Consistency monitoring of Adaptive Controllers using Bayesian linear regression

A method based on Bayesian linear regression for output monitoring of an adaptive controller is presented. As a basis, a feedback linearized system is augmented by a Model Reference Adaptive Controller. The application of Bayesian linear regression with online recorded data allows the prediction of the adaptive control output and the detection of off-nominal behavior. Additionally, the presented algorithm explicitly computes the probability of the observation error lying within a predefined error interval. The concept is shown in numerical simulation of an aircraft subject to wingrock dynamics.