meSO: simulation optimization using a multicriteria process capability index and evolutionary algorithms

Abstract

A successful implementation of a simulation-optimization (SO) methodology is presented. Based on evolutionary algorithms with a multicriteria fitness function, the new SO is used to developed weekly schedules at a ship building factory that manufactures around 60 jobs per week. Simulation modeling is used to account for randomness on the input data, as well as to correctly abstract the complex operations carried out in the real system. A variant of genetic algorithms is used to search for the appropriate schedule. Its fitness function is a multicriteria process capability index that aggregates three individual criteria, namely, makespan, line blockage and idleness of resources. The index is based on the satisfaction of thresholds for each and every criterion, thresholds that are tightened as improved schedules are found. The thresholds are also used to reject non-promising alternatives without having to perform the same number of runs as for the candidates that stand out for implementation. The name of the methodology is meSO: multicriteria evolutionary SO.