Research On Fault Diagnosis Method Of Rolling Bearing Based On Deep Learning

With the rapid development of industry,Rotating machinery and equipment gradually tends to be complex,precise and intelligent,and the monitoring data of mechanical equipment has gradually entered the era of "big data".Rolling bearing as a key component of rotating machinery,once the failure occurs,it will lead to equipment operation disorder,bring property losses,or affect people’s life safety.Therefore,the research on fault diagnosis of rolling bearing has very important theoretical significance and research value.Traditional fault diagnosis methods often need complex expert knowledge and manual extraction process,which can not meet the analysis requirements of big data.Intelligent diagnosis based on deep learning has natural advantages in the context of big data.Using the powerful feature extraction ability of deep learning,the rolling bearing fault signal can be better diagnosed.This paper mainly studies the fault diagnosis method based on deep learning:(1)The model of automatic encoder is applied to fault diagnosis of bearing,and the applicability of learning feature information from original signal by three kinds of autoencoders,namely incomplete,complete and over complete,and the influence of network structure on fault diagnosis accuracy are explored.At the same time,an improved deep autoencoder model is proposed,which can solve the problem of large reconstruction loss in self encoder,extract features better and have higher fault recognition accuracy.(2)Aiming at the problem of rolling bearing fault diagnosis under small sample data set,a convolution neural network fault diagnosis method based on time-frequency graph input is proposed by improving convolution neural network.This method uses the short-time Fourier transform theory to obtain the time-frequency image as the input,and uses Se LU activation function,hierarchical regularization method to optimize the convolutional neural network to obtain better training effect.The experimental results show that the proposed method can significantly shorten the training time and has higher fault diagnosis accuracy,which greatly reduces the sample requirements for training deep network model.(3)In order to solve the problem that a single sensor contains limited fault information and the ability of traditional convolution neural network to extract multi-scale fault information is limited,a multi-scale one-dimensional convolution neural network fault diagnosis method based on information fusion is proposed.After fusing the data of three direction sensors,a multi-scale one-dimensional convolution neural network is designed for fault diagnosis.The effectiveness of the proposed method is verified by experiments,and compared with single sensor multi-scale convolution neural network and traditional convolution neural network.