Motor Bearing Fault Detection Based On Deep Learning

In modern society,the motor is a tool to convert energy,people use a lot of motor in their daily life.Once the motor fails,it will have a great impact on human production and life,and even cause serious accidents and casualties.Therefore,it is of great significance to detect and eliminate the possible accidents before the motor failure to ensure the normal operation of production and living system.In this paper,two methods based on deep learning are studied for bearing fault detection.Firstly,convolution neural network is applied to motor bearing fault detection.Convolution neural network can learn the features of the original signal directly,so as to complete the task of classifying a large number of data.The output after classification is regarded as the input of softmax,and the data after classification can be detected by softmax classifier.The regularization method is analyzed to optimize the network and avoid over fitting.The convolution neural network is initialized and trained,and the advantages of convolution neural network compared with traditional motor bearing fault detection model are verified through experiments.Then,the self coding network is applied to the motor bearing fault detection,and the stack type self coding network and softmax classifier are combined to construct the motor bearing fault detection model,and the model is initialized and trained.Through the simulation experiment,the sparse noise reduction trestle self coding network with improved dropout rate and convolutional neural network are compared,and the advantages of the improved trestle self coding network method for detecting motor bearing fault are verified.Finally,the motor bearing fault detection model based on convolutional neural network and the motor bearing fault detection model based on improved trestle self encoder and softmax classifier are compared.The experimental results show that the improved motor bearing fault detection method based on trestle self coding network can improve the accuracy of motor bearing fault detection.The traditional motor bearing fault detection method is compared with convolutional neural network and sparse noise reduction trestle self coding network with improved dropout rate,which shows that the two deep learning methods have greater advantages than the traditional motor bearing fault detection method in motor bearing fault detection.