A Decomposition-Based MILP Framework for Short-Term Scheduling in High-Mix Cork Composite Manufacturing

Abstract

Contemporary manufacturing environments face a persistent tension between product customization and operational efficiency. As the industry shifts toward Make-to-Order models, traditional inventory buffers are reduced, making short-term production scheduling a critical determinant of performance. However, allocating finite resources across diverse production orders with sequence dependent setups constitutes a strongly NP-hard optimization problem. Industrial practice often relies on manual, experience-based heuristics that struggle to capture complex shop-floor inter dependencies, frequently resulting in schedules that fail at execution. This dissertation addresses the gap between academic theory and applied decision-support by designing a mathematical optimization model for a complex, high-mix manufacturing environment, specifically within the cork composite industry. A Mixed-Integer Linear Programming formulation is developed within the General Lot Sizing and Scheduling Problem framework. To ensure computational tractability for industrial-scale instances, a staged decomposition strategy spanning a multi-week horizon is introduced, alongside a two-pass architecture that decouples setup sequencing from production timing. Computational validation on real historical data confirms the framework's viability, demonstrating its ability to generate high-quality, feasible schedules within operational time limits. By translating tacit knowledge into a quantified mathematical framework, this work replaces intuition with an auditable decision-support structure. It establishes a reproducible workflow that exposes operational trade-offs, enabling data-driven execution on the shop floor.