Convolutional Neural Network-based Fault Diagnosis for Spacecraft Attitude Multiple Components

With the growing number of satellites in spacecraft cluster, the pressure of ground-based monitoring has increased seriously. To alleviate this pressure, many intelligent fault diagnosis methods have been developed. However, the existing algorithms are hard to apply to real data. This paper proposes an intelligent fault diagnosis method for spacecraft multi-components based on the convolutional neural network, which provides a feasible idea for intelligent diagnosis of spacecraft in orbit. Firstly, the fault data signals of multiple components are generated by semi-physical simulation. Secondly, a double normalization method is proposed to process the signal data. After that, the simple convolutional neural network model is established to train and test for fault diagnosis. Finally, the fault diagnosis of real data is performed to verify the capability of on-orbit application. Simulation results show that the proposed intelligent diagnosis algorithm of spacecraft multi-mechanism fault based on convolutional neural network can quickly and accurately locate the fault location in multiple components. The model trained on the ground can be directly applied to on-orbit spacecraft.