How simplified models of different variability affects performance of ordinal transformation

Abstract

Ordinal transformation is a technique of ordinal optimization that utilizes a simplified model for performance evaluation and ranking to further reduce computational effort. This presentation-only paper will be focused on investigating how simplified models of different variability levels affect ranking. The simulation-based study investigates capacity allocation of a re-entrant line in the context of semiconductor manufacturing by using two queuing network approximation models, Jackson network approximation (JNA) and queuing network analyzer (QNA). Both are based on parametric decomposition method and JNA is a special case of QNA with a unity squared coefficient of variation because of the exponential assumptions. Mean cycle time (MCT) is the performance index. Simulation studies of a five-station re-entrant line demonstrate that QNA capture of heterogeneous variability greatly improves the MCT ranking correlation of top-10 allocations out of 415 designs by almost 8 times over JNA at the cost of less than 3% computation time increase, i.e., the value of keeping a good model of variability from simplification.