ESA GNC Conference Papers Repository
Lessons Learned of Satellite On-board Reaction Wheel Friction Estimation
This paper describes lessons learned from the anomaly of the reaction wheel friction estimator on FORMOSAT- 2, an agile remote sensing satellite. To achieve the agility requirement, first, the Attitude and Orbit Control System (AOCS) is equipped with a cluster of four reaction wheels. In the four-wheel cluster configuration, the wheel rate zero-crossing is avoided through the additional of degree of freedom compared to the lowest requirement, three wheels. Each wheel operates at a different kernel speed in hot redundancy configuration while overall wheel momentum is nullified. Then, to increase the wheel-loop control bandwidth, a wheel friction torque estimator is designed based-on Kalman filtering theory, to work with the inner wheel-loop. Theoretically, the aforementioned AOCS design is able to accommodate the failure of one wheel. The satellite has been operational in the four-wheel configuration for many years after its launch. However, the reaction wheel hardware failure forces satellite to operate in a degraded three-wheel configuration (around the end of 2009). When satellite operated in the three-wheel configuration, the wheel friction estimation diverged in an oscillating manner which eventually prevented the satellite from normal operation. This paper proposes a high fidelity reaction wheel simulation model which is the key to the reconstruction of the friction divergence phenomenon. With the simulation model and results, resolution to resume operation in the degraded configuration, including algorithm modification and AOCS parameter tunings, may be fully tested before real flight test campaign. Also, this paper proposes a robust sliding-mode wheel friction estimator design along with its simulation results as an alternative to the linear Kalman filter for the friction estimation purpose. Finally, the approach using close-to-zero wheel momentum bias is considered to resolve such an anomaly. Real flight data is then presented where satisfactory results are observed.