Fault Diagnosis Method Of Rolling Bearing Based On Deep Learning

Rolling bearings are an important part of contemporary large-scale machinery and equipment,and are widely used in aerospace,marine engineering and seismic exploration.The healthy of rolling bearings is important to the safety of production activities and product quality.However,in view of the undiscoverability and potential serious loss of faults,it is of great necessary and challenge to accurately and effectively complete fault identification.Modern fault monitoring technology relies on sensor data,with unbalanced,noisy,large amount of data and other characteristics,based on the original data for fault classification is difficult,low efficiency,so this paper proposes a feature-driven fault classification algorithm,and for small samples and uneven distribution of the scene and large data volume scenarios put forward the corresponding model,the model based on different scene characteristics,balance calculation resources and monitoring accuracy,based on the CWRU fault database,improve the rolling bearing fault diagnosis efficiency,as follows:（1）For the scene with uneven distribution of fault data,the three-layer algorithm architecture of feature extraction,fault classification and model optimization is designed,feature mining is carried out based on sparse autoencoder,and the support vector machine is cited as the classification recognizer,the algorithm architecture has the characteristics of lightweight model and high computational efficiency,and can classify faults for scenes with small samples and uneven distribution.（2）Design a three-layer algorithm architecture for scenes with large amount of fault data,build a self-encoder based on a fusion convolutional neural network to achieve feature extraction,and classify data based on a support vector machine,the algorithm architecture has the characteristics of high feature extraction accuracy and fast iteration speed,and can mine hidden information for fault classification under large samples of massive data.（3）In view of the difficulty of selecting the hyperparameter of the support vector machine classifier,the heuristic algorithm Sparrow search algorithm is proposed for the optimization of hyperparameters,including the chaotic optimization of the heuristic algorithm,and applied to the actual project,the algorithm can quickly find the optimal hyperparameter that makes the support vector machine perform the best within a certain range.