AI-Driven Real-Time Defect Detection for Paper Production: A Scalable Smart Manufacturing Approach

Industrial paper production poses unique quality control challenges due to high-speed operations and frequent surface defects. Manual inspection methods are insufficient for real-time monitoring, resulting in increased waste and operational inefficiencies. This study proposes an AI-driven defect detection system that combines deep learning-based computer vision with high-speed imaging and edge AI deployment. Lever- aging YOLOv7 for real-time detection and SSD for small-defect sensitivity, the system achieves 91–98% accuracy across various defect types. Explainable AI techniques such as Grad-CAM enhance transparency and support root-cause analysis. Field implementation led to a 9% reduction in unplanned downtimes, 12% lower material waste, and 1.5 million kWh annual energy savings. The results demonstrate a scalable, high-performance solution for automated quality assurance in industrial manufacturing. Future work will explore model generalization via federated learning and predictive maintenance integration.