Reliability Estimation For Momentum Wheels Based On Information Fusion

The Momentum Wheel is one of important inertial actuators,because of its advantages of high-precision,less-consumption and no-pollution,it has been widely used in attitude control system of satellites.Momentum Wheels provides driving force for adjustment of the satellite attitude and resistance to external interference,the failure of which is fatal to the satellite.To a great degree,the reliability of Momentum Wheels determine the reliability,safety and effective service time of the whole satellite.On the one hand,the reliability estimation of Momentum Wheels can make full use of space resources and guide the planning of the spacecraft mission.On the other hand,it can provide theoretical basis for decision-making of on-orbit maintenance for spacecrafts.Because Momentum Wheels have the characteristics of small sample,long lifespan and high cost,the traditional statistical inference method based on a large number of lifetime data is no longer appropriate.The methods based on performance degradation and information fusion provide new solutions to the problem of reliability evaluation for it.Taking Momentum Wheels as the research object,the dissertation explores how to carry out reliability estimation by using multi-source information reasonably.Reliability assessment methods by fusing expert knowledge and lifetime data,by using performance degradation data,and by fusing degradation data and other types of data are studied respectively.The main contents of the dissertation are summarized as follows:(1)Reliability estimation for Momentum Wheels by fusing lifetime data based on Dempster-Shafer evidence theory and Bayesian theory.According to the investigation and analysis,the Weibull distribution is selected as the life distribution type of Momentum Wheels.Dempster-Shafer evidence theory is applied to fuse expert knowledge,the result of which is converted to be the prior distribution of the unknown parameters.The likelihood function is used to construct a mathematical model integrating failure lifetime data and right-censored lifetime data.Zeros-ones trick is employed to convert the complex model to a simpler one.Finally,the unknown parameters are evaluated based on Markov Chain Monte Carlo(MCMC)simulation method.The proposed method is compared with the non-information prior distribution method,and the result shows that the proposed method is helpful to improve the precision of the evaluation.(2)Reliability estimation for Momentum Wheels based on performance degradation.The failure mechanism of Momentum Wheels is analyzed,and the lubrication dosage of lubrication system is selected as the key performance parameter.The standard Wiener process and random effect Wiener process are applied to model the degradation data respectively.Then,the Deviance Information Criterion(DIC)is used to select the best model,which is the random effect Wiener process.MCMC simulation method is employed to estimate the unknown parameters.The case study shows that the reliability estimation for Momentum Wheels based on performance degradation can solve the problem of small samples and no failure lifetime data.(3)Reliability estimation for Momentum Wheels based on multi-source information fusion.Considering the characteristics of Momentum Wheels of small sample size,multisource reliability information and insufficient data from a single information source,an estimation method based on multi-source information fusion is proposed.Reliability estimation methods for Momentum Wheels based on multi-source information fusion in three cases are studied respectively,including fusing performance degradation data with Bernoulli data,fusing performance degradation data with lifetime data,and fusing performance degradation data,lifetime data with Bernoulli data.The construction of fusion models and the methods to evaluate the unknown parameters based on Bayesian theory in the three cases are given.Comparing the proposed method with the reliability evaluation method considering only performance degradation data,the result shows that the proposed method can improve the accuracy of the estimation.Considering the difference of reliability between individual and population,the reliability characteristics of an individual product can not be accurately represented by the reliability characteristics of the population.A real-time reliability evaluation model of an individual Momentum Wheel is proposed.Firstly,a degradation model based on degradation data of an individual Momentum Wheel is constructed.Then,Multi-source information of Momentum Wheels and expert experience is employed to construct the prior distribution of the unknown parameters of the degradation model.According to Bayesian theory,the real-time updating of unknown parameters can be completed by collecting the on-orbit monitoring data of the individual Momentum Wheel,and then the real-time reliability evaluation of the individual Momentum Wheel is completed.