A goal directed simulation method using fuzzy cognitive mapping

Abstract

A goal directed simulation method using fuzzy cognitive mapping (FCM-GDS) is partially developed. The FCM-GDS system can be used for the analysis and experimental design associated with traditional manufacturing studies. The FCM-GDS system was constructed using the C++ programming language and the EXSYS expert system shell. All simulation models were constructed and analyzed using ARENA simulation software. Additional analyses were conducted using Excel spreadsheets. A goal directed system is an intelligent back-end to a simulation language that suggests specific changes to variables in the model according to a knowledge-based analysis of the previous output. Fuzzy cognitive mapping is a relatively new method of collecting and standardizing expert opinion about a problem domain. The information in a cognitive map is stored as a bidirectional graph of nodes (concepts) and arcs (causal connections). The causal connections in a fuzzy cognitive map are fuzzy values or qualifiers. The research was conducted in four steps. First, a set of important manufacturing concepts was identified and the directions of their causal interactions were developed. Next, a logical procedure for manipulating these causal weights in the context of an experimental design framework was formulated. Third, detailed experiments were run to ascertain estimates of the relevant causal weights in both a typical flow line environment and a typical job shop environment. Finally, the constructed FCM-GDS system was compared to a traditional response surface methodology in two similar experiments. The two simulation models tested were a lightly congested flowline and a highly congested job shop. The suggested goals dealt with throughput, cycle time, utilization, and cost. The allowable changes involved decreasing the arrival rate and adding additional machines. The specific results of this research were favorable. Over the limited tests performed, the FCM-GDS system decreased the number of simulation runs required to reach the neighborhood of a predetermined set of goals by 75% when compared to the response surface method. Additional tests are required to establish the procedure's usefulness over a wide variety of manufacturing situations.