THE DEVELOPMENT OF AN ARTIFICIAL INTELLIGENCE SYSTEM FOR INVENTORY MANAGEMENT USING MULTIPLE EXPERTS

he artificial intelligence technology of expert systems has emerged from the research laboratory and demonstrated quite successful results. Practical business applications of expert systems have been in daily use since 1979. Despite this record of success, expert systems have seen limited practical application in the logistics discipline.;The purpose of this research was to investigate the potential application of expert systems to the management of inventory. Inventory management was believed to offer a pervasive and potentially rewarding area in which to apply expert systems. The NCPDM study, Measuring and Improving Productivity in Physical Distribution 1984 reported that the estimated U.S. logistics expenditures for inventory carrying cost for 1982 were ;The primary objective of this research was to discover the expert decision heuristics for a limited inventory management task. A second objective was to incorporate these heuristics into an expert system and measure the performance level of this expert system, both in terms of the effectiveness and the efficiency of the decisions which resulted.;A fourteen-step approch to the development and testing of expert systems which incorporates many of the lessons learned from past developmental efforts was used. The approach included the use of the Delphi Technique for task and expert selection, the use of multiple experts as a group, and use of the Nominal Group Technique to achieve consensus. Seven experts worked as a group with the researcher to determine the decision heuristics. The expert system which was developed, the Inventory Management Assistant (IMA), contained 441 rules and recommended advice to the user on ten separate elements. The expert system was then tested to determine the effect of the expert system on decision effectiveness and efficiency.;Inventory managers who were assisted by the expert system exhibited significant performance improvements on a complex problem. Improvements of 15.1 to 17.73 percent were demonstrated. In addition, during a single pretest a...