Self-testing and self-learning fuzzy expert system for technological process control

Abstract

The application of expert systems in technological processes control is usually limited. One problem is how to respond to changes in the environment surrounding system. A way to overcome the above shortcoming is to use fuzzy logic for representation of the inexact information. Still more the fuzzy system could be responsible for the changes in the environment if it is self-learning by automatic fuzzy rules generation. Another problem is how to balance the conflicting situations concerning facts and rules. The certainty factor formalism in classical expert systems is rather subjective assessment to overcome this. In the present work a self-testing method is proposed, based on fuzzy certainty factor. The system is designed to find the contradictory facts and incompleteness of the fuzzy rule-base. A self-learning and self-testing fuzzy expert system is presented, where the control actions of a complex technological process could be automatically generated. The system can operate in two modes, expert and fuzzy ones, that corresponds to implicit and explicit use of human knowledge during the control process. The fuzzy rule base can be automatically tested for completeness and correctness using production rules. A fuzzy module allows automatically fuzzy rule generation and fuzzy inference that allows permanent model testing for completeness. The production system is a mechanism for creating an expert module that trains the fuzzy system artificially with new learning sequences, like as states and control actions.