Fault Diagnosis And Prognosis Of Nonlinear Mechatronic System Based On Particle Filter And Extreme Learning Machine

In modern industrial production,with the continuous improvement of automation degree and structural complexity of mechatronic systems,the possibility of mechatronic system fault in the process of industrial production and application is significantly increased.The mechatronic system fault usually leads to production outages or equipment damage,which may cause serious economic losses and heavy casualties.Therefore,it is imperative to design a reliable and effective scheme for fault diagnosis and remaining useful life estimation of the mechatronic system.To this end,the main contents of fault diagnosis and RUL estimation of the nonlinear mechatronic system proposed in this thesis can be concluded as follows:1)The basic principle of bond graph is introduced for modeling,based on which the bond graph model of nonlinear mechatronic system is established.Based on the bond graph model,the bi-causal bond graph model of the system is developed by separating the causality.Then,through analyzing the causality of each terminal node of bi-causal bond graph model,the analytical redundancy relations and fault signature matrix can be constructed,which enables fault detection and isolation.Compared with the analytical redundancy relations derived from bond graph model,the additional analytical redundancy relations can be obtained from bi-causal bond graph model,which may improve the fault isolation ability to some extent.On the basis of analytical redundancy relations and fault signature matrix,a set of possible fault candidates can be obtained by implementing fault detection and isolation procedure.Finally,the effectiveness of the proposed fault detection and isolation method is verified by simulation and experiment results.2)When the possible fault candidates is isolated by fault detection and isolation,an improved fault identification algorithm called biogeography-based optimization-particle filter is proposed to determine a set of true faults from possible fault candidates.In the proposed biogeography-based optimization-particle filter algorithm,introducing the migration and mutation operations of biogeography-based optimization into the particle filter effectively increase the particle diversity and suppress the particle impoverishment phenomenon,which can improve the parameter identification accuracy.The validity and superiority of the proposed method of fault identification are demonstrated by experimental research and comparative analysis.3)In order to address the remaining useful life estimation problems for faulty components,a method of fault degradation trend prediction and remaining useful life estimation based on optimized extreme learning machine is proposed.In traditional extreme learning machine,the input weights and offsets of the hidden layer are given randomly,which leads to poor prediction performance.While optimized extreme learning machine optimizes the weights and offsets by biogeography-based optimization algorithm to improve the prediction accuracy.The simulation and experiment results verify the validity and feasibility of the degradation trend prediction and the remaining useful life estimation methods based on optimized extreme learning machine.