Study On Data-based Lifetime Prediction Methods For Satellite’s Momentum Wheel

Momentum wheel(MW) is a critical component in satellite attitude control systems. Its remaining lifetime has a direct impact on the reliability and lifetime of an on-orbit satellite. Howerver, MW is a type of aerospace product designed with long life and high reliability. It’s difficult to obtain a large number of failure data in a certain period of time. Therefore, remaining lifetime prediction techniques based on the traditional methods(such as the accelerated life test) are difficult to use.On the other side, a lot of performance degradation data can be acquired in an operating MW, which contain rich information about the remaining lifetime of MW.Thus, this thesis focuses on the study of data-driven lifetime prediction methods for MWs equipped in on-orbit satellites. Major contributions are summarized as below. 1. Lifetime prediction based on single degradation variable is studied. Firstly, key factors connecting to MW’s lifetime are analyzed; the lubricant remaining amount and current are selected as two variables representing the degradation of an in-service MWs. Then, two remaining lifetime prediction models are developped and verified using experimental data and simulation data. 2. A lifetime prediction method based on multiple degradation variables is proposed. It uses Copula functions to describe the correlation among multiple degradation variables. The joint distribution function is derived by fusing the two marginal distribution functions, based on which a remaining lifetime prediction model with two degradation variables is obtained. Since there are several Copula function families, the best suitable function will be selected by AIC(Akaike Information Criterion) criterion. 3. As the telemetry data may be incomplete, the missing data should be properly filled for the lifetime prediction method. Thus, the pattern of missing MW current data is analyzed. Regression and maximum likelihood methods are used to fill the missing data. Comparison of the two missing-data-filling methods has also been done in this thesis.