HANDLING THE NONLINEARITY OF A FUZZY-LOGIC CONTROLLER AT THE TRANSITION BETWEEN RULES

A fuzzy logic controller approximates a desired control surface by using the outputs of its fuzzy control rules. The shape of this control surface is mainly influenced by the linguistic variables of the rule base, the basic operators of the fuzzy sets, the implication, the inference and the defuzzification method. The linearity/nonlinearity of this control surface is discussed in many studies, however none of those studies has explicitly considered the linearity/nonlinearity at the transition between fuzzy logic rules. In this paper a simple approach to control the linearity/nonlinearity at this transition by a modified center of gravity defuzzification method is proposed, namely by introducing the so-called defuzzification weights to the overlapping areas of the consequent. Those defuzzification weights can be determined heuristically or automatically. Both cases will be demonstrated, for the latter a feedforward neural network is employed.