Integration of Simulation Modeling and Inductive Learning in An Adaptive Decision Support System

Abstract

This paper presents a decision support system (DSS) with inductive learning capability for model management. Simulation is used as the primary environment for modeling manufacturing systems and their processes. We propose an adaptive DSS framework for incorporating machine learning into the real time scheduling of a flexible manufacturing system and flexible flow system. The resulting DSS, referred to as pattern directed scheduling (PDS) system, has the unique characteristic of being an adaptive scheduler. While the bulk of previous research on dynamic machine scheduling deals with the relative effectiveness of a single scheduling rule, the approach presented in this study provides a mechanism for the state-dependent selection of one from among several rules. We address the PDS approach in the context of a model management system (MMS), with built-in simulation and inductive learning modules for heuristic acquisition and refinement. These modules complement each other in performing the decision support functions. Computational results show that such a pattern directed scheduling approach leads to superior system performance. It also provides a new frame-work for developing adaptive DSS.