Research On Mechanical Failure Prediction Based On Cointegration Theory And Convolutional Long And Short Time Memory Networks

Rotating machinery is widely used in modern manufacturing.As an important part of rotating machinery,the stable operation and reliable work of rolling bearings are of great significance to the stability and reliability of rotating machinery.Bearings are also frequent failure parts in rotating machinery.Carrying out fault trend diagnosis and remaining useful life prediction on them is inestimable in terms of avoiding major production accidents,protecting equipment and personnel safety,reducing equipment maintenance costs,and improving enterprise management How to correctly diagnose the fault trend of the bearing,and then accurately predict its remaining useful life is still the hot and difficult point of the current fault diagnosis discipline.Based on the vibration signal and its time-domain characteristics,this paper uses the time series model and deep learning theory to predict the performance degradation trend and remaining service life of rolling bearings.The main contents of this article are as follows:Firstly,introduce the overall framework of mechanical failure methods briely,and then summarize the research status of domestic and foreign rotating machinery fault prediction technology,and finally point out the advantages and application prospects of data-driven methods in mechanical failure prediction.Secondly,three traditional time series analysis models about moving average autoregressive model(ARMA)are described.These time series models can only handle single-channel stationary time series.In order to improve the accuracy of the prediction results,this chapter constructs the ARIMA-KF model after differentially processing the non-stationary time series.The bearing life cycle vibration data collected by the Intelligent Maintenance System(IMS)bearing test platform proves that the proposed method improves the prediction accuracy of the traditional ARIMA model.Thirdly,in order to deal with multiple non-stationary time series variables at the same time,this chapter proposes a mechanical fault prediction method based on cointegration theory and vector error correction model.To establish a vector error correction model,it is required that multiple characteristic variables have non-stationary characteristics and have a cointegration relationship.This chapter first extracts three time-domain features from the vibration signal,and verifies its non-stationarity and cointegration relationship,then establishes a vector error correction model,and analyzes and explains the reasons for the prediction results.By comparing with the ARIMA-KF model,the vector error correction model simplifies the modeling process while improving the prediction accuracy.Finally,due to the powerful feature extraction capability of convolutional neural networks,long-term and short-term memory networks can "remember" the dependence of sequence data for a long period of time and can avoid the relationship of gradient disappearance or gradient explosion to a certain extent.Stacked convolution long and short-term memory network model to predict the degradation trend of bearing performance.Experimental results show that,compared with a single convolutional neural network or long-term and short-term memory network,this method improves the prediction accuracy of bearing performance degradation trends.