Adaptive quasi-optimal gains tuning of PI-fuzzy controllers

This paper proposes a new methodology for tuning the scaling factors, or gains, of fuzzy proportional-integral controllers, where the closed loop system performance is explicitly taken into account. The tuning problem is formulated as an online constrained non-linear optimization problem using a linear model structure to capture the system's dynamics. The model parameters are updated online using the most recent information collected from the plant. In the case of stringent sampling times the optimization routine is prematurely stopped before a local minimum is reached and the obtained gains subsequently provided to the controller. Experimental results from a testbed illustrate the relevance of the proposed tuning technique.