Research On Fault Diagnosis Algorithm Based On Deep Learnin

Recently,with the complication of industrial equipment mechanisms and the rapid development of computer hardware,deep learning has been widely used in the field of industrial fault diagnosis.However,the diagnosis accuracy is affected by various factors such as noise interference and model uncertainty.This article is based on deep learning to study industrial fault diagnosis algorithms.The following parts are the main work of this paper:First of all,to solve the problem of fault diagnosis for noisy signals.A fault diagnosis algorithm based on mutual information analysis and deep learning is proposed.Use mutual information analysis to compute the similarity between each IMF decomposed by CEEMD and the original signal to choose out the noise-dominant IMFs which will be denoised by the wavelet soft-thresholding method,which solves the problem of excessive denoising caused by simply treating high-frequency components as noise-dominant components.Then multi-scale convolutional layers and residual modules is introduced to optimize the deep learning model,and the denoised signals are extracted with different scales of deep features.The improved 1D multi-scale residual network effectively improves the feature extraction capability and the accuracy of fault diagnosis.Secondly,to solve the problem that only using single sensor data for fault diagnosis cannot guarantee the extraction of key features,which results in the decline of fault diagnosis accuracy,a data layer fusion algorithm based on VMD-HT is proposed.The algorithm introduces the VMD algorithm to improve the HHT algorithm to convert the time-frequency graph and the marginal graph from one-dimensional signal,eliminating the modal mixing problem caused by EMD,then combine the two graph with the two-dimensional monitor graph to perform Decomposition and fusion of the-RGB-channel layer and synthesize three-channel twodimensional images containing time-frequency diagram,marginal diagram and waveform diagram information,completing the data layer fusion;considering the advantages of processing three-channel images,I choose the VGG16 network as the Fault diagnosis model,and use bottleneck layer and stretch layer to improve its structure.The proposed algorithm effectively improves the accuracy rate.Third,in the field of multi-source data fusion fault diagnosis,in view of the failure to select the optimal data fusion level based on specific fault diagnosis problems,this chapter proposes a fault diagnosis algorithm based on improved D-S evidence theory and multi-level data fusion.By taking the fault diagnosis result based on single source data and the fault diagnosis result based on data layer fusion as the basic probability function of D-S evidence theory to complete the decision-making layer fusion,the fault diagnosis framework of multi-level data fusion is completed;when the D-S evidence theory deals with the high conflict evidence,the fusion effect is not good.By introducing the distance measurement and the conflict factor,the credibility of each evidence is calculated,and combined with the dempster synthesis rule proposed by Sun Quan to calculate the weighting of each evidence,which improves the D-S evidence theory and makes it more accurate when dealing with highly conflicting evidence.Finally,this paper selects the bearing fault diagnosis data set of Case Western Reserve University,and conducts simulation and comparative experiments on the above-mentioned algorithms and the commonly used algorithms in the field,and verifies the effectiveness of the algorithms proposed in the three parts of this paper.