Data-based incipient fault detection and prediction for satellite's attitude control system

Incipient fault detection (FD) and prediction are crucial for the safe operation of in-orbit satellite's attitude control system (ACS). In this paper, a locally linear embedding (LLE) model combined with exponentially weighted moving average (EWMA) technique is proposed in FD for ACS, which is more suitable when the magnitude of the fault is small. After that, fault trend prediction with multi variables is conducted. Firstly, a preprocessing for high-dimensional telemetry data from the satellite ACS is conducted. Considering that there exists non-linear correlation relationship among telemetry parameters in ACS, LLE is used for online FD, while EWMA is used to accumulate the fault value. Based on the results of fault detection, vector autoregressive integrated moving average model (VARMA) is used for tracking the trend of fault. The case study on a simulated satellite ACS demonstrates the effectiveness of the proposed method.