Optimization versus Robustness in Simulation: a Practical Methodology, with a Production-management Case- Study

In practice, a robust solution is more appealing than an optimal solution. The methodology adds risk analysis and bootstrapping. The case study concerns pull production-control systems.Abstract Whereas Operations Research has always paid much attention to optimization, practitioners judge the robustness of the 'optimum' solution to be of greater importance. Therefore this paper proposes a practical methodology that is a stagewise combination of the following four proven techniques: (1) discrete-event simulation, (2) heuristic optimization, (3) risk or uncertainty analysis, and (4) bootstrapping. This methodology is illustrated through a case study on production control systems. That study defines robustness as the system's capability to maintain a short-term service measure, in a variety of environments (scenarios). More precisely, this measure is the probability of the short-term fill rate remaining within a prespecified range. Besides satisfying this probabilistic constraint, the system should minimize long-term work-in-process. Actually, the case study compares four systems: Kanban, Conwip, Hybrid, and Generic. These systems are studied for a well-known example, namely a production line with four stations and a single product. The conclusion of this case study is that Hybrid is best when risk is not ignored, but otherwise Generic is best: risk considerations do make a difference.