Research On Remaining Useful Life Prediction Of Mechanical Equipment Based On LSTM

Due to the rapid development of industrial technology,mechanical equipment is more complex and intelligent,and the data collected is becoming larger and larger.If the mechanical equipment fault may cause serious economic losses,endanger people’s lives and health and threaten the ecological environment.Therefore,the prediction of remaining useful life of mechanical equipment is of great significance.Traditional remaining useful life prediction method requires to extract feature manually,which reduces the accuracy of remaining useful life prediction.With the rapid development of computer technology,researchers applied deep learning to remaining useful life prediction of mechanical equipment,and propose a remaining useful life prediction method based on data-driven.Long short-term Memory network(LSTM)in deep learning has been widely used in time series fields such as natural language processing and machine translation due to its advantages in extracting temporal feature information.Aiming at the difficulty of time series feature extraction in remaining useful life prediction,this thesis studied the remaining useful life prediction of mechanical equipment based on LSTM.LSTM network and its variants are used to extract features and improve the accuracy of remaining useful life prediction results.The specific research contents of this thesis include:(1)Research on remaining useful life prediction based on LSTM EncoderDecoder.A remaining useful life prediction method based on LSTM EncoderDecoder is designed and implemented.The multi-time series of mechanical equipment is taken as the input,and the intermediate representation containing state information is encoded by LSTM Encoder,and the remaining useful life is predicted by LSTM Decoder.This method combines the advantages of LSTM and EncoderDecoder.This method makes up for the problems of manual extraction of remaining useful life prediction methods and insufficient extraction of time series feature information.(2)Research on remaining useful life prediction based on Bi LSTM and attention mechanism.A remaining useful life prediction method based on Bi LSTM and attention mechanism was proposed.Firstly,Bi LSTM was used to extract degradation feature information,and attention mechanism was used to characterize the relationship between feature information,highlighting the influence of key information on remaining useful life prediction results.The effects of three different attention mechanisms on remaining useful life prediction were compared,and the comparison experiment was conducted on C-MAPSS data set.The experimental results show that Bi LSTM combined with attention mechanism can improve the accuracy of remaining useful life prediction.(3)Research on remaining useful life prediction of mechanical equipment based on Conv GRU-Attention.A method for remaining useful life prediction of mechanical equipment based on Conv GRU-Attention was designed and implemented.The spatial and temporal characteristics of equipment state were extracted by Conv GRU to better characterize the equipment degradation process.The feature information is assigned different weights by the attention mechanism,and the feature output of the intermediate network layer is visualized.Experiments on C-MAPSS and PHM2012 datasets demonstrate the effectiveness of Conv GRU-Attention method in feature extraction.