Wear Fault Diagnosis For Engineering Machinery Based On Evidence Reasoning And Belief Rule-based Inference

Most engineering machines are composed of a large number of tribological systems.Friction and wear between mechanical parts and the mechanical movements occurs continuously,and therefore friction and wear faults are the main causes of engineering machinery failure.In order to avoid the serious accidents caused by abnormal wear of mechanical equipment,it is necessary to study the intelligent fault diagnosis of engineering machinery,determine the severity of wear faults,and identify the wear failure mode.As a result,the early and rapid maintenance of the equipment can be achieved by the above actions.This paper takes two kinds of complex engineering machines as the research objects,which are industrial robots and diesel engines.Considering the insufficient usage of the of quantitative and qualitative information for wear fault diagnosis,the uncertainty of information,the intelligent wear fault diagnosis for engineering machinery is studied based on belief rule-based inference(BRB)and evidence reasoning(ER)rule.The main work are as follows:(1)Wear fault diagnosis for crankshaft of industrial robot based on BRB.Aiming at mapping the nonlinear relationship between the wear characteristics and wear severity of industrial robots crankshaft,this paper comprehensively uses qualitative and quantitative information to establish a BRB-based fault diagnostic model.The fuzzy uncertainty and unknown uncertainty in the wear fault information are expressed in the form of belief degree,and the analytic evidence reasoning is used for the uncertain information fusion and reasoning.Compared with the methods by using quantitative data only,the wear fault diagnosis model based on BRB has higher accuracy and better interpretability.(2)Wear fault diagnosis for diesel engine based on improved ER rules.The traditional diagnostic models are failed for the incomplete data and cannot fit the relationship between fault characteristics and fault modes by using the linear weighted fusion.To solve these problems,this chapter establishes a wear fault diagnostic model for diesel engines based on improved ER rules.The ER algorithm is used to nonlinearly fuse the activated sub-evidence,replacing the original linear weighted summation method.This fusion method can fit the nonlinear relationship between the wear fault feature and the fault mode more accurately.In order to further effectively determine the reliability factor and importance weight factor of evidence,this chapter also proposes an new method to determine the initial importance weight factor of evidence based on information entropy and a new method to determine the reliability factor of evidence based on principal component analysis(PCA).The wear fault diagnostic model for diesel engines based on improved ER rule method has better performance compared with other diagnostic models,and the methods to determine the importance weight factor and reliability factor of evidence are more objective and accurate.(3)Wear fault diagnosis for diesel engine based on multi-classifier fusion.To solve the problems in the wear fault diagnosis by a single diagnostic model,such as low diagnostic accuracy and poor fault tolerance,this chapter constructs a wear fault diagnosis model based on multi-classifier fusion,in which the improved ER rule based diagnostic model,the BRB based diagnostic model and the ANN based diagnostic model are integrated in the decision-making layer by ER rule.In the process of fusion,the reliability factor of every model is determined by considering the accuracy and stability of the individual model.The importance factor of each model is optimized by GA.As the result,the diagnostic accuracy of the fused model is not lower than the most inferior single diagnostic model which is fused.The diagnostic accuracy and fault tolerance are greatly improved by the fused diagnostic model.