Comparative analysis of lot-sizing models for multi-stage systems: a simulation study

Abstract

In the past few years researchers have given considerable attention to various aspects of lot-sizing for single-stage and multi-stage systems in material requirements planning (MRP). Numerous models have been developed and tested on problems with finite horizons and deterministic time-varying demand. A real production system is, however, so complex that no model can capture all the elements under consideration. Instead of the static horizon commonly assumed by researchers, real planning is usually carried out on rolling horizons with different lengths. Such performance measures may include minimization of the total cost, inventory level, and schedule instability caused by rolling horizon and the number of setups. For large and complex product structures, the conventional approach is to apply a single-stage lot-sizing rule once to every stage of the multi-stage system. The Wagner–Whitin (WW) algorithm does provide a solution to this problem, but the length of the necessary calculations precludes its use in practice. As a result, many researchers have proposed numerous other lot-sizing procedures. The purpose of this paper is to examine the performance of various multi-stage lot-sizing procedures under rolling horizon environment using simulation. Cost and schedule instability were used as the performance criteria and product structure, demand variabillity, cost structure and planning horizon were considered as independent variables. It has been shown that Silver–Meal (SM) procedure outperforms other procedures such as WW and incremental approach (ICA) in most of the cases. The performance of ICA and SM are better than their respective counterparts Gaither’s rule (GA) and modified Silver-Meal (MSM). The results also indicate that there are differences in the performance of various lot-sizing procedures when applied to two different product structures.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 17 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.