Design Of Multi-signal Model Based Fault Diagnosis And Their Application To Hydraulic AGC System

With the increasingly complicated structure and function of industrial systems,faults are inevitable.Efficient and effective fault diagnosis and maintenance has become the focus of system testability and health management.The purpose of optimal design of fault diagnosis strategy is to construct an optimal test sequence that can effectively isolate various faults in the system with low test cost.Fault diagnosis strategy optimization technology involves not only the product design phase but also the usage and maintenance phases.It plays an important role in improving the performance of fault detection and isolation,reducing the false rate and the missed rate of fault detection and diagnosis,and reducing the life-cycle cost of equipment.Aiming at the optimal design of fault diagnosis strategy,this thesis has carried out some fundamental research work,and the key research results obtained are summarized as follows.Firstly,the thesis has investigated the background and the state of the art of fault diagnostic strategy design.For the modeling problem of diagnostic strategy design,the commonly used modeling methods are summarized with detailed analysis on their distinctive characteristics.Then,the algorithms commonly used in fault diagnostic strategy design are reviewed,and their application limits and the strengths and weaknesses are discussed.Uncertainty is a key but difficult issue in the design of fault diagnostic strategy.This thesis has summarized the sources of uncertainty and pointed out the shortcomings of the current researches on the fault diagnosis oriented testability analysis with uncertainties.Then,aiming at an Automatic Gauge Control(AGC)system,a key subsystem in steel rolling processes,the thesis has studied the multi-signal modeling of the AGC system.Typical faults were simulated and a group of tests were set up to get the dependency matrix between faults and tests.In order to reduce the occurrence of fault ambiguity group,the traditional binary dependency matrix is improved to be a multi-valued matrix,based on which better fault diagnosis results can be achieved.Next,a novel diagnostic strategy generation method based on information entropy with one-step backtracking is proposed.Traditional diagnostic strategy generation methods are based on binary dependency matrices,which often encounter the fault ambiguity group problem.For multi-value systems,a diagnostic strategy generation method is proposed based on a multi-value dependency matrix and an information-entropy based heuristic function.The simulation results on the AGC system can show that the proposed method can generate the optimal test sequence with less test cost.Finally,the optimal test selection problem under unreliable tests is studied.Bayesian Network(BN)is used to describe the relationship between faults and tests with uncertain test information.In order to deal with the uncertainties caused by missed fault detections,the Maximum Likelihood(EM)algorithm is applied to estimate the parameters of BN.Then the test optimization problem with uncertainty is transformed to be an optimization problem to minimize the test cost where the fault detection rate,the isolation rate and the false alarm rate are considered as the constraints to the test cost function.Then,a searching algorithm that combines Simulated Annealing Algorithm and Genetic Algorithm is proposed to solve the optimization problem.The simulation results on the AGC system has shown that,the proposed optimal test selection algorithm with unreliable tests can greatly reduce the test cost;also can ensure good fault detection and isolation performance.