TRAINABLE TRANSPARENT UNIVERSAL APPROXIMATOR FOR DEFUZZIFICATION IN MAMDANI-TYPE NEURO-FUZZY CONTROLLERS

A novel technique of designing application specific defuzzification st rategies with neural learning is presented. The proposed neural archit ecture considered as a universal defuzzification approximator is valid ated by showing the convergence when approximating several existing de fuzzification strategies. The method is successfully tested with fuzzy controlled reverse driving of a model truck. The transparent structur e of the universal defuzzification approximator allows to analyze the generated customized defuzzification method using the existing theorie s of defuzzification, The integration of universal defuzzification app roximator instead of traditional methods in Mamdani-type fuzzy control lers can also be considered as an addition of trainable nonlinear nois e to the output of the fuzzy rule inference before calculating the def uzzified crisp output, Therefore, nonlinear noise trained specifically for a given application shows a grade of confidence on the rule base, providing an additional opportunity to measure the quality of the fuz zy rule base, The possibility of modeling a Mamdani-type fuzzy control ler as a feed-forward neural network with the ability of gradient desc ent training of the universal defuzzification approximator and anteced ent membership functions fulfill the requirement known from multilayer preceptrons in finding solutions to nonlinear separable problems.