A Design and Analysis Method for Estimator-based Multiple Model Adaptive Control

International audienceThis paper aims to present a fault-tolerant control architecture based on the Multiple Model Adaptive Control (MMAC). The approach decouples the development of a bank of local controllers together with the estimation process consisting of a bank of estimators from the switching logic. A major contribution is the presentation of a promising framework based on µ-analysis for the stability analysis of the global control law. This analysis technique provides under certain assumptions, a domain of candidate controller combinations where stability robustness is guaranteed. Afterwards, the paper focuses on supervisory unit tuning. A key proposal is to use Luenberger observers and link their designs using Linear Matrix Inequality (LMI) with an online validity computation following a Bayesian law. A systematic strategy involving stability analysis through characteristic criterion to be minimized is described. The viability of the design methodology is illustrated for the inverted pendulum on a cart subject to parametric variations. Results demonstrate the successful stability and performance of the proposed Estimator-based MMAC