<![CDATA[Conveyor belts play a critical role in the cement industry as continuous material transportation systems. Undetected damage such as tear, hole, patch work, and puncture can lead to production downtime, increased maintenance costs, and safety risks. Conventional manual visual inspections are limited in terms of time efficiency, accuracy, and operator subjectivity. This study aims to develop a real-time conveyor belt damage detection system based on deep learning that can be deployed on resource-constrained edge devices. The proposed system employs a YOLOv11 model trained on a combined dataset consisting of field inspection data, public datasets, and automatically collected images, resulting in more than 3,000 augmented training samples. The trained model is converted into TensorFlow Lite format and compiled for Google Coral EdgeTPU to enable efficient deployment on a Raspberry Pi 5 integrated with an RTSP-based IP camera. Training results demonstrate promising performance, achieving a precision of 0.742, recall of 0.730, mAP@50 of 0.764, and mAP@50–95 of 0.553. Implementation in an industrial environment shows stable real-time inference with low latency and reliable detection performance, particularly for the patch work class. The developed system enables continuous and automated conveyor belt condition monitoring, offering potential benefits in reducing downtime and improving maintenance efficiency in industrial applications.]]>
Copyrights © 2026
