Low-cost neuro-fuzzy control solution for servo systems with variable parameters

Abstract

This paper treats the design and implementation of a low-cost neuro-fuzzy control solution for a class of servo systems with an integral component and variable parameters. A hybrid Takagi-Sugeno PI-neuro-fuzzy controller (T-S PI-N-FC) is proposed and presented along with its relatively simple design approach. The solution carries out the on-line adaptation of a single parameter of the input membership functions of a Takagi-Sugeno PI-fuzzy controller with input integration (T-S PI-FC-II) by a single neuron trained by back propagation with momentum factor in the framework of a model reference adaptive controller structure. The second parameter of the input membership functions is tuned by the modal equivalence principle. Linear matrix inequalities are proposed as sufficient stability conditions to be fulfilled by the parameters of the rule consequents of the T-S PI-FC-II in order to guarantee the stable design of the hybrid T-S PI-N-FC. The solution is validated by a case study using a set of three process parameters that correspond to a strip winding system laboratory equipment. Digital simulation results and experimental results are given.