State monitoring of systems augmented by model reference adaptive control using analytic time-series forecasting

Certification of adaptive control algorithms for use on aerospace applications has not yet been accomplished in the aerospace industry. According to an emerging consensus between various authors, online monitoring and health assessment will play an integral role in closing this gap. In this paper we propose a monitoring system for Model Reference Adaptive Controllers, which enables online detection of future state requirement violations. We achieve this by employing Gaussian Process regression, which leads to a belief on the uncertainty in the system dynamics. Using analytic time-series forecasting, the system dynamics can be projected into the future, thus allowing for a statistical assertion whether a state requirement will be violated during the prediction horizon. We show the concept in numerical simulation.