Research On Motor Bearing Fault Diagnosis Based On Feature Fusion And Relevance Vector Machine

Bearings are indispensable parts of the motor and other rotating machinery,and they are also vulnerable parts.Once the bearing fails,it is easy to cause the equipment to be paralyzed,which hinders normal production activities and paralyze certain economic losses.Therefore,the research on bearing fault diagnosis is of great significance.Focusing on the study of the vibration mechanism of motor bearings,in order to obtain better fault diagnosis results,in this paper,the feature fusion and intelligent diagnosis algorithms,the wavelet packet energy entropy,local linear embedding(LLE),Artificial bat algorithm(BA)and relevance vector machine(RVM)are combined in bearing fault diagnosis,based on the analysis of the current status of fault diagnosis research at home and abroad,corresponding improvement algorithm is proposed,and a new diagnosis model is established.First,the original vibration signal of the bearing is de-noised and decomposed using wavelet packet theory,the frequency band signal generated after the decomposition is extracted in the form of energy entropy,and then the signal is reconstructed to extract the time domain and frequency domain features.The information contained in a single feature is often limited,and the high-dimensional feature vectors based on the above three forms can reflect the feature information more comprehensively,but there are problems such as large calculation and low efficiency.Therefore,the LLE algorithm is used to fuse three types of high-dimensional feature reduction into low-dimensional feature vectors,and use it as the input of the diagnostic model.Secondly,a motor bearing fault diagnosis model based on RVM theory is built.In order to realize the multi-class diagnosis of bearings,the binary tree method is selected to construct the RVM multi-class diagnosis model.The choice of nuclear parameters directly affects the accuracy of diagnosis.To ensure the optimal selection of nuclear parameters,and to solve the problem that the optimization algorithm is prone to fall into local optimization,the improved bat algorithm(IBA)based on chaos theory is introduced to optimize the selection of nuclear parameters,and the IBA-RVM Multi-class motor bearing fault diagnosis model is built.Finally,a comparative experiment is designed.The experiment consists of two parts: first is the comparison experiment of single feature diagnosis and fusion feature diagnosis;secondly,in the case of fusion features as input,the comparison experiment is constructed by IBA-RVM and PSO-RVM,IBA-SVM and PSO-SVM.The results show that the fusion feature has better feature expression than the single feature;compared with other diagnostic models,the IBA algorithm has better global search capabilities,which improves the accuracy of RVM in bearing fault diagnosis.