LMI-based boundedness analysis of neuro-adaptive controllers

Control systems for safety critical applications, including the ones relying on adaptive elements have to be certified against strict performance and safety requirements. This paper presents an approach for verifying worst-case tracking performance of neuro-adaptive systems in presence of bounded uncertainties. In this approach the boundedness of the tracking error vector is quantitatively investigated by applying robust invariant set analysis. In this framework it was possible to specify componentwise worst-case tracking error requirements via a set of LMIs, and to systematically verify the specifications using a numerical LMI solver. The proposed method was employed to analyze and compare the worst-case performance of two neuro-adaptive controllers.