Optimal formation of fuzzy rule-base for predicting process's performance measures

Abstract

For various physical processes, especially those demanding high cost or operational time, it becomes crucial to have accurate predictions of their key performance measures based on given settings of different input parameters. Among other artificial intelligence based tools, fuzzy rule-based systems have also been widely used for this purpose. Widespread applicability of the rule-based systems has been restricted by lack of accuracy in the prediction results and inherent difficulties in different approaches that have been utilized for improving their prediction capabilities. The paper presents a two-stage approach for enhancing accuracy of prediction results. The first stage seeks best possible assignment of fuzzy sets of a response variable to the rules of a fuzzy rule-base, while the second stage looks for further improvement by adjusting shapes of the fuzzy sets of the response variable. For accomplishment of both of the stages, simulated annealing algorithm has been utilized and the approach has been practically applied on experimental data related to a turning process. The process has resulted in development of a rule-base that predicts with highly acceptable levels of accuracy. (C) 2010 Elsevier Ltd. All rights reserved.