Transmission Failure Prediction Based On Dynamic Principal Component Analysis And Hidden Semi-markov Model

In order to reduce the filtering interference of vibration data,it is a more accurate and effective method to use the oil data of gearbox for system fault prediction.The VAR model is used to fit the oil historical data,and dynamic principal component analysis is carried out.A large number of original data are reduced in dimension to obtain the observation sequence that can express the characteristics of the original data to the greatest extent.Then the tri-state hsmm of the gearbox system is established to identify the state of the historical data,calculate the reliability function and the average remaining service life of the gearbox,and predict the system failure.Compared with previous prediction models,the accuracy and applicability of this method are better.The main research contents of this thesis are as follows:(1)Introduce the theories and algorithms related to Hidden Markov Models(HMM)and Hidden Semi-Markov Models(HSMM).Based on the theories,the residence time follows the general Erlang distribution,and the parameters can approximate continuous time.Establishing a three-state hidden semi-Markov model for the degradation process of the gearbox,identifying and judging the degradation process,designing an algorithm for the degradation state model of the gearbox,and finally calculating the conditional average residual life(CMRL)and conditional reliability function(CRF)of the gearbox system;(2)The dynamic principal component analysis(DPCA)is used to preprocess the oil data of the gearbox,fit the history of health data into the vector autoregression(VAR)model,reduce the data dimension by DPCA,and extract as many reaction original information as possible from fewer comprehensive variables according to actual needs,thus providing observation variables for the establishment of the hidden semi-Markov model;(3)The residual error calculated in the vector autoregressive model is fitted by the health history data,and the HSMM observation process of the historical oil data is constructed,the parameters in the model are estimated,and the remaining service life of the system is calculated.The fault history data and pause history data from actual samples are extracted for testing,and the calculation results of HSMM and HMM are compared to prove the superiority and accuracy of the model.