Optimization of Sourcing Decisions in Supply Chains

Abstract

Sourcing decisions in supply chains traditionally are made solely based on cost considerations. Typically sourcing models or decision-making procedures are either ad hoc or proprietary. In this dissertation a meta-heuristic-based optimization approach is developed and applied to sourcing decisions in several different apparel supply chains. To solve this kind of a large combinatorial problem, two issues need to be addressed: the solution methodology and the performance measure to be used. We develop a simulation-optimization methodology based on genetic algorithms (GA). Deficiencies of some of the current supply chain performance measures are addressed and a new performance measure, GMROILS, is developed. The GA with complex evaluations is used to determine optimal sourcing decisions for a seasonal item in an apparel supply chain with GMROILS as the performance measure. The Sourcing Simulator® is used for the simulation. Results show that the proposed methodology produces very good results. Based on the success of the proposed method, a large experimental design is carried out. Parameters in the experimental design include number of SKUs, SKU mix error, lead-time, seasonality pattern, min. order quantity per SKU, demand volume error. Multivariate regression, ANOVA and neural network analysis are performed on the results and we determine that best GMROILS is achieved generally when number of reorders is maximized. Next, we develop a constrained GA with complex evaluations. First, the constrained GA is used with the same experimental design using GMROI as the main performance measure with a 95% service level constraint. Results from the experimentation are in agreement with the previous results, i.e. the leanest and most flexible and responsive supply chain yields the best performance. We also conclude that GMROILS is an appropriate performance measure to be used in these cases. The constrained GA with complex evaluations was applied to a industry-supplied case study. We look at the sourcing of workpants to large, discount retail chain. The problem is to find the best inventory policy for the supplier of the workpants in order to provide a 95% service level to the retailer while maximizing GMROI (the objective defined by the supplier). We use the constrained GA with complex evaluations to find the best combination of presentation stock levels and target week of supply for the 420 SKUs so as to maximize GMROI while keeping 95% service level. We compare results from the optimizer to some classical alternatives including the existing policy. In the last part of the dissertation, a multiobjective GA with complex evaluations is developed. Multiobjective GA is used to maximize GMROI and SL at the same time for the seasonal item scenario. A pareto-optimal frontier is generated, which can be used by the decision-maker to make better and more robust decisions. In conclusion, the GA approach with complex evaluations performed successfully, as well as the new performance measure, GMROILS. Insights into optimal sourcing strategies are obtained. The optimization methodology is believed to have great potential for providing good solutions to a larger class of supply chain and logistics problems.

Description

Keywords

non-dominating sorting, NSGA-II, pareto optimality, multi-objective genetic algorithms, genetic algorithms, supply chain optimization, optimal sourcing decisions, simulation-optimization, neural networks, multivariate regression analysis, constrained genetic algorithms, Bean's adaptive penalty method, reordering, replenishment strategy, apparel supply chain, sourcing simulator, mex

Citation

Degree

PhD

Discipline

Industrial Engineering

Collections