Simulation-based Variable Neighborhood Search for Optimizing Skill Assignments and Priorities in Service Queues

Abstract

This paper studies optimizing cross-training policies and static priorities in a multi-skill, multi-server repair facility with one inventory for ready-to-use spare parts. If available, the failed spare parts are immediately replaced with new ones from inventory. Otherwise, the spare parts are backordered with penalty costs. First, a Simulation-based Variable Neighborhood Search (VNS) framework is developed to optimize skill assignment with no static priorities. This study aims to minimize the repair facility’s total cost, including servers, training, holding, and backorder costs. The performance of our proposed framework is tested by comparing its results with optimal solutions for small-size instances obtained using brute-force optimization. Also, the study compares the performance of the proposed VNS algorithm to GA. The VNS-based framework obtains better results in 94.5% of the cases. The study also compares the convergence of both frameworks with the same number of iterations. The VNS has a better solution than GA in 74% of the cases with an average of 2.94%.

Second, This study optimizes the static priority rules while optimizing skill assignments. Two simulation-based frameworks are developed for optimizing skill assignments and priorities. First, a two-phase simulation-based Variable Neighborhood Search framework is developed. The first phase optimizes the skill assignments to different servers with no priorities, While the second stage optimizes the assignments of priorities to different tasks. The second framework is a one-phase simulation-based Variable Neighborhood Search approach for optimizing both priority classes and skill assignments of different tasks to servers. Both frameworks aim to optimize the system’s total cost, including server costs, skill costs, and holding and backorder costs. The cost savings of both frameworks are compared, and the results of the test experiments show that the one-phase framework produces better results in 86% of the test instances.

Date of Award	Dec 2022
Original language	American English
Supervisor	Andrei Sleptchenko (Supervisor)