A methodology for using simulation to create manufacturing control expert knowledge

Graduation date: 1996Expert systems is an area of artificial intelligence that shows promise for a wide\ud variety of applications, particularly in solving problems that have always been considered\ud too large and complex for solution by conventional approaches. However, the process of\ud eliciting expertise from an expert, knowledge acquisition, is recognized as the principal\ud bottleneck in expert system development.\ud This research explores the use of simulation for identifying knowledge for expert\ud systems. With its capability for allowing experimentation on a model of a real system,\ud computer simulation can be used to search for the best knowledge (strategies, constraints,\ud scenarios, etc.), from among a set of alternatives, that optimizes system performance.\ud Although developing high-level abstractions for decision schema that would be applicable\ud to a large set of problem domains is desirable, it is extremely difficult. The focus of this\ud study is the sawmill industry.\ud A moderately complex and typical sawmill simulation system is modeled and used\ud as a basis for developing a knowledge base to improve the material flow for the sawmill.\ud The system components included in the system model are the bucksaw, debarker, headrig,\ud gangsaw, flitch edger, and resaw. The inputs to the system are raw material mix, order\ud file, product price, and operating times. The expert system's knowledge base is a set of\ud operating policies which relate the constraints and actions in the form of production rules.\ud Factors considered in the knowledge base involve material flow, machine breakdowns,\ud queue capacities, and alternative routings.\ud The methodology of using simulation for identifying knowledge for expert systems\ud is developed. Simulation experiments are conducted and rules are generated based on the\ud simulation system performance. The rules are recorded in an initial knowledge base and\ud tested for interaction effects. Every combination of rules in the initial knowledge base is\ud checked for its ability to improve the system performance using analysis of variance. A\ud combination of rules that improves the system performance the best is selected as the final\ud knowledge base