Research On Remaining Useful Life Prediction Method Of Rotary Equipment Based On Machine Learning

With the continuous development of economy and technology,rotating mechanical equipment is gradually developing towards large-scale,continuous,automated,and complex directions.Moreover,key components have been working in high-temperature,high-pressure,and heavy-duty environments for a long time,posing great challenges to the operation and maintenance of equipment.Prognostics Health Management(PHM),which focuses on machine learning,is an important way to ensure the safe operation of rotating machinery and equipment,while remaining useful life prediction is an indispensable part of PHM.This paper studies the entire process of predicting the remaining service life of rotating mechanical equipment.The research is carried out from three aspects,namely,health indicator construction,early fault detection,and remaining service life prediction,using machine learning methods such as Auto Encoder(AE),Long Short Term Memory Networks(LSTM),and other methods.The main research contents of this paper are as follows:(1)Research on the construction method of health indicators for rotating machinery and equipment.Construction of health indicators based on improved multiscale entropy: Based on power spectral information entropy and morphological gradient operators,combined with multiscale analysis,an innovative calculation method of average multiscale morphological gradient power spectral information entropy is proposed.The validity of the two health indicators constructed was verified using laboratory bearing data and centrifugal pump engineering case data.At the same time,an innovative health indicator processing algorithm is proposed: based on HP filtering,the upper and lower boundary lines are adaptively divided for the health indicators calculated in the previous step,forming a confidence interval based on the health indicators themselves.(2)Research on the construction of early warning models for rotating machinery and equipment.Based on the vibration signals collected by mechanical equipment,Auto Encoder neural network is built by using the health indicators of normal data input,which is a kind of artificial neural network used in semi supervised learning and non supervised learning.It can self learn the early warning threshold value,and finally achieve fault early warning.The model is verified using the bearing data set published by the network and actual engineering data.(3)Research on the prediction method of the remaining useful life of rotating mechanical equipment.Combining the current popular population intelligence optimization algorithm to optimize the particle filter algorithm,and combining the Pairs-Erdogan crack growth model or short-term memory network(LSTM),recurrent neural network(RNN),using known data to learn the performance degradation equation of the data to be tested,which serves as a state space model,Finally,the proposed improved particle filter prediction model is used to predict the remaining useful life of the rotating equipment to be tested.The validity and practicality of the prediction model are verified using the full life cycle data of the rotating equipment or components in the open dataset and actual project case data.