Computation of basic fuzzy controllers by approximation of time-optimal controllers

This paper presents a method to compute robust fast nonlinear controllers for real plants for which a/spl minus/not necessarily precise/spl minus/mathematical model is available. The development is based on ideas from fuzzy control, but in contrast to usual fuzzy controller designs, most of the rules are not derived from heuristics but rather are mathematical formulae which, together with the standard fuzzy quantization of the system's variables, approximate a time-optimal control law. This basic fuzzy controller can then be improved by adding further heuristic rules gained from the observation of the behaviour of the controlled plant. Our approach is illustrated by the example of the control of a laboratory two-tank system. The fuzzy controller (basic controller plus heuristics) exhibits a performance which is not attainable with standard linear control nor with classical time-optimal control.<>