Satellite Component Reliability Prediction Based On Information Fusion

As an important spacecraft,satellites have features such as high system integration,complicated component functions,high construction cost,long life,and difficulty in maintenance.Compared with ordinary products,satellites work in harsh outer space environments for a long time.Various factors,such as high-intensity radiation,the impact of space debris and single-particle effects,cause the inevitable degradation of satellite components,resulting in various failures.Momentum wheels,gyroscopes and other components are the key components that are most likely to fail and degrade on satellites,and their failure often leads to serious failures.Guaranteeing the safe and reliable operation of satellites in orbit is the prerequisite for ensuring high-quality completion of follow-up scientific research and other missions.Therefore,it is of great significance to predict the residual life of key satellite components.This paper analyzes a large number of failure modeling and residual life prediction data including overseas and domestic research,and summarizes various methods.It is considered that a single measurement data,a single degradation parameter,and a single degradation model are not sufficient to accurately describe the degradation process of complex products.In view of the above insufficiency,the information fusion theory is used to predict the residual life,which is summarized as follows:First of all,aiming at the problem that outer space data is easily contaminated by noise,a 1D signals denoising method for fusion of non-local information is proposed.For missing data,a missing data filling algorithm based on fusion of different models is proposed.The experiments show that the two methods are feasible and effective.Secondly,based on the Wiener process,for the problem of large degradation modeling error of single-sensor system,the multi-source sensor fusion theory is used to model the degradation data,and a distributed Kalman filter is used to fuse the state information of multiple sensors.The EM algorithm is used to update the system parameters,and the residual life prediction model of synchronous multi-sensor is obtained.Compared with the residual life prediction model of single sensor,a more accurate estimation result is obtained.At the same time,for some systems where the measurement time of the sensors is inconsistent,a method is used to convert the asynchronous indirect observations in the fusion cycle into the fusion time,establish a corresponding state space equation.The Kalman filter is used to estimate the state,and the EM algorithm is used to update the model parameters,which solves the problem of lifetime prediction for asynchronous multi-sensor systems.In addition.multi-sensor modeling for Weibull distribution products was conducted to compare the forecast results of distributed fusion strategy and centralized fusion strategy.Thirdly,by studying the degradation process of the momentum wheel,it is found that the current,the residual amount of lubricant,the bearing temperature,and other degradation can be used as the deteriorating parameters for measuring the momentum wheel failure process.Therefore,this paper uses the data collected by these three heterogeneous sensors to perform fusion modeling and uses the Copula theory to fuse the failure distribution models based on current,lubricant,and temperature respectively.Then the joint probability distribution and its reliability function are obtained.Then,according to the analysis of strictly nonlinear degradation products such as gyroscope,it is considered that a single degenerate model is difficult to accurately depict its degenerate trajectory.Therefore,a new degradation model incorporating multiple nonlinear degradation models is proposed,the residual life probability density function expression is derived.The estimation of the real time residual life is obtained by calculating its mathematical expectation,and the degradation data of a certain type of gyroscope are verified experimentally.The results show that the new nonlinear degradation model can describe the degradation process well.Finally,the degradation process of lithium battery is studied,and the degradation parameter of lithium battery capacity is obtained through correlation analysis.A residual life prediction algorithm for lithium battery based on long short term memory(LSTM)and sparse denoising autoencoder is proposed.The Stacked Denoising Autoencoder(SDA)is used to train the correlation model between degradation and battery capacity,and then the sequence prediction model of degradation is established by LSTM.Finally,the degradation prediction results of LSTM are entered into SDA,and then the prediction of lithium battery capacity is realized,and the residual life results are finally analyzed.