Maziyar Khadivi

Topic

Optimization of parallel machine scheduling problems with resources and complex side constraints

Department of Mechanical Engineering

Date & location

• Thursday, August 27, 2026

• 9:00 A.M.

• Virtual Defence

Reviewers

Supervisory Committee

• Dr. Homayoun Najjaran, Department of Mechanical Engineering, University of Victoria (Supervisor)

• Dr. Keivan Ahmadi, Department of Mechanical Engineering, UVic (Member)

• Dr. Adel Guitouni, School of Business, UVic (Outside Member) 

External Examiner

• Dr. Abraham Punnen, Department of Mathematics, Simon Fraser University 

Chair of Oral Examination

• Dr. Marzieh Mosavazadeh, Department of Curriculum and Instruction, UVic

   

Abstract

The resource-constrained parallel machines scheduling problem (RPMSP) is common in the manufacturing and service sectors, requiring the optimal assignment of jobs to machines supervised by limited additional resources. This dissertation investigates how to efficiently solve the RPMSP considering manufacturing constraints such as job release times, due dates, machine eligibility, and resource transition limits per machine. Previous research has not addressed this combination of real-world industrial constraints, leaving limitations on practical and scalable scheduling algorithms for the RPMSP. 

To determine the most effective optimization methodology, this research first provides a comprehensive review of deep reinforcement learning (DRL) applications in machine scheduling. While state-of-the-art DRL approaches can match or exceed exact methods, heuristics, and metaheuristics in speed and solution quality for certain scheduling problems, our review reveals their current limitations. Specifically, compared to conventional techniques, DRL struggles with generalizability, navigating large state-action spaces, and adhering to complex operational constraints.

Given these limitations of DRL and supported by the RPMSP literature, this dissertation relies on the mathematical programming and metaheuristics to solve RPMSPs with operational constraints. We develop mixed-integer linear program ming (MILP) models, followed by the introduction of a novel biased random-key genetic algorithm (BRKGA). We demonstrate the scalability of BRKGA compared to the MILP model on synthetic datasets based on a real-world case study of a food processing plant. Furthermore, we demonstrate that BRKGA either outperforms or matches state-of-the-art metaheuristic and exact methods on benchmark datasets from the literature. This research provides a broadly adaptable, scalable, and license free scheduling software tool, enabling practitioners and researchers to tackle similar resource-constrained parallel machine scheduling problems effectively.