Expert system with an adaptive fuzzy inference module

Abstract

An adaptive fuzzy expert system (AFES) is constructed as a hybrid in which an adaptive fuzzy inference module is combined with a neural network and equipped with a preprocessor of input data, user interface and a knowledge acquisition and modification unit. The adaptive fuzzy inference module (AFIM) is based on generalized Takagi-Sugeno fuzzy "If-Then" rules, forms of which have fuzzy sets involved only in premise parts, while consequent parts (i.e. the output of each rule) are functions of input variables. The final output of the module is the weighted sum of all the rule's output. The basic idea of AFIM is to realize a process of fuzzy reasoning and to express parameters of fuzzy reasoning by connection weights of a neural network and by forms of 4-parameter membership functions of fuzzy sets. The system is constructed for the needs of an opto-computer system for diagnosis of surface imperfections of technological elements. Similar systems can be useful in other situations, for example in the case of experimental results in which the data are imprecise and a unique functional relation between inputs and outputs is not reachable by means of classical methods.