Short-term robustness of production management systems

Short-term performance of a production management system for make-to-stock factories may be quantified through the service rate per shift; long-term performance through the average monthly work in process (WIP). This may yield, for example, that WIP is minimized, while the probability of the service rate being higher than 95% still exceeds 0.90. By robustness the Japanese quality guru Taguchi meant that controllable factors (inputs) are selected such that good performance results, while the undesirable effects of environmental inputs are minimized. This paper, however, is the first to investigate robustness through risk analysis, which uses Latin hypercube sampling (LHS) to estimate the probabilities of specific system outputs for different environmental scenarios. To derive a confidence region for the resulting system performance, this paper uses bootstrapping. Short-term robustness is illustrated for a four-stage production line and several productioncontrol systems, namely Kanban, Conwip, Hybrid, and Generic. In this example, Hybrid turns out to be best. However, when risk is ignored, then Generic is best; so risk considerations do make a difference! The methodology can be easily applied to any practical production management situation, as it combines the standard techniques of simulation, heuristic optimization, risk analysis, and bootstrapping.