An optimization-based hybrid decision support framework for supplier selection and order allocation

restricted_to_institution
Restricted (Penn State Only)
Author:
Chuang, Shu Wei
Graduate Program:
Industrial Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
June 09, 2026
Committee Members:
Jingjing Li, Major Field Member
Guhaprasanna Manogharan, Major Field Member
Robert Voigt, Chair & Dissertation Advisor
Chetan Chimate, Special Member
Ling Rothrock, Program Head/Chair
V Guide, Outside Unit & Field Member
Keywords:
Supplier selection and order allocation
Optimal near-neighbor solution
Mixed-integer linear programming
Analytic network process
Fuzzy technique for order preference by similarity to ideal solution
Remanufacturing
Abstract:
Supplier selection and order allocation (SSOA) influence purchasing costs, operational efficiency, and resilience in supply chain management. Strategically sourcing components and sub-assemblies is critical to reducing overall costs and improving resilience. Large organizations typically maintain multiple supplier options for proprietary products and multiple warehouses across various geographic locations. Companies need to relocate non-moving inventory between warehouses when it is economically viable to ensure internal inventory is utilized before procuring from external suppliers. Therefore, companies must identify qualified suppliers and allocate the appropriate order quantities to balance multiple criteria under operational constraints. However, most existing approaches identify only an optimal solution, limiting decision flexibility when unmodeled criteria and decision-maker preferences are considered. To address this limitation, this dissertation develops an optimization-based hybrid decision support framework that integrates mathematical optimization with human decision-making for SSOA problems. To demonstrate the applicability of the proposed framework, this dissertation focuses on the remanufacturing supply chain. Remanufacturing restores failed or used products to like-new condition through a series of industrial processes. Remanufacturing companies have specialized facilities, suppliers, and customers worldwide. Selecting the best locations for sourcing subsystem remanufacturing and final product assembly is critical for reducing overall costs. Compared with traditional manufacturing, remanufacturing introduces greater uncertainty, longer lead times, and more complex sourcing decisions, making SSOA more challenging. Therefore, the proposed framework is evaluated on a remanufacturing dataset validated by subject matter experts (SMEs). The proposed framework includes two stages. In the first stage, it integrates a robust method called Optimal Near-Neighbor Solutions (ONNS) with mixed-integer linear programming (MILP). The model is formulated to minimize the total cost, including purchasing, transportation, tariff, and fixed order costs, and is subject to demand and supplier capacity constraints. Its performance is evaluated using a remanufacturing dataset with 23 worldwide supplier locations, 12 products, various supplier capacities, and product demand. After obtaining the baseline cost-optimal solution, an ONNS strategy using heuristic processes is applied to generate a set of solutions that remain close to the optimum. In the second stage, multi-criteria decision-making (MCDM) is employed. The weights of the criteria are determined by SMEs using the Analytic Network Process (ANP) to account for interdependencies among criteria. To reduce linguistic ambiguity, the ONNSs are rated by using the fuzzy Technique for Order Preference by Similarity to Ideal Solution (TOPSIS). A web-based user interface (UI) has been developed to visualize the evaluation process and support interactive decision-making. For sensitivity and robustness analyses of the ONNS framework, this dissertation employs a two-layer design of experiments. The outer layer uses a three-factor, two-level design to generate eight structural scenarios based on product value level, pricing structure, and buyer location. The corresponding datasets for these scenarios are validated by SMEs in the remanufacturing industry. The inner layer applies a three-factor, three-level perturbation design, in which tariff, unit price, and supplier capacity are varied by -20%, baseline, and +20%. 216 perturbation cases are evaluated. Relative cost gaps are used to compare ONNSs with the baseline optimal solutions, and Hamming distances are used to evaluate diversity between ONNSs and the optimal supplier patterns. Based on the ONNSs obtained in the first stage, the final rankings for the eight scenarios are determined based on all criteria, as shown in the UI. The results show that ONNSs are generated across the structural scenarios and perturbation cases. The top four ONNSs are generally robust and remain close to the cost-optimal solution with different supplier selection patterns. The structural diversity analysis confirms that ONNSs provide practical flexibility without cognitive overload. The perturbation analysis further shows that the ONNS framework remains robust under uncertainty. Additionally, the cost-optimal solutions are not necessarily the overall best solutions when trade-offs, such as quality and on-time delivery, are considered across all criteria. The hybrid decision support framework enables decision-makers to leverage their experience and knowledge to systematically select acceptable solutions from near-neighbor sets. Decision-makers can dynamically evaluate and choose optimal solutions by using modeled and unmodeled criteria. By integrating human-centered decision-making with mathematical optimization, this dissertation advances rigid selection models to an adaptive decision process. The framework improves flexible, robust, and dynamic decision-making, supporting operational efficiency in the remanufacturing industry and other industries where complex sourcing decisions must be made under uncertainty.