Study On One-dimensional Residual Convolutional Auto-encoder Feature Learning And Transfer Towards Bearing Fault Diagnosis

Modern mechanical equipment is developing in the direction of high speed,high load and high reliability.In order to ensure the safe operation of mechanical equipment,mechanical fault detection and diagnosis is particularly important.In the research of bearing intelligent fault diagnosis,data-driven fault diagnosis method is widely used.In this paper,the deep convolutional autoencoder network and feature knowledge transfer theory are introduced to the rolling bearing fault diagnosis task,which provides a new idea for the effective application of bearing intelligent fault diagnosis method in practice.The main research contents are as follows:Firstly,the research status of intelligent fault diagnosis method is investigated deeply,and the application of transfer learning theory in fault diagnosis is further understood.In this paper,the process of intelligent fault diagnosis,the principle and application of deep learning and transfer learning are described.Secondly,a fault diagnosis model based on one-dimensional residual convolutional auto-encoder is proposed to solve the problem of missing training data labels and the difficulty of deep neural network training.In this method,one-dimensional vibration signal is used as input,and residual learning is introduced into the stacked one-dimensional convolutional auto-encoder network for feature extraction.Finally,the fault features are analyzed through the softmax classification model to complete the fault classification.One dimensional residual convolutional auto-encoder can reveal the inherent properties and laws of the data by learning the unlabeled training data,and automatically learn the fault details of the original vibration data,so as to overcome the dependence of deep neural network training on labeled data and reduce the cost of data annotation.Finally,in practical application,due to the complex and changeable operation conditions,it is difficult to obtain a large number of real fault samples and other factors,the actual diagnosis effect of most theoretical methods is not ideal.Therefore,this paper studies an unsupervised domain adaptive transfer learning fault diagnosis method,which uses onedimensional residual convolutional auto-encoder network for feature extraction,and considers that the feature distribution learned by the feature extractor will change with the updating of network parameters,multi-layer and multi-core probability distribution adaptation is integrated to constrain the invariant features of network learning domain,and the unsupervised domain adaptive transfer learning fault diagnosis is realized.The method in this paper is verified on the bearing data set of Case Western Reserve University in the United States and Paderborn University in Germany.The important parameters of the method are also analyzed and compared with the commonly used methods.The experimental results show the effectiveness of the research method in this paper.