Research On Fault Diagnosis And Remain Useful Life Prediction Of Rolling Bearing Based On Novel Deep Learning Algorithm

Rolling bearings are casually used in industrial production,Large-scale rotating machinery shaft systems,such as wind turbines,have large spans,complex structures,and variable working conditions.There are conducive to timely control measures and avoid serious accident using remaining useful life(RUL)prediction and fault diagnosis of rolling bearing.Currently,the small and medium-sized rotating machinery often uses rolling bearings as a support.It is meaningful to explore the fault diagnosis and the RUL prediction methods of rolling bearing,and also can ensure equipment safe and smooth production.Different from the perspective of vibration signal analysis and processing,this paper uses deep learning and domain adaptation methods to implement fault diagnosis and RUL prediction of rolling bearings.The main contents are as follows:In the view of the non-stationary conditions and the difference between the distribution characteristics of training data and target data in the working process of rolling bearings.At the first,a domain adaptation learning framework based on deep JDA network is proposed.In addition,a fault diagnosis algorithm for rolling bearings based on deep JDA network is designed,which solves the problem that the training set(source domain)and the test set(target domain)cannot obey the same distribution in fault diagnosis.The proposed algorithm makes the use of the classifier trained with the source domain which can classify the faults in the target domain.Experiments show that the proposed bearing fault diagnosis algorithm based deep JDA has good domain adaptive learning ability and fault classification effect.Due to the imbalance problem of samples in the target domain and the target domain,traditional domain adaptation learning method using the pre-defined domain difference distances lacks domain adaptation ability.At the beginning,a deep domain adaptation learning framework based on generative adversarial networks is proposed.Meanwhile,a rolling bearing fault diagnosis algorithm is designed,which solves the problem that the amount of data between the source and target domains is mismatch in the fault classification situation.The proposed algorithm makes the use of the classifier trained with the source domain which can classify the faults in the target domain.The experiment proves that the proposed bearing fault diagnosis algorithm using deep adversarial domain adaptation can maximize the use of limited target domain data to achieve domain adaptation.Aiming at the mismatch of the sample space types in the source and targetdomains,a partial adversarial domain adaptation network is used to construct the partial adversarial domain adaptation(PADA)framework.Firstly,a PADA framework based on generative adversarial network is proposed.At the same time,the proposed framework is used in rolling bearing fault diagnosis,and an PADA algorithm is designed to solve the problem that the types of the sample space between the source domain and the target domain is mismatch.The PADA network can classify the faults in the target domain.Experiments prove that the proposed PADA bearing fault diagnosis algorithm can weight the source domain samples and select the common sample types.In order to solve the problem of inaccurate prediction caused by the accumulation of traditional remaining life prediction errors,a remaining life prediction framework based on generative adversarial network is proposed.At the same time,the proposed framework is used in rolling bearing fault diagnosis.The LSTM-GAN rolling bearing remaining life prediction algorithm is designed to solve the problem of error stacking in the rolling bearing remaining life prediction.The experiment proves that the proposed LSTM-GAN algorithm has a good predictive performance of rolling bearing remaining life.