Motor Bearing Fault Diagnosis Based On Current Signal

Because of the hostile working environment and high load,it is a frequent phenomenon for asynchronous motor to have mechanical and electrical faults,at the same time,rolling bearing is a high risk part for asynchronous motor to failure.Rolling bearing fault occurs in the early stage of motor fault,If the abnormal state of motor internal bearing can be accurately judged in time,it is possible to prevent further deterioration of the incident.By analyzing the current signal,we can avoid adding sensors to the motor,and has little effect on the environment.Therefore,the core of this paper is to analyze and diagnose the fault current of the internal bearing of an asynchronous motor,and to classify the fault types quickly and accurately in order to repair them in time.Combining convolutional auto-encoder with residual network,the thesis introduces rolling bearing fault diagnosis theory and classifies the internal bearing fault of asynchronous motor.Firstly,the structure principle of rolling bearing based on asynchronous motor,the mathematical model of fault bearing is created,the correlation between current and fault is derived,and the variation between three-phase current,bus current and fault is supported by experiments.Secondly,the mathematical principle of convolution neural network propagation process is introduced.On the basis of this,the convolution self-coding network model structure is adopted to enhance data feature extraction,and the residual blocks and jump connections are introduced to prevent feature loss in convolution and pool processes,deepening the number of network layers,increasing the accuracy of the algorithm while reducing computation volume,and then using softmax classification model to classify fault features.Finally,the collected experimental signals are input into the residual convolutional auto-encoder network after data processing for bearing fault classification and diagnosis.By building the information acquisition platform of the asynchronous motor,by adjusting the torque of the load to receive the data of the motor under different operating conditions,and according to the data of the bearing fault diagnosis of the asynchronous motor under different operating conditions,the training is carried out through the training process of the model and avert the problem of over-fitting,the BN,Adam optimization and Dropout were applied to optimize the model,and the diagnostic performance of the algorithm was confirmed by comparison experiments based on the indices of accuracy and loss function.