Research On Bearing Fault Diagnosis Method Based On Improved Stacked Sparse Denoising Auto-encoder

As a general component of mechanical equipment,the normality of rolling bearing’s state directly affects the safe operation of the equipment.If the rolling bearing fails during the operation of the equipment,it will not only cause huge economic losses,but also cause a series of disasters.Therefore,it is of great significance to study the fault diagnosis technology of rolling bearings.The process of fault diagnosis can generally be divided into four steps: signal acquisition,feature extraction,feature selection and fault identification.Among them,fault feature extraction and fault identification are the important steps to ultimately determine whether the fault category of rolling bearings can be accurately judged.Therefore,this thesis also conducts in-depth research from these two aspects.The main work of this thesis is as follows:(1)Research on feature extraction method of rolling bearing fault signal: First,for the problems of high computational complexity,difficult parameter adjustment and slow training convergence speed in the feature extraction process of SDA,the loss function of the SDA network was marginalized to construct a MSDA.Then combined with layer-by-layer greedy algorithm to get SMSDA which has deep neural network properties.Then combined the SMSDA network with the Softmax classifier to obtain the SMSDA-Softmax feature extraction model,and finally used the QPZZ-Ⅱ rotating machinery fault simulation experiment platform to simulate multiple fault types.The comparison experiments show that the proposed method has greatly improved the convergence speed and learning accuracy compared with the traditional SSDA network.(2)Research on the fault identification method of the rolling bearing fault signal: First of all,this thesis introduced the classification principle of SVM in depth,and used the improved PSO algorithm to optimize the parameter selection of SVM.Then,the SMSDA-Softmax feature extraction model was combined with the SVM classifier optimized by the PSO algorithm to construct a fault diagnosis model based on SMSDA-PSO-SVM.Then made the ten fault signals collected by the Western Reserve University Bearing Data Center as the research object,and analyzed the influence of the sparseness parameter,the weight of the sparse penalty item and the weight attenuation item on the model classification performance through comparative experiments,and to determine the parameters of the SMSDA-PSO-SVM model to obtain the fault diagnosis accuracy rate of the model.Finally,the diagnosis results were compared with SVM,PSO-SVM and SMSDA algorithms.The experimental results show that the applied method has higher diagnosis accuracy and can be effectively applied to the fault diagnosis of rolling bearings.