A Health Index Construction Method for Control Moment Gyroscopes Based on Physics-Inspired Deep Learning Approach

The attitude control system is crucial for spacecraft stability, with the Control Moment Gyroscope (CMG) as a key component. As spacecraft deployment expands, CMG failures have become more frequent, highlighting the importance of health monitoring. This paper presents a health index (HI) construction model based on thermal balance principles, which integrates deep learning with physics-informed priors for effective feature extraction across parameter and physical spaces. Local features are extracted using a one-dimensional (1D) Convolutional Neural Network (CNN), followed by a multi-layer Transformer encoder to capture global temporal dependencies and construct the parameter space. The temperature and current derivatives, along with their coupling terms, define the physical space. The fusion of both spaces is achieved through a two-dimensional (2D) CNN, generating the final HI and improving model interpretability. Validated with real aerospace telemetry data, the model demonstrates high precision and robustness in distinguishing between different health states. The proposed approach offers a novel and efficient solution for monitoring CMG health with significant practical implications.