Research On Mechanical Equipment Degradation State Modeling And Remaining Useful Life Prediction Based On Deep Learning

With the development of science and technology,analyzing the huge amounts of data produced by mechanical equipment to provide some useful decision-making information for the independent health management and maintenance has become one of the important industrial tasks.The degradation state modeling and remaining useful life(RUL)prediction are core contents.Traditional methods rely too much on signal processing technologies and expertise,and the prediction accuracy needs to be improved when dealing with complex time series,which cannot meet the accurate and efficient equipment health management requirements gradually.Thus,looking for better mechanical equipment degradation state modeling and RUL prediction methods becomes more and more important.As a new algorithm in recent years,deep learning has made brilliant achievements in many fields with its powerful high-level feature extraction and nonlinear mapping ability,but the application in the field of health management of mechanical equipment is still to be further explored.In order to solve the problems in traditional methods,the degradation state modeling method based on deep learning is deeply studied in this thesis,and the RUL prediction for mechanical equipment is further researched based on obtained degradation curves.This thesis starts from two conditions of one-dimensional and multi-dimensional monitoring data of mechanical equipment,and the bearing and turbofan engine are taken as examples to carry out the data-driven degradation state modeling and remaining useful life prediction.Firstly,the typical bearing vibration data are studied.On the basis of the advantages and disadvantages of existing deep learning models,the one-dimensional monitoring data degradation state modeling method based on stack denoising autoencoders(SDAE)and self-organizing mapping network is proposed,and the performance verification for the algorithm is conducted on PHM2012 dataset.Then this thesis takes the turbofan engine as an example and the multi-dimensional monitoring data degradation state modeling method based on SDAE is proposed,and the CMAPSS dataset is used for performance assessment.Finally,this thesis compares the existing RUL prediction models and ultimately chooses the Long Short-Term Memory(LSTM)network for mechanical equipment RUL prediction.The experimental verification is conducted on the basis of obtained degradation curves of bearing and turbofan engine.The results show that,compared with traditional mechanical equipment degradation state modeling methods,the proposed method based on SDAE is able to construct more smooth degradation curves with smaller noise on the one-dimensional and multi-dimensional monitoring data,and the degradation curves have better time correlation and monotonicity.The degradation state modeling method has less dependence on artificial participation,and the whole procedure is in an unsupervised manner,with better generality.On the basis of constructed degradation curves,the way that LSTM model maps the health values of bearing and turbofan engine directly into the RUL achieves excellent results,which further proves the necessity of degradation state modeling and effectiveness of the prediction model.