Determination of design regions for fuzzy gain-scheduled robust controllers

Addresses the problem of choosing suitable representative regions in the operational envelope of a nonlinear plant for designing gain scheduled control laws. A radial basis function network (RBFN) is used to specify the size and shape of the design regions based on the behavior of parameters in linearized approximation models. Based on the training results of the RBFN, robust LTI control laws are synthesized to meet design goals throughout each region. Without modifying the basis functions, a least-squares fit is used to encode the controller gains into the scheduling function. The resulting scheduled control law is evaluated using nonlinear simulation and problem areas are used to refine the RBFN topology and the process is repeated until all design goals are satisfied over the full envelope. To illustrate the methodology it is employed to design a balancing controller for a rotary arm inverted pendulum system