Research On Data-driven Incipient Fault Detection And Prediction Methods For Satellite Attitude Control System

Satellites,which are space equipment with high cost,have great application value both in the military field and in the civil field.With the great development prospect,the satellite safety and reliability have always been a hot issue in the field of aerospace engineering in the world.However,there are some factors that could pose a huge threat to the safe operation of satellites,such as the complex and changeable space environment,the large number of satellite components,the limited resources on the planet and the various unknown disturbances.Once a serious satellite fault occurs,it will bring tremendous economic losses and adverse social impacts to the country.Therefore,it is a challenging and meaningful issue to design the useful fault detection and diagnosis programs.And it is helpful to detect the abnormal signs in time,predict the trend of incipient fault development and take preventive measures to reduce the possible injury.In this paper,the unique characteristics of the satellite telemetry data from attitude control systems are analyzed.On that basis,the incipient fault detection and prediction methods based on the telemetry data are researched.This study provides a theoretical support for the on-orbit maintenance of satellites.The main work of this paper is summarized as follows.First of all,to improve the utilization of telemetry data,the periodicity and fragmentation of telemetry data resulted from orbit characteristics are analyzed.Due to the information redundancy between high-dimensional telemetry data,it is difficult to detect the fault for satellite.Based on this,the fault detection scheme combining principal component analysis and statistics is proposed to realize the on-orbit satellite anomaly monitoring function based on telemetry data.Secondly,aiming at the problem that the principal component analysis method is useless to describe the nonlinear relationship between telemetry data,a fault detection method based on NPE which belongs to nonlinear manifold learning method is studied.Firstly,the way to select the neighborhood parameters for NPE is improved,as a result,the number of neighborhoods is dynamically selected according to the density of samples.Then the improved incipient fault detection method called EWMA-DNPE is proposed,which is sensitive to the incipient fault by the EWMA control chart.Finally,considering the characteristics of telemetry data coupling,the fault prediction method based on multivariate time series is studied.First of all,aiming at the correlation between telemetry parameters,multivariable time series prediction model(VAR)is applied to the field of satellite attitude control system fault prediction.Then,the prediction scheme based on the improved dynamic VAR model is proposed when involving multi-step prediction error accumulation problem.By introducing dynamic weighting factors,the accuracy of incipient fault trend prediction is improved.