Support Matrix Machine And Its Application Research In Rolling Bearing Fault Diagnosis

Rolling bearing as one of the key components of rotating machinery,the working state of rolling bearing plays an important role in the normal operation of mechanical equipment.In the actual working conditions,the rolling bearing is often affected by various alternating loads,and there may be problems such as machining error and improper assembly,so the rolling bearing has become one of the mechanical parts prone to failure.Therefore,it is of great significance to carry out the condition monitoring and intelligent diagnosis of rolling bearing.In the intelligent recognition of rolling bearing fault mode,feature extraction of fault signal is a key link.There are two problems in feature extraction:(1)when the rolling bearing is in early fault or strong background noise,the limited feature parameters can not accurately express the fault information of the bearing;(2)When extracting features,the structural information between data will be destroyed,and it is impossible to use this feature to establish an effective prediction model.As a pattern recognition method with solid theory,support matrix machine(SMM)can fully mine the structural information between matrix rows and columns,and use the spatial information contained in the matrix to establish a complex nonlinear mapping relationship between input and output.Aiming at the problems of difficult parameter selection,poor robustness and slow solution speed of SMM,this paper carries out relevant research respectively,and puts forward the corresponding improvement methods.The effectiveness and superiority of the proposed method are verified by relevant fault diagnosis experiments..The main research contents and innovations of the thesis are as follows:(1)Aiming at the problem of super parameter selection of SMM,this paper proposes whale optimization algorithm-support matrix machine(WOA-SMM).First,the high-energy aggregation time spectrum of the original signal is obtained by using the multisynchronous queezing transform(MSST)to extract rich bearing state information.Then,the whale optimization algorithm(WOA)is used to adaptively select the model parameters to avoid the defect of subjective parameter adjustment.Finally,a series of experiments on the measured data set show that WOA-SMM has better classification performance and more stable recognition results than extreme learning machine(ELM),bilinear support vector machine(BSVM)and SMM.(2)Aiming at the problem that SMM is easy to be disturbed by outliers,resulting in poor model generalization performance,this paper proposes a ramp sparse support matrix machine(RSSMM).The core of RSSMM is to construct the loss term of the model through ramp loss,and establish regularization by using sparse constraints and low rank constraints.Because ramp loss can limit the maximum loss of outliers,it reduces the sensitivity of the model to outliers and weakens the influence of outliers on the decision hyperplane.In order to verify the classification performance of RSSMM,a series of related experiments were carried out.The experimental results show that the proposed method is feasible in rolling bearing fault diagnosis.(3)Aiming at the limitations of SMM objective function,this paper proposes smooth support matrix machine(STSMM)and applies it to the fault diagnosis of rolling bearing.STSMM uses the globally smooth bell loss to construct the loss term of the model,and further expands the binary bell loss into a loss function with multi classification ability.Bell loss function is globally smooth.When using bell loss to construct the loss term of STSMM,it is conducive to the solution of the overall optimization problem because of its continuous differentiability.In addition,bell loss can limit the margin loss of outliers and improve the robustness of the model.Two sets of rolling bearing data are used to verify the effectiveness of STSMM.The analysis results show that this method has good application effect on rolling bearing fault diagnosis.