Remaining Useful Life Prediction Of Rolling Bearings Using Recurrent Neural Network

Predicting the remaining useful life of rolling bearings is a crucial component of diagnostics and maintenance of rotating equipment.As an essential part of rotating machinery,the working condition of rolling bearings has an impact on the machinery’s health at all times.The management of the rolling bearings’ health and state during their operation will help to arrange the maintenance plan of the equipment reasonably,decrease equipment’s repair expenses and enhance the quality of equipment’s economic benefits,and prevent emergencies and avoid major safety accidents.In order to realize the prediction of the remaining useful life of rolling bearings,it is necessary to obtain the operating condition data of the entire life cycle,extract effective degradation characteristics from them,and predict the future degradation trend of rolling bearings through historical information.Traditional remaining useful life prediction methods have certain limitations in the division of degradation stages and the determination of failure thresholds,and cannot adapt to the change of operating conditions of rolling bearings,resulting in low prediction efficiency and inaccurate prediction results.In the last decade,data-driven method has become the common technique for remaining useful life prediction of equipment,and deep learning technology has been widely used in mechanical equipment fault diagnosis and health management due to its powerful data processing capabilities.As a common deep learning model,recurrent neural network has great advantages in time series prediction and is often used as a rolling bearing life prediction model.However,the memory capacity of traditional recurrent neural network is limited,and long-term historical information cannot be effectively used for remaining useful life prediction.Meanwhile,there are problems of gradient disappearance and gradient explosion,which affect the accuracy of remaining useful life prediction of rolling bearings.In view of the above problems,this paper builds the foundation for the remaining useful life prediction method by using rolling bearings as the study object based on recurrent neural network.The following are the primary study contents:(1)Targeting the issues with poor prediction performance and significant manual measurement error of the standard prediction method,a method of automatically determining the early failure point and adaptively determining the failure threshold is proposed,and the entire life cycle is divided into different stages according to the operating state of rolling bearing.Effective degradation information can be quickly obtained,which can be input into the deep learning model for prediction.The starting point of prediction and complete failure point are automatically determined through early failure point and failure threshold.Research shows that this method can adapt to the change of operating conditions of rolling bearing and reduce prediction errors.(2)To address the issue that the conventional recurrent neural network is limited to short-term memory ability and is unable to effectively use long-lasting past information for prediction,a remaining useful life prediction method based on long short-term memory network is proposed,using the sliding window to prediction the future degradation state,to estimate the failure time point and remaining useful life of rolling bearing.Research shows that long short-term memory network has high accuracy in remaining useful life prediction of rolling bearing.(3)Targeting the issues with breakout and elimination of gradient in the prediction of conventional recurrent neural network,a remaining useful life prediction method of rolling bearing based on deep autoregressive model is proposed,which combines recurrent neural network with autoregressive model,using historical degradation information to predict the future degradation trend directly.Considering the randomness in the degradation process of rolling bearing,the method can predict the failure probability interval while predicting the failure time point.Research shows that the deep autoregressive model has strong stability and can improve the accuracy and efficiency of remaining useful life prediction of rolling bearing.