A Combined Residual Detection Method of Reaction Wheel for Fault Detection

Abstract

A fault detection method of combined residual is proposed to effectively master the health state of the reaction wheels of in-orbit satellite according to the telemetry data. Based on the characteristics of in-orbit telemetry data, in the proposed method, the XGBoost regression model is used to predict the rotation speed and generate residual with available data of the closed-loop controlled reaction wheel. The sensitivity of viscous friction coefficient to the sudden change of friction torque is also combined with the method. In addition, the fault detection method is verified in view of the incomplete telemetry data in orbit. The result shows that the proposed method has an excellent detection effect on common reaction wheel faults. The combined residual fault detection method does not rely on fault samples and has low requirements for samples, so it has a certain application value in practical fault detection system.

Keywords

• closed-loop system
• reaction wheel
• fault detection
• incomplete data
• machine learning