An adaptive digital twin framework for flow shop scheduling and production planning with machine learning integration

Demand fluctuations, machine downtime, and execution plans’ uncertainties affect production planning and scheduling in flow shops; therefore, these factors must be considered during planning to prevent operational disruptions. Unlike existing static integrated planning and scheduling models or heuristic-based dynamic approaches, this work introduces a closed-loop Digital Twin (DT) framework that preserves the mathematical optimality of a Mixed-Integer Nonlinear Programming (MINLP) model while enabling real-time adaptability. Our primary innovation is the systematic and real-time integration of Support Vector Machine (SVM)-based demand forecasting and Association Rule Mining (ARM)-based proactive downtime prediction. We develop a data-driven DT that synchronises the MINLP with live shop-floor data. The framework integrates SVM for demand forecasting and ARM for proactive machine downtime prediction. Through event-driven re-optimisation, the DT continuously updates key parameters and triggers schedule revisions in response to operational deviations. Computational experiments demonstrate that the proposed approach reduces total operational costs by 13%, significantly decreases machine waiting times and unscheduled downtime, and improves schedule stability compared to static and data-driven baselines. The results underscore the unique value of closed-loop adaptive systems in enhancing resilience and efficiency. This study bridges the gap between traditional optimisation models and real-world execution to offer more responsive and robust production systems.