Research On Prognostics And Health Management System Of Docking Mechanism By LSTM

With the progress of science and technology,many new technologies have been introduced into equipped vehicles,which makes the information technology of equipped vehicles increasingly perfect and the degree of automation gradually improved.This also leads to the complication of equipping vehicles.The docking mechanism is one of the complex new-type equipment vehicles,and the reliable and safe operation of the docking mechanism is particularly important.At present,at the initial stage of research and application of fault prediction and health management system in China,there is no general software and hardware platform for the system,and a unified and perfect system can not be formed.Therefore,this paper will establish a set of fault prediction and health management system to meet the stable and reliable operation of docking mechanism.Taking docking mechanism as the research object,this paper designs a fault prediction and health management system based on Long Short-Term Memory.Firstly,the characteristics of Long Short-Term Memory(LSTM),the theoretical basis of fault prediction and health management system,are analyzed.LSTM cell structure is improved and its learning ability is verified by simulation.Secondly,the overall scheme of fault prediction and health management system for docking mechanism is designed,and the experimental acquisition platform for key components of docking mechanism is built to collect and analyze real-time data.In order to prove the diagnostic and predictive ability of the model more convincingly,sensor data at different speeds,loads and sampling frequencies were collected.Thirdly,a fault diagnosis model of docking mechanism based on LSTM is established.The main faults of docking mechanism are analyzed,and the time-frequency characteristics of sensor data are extracted by Hibert-Huang analysis of sensor data,and the feature components are used as input sequence of fault diagnosis model.By training the fault diagnosis model based on LSTM,the fault detection rate can reach 99.9% and the false alarm rate is 0.2% under different loads,different sampling frequencies and different fault data.Finally,the health management model of docking organization based on LSTM was established.Health management model can be divided into fault prediction model and residual life model.In the fault prediction model,Encode-Decode model is introduced into LSTM,which takes the time series of the current time as input and the time series of the next time as output to conduct unsupervised learning.It has good data prediction ability for normal data and fault data.The average root mean square error of the final fault prediction model can reach 0.17.Fault prediction can be realized by inputting prediction data into fault diagnosis model.Residual life prediction model uses Attention mechanism to select and learn the intermediate output sequence of LSTM output,so as to realize the prediction of residual life based on real-time state.The mean square root error of the residual life model can reach 6.4.