Observer-Based Output Feedback Control Using Invariant Polyhedral Sets for Fuzzy T-S Models Under Constraints
Fuzzy Takagi-Sugeno models, control under constraints, output feedback, invariant sets, reference tracking, bilinear programming.
In this work, a numerical method for the computation of observer-based output feedback controllers is proposed for fuzzy Takagi-Sugeno (T-S) systems subject to constraints, based on set-invariance theory. Positively Invariant (PI) polyhedral sets are used to ensure that state and control constraints are satisfied at all times. Sufficient conditions are established for a polyhedron defined in the augmented state space (state + estimation error) to be PI. From the invariance conditions, a bilinear optimization problem is formulated to simultaneously calculate the controller and observer gains and the positively invariant polyhedron that guarantee the satisfaction of the constraints. The two types of observers found in the literature of fuzzy T-S systems are considered: the first considers the membership functions dependent only on the system output, while the second refers to the general case, where these functions can be associated with any state variable. In the simplest case, although the membership functions depend only on the output, the estimated state feedback results, in general, in controllers with better performance and with larger sets of admissible states associated with them than the output static feedback control. For the general case, as membership functions depend on non-accessible states, an estimation mechanism is needed to calculate these variables. In both cases, this role is played by the fuzzy observer T-S. The problem of tracking a constant reference signal is also considered, for which the concept of robust positive invariance is used in conjunction with an Integral-Proportional (I-P) controller. Sufficient conditions are established for a polyhedron defined in the augmented state space (state + estimation error + tracking error integral) to be PI in the presence of a constant reference signal. Several numerical experiments illustrate the effectiveness of the proposed approach.