This study proposes an integrated production and maintenance planning framework that combines machine learning-based failure prediction with mathematical optimization. Sensor data collected from an industrial system are used to develop a machine learning model that estimates failure probabilities and generates risk indicators reflecting machine health conditions. Unlike conventional approaches that rely on deterministic assumptions, the proposed method transforms predictive outputs into probabilistic risk measures and integrates them into a mixed-integer linear programming (MILP) model. This enables dynamic and condition-based decision-making under uncertainty. In addition, machine health-dependent processing times are incorporated into the model to capture performance degradation effects. The optimization framework simultaneously determines production sequencing and maintenance decisions by balancing makespan, maintenance cost, and failure risk. To evaluate the effectiveness of the proposed approach, two benchmark strategies are considered: reactive maintenance and periodic preventive maintenance. The results show that the proposed model reduces makespan by up to 12% and total cost by approximately 30% compared to reactive maintenance, while also achieving significant improvements over periodic maintenance with a 50% reduction in maintenance interventions. The findings demonstrate that integrating machine learning outputs into optimization models provides a more efficient, adaptive, and reliable production planning framework. The proposed approach offers a practical solution for Industry 4.0 applications, where real-time data-driven decision-making is essential.

Keywords

Predictive Maintenance, Remaining Useful Life, Machine Learning, Integrated Production and Maintenance Planning, Mixed Integer Linear Programming