Research And Application Of Remaining Useful Life Prediction Method Of Rolling Bearing Based On Deep Learning

Under the new situation of industrial Internet,the rise of artificial intelligence,simulation interaction and other technologies has driven industrial enterprises to transform and upgrade from traditional product production type to manufacturing service output type.Supported by the research and development of key technologies and platforms of cloud manufacturing based on open architecture,this paper carries out the research on Digital Twin construction and optimization technology of production line based on manufacturing big data,focuses on the research object of rolling bearing,carries out application research on the remaining life prediction of key components in industrial production line,and finally applies the prediction model to the Digital Twin data interaction platform.The specific research work carried out in this paper is as follows:(1)Aiming at the problem that it is difficult to effectively monitor the life of key components in complex industrial production line,this paper takes rolling bearing as the research object and carries out the research on the remaining life prediction of key components in industrial production line based on data-driven deep neural network algorithm.Based on XJTU-SY rolling bearing data set,five neural network prediction models are built by using cyclic neural network and convolutional neural network respectively to predict the remaining life of rolling bearing.The experimental results show that CNN-LSTM has better RMSE evaluation score than the other four prediction models.(2)On the basis of experiment(1),using cuckoo search(CS)optimization algorithm and CNN-LSTM neural network prediction model,a model optimization training method for rolling bearing residual life prediction is constructed to automatically optimize the super parameter combination of the model.The experimental results show that compared with the original CNN-LSTM model,the CNN-LSTM neural network prediction model based on CS optimization not only accelerates the convergence of the model,but also effectively reduces the prediction error and improves the fitting degree of the model for the prediction of the remaining life of rolling bearings.(3)Based on the background of the project,combined with the above neural network model of rolling bearing residual life prediction,this paper designs and implements the development of Digital Twin data interactive application platform.For the need of Digital Twin physical objects to represent their history and real-time state at the same time,the platform designs database service cluster,so that the prediction model can be optimized offline and incrementally based on the accumulated data.Finally,based on Docker technology,the platform is deployed on Casicloud to provide users with model cloud services.