Rolling Bearing Fault Diagnosis Based On Deep Learning And Transfer Learning

Rolling bearings are widely used in power,transportation,aerospace and other fields.They are indispensable parts of machines,and play an important role in the operation of equipment.The working environment of the bearing is relatively harsh,and the working conditions are very complicated.So many kinds of failures will occur after long-term work,which may lead to big accidents.Not only will they cause economic losses,but they may also threaten the safety of human lives.The vibration signal generated during the operation of the bearing contains a wealth of information.How to use the vibration signal data and parameters of the bearing to distinguish the health status of the bearing in a timely and accurate manner is of great significance to improving the reliability and stability of the bearing operation.This paper proposes different solutions based on different bearing fault diagnosis problems,and proves the effectiveness of the method through a large number of experiments.The main research work is as follows:Firstly,when the signal data comes from complex working conditions and the size of samples is small,in order to diagnose the health status and fault location of the bearing easily and quickly,this paper propose a fault diagnosis research method based on deep learning.In the experiment,it need to convert the vibration signal of the bearing into an RGB images at first,and then directly use them for SE-Res Net-50 model training and detection.The classification result of the model is the bearing fault diagnosis result.The experiment does not require feature engineering,and the average accuracy of the experiment reaches 98.28%,which fully proves the fault recognition ability of this method.Secondly,in order to detect the type and degree of bearing faults at the same time,a fault diagnosis method that combines signal processing and deep learning is proposed.In the process of fault diagnosis,if the data is not signal-processed,the deep learning method is directly used to identify the fault of the bearing,which is easy to ignore the deep features of the signal.In this paper,Fourier transform is used to process the vibration signal data,and then the processed signals are converted into RGB images for fault identification of the SERes Net-50 model.After signal processing,the model’s fault diagnosis accuracy rate is increased from 88.8% to 95.9%,which proves the feasibility of this method.Finally,in order to solve the problem of bearing fault diagnosis under variable conditions,a research method based on transfer learning and deep learning is proposed.Deep learning classification tasks require labeled data when training the model.However,in practical applications,the data collected in some working conditions does not have the same-origin labeled data,which brings difficulties to fault diagnosis.The SRDAN model proposed in this paper first uses the labeled data of different working conditions to train the model,and then it is used for fault diagnosis of target working condition data.Experiments have proved that the model can not only use its deep learning part to extract the deep features of the data,but also reduce the gap of samples of the same kind between the two working conditions and improve its transfer learning ability through domain adaptation.