A method to build membership functions

Abstract

During the last years, the development of knowledge based systems has taken an increasing expansion in the control of complex industrial processes. The control knowledge is often vague, and it is necessary to create a tool to define the words of the language that are used by the expert when this one is reasoning. We propose a method to acquire and to represent the vague knowledge of the expert.

This method uses membership functions which are coming from the fuzzy set theory. This method is defined in a semantic definition module that allows to build the membership function of three dependent predicates which are defined on the same universe of discourse. Thus, the membership functions permit to obtain a representation which is conformable to the meaning that the expert gives to each predicate that corresponds to the terms of his domain. We determine the parameters of these functions from two kinds of knowledge : first, the knowledge about semantic links joining the three predicates we want to represent on the same universe of discourse, and next, the expert knowledge (heuristics) about the meaning the expert wants to give to each predicate.

The semantic definition module is used in a decision making system assisting the operators which are controlling the process during disturbing conditions. This module is a "numeric / symbolic" interface that, from each numerical datum issuing from the process, interprets it in symbolic information. Then, these symbolic information is in the formalism of the decision making system. The collected information is in harmony with the knowledge of the operators, and leads to a best exploitation of the decision making system.