Prediction Of Remaining Useful Life Of Rolling Bearing Based On TCN Combined With AGRU

As a common and key rotating support component in mechanical equipment,the safe and effective maintenance and replacement of rolling bearings is one of the important tasks of modern PHM technology,and the industrial application of real-time condition monitoring and remaining life prediction of rolling bearings will greatly improve the safety,reliability and economy of relevant mechanical equipment.At present,the research on the remaining life of rolling bearings at home and abroad is still inadequate in terms of the whole life cycle prediction and the generalization performance of the prediction model,for this problem,this dissertation proposes a method of remaining useful life prediction of rolling bearing based on TCN combined with AGRU,the main research content is as follows.Aiming at the problem that the rolling bearing vibration signal contains a large number of timing features and the traditional deep learning model performs poorly in timing feature extraction,through theoretical analysis and research on the principle of rolling bearing life prediction,TCN is proposed as the basic network structure of this dissertation as a timing feature extraction tool for bearing vibration data.In order to obtain more and different types of model training samples for model generalization optimization,and in view of the lack of running data of rolling bearings at home and abroad,this dissertation designed and implemented a series of life cycle acceleration tests of rolling bearings,and obtained a large number of relevant data samples that can be used for fault diagnosis and life prediction of rolling bearings.The quality of test data is verified by data evaluation,which provides more materials for related research of rolling bearing.In view of the lack of rolling bearing operation data at home and abroad,this dissertation designs and implements a series of rolling bearing whole life cycle acceleration tests to obtain a large number of relevant data samples that can be used for rolling bearing fault diagnosis and life prediction,which provides more fresh blood for rolling bearing related research.The quality of test datas are verified by data evaluation.In order to improve the comprehensive performance of the model,the optimization and exploration of the model structure,parameters,optimizer,data set and data preprocessing module are carried out in this dissertation.The residual block structure is used to replace the traditional neural network layer stack.A series of comparative tests were carried out to obtain the optimal parameters of the model.The influence of data set structure and size on model performance and data preprocessing methods are explored.Finally,by comparing with the LSTM-based and Res Net-based models,it is verified that the TCN-AGRU rolling bearing RUL prediction model proposed in this dissertation can not only achieve the RUL prediction of rolling bearings throughout their life cycle,but also has good performance in terms of prediction accuracy and generalization.