A Fault Diagnosis Method for Satellite Reaction Wheel Based on PSO-ELM

In this paper, the high-precision mathematical model of satellite reaction wheel is built, and the common fault types of reaction wheel are analyzed and simulated. In order to improve the efficiency of the computation in satellite fault diagnosis, a particle swarm optimization extreme learning machine (PSO-ELM) algorithm is introduced in this paper. Compared with the classical feedback neural network, ELM uses random weights and thresholds to reduce the amount of calculation by at least half. Due to the existence of the controller, the change of satellite observation signal is very small when some faults occur. Compared with the simple ELM algorithm, this new algorithm uses PSO to select the optimal output weight and threshold. So that when the training set has relatively close data, it still maintains high regression accuracy. In this paper, a reaction wheel fault diagnosis model based on PSO-ELM is established, and the data obtained from the fault simulation model are trained and tested. Compared with ELM, the regression accuracy has been significantly improved. This new fault diagnosis model provides a new research direction for the fault diagnosis of satellite reaction wheel.