Research On Online Fault Diagnosis Of Rolling Bearing Based On Deep Transfer Learning

With the continuous development and progress of science and technology in modern society,all walks of life have gradually entered the era of big data.Rolling bearings play a key supporting role in mechanical equipment,and are one of the most prone to failure in actual production.And its running condition detection and fault diagnosis are of great practical significance.Data-driven deep learning fault diagnosis autonomously learn bearing condition information from large amounts of data.As a commonly used method in deep learning,transfer learning can effectively learn the common features between two different but related data domains.While the combination of online learning and transfer learning can effectively solve the problem that data has strong temporal attributes.Based on deep transfer learning,this paper conducts online fault diagnosis of rolling bearings,and the main research contents include:(1)Aiming at the problem that the original vibration signal of rolling bearing operating state is difficult to provide a large number of effective deep layer features,a deep transfer fault diagnosis method based on sparse auto-encoder network is proposed.Firstly,sparse restriction is introduced in the auto-encoder network to improve the feature extraction capability of the model.The one-dimensional original vibration signal of rolling bearing is used to optimize the parameters and structure of the network model to improve the accurate classification performance of the model.The established model is transferred in different data domains to further verify and improve the expression ability of the original vibration signal to different data feature,and reduce the dependence of transfer model on historical knowledge.(2)Aiming at the domain adaptation problem of transfer model in different data domains,a deep transfer fault diagnosis model based on improved convolutional neural network is proposed.In order to improve the feature expression ability of vibration signal,the time-frequency analysis method is used to convert it into a two-dimensional time-frequency image.And the convolutional neural network model is improved,the two-dimensional signal is input into the improved network,and the input data size is expanded.So that the model can obtain stronger data feature extraction capability.For two data domains with different data distributions,multi-layer domain adaptation is introduced into the model.And the difference between the two data domain distributions is continuously narrowed in the process of model training,so as to better improve the transfer ability and generalization performance of the model.(3)Aiming at the problem that the transfer model has real-time requirements in actual production applications,an online fault diagnosis model of rolling bearings based on deep transfer network is proposed.In order to make the small sample data with temporal features have powerful feature expression ability,a multi-channel convolutional neural network is constructed.And a three-channel dataset is constructed as the input of the model.The model contains offline and online parts.And the domain invariant features of different data domains are fully learned in the offline stage to ensure that the model has strong transfer ability.In the online stage,the model is fine-tuned in real time according to the dynamic data.So that the model has the ability to train new data in real time,and finally achieve the online fault diagnosis goal of the deep migration model.