is especially true for engineering systems,whose complexity is permanently growing due to the inevitable development of modern industry as well as the information and communication technology revolution. Indeed, the design and operation of engineering systems require an increased attention with respect to availability, reliability, safety and fault tolerance. Thus, it is natural that fault diagnosis plays a fundamental role in modern control theory and practice. This is re?ected in plenty of papers on fault diagnosis in many control-oriented c- ferencesand journals.Indeed, a largeamount of knowledgeon model basedfault diagnosis has been accumulated through scienti?c literature since the beginning of the 1970s. As a result, a wide spectrum of fault diagnosis techniques have been developed. A major category of fault diagnosis techniques is the model based one, where an analytical model of the plant to be monitored is assumed to be available. |

### What people are saying - Write a review

We haven't found any reviews in the usual places.

### Contents

Introduction | 1 |

11 Organization of the Book | 3 |

Modelling Issue in Fault Diagnosis | 7 |

21 Problem of Fault Detection and Fault Diagnosis | 8 |

22 ModelsUsedinFaultDiagnosis | 11 |

221 Parameter Estimation | 12 |

223 Observers | 13 |

224 Neural Networks | 14 |

516 System Identiﬁcation Based on Real Process Data | 92 |

517 Convergence of Network States | 93 |

52 Stability Analysis Layers Networks with Two Hidden | 96 |

521 Second Method of Lyapunov | 97 |

522 First Method of Lyapunov | 106 |

53 Stability Analysis Cascade Networks | 110 |

54 Summary | 111 |

Optimum Experimental Design for Locally Recurrent Networks | 113 |

225 Fuzzy Logic | 15 |

23 Neural Networks in Fault Diagnosis | 16 |

232 Radial Basis Function Network | 18 |

233 Kohonen Network | 20 |

234 Model Based Approaches | 21 |

235 Knowledge Based Approaches | 23 |

24 Evaluation of the FDI System | 24 |

25 Summary | 26 |

Locally Recurrent Neural Networks | 28 |

31 Neural Networks with External Dynamics | 30 |

32 Fully Recurrent Networks | 31 |

33 Partially Recurrent Networks | 32 |

34 StateSpace Networks | 34 |

35 Locally Recurrent Networks | 36 |

351 Model with the IIR Filter | 40 |

352 Analysis of Equilibrium Points | 43 |

353 Controllability and Observability | 47 |

354 Dynamic Neural Network | 49 |

36 Training of the Network | 52 |

362 Adaptive Random Search | 53 |

363 Simultaneous Perturbation Stochastic Approximation | 55 |

364 Comparison of Training Algorithms | 57 |

37 Summary | 62 |

Approximation Abilities of Locally Recurrent Networks | 65 |

41 Modelling Properties of the Dynamic Neuron | 66 |

411 StateSpace Representation of the Network | 67 |

43 Approximation Abilities | 68 |

44 Process Modelling | 72 |

45 Summary | 74 |

Stability and Stabilization of Locally Recurrent Networks | 76 |

51 Stability Analysis Networks with One Hidden Layer | 78 |

511 Gradient Projection | 82 |

513 Strong Convergence | 86 |

514 Numerical Complexity | 88 |

515 Pole Placement | 90 |

61 Optimal Sequence Selection Problem in Question | 114 |

612 Sequence Quality Measure | 115 |

613 Experimental Design | 116 |

62 Characterization of Optimal Solutions | 117 |

63 Selection of Training Sequences | 118 |

64 Illustrative Example | 119 |

642 Results | 120 |

65 Summary | 121 |

Decision Making in Fault Detection | 123 |

71 Simple Thresholding | 124 |

72 Density Estimation | 126 |

722 Density Estimation | 127 |

723 Threshold Calculating A Single Neuron | 130 |

724 Threshold Calculating A TwoLayer Network | 131 |

73 Robust Fault Diagnosis | 132 |

731 Adaptive Thresholds | 133 |

732 Fuzzy Threshold Adaptation | 135 |

74 Summary | 140 |

Industrial Applications | 141 |

811 Instrumentation Faults | 143 |

813 Experiments | 146 |

814 Final Remarks | 160 |

82 Fluid Catalytic Cracking Fault Detection | 161 |

821 Process Modelling | 163 |

822 Faulty Scenarios | 164 |

824 Robust Fault Diagnosis | 168 |

825 Final Remarks | 172 |

831 AMIRA DR300 Laboratory System | 173 |

832 Motor Modelling | 176 |

834 Robust Fault Diagnosis | 181 |

835 Final Remarks | 182 |

Concluding Remarks and Further Research Directions | 187 |

References | 190 |

203 | |

### Other editions - View all

Artificial Neural Networks for the Modelling and Fault Diagnosis of ... Krzysztof Patan Limited preview - 2008 |

Artificial Neural Networks for the Modelling and Fault Diagnosis of ... Krzysztof Patan No preview available - 2009 |

### Common terms and phrases

actuator adaptive thresholds algorithm application approximation abilities architecture artiﬁcial neural networks based approaches based fault diagnosis constant thresholds constraints Control Engineering Practice convergence cumulative distribution function decision deﬁned detection and isolation diﬀerent dynamic neural network dynamic systems EDBP eﬀective equation false alarms fault detection fault detection system fault isolation faulty scenarios feedback feedforward networks fuzzy fuzzy sets globally hidden layer hyperbolic tangent i-th identiﬁcation IIR ﬁlter input iterations learning linear Lipschitz continuous locally recurrent networks locally recurrent neural Lyapunov matrix methods model based fault model error modelling multi-layer perceptron network parameters neural model neuron number of false observed obtained optimal optimisation order ﬁlter output Patan performed poles presented in Fig probability density function problem proposed Re(z recurrent neural networks satisﬁed Section selected sensor sequences SPSA stability conditions state-space Step Stochastic Approximation structure testing set Theorem training process uncertainty variables vector weight