Research On Rolling Bearing Fault Feature Recognition Method Based On Deep Learning

The safe operation of large rotating machinery equipment is inseparable from the support of bearings and other key components.It is of great significance to identify the fault types of support bearings through intelligent diagnosis technology.In recent y ears,deep learning technique has become research focus in the field of fault diagnosis,based on the key support for transit subsystems of rotating machines parts,rolling bearing as the research subject,aiming at the shortage of the current bearing faul t recognition algorithm for learning and feature fusion based on the migration of bearing fault identification method,the main research content s are as follows:(1)Aiming at the problem that traditional fault identification methods rely on manual experience and have low diagnostic accuracy,a fault diagnosis model based on deep residual network is designed.The residual network structure is optimized by network layer sorting and attention mechanism is introduced to enhance the feature extraction ability of the fault diagnosis model.Then,by means of transfer learning,fault diagnosis under cross-load and small-sample data conditions is realized.(2)Considering that some important features may be lost in the process of converting original time domain signal data into two-dimensional image data,a fault recognition network model integrating features of different dimensions is designed by using the strong feature extraction ability of convolutional neural network.Onedimensional convolutional neural network and two-dimensional convolutional neural network were used to extract the corresponding features from the original vibration signal and time-frequency dimension data respectively.Then the two groups of feature vectors are fused to realize multi-dimensional feature fusion and reclassification of fault signals.Based on the fault data of deep groove ball bearing from Case Western Reserve University(CWRU),the performance of the proposed bearing fault identification model is better than that of the traditional algorithm.