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

With the continuous development of industrial technology and the increasing number of equipment monitoring sensors and monitoring data,the diagnosis of mechanical equipment has entered the "big data era" of intelligent diagnosis.The more traditional intelligent diagnosis method can get rid of the difficulty of artificial feature extraction and feature selection to a large extent.As one of the essential components of equipment,rolling bearings are widely used in various fields.Therefore,rolling bearings are particularly important in the safety,efficiency and reliability of mechanical equipment.This article takes rolling bearing as the research and analysis object,and conducts research and analysis on its problems in the field of deep learning.The research work of this article is summarized as follows:(1)Aiming at the problem of the lack of labeled data sets in actual working conditions,a small-sample deep generative confrontation network(DCGAN)is designed.This method uses the powerful feature extraction capabilities of neural networks to improve the generation effect and generate labeled Sample set of failures.The model is composed of a Generative Model and a Discriminative Model.The generative model uses a deconvolutional network to generate false fault samples so that the discriminant model cannot distinguish between true and false.The discriminant model uses convolutional networks to extract sample features for true and false The samples are accurately classified.In the training process,the two confront each other,so as to achieve a kind of Nash balance,so that the generative model generates samples with similar distributions of real sample data.In the model training process,batch normalization(BN)is used to normalize the data,and the Adam optimization algorithm is used to adjust the weight and bias of the model.Experiments verify that this method can effectively solve the problem of the lack of labeled data sets for rolling bearings,which leads to the difficulty of deep learning model training.(2)Aiming at the problem of sufficient data sets but no labels in actual working conditions,a deep adversarial transfer learning diagnosis model(DATL)is further proposed.This method utilizes the abundant fault data set migration of the laboratory to be applied to the diagnosis of actual working conditions.The model is composed of a feature extraction layer,a fault classification layer and a domain discrimination layer,and the original vibration signal of the rolling bearing is used as input.In the feature extraction layer,the Res Net network is used to deeply extract signal features,and the fault classification layer is composed of a fully connected layer to achieve accurate classification of source domain data.There is a gradient reversal layer(GRL)between the domain confrontation layer and the feature extraction layer to enable the network to achieve identity transformation during the forward propagation process.During the back propagation process,the gradient direction is automatically reversed to achieve confrontation learning.The domain confrontation layer is composed of a global domain confrontation layer and a local domain category confrontation layer.The edge probability distribution and conditional probability distribution of the source domain and the target domain are aligned in the feature extraction layer,thereby reducing the difference in the feature distribution between the source domain and the target domain.Unsupervised learning using the classifier trained by source domain samples to accurately identify the target domain samples.Through experimental comparison and analysis,the proposed method further improves the accuracy of the migration learning diagnosis of rolling bearing faults under certain noise conditions.