Research on Satellite Momentum Wheel Bearing Fault Detection Based on Variational Mode Decomposition and Deep Learning

The satellite momentum wheel which has high torque accuracy can exchange angular momentum with the spacecraft by changing the size of angular momentum to achieve attitude control. It is widely used in communication, remote sensing and other satellite platforms, and is an important control component of the spacecraft. When the momentum wheel is in orbit, there is a need for long-term high-speed operation, so it is very prone to malfunctions. The traditional fault detection method is limited to relying on the knowledge and experience of experts and relevant technical personnel, as well as the interpretation of the threshold value of the single machine working state. If the key telemetry values of the momentum wheel do not exceed the interpretation range, it is difficult to detect the fault in a timely manner and miss the optimal disposal time, which may lead to abnormal satellite attitude. The friction torque of the momentum wheel is the main characteristic that affects the performance of the momentum wheel bearing. In this context, this article mainly analyzes the momentum wheel motor current telemetry that is strongly related to the friction torque. The Variation-al Mode Decomposition is used to extract the signal amplitude characteristics of motor current telemetry, and then the signal amplitude characteristics of motor current telemetry are learned by using auto-encoder and variational-auto-encoder respectively to get cross entropy. Finally, it is concluded that due to the inherent constraints of the variational-auto-encoder, it is more suitable for extracting fault features of momentum wheel bearing.