Multi-objective Robust Optimization for Remanufacturing Systems

Supervisor: Dr. Ahmad Makui

Overview

Growing population, shorter product life cycles, and tightening environmental regulation have made remanufacturing — returning used products to original quality standards rather than discarding or merely recycling them — economically and ecologically essential. Unlike simple recycling (material recovery), remanufacturing recovers value-added functionality, significantly reducing energy consumption and production costs.

This thesis, completed at Iran University of Science and Technology under Dr. Ahmad Makui, develops a mathematical model for planning hybrid manufacturing–remanufacturing operations under the three sources of uncertainty that make the problem genuinely hard.

Problem and Uncertainty

A hybrid system produces both new products and remanufactured ones using the same facilities, competing for the same capacity. Three uncertainty dimensions are modelled simultaneously:

1. Return quality — the condition of returned products is unknown until inspection; poor-quality returns require more reprocessing time and may be non-remanufacturable.
2. Processing and reprocessing times — variability in disassembly, refurbishment, and assembly operations.
3. Demand — market demand for both new and remanufactured products fluctuates across the planning horizon.

Ignoring any one of these dimensions produces plans that frequently become infeasible or sub-optimal in practice.

Model

The thesis formulates a multi-period, multi-objective robust MILP with a budgeted uncertainty set. The two competing objectives are:

• Minimize total cost (production, reprocessing, inventory, holding, and disposal costs)
• Minimize environmental impact (emissions from manufacturing and reprocessing operations)

Because the problem is NP-hard and exact solvers (GAMS/Gurobi, ε-constraint method) cannot solve large instances in reasonable time, the thesis develops and benchmarks a metaheuristic solution:

Algorithms

Sixteen benchmark test instances of varying size (small to large) were solved using three methods and compared on three multi-objective performance metrics:

Performance was evaluated using:

• NPS (Number of Pareto Solutions): richness of the Pareto frontier
• CM (Coverage Metric): dominance of one algorithm’s frontier over another
• DS (Diversity Spread): spread of solutions across the objective space

NSGA-II outperformed MOPSO across all three metrics for small and medium instances, while offering acceptable performance at larger scales. The ε-constraint method provided the exact Pareto frontier for small instances, confirming algorithm quality.

Case Study

The validated model was applied to a real Iranian manufacturing company, providing production and remanufacturing plans across multiple periods. The robust plan achieved a 14% reduction in total cost and a 10% reduction in environmental impact relative to the deterministic baseline. Sensitivity analysis on key model parameters generated managerial recommendations for supply chain planners under changing uncertainty levels.

Managerial Insight

The robust model allows operations managers to explicitly trade off cost and environmental performance against schedule reliability. Modest increases in Γ significantly improve feasibility under disruption with small degradation in nominal objectives — a favorable exchange in industries where product return quality is genuinely unpredictable.