A Continuous Learning Method for Fault Location in Spacecraft Attitude Control System

In today's information age, mankind is faced with a huge speed of knowledge and technology update. The swift progress renders the conventional learning model inadequate for addressing human needs anymore. In the field of machine learning, since data in the real world is always changing, traditional machine learning systems need to be constantly retrained to adapt to new data, which is very demanding in terms of computational resources and time. Aiming at the problem that spacecraft attitude control system faults change with the change of on-orbit time leading to the decrease of localization accuracy, this paper proposes a continuous learning method for spacecraft control system fault localization. Aiming at the spacecraft attitude control system fault localization problem, a spacecraft attitude control system fault localization model is constructed, first the spacecraft attitude control system is simulated, then the fault localization system is simulated by neural network, the different fault data images are generated by residual generation, and the classification is performed by using the neural network. In response to the change of fault types and amplitudes over time in spacecraft operation, the optimized continual learning method is used to solve the situation that the model forgets the knowledge of old faults, which improves the model's ability to cope with and memorize a variety of different faults.