Data-Driven Multi-Objective Predictive Maintenance Optimization: Application to Bushings in Fiberglass Manufacturing

Poor maintenance practices in fiberglass manufacturing cause downtimes, quality defects, and increased costs. This study proposes a predictive maintenance (PdM) optimization framework combining a data-driven system, employing a convolutional neural network (CNN) model to predict bushing operational efficiency (OE), and a model-based system that operates a multi-objective optimization approach for scheduling maintenance. The data-driven system identifies critical bushings by predicting OE zones, while the model-based system prioritizes tasks based on criticality levels, minimizing maintenance completion time and servicing costs hierarchically. Results demonstrate criticality-based scheduling where higher-priority bushings are serviced earlier, while lower-priority bushings face longer waiting times. Operator utilization is optimized with balanced task allocation and sequential execution, ensuring efficient resource use and minimized downtime. The integrated framework improves operational efficiency, reduces delays, and addresses urgent tasks, which offers a robust solution for predictive maintenance in fiberglass production.