Fault Diagnosis for Momentum Wheels of Communication Satellite Based on Artificial Neural Network

Abstract

With the advent of several fault detection techniques in modern control systems design, this paper adopted the Artificial Neural Network (ANN) Fault Detection scheme for the Fault Detection of the Attitude Control System for a Communication Satellite. In satellite applications, telemetry data can be very large, and ANN is best suited for network modeling involving large sets of data. The availability of real satellite data from Nigcomsat-1R communication satellite provided a practical platform to assess the fault detection algorithm. Results obtained showed a good correlation between raw satellite telemetry data and Neural Network model-generated results for subsequent fault detection. The fault detection models were able to detect faults, log them and provide a notification to enhance subsequent isolation and rectification. Momentum Wheel Speed and Torque were used to investigate the performance of the wheels while the Momentum Wheel Voltage and Current helped to monitor the wheel’s health state. A fault is detected if the absolute difference between original output (MW Torque) and the NN Torque output is greater than 0.012. With this, an accuracy of 100% and mean squared error of 9.8489e-6 were achieved.