Application of Fuzzy CMAC to Filament Tension Control

Abstract

Filament winding is a complex nonlinear time-varying process. Traditional PID controller can not ensure stable filament tension. Parallel implementation of fuzzy cerebellar model articulation controller (CMAC) and PID controller is proposed to control the plant. The adoption of Gaussian curve as activation function to the receptive fields can eliminate the undesired binary behavior of conventional CMAC. The modification to the learning algorithm of fuzzy CMAC greatly improves the learning efficiency. The creditability, which is defined by combining the exciting degree and learning times, is assigned to each hypercube. Error correction is not equally distributed among hypercubes and weights updating is performed by the creditability. To adjust the learning rate dynamically can further improve the learning speed and avoid the learning interference. The online learning ability of fuzzy CMAC makes it suitable to the application of time-varying process. The experimental results have shown that accurate and stable filament tension control can be obtained by the proposed control scheme.