Akademik Veri Yönetim Sistemi
ARABIAN JOURNAL FOR SCIENCE AND ENGINEERING, 2025 (SCI-Expanded, Scopus)
The detection of defects emerging in manufacturing processes with high accuracy is crucial, and accurate classification of these defects will optimize the quality control processes and overall efficiency. In this study, a real-time prototype cable production line was first developed to create a dataset consisting of the defect types Fisheye, Dent, and Slit. Afterward, a deep learning (DL)-based detection and classification algorithm was optimized for real-time detection and classification of defects occurring on the surfaces of energy cables during the production process. The proposed approach stands out by incorporating a genetic algorithm (GA)-based hyperparameter optimization of the You Only Look Once (YOLOv11n) object detection algorithm. The performance of the optimized YOLOv11n was compared to that of the basic YOLOv11n and other YOLOv11 algorithms in terms of precision, recall, mAP@50, mAP@(50-95), F1-score, GFLOPS, and inference time metrics. The results obtained show that optimized YOLOv11n was able to provide better recall, mAP@50, F1-score, and inference time values. Optimized YOLOv11n produced the highest accuracy rates with the recall value of 91%, mAP@50 value of 94.7%, F1-score value of 90.9%, and inference time value of 1.2 ms. These values correspond to approximately a 3% increase in recall value, a 0.5% increase in mAP@50 value, a 1% increase in F1-score value, and a 0.1 ms decrease in inference time value. Consequently, it can be stated that GA-based hyperparameter optimization enhances the accuracy of DL-based defect detection.