Fault Diagnosis Of Three Phase Induction Motor Based On Convolution Neural Network

As a device which converts electrical energy into mechanical energy,three-phase asynchronous motor is widely used in various fields of daily life.The fault diagnosis of three-phase asynchronous motor can detect the early fault of the motor in time,reduce or avoid the loss caused by the motor failure.This paper takes three-phase asynchronous motor as the research object,and studies the fault diagnosis method of motor by using the traditional machine learning,deep learning and transfer learning.The method of VMD BP neural network and convolutional neural network(CNN)is applied to the fault diagnosis of three-phase asynchronous motor.The effectiveness and superiority of the above-mentioned fault diagnosis method are verified by experiments.At the same time,based on CNN,the method of moving CNN is applied to the fault diagnosis of three-phase asynchronous motor in different working conditions and verified.Firstly,the paper analyzes the structure,working principle and causes of the motor,selects the reasonable experimental equipment and sets up two kinds of motor failure modes:rotor strip break and bearing fault.After the motor fault diagnosis test platform is built,the vibration data of motor fan end under 600r/min,900r/min and 1200r/min conditions are collected.Aiming at the complexity of motor fault and the non-linear and non-stationary characteristics of motor vibration signal,the VMD BP neural network method is applied to fault diagnosis of three-phase asynchronous motor.The feasibility and effectiveness of the method are verified by comparing EMD BP method.Secondly,CNN is applied to the fault diagnosis of three-phase asynchronous motor,which is affected by the human feature extraction and feature selection,and there are many problems such as the complex diagnosis process,the dependence on expert knowledge and the lack of shallow structure feature learning ability.The original vibration data is directly input into the designed CNN model,the deep features are extracted by the convolution layer,the gradient dissipation problem is solved by the relu activation function,the calculation amount is reduced and the fitting is controlled by the maximum pool layer,and then the output is transformed into one dimension through the global average pool layer and input to the full connection layer,The degree of over fitting is reduced by dropout and input to softmax classifier to identify the motor state.The advantages of the method in feature extraction are proved by comparing VMD BP neural network,and the end-to-end fault diagnosis of motor can be realized.Finally,in view of the problem that the data quantity is small and the data distribution is different,it is difficult to establish a good CNN model,this paper applies the method of moving CNN to the fault diagnosis of three-phase asynchronous motor.The transfer learning and CNN model are combined to form the model of migration CNN.A large number of labeled data are used to train and verify CNN network under condition a,and the trained model is transferred to fault diagnosis task of condition B.The pre training model is fine tuned by using the label data in a small amount of condition B.finally,the trained model is applied to the condition identification under condition B to realize the fault diagnosis of motor under changing working condition.The necessity and feasibility of the method are verified by experiments,and the superiority of the method is verified by comparing with other methods.