Research On Fault Diagnosis Based On Intelligent Computing

With the development of science and technology,the structure of industrial system is becoming more and more complex,and the possibility and complexity of failure are increasing.In order to improve the safety and reliability of the system,it is very important to detect,isolate and identify the faults in time.According to the nature of fault,fault can be divided into structural fault(hard fault)and parametric fault(soft fault).Because structural faults can bring catastrophic consequences to the system,the theory and method of structural faults diagnosis are the focus of early fault diagnosis research.In recent years,with the development of structural fault diagnosis theory,the focus of fault diagnosis research has gradually shifted to parametric fault diagnosis.Compared with structural faults,parametric faults do not immediately cause changes in the topology of the system,but may degrade or even fail the overall performance of the system.Therefore,parameter-based fault diagnosis is the premise and key for early fault determination,fault root analysis,residual life estimation and fault tolerant control.Compared with structural faults,the fault diagnosis of parametric faults is more difficult and the related research is more active.At present,there are two kinds of parametric fault diagnosis methods: model-based and data-driven.Most of the fault diagnosis methods based on model are to extend the fault diagnosis method of linear system to nonlinear system,and then use the fault diagnosis theory of linear system to study.When the system contains model error,random noise and external interference,this linearization method is likely to produce large errors,or even lead to wrong conclusions.In data-driven methods,the diagnostic performance often depends on the quantity and quality of training data.Because the actual system has many fault modes and limited fault samples,the data-driven diagnosis method has many problems,such as high computational complexity and poor accuracy of fault approximation.In order to overcome the disadvantages brought by a single method,this paper adopts the integration of model based,data-driven and intelligent optimization methods to carry out fault diagnosis.In this paper,the intelligent observer,the intelligent adaptive observer,the robust intelligent adaptive observer and its application in fault diagnosis of nonlinear systems are discussed.In response to the above questions,the main research contents and achievements of this paper are as follows:1.This paper systematically summarizes the advantages and disadvantages of the existing diagnostic methods.Focusing on the problems such as how to improve the identification accuracy of multi-dimensional parametric faults and how to enhance the robustness of fault estimation,this paper mainly analyzes the conservatism in fault diagnosis of uncertain systems,the limitations of mathematical models,and the advantages of the fusion of multiple diagnostic methods.2.An intelligent observer is proposed.In this paper,the problem of multi-parameter fault diagnosis is transformed into a multi-parameter optimization problem,which is difficult for traditional observer to identify multi-parameter of nonlinear system and to solve the problem of observer gain.Under the premise that the initial state of the system is known and the state can be measured,the basic differential evolution(DE)algorithm is used to optimize the system parameters.By substituting the parameters estimated by the optimization algorithm into the system model,an intelligent observer is constructed to realize the identification of nonlinear multi-dimensional parameters.3.An intelligent adaptive observer is proposed.For the system whose initial state is difficult to determine,and the state is not completely observable,this paper designs an intelligent adaptive observer that can perform the joint estimation of parameters and states.Then the sufficient conditions for the convergence of the observer are given from three aspects of system structure,optimization and intelligent algorithm.In order to improve the robustness of the intelligent adaptive observer,the consistency analysis and feature extraction of the output signals are carried out by constructing different fitness functions for noise in the output data.4.In order to improve the search stagnation and premature convergence of DE,Cubic chaotic map was introduced in this algorithm to enhance the diversity of the initial population,and an adaptive shrinkage factor was introduced in the mutation link to ensure the ergodicity and convergence speed of the algorithm.5.For the time-varying parameter faults of nonlinear dynamic systems,this paper introduces the sliding time window based on the "time freezing" theory to realize the time-segment identification of model parameters.Through several simulation cases,it is proved that the proposed observer has good performance for both single parameter and multi-parameter fault diagnosis.