Proceedings Article | 6 June 2024
KEYWORDS: Satellites, Tunable filters, Error analysis, Bismuth, Control systems, Complex systems, Actuators, Switching, Covariance, Systems modeling
An inherent property of dynamic systems with real applications is their high degree of variability, manifesting itself in ways that are often harmful to system stability and performance. External disturbances, modeling error, and faulty components must be accounted for, either in the system design, or algorithmically through estimation and control methods. In orbital satellite systems, the ability to compensate for uncertainty and detect faults is vital. Satellites are responsible for many essential operations on Earth, including GPS tracking, radio communication/broadcasting, defense, and climate monitoring. They are also expensive to design and fabricate, to deploy, and currently impossible to fix if suddenly inoperable. In being subjected to unforeseen disturbances or minor system failures, communications with Earth can cease and valuable data can be lost. Researchers have been developing robust estimation and control strategies for several decades to mitigate the effects of these failure modes. For instance, fault detection methods can be employed in satellites to detect deviations in attitude or actuator states such that error or incorrect data does not propagate further across its long life cycle. The Kalman Filter (KF) is an optimal state estimation strategy with sub-optimal nonlinear variations, commonly applied in most dynamic systems, including satellites. However, in the presence of aforementioned uncertainties, these optimal estimators tend to degrade drastically in performance, and must be replaced for more robust methods. The newly developed Sliding-Innovation Filter (SIF) is one such candidate, as it has been demonstrated to perform state estimation robustly in faulty systems. Using an in-lab Nanosatellite Attitude Control Simulator (NACS), an adaptive hybrid formulation of the SIF and EKF is applied to a satellite system to detect faults and disturbances in experiments, based on the Normalized Innovation Squares (NIS) metric. This strategy was demonstrated to improve state estimation accuracy in the presence of multiple faults, compared to conventional methods.
