Diagnostic Method And Experimental Study Of Typical Faults Of Asynchronous Motor Based On Deep Belief Network

The real-time detection and fault diagnosis of the motor can analyze and identify the characteristic changes in the initial stage of the motor fault to avoid further deterioration of the motor fault.In addition,it provides data and experience for the improvement of motor performance,which is beneficial to further improve the reliability of the motor.Therefore,research and development of better motor fault diagnosis techniques and systems not only have important theoretical significance,but also significant economic value and practical value.In the traditional motor fault diagnosis,due to the complicated mechanical structure of the motor,the non-stationary signal and other factors,and the human factors in the process of signal feature extraction,the recognition rate of fault diagnosis is limited.This paper introduces the concept of deep learning and applies the Deep Belief Network(DBN)to the fault diagnosis and identification of bearings and motors.Using the power spectral density of the vibration signal of the bearing as the input of the network,the different fault types and fault degrees of the bearing are classified and identified.The results show that the recognition accuracy of the deep belief network in bearing fault diagnosis is close to 100%,indicating the deep belief network.It is highly feasible and practical in the field of fault diagnosis and identification.The Deep Belief Network is applied to the fault diagnosis of the motor.The mechanical fault comprehensive simulation platform of SQI Company is used to simulate the five motor states.The power spectral density of the original vibration signal of the motor is used as the training and test data of the network.The effects of forward training times,fine tuning training times,batch processing volume,network depth and sample length on network identification accuracy and time provide a basis for determining optimal network parameters.The DBN is compared to a traditional BP network,a stacked automatic encoder(SAE),and a convolutional neural network(CNN).The results show that the recognition accuracy of DBN in motor fault diagnosis can reach 100%,which is much better than traditional BP network.Compared with other deep learning networks,it is optimal in training time.