Research On Manifold Learning-Based Fault Detection For Satellite Attitude Control Systems

Quite a number of telemetry data transmitted back to ground stations from in-orbit satellites directly indicate the working status of satellites and their payloads. Not only the reliability but the anomaly detection ability can be improved by mining the embedded information in high dimensional data. In the paper, the feature extraction(FE) and fault detection(FD) scheme for satellite attitude control systems(SACS) based on locally linear embedding(LLE) is developed and demonstrated by Satellite “TX-I” telemetry. Furthermore, a semi-physical platform is built up to validate the scheme.Telemetry data features of SACS are studied intensively by studying the features and classifications of general telemetry. And a FD scheme based on principle component analysis(PCA) for SACS is developed in consideration of the difficulty in analyzing of high dimensional data. Basic features of the SACS telemetry are presented according to the satellite working modes and data influencing factors. The dimensionality reduction(DR) and FE based on PCA are performed on high dimensional data. And anomaly detection for the extracted data can be achieved with the help of statistics.Given the difficulty in extracting the features for high dimensional telemetry thought linear methods, the DR and FD based on LLE are designed. DR and FE for high dimensional data are executed with the LLE in consideration that general linear methods hardly deal with the embedding local geometrical structure in high dimensional data. SPE and2 T are used to detect faults for the extracted data. Finally, telemetry of “TX-I” are used again to demonstrate the proposed FD scheme.The conventional LLE operates in a batch mode, put it another way, the original database is set even the online samples are updating. While, an Incremental LLE(I-LLE) version is proposed here to implement FE and FD task, which is achieved by updating the weight matrix with the online samples. Similarly, telemetry of “TX-I” are used to validate the scheme..In view of the fact that telemetry data in ground stations are fragmentary and short of data about fault mode, a FD rapid simulation for SACS is designed.The system is consisted of a control computer using PC104, a simulation computer using AD7011-EVA and the reaction wheels. The fault simulation can be performed through the fault injection simulation. With the simulation data obtained from the platform can be used to demonstrate the FD methods presented in this paper.