Research On Fault Diagnosis Method Of Asynchronous Motor Based On Deep Learning Theory

Asynchronous motors are widely used in daily life and social industrial production.Due to the long-time,high-speed and high-voltage operation of the motor and the harsh environment,the probability of motor fault is very large.The abnormal operation of the motor directly affects safety and economic benefits in industrial production.Therefore,it is of great significance to use advanced science and technology in motor condition monitoring and fault diagnosis,it can achieve the early detection of equipment fault,which is beneficial to improve the reliability of industrial production and reduce maintenance costs.This paper mainly studies the deep learning algorithm and its application in the common fault diagnosis of asynchronous motors.The main research works are as follows:This paper first introduces the common fault types of asynchronous motors,including stator,rotor and bearing faults,and analyzes the reasons of these faults in the motor,and then the basic principles and methods of motor fault diagnosis are introduced.Secondly,the basic theory of deep learning and the basic models are studied.The convolutional neural network(CNN)deep learning model is proposed.The method can extract the deep abstract feature directly from the motor original data,realize the big data feature mining and fault diagnosis of the end-to-end.The rationality of the traditional CNN applied in the field of motor fault diagnosis is verified by experiments.Then,for the problem of the feature understanding of traditional CNN methods is lacked,the feature extraction efficiency and diagnostic accuracy are low,and the model robustness is poor,a new fault diagnosis method based on Gate-structure Dilated Convolutions Capsule Network(GDCCN)is proposed.The method first introduces the input gate structure of LSTM and dilated convolution,which enlarges the receptive field of the filter and improves the classification performance and robustness of the model.Secondly,the extracted feature values are sent to the primary neural capsules(lower layer)and the digit neural capsules(higher layer).The digit capsules get the final output vector neurons,which can be extracted to a more detailed feature representation through the Squash activation function and dynamic routing protocol.The diagnostic accuracy of the GDCCN model is 99.75%.This model automatically extracts the effective discriminative features of the motor signal and implements fault classification.Compared with the traditional deep learning method,its diagnostic accuracy is higher,the ability of robustness and generalization is stronger,it significantly reduces the error recognition rate and provides an efficient new intelligent diagnosis method for motor fault diagnosis.Finally,the asynchronous motor fault diagnosis experimental platform is built,and the inter-turn short circuit in stator winding and rotor bar broken,air-gap eccentricity,bearing abrasion and cage broken faults are artificially set.The vibration signals of the normal and fault operation of the motor are collected through the data acquisition industrial computer Yanhua-IPC-610 L.In order to facilitate human-computer interaction,an asynchronous motor data acquisition system interface is created,and the collected signal can be displayed on the interface of the data acquisition system.An intelligent fault diagnosis interface is also created,and the trained GDCCN model can be directly called to complete the online prediction of the motor running state,and displays the satisfactory diagnosis results on the fault diagnosis interface.