A Data-Driven, Event-Based Dynamic Rescheduling Framework for Project-Type HVAC Manufacturing Under Supplier Uncertainty

Project-type production in the Heating, Ventilation, and Air Conditioning (HVAC) sector is highly vulnerable to scheduling disruptions caused by high product variety, complex routing structures, and uncertainty in the delivery of critical externally sourced components. In the investigated manufacturing environment, supplier-related delivery variability, rather than internal capacity limitations, constitutes the dominant source of schedule instability. This study develops a data-driven, event-based dynamic rescheduling framework that integrates supplier delay characterization with a Mixed-Integer Linear Programming (MILP) model. Historical procurement records obtained from the enterprise resource planning system were analyzed using K-Means clustering, identifying four representative delay classes of 5, 26, 51, and 127 days. These empirically derived patterns were incorporated into the rescheduling logic. The framework is activated at the disruption information time, t_info, when completed and ongoing operations are fixed to preserve schedule feasibility. A rolling-horizon mechanism freezes the subsequent five-day production window while reoptimizing the remaining horizon. Delays exceeding 40 days trigger an escalation rule that enforces the fastest available logistics mode. The multi-objective formulation simultaneously minimizes service-level agreement penalties, logistics expenditures, and operational energy costs. Implemented in Python 3.9 and solved with Gurobi 11.0, the model attained a 0.00% optimality gap within 1.2-2.4 s across the tested instances. Results show that early disruption information limits logistics costs to approximately 2,397 USD, whereas delayed information increases them to approximately 2,723 USD. The proposed framework provides a computationally efficient and transferable decision-support mechanism for coordinating production rescheduling and logistics expediting under supplier uncertainty, while indicating that expediting is economically justified when its marginal cost per recovered day remains below the corresponding dynamic delay penalty.