Research On Motor Bearing Fault Diagnosis Based On Deep Learning

As one of the main power sources in modern industrial production,electric motors are playing an increasingly important role in industrial production.As one of the important components of the motor,the working state of the bearing has a huge impact on the motor operation and the entire industrial production activities.How to effectively and reliably predict and diagnose the current running state of motor bearings is one of the hottest research hotspots in the field of fault diagnosis.At present,motor bearing fault diagnosis methods often combine several algorithms to realize the method of extracting effective bearing fault characteristics from small sample signal data,and then input to the neural network for training and classification.But at the same time,as the working environment of motors becomes more diverse and complicated,traditional fault diagnosis methods of motor bearings often fail to effectively extract fault characteristic information.However,the shallow classification neural network has the situation that the classification accuracy cannot be increased with the increase of the sample number,which limits the further improvement of the fault diagnosis efficiency.Aiming at the shortcomings of traditional motor bearing fault diagnosis methods,combined with the advantages of Convolutional Neural Network(CNN)in deep learning in processing large sample signal data and feature extraction,this paper first adopts the CNNbased motor bearing fault diagnosis method.Select the appropriate model structure and network parameters through experimental comparison,and directly use the original vibration data as input to the network model to train it to exert its strong self-learning ability,reducing the raw data preprocessing work in current motor bearing fault diagnosis methods.By using Case Western Reserve University's open bearing database data for training classification,the accuracy of the model for bearing fault classification can reach 99.8%,achieving an ideal classification effect.At the same time,the bearing fault data collected by the laboratory QPZZ-II motor experimental bench was used as a training sample for the model to train,and finally achieved a classification accuracy of 99.7%,which verified that the model has high reliability and practicability.At the same time,in order to further improve the efficiency of motor bearing fault diagnosis,and reduce the shortage of too many training samples in the method of CNN-based motor bearing fault diagnosis.Combining the characteristics of the collected motor bearing fault signals as time series data,and the advantages of Gated Recurrent Unit(GRU)to fully exploit the time sequence information in the fault signal,the motor bearing fault diagnosis method based on CNN and GRU is adopted.In the case where the training sample is reduced by half,by combining the advantages of CNN feature extraction on large sample data with the advantages of GRU processing time series data,using the open bearing database data of Case Western Reserve University and the bearing fault data collected by the laboratory QPZZ-II motor experimental bench as training samples for this model,respectively,the classification accuracy of 99.7% and 99.6% was achieved,and CNN-based methods for fault diagnosis of motor bearings are similar.In the case of reducing the training sample,the method can dig through fault signal timing information inherent to ensure classification accuracy of fault diagnosis model,to further improve the efficiency and reliability of your diagnosis motor bearings.