Research On Fault Diagnosis Method Of Rolling Bearing Under Different Types Based On Deep Adaptation Network

Rolling bearings are widely used in production,transportation,aerospace and other fields.There are many types of bearings in practice,and the vibration signals of different types of bearings are quite different.The traditional fault diagnosis methods under fixed type have some shortcomings.Therefore,it is very important to accurately identify the running state of bearings under different types and replace the faulty bearing in time.Two fault diagnosis methods are proposed for rolling bearings under different types.(1)Based on the deep adaptation network(DAN),a fault diagnosis method for rolling bearings under different types is proposed.Firstly,this method uses time-frequency analysis and recurrence plot to transform the time-domain vibration signals of bearings under different types into two-dimensional images.Then,the image data of a certain type is regarded as the source domain(labeled),and the image data of other types is regarded as the target domain(unlabeled).The image data is input into the deep adaptation network which is improved by using a weighted mixed kernel function and a weighted loss curve.Deep neural network is used to extract the deep features of image data,and combined transfer learning to further reduce the distribution differences between different types,thereby achieving fault diagnosis for rolling bearings under different types.The experimental results show that the proposed method can obtain higher accuracy for fault diagnosis of bearings under different types.The process of image transformation is complicated in the above method.In addition,the parameter setting of the image transformation method is greatly affected by human factors.Therefore,another fault diagnosis method under different types is proposed.(2)Based on the transfer neural tree(TNT),a fault diagnosis method for rolling bearings under different types is proposed.In this method,the time domain vibration signals data of a certain type is regarded as the source domain data and the time domain vibration signals of other types is regarded as the target domain data.Among them,the source domain data is all labeled,and the target domain data contains a small number of labels.First,Fourier transform is used to preprocess the target domain data and obtain the corresponding frequency domain amplitude signal.Then the source and target domain data are input into the TNT.A large amount of labeled source domain data is used to assist a small number of labeled target domain data for establishing a domain adaptation model,which is used to classify a large number of unlabeled target domain data and obtain diagnosis results.When the source domain data and target domain data are selected from different types of rolling bearing,the experimental results show that the proposed method can obtain higher accuracy.Compared with method(1),the method is more suitable for occasion with higher fault diagnosis accuracy requirement.