چکیده مقاله
Visual inspection of welds in many production lines faces limitations such as being time consuming, dependence on inspector expertise, and sensitivity to variations in illumination and viewing angle In this study, a deep learning based machine vision system was developed for automatic weld quality assessment and defect identification using the YOLOV8 object detection network The dataset consisted of weld images annotated in the YOLO format, and the model was trained using pretrained YOLOv8m weights with an input resolution of 640×640 To improve generalization, in addition to the baseline dataset, an extended version was created by incorporating supplementary images and applying advanced data augmentation techniques including HSV augmentation, Mosaic, and MixUp Performance was evaluated on an independent test set using precision, recall, F1 score, and mAP@0 5 The results indicate that the extended version achieves a significant improvement over the baseline, increasing mAP@0 5 from 66% to 75% and recall from 61% to 74%, while maintaining inference latency on the order of tens of milliseconds Therefore, the proposed method is suitable for real time weld inspection scenarios
کلیدواژهها
نویسندگان
Amir Bizheh
M Sc student in Materials and Metallurgical Engineering Welding specialization , Amirkabir University of Technology
Seyed Mohammad Mahdi Zamani
Postdoctoral researcher of Faculty of Mechanical Engineering, Amirkabir University of Technology
Islam Ranjbarnoodeh
Faculty member of Faculty of Materials and Metallurgical Engineering, Amirkabir University of Technology
شیوه ارجاع
Bizheh, Amir and Zamani, Seyed Mohammad Mahdi and Ranjbarnoodeh, Islam,1404,Automatic Weld Defect Detection Using YOLOV8 for Real-Time Inspection,8th International Conference on Welding and Non Destructive Testing & 26th National Conference on Welding & Inspection & 15th National Conference on NDT & 4th National Conference on Additive Manufacturing,Tehran
ارائهشده در
مجموعه مقالات هشتمین کنفرانس بین المللی جوشکاری و آزمایش های غیرمخرب، بیست و ششمین کنفرانس ملی جوش و بازرسی، پانزدهمین کنفرانس ملی آزمایش های غیرمخرب و چهارمین کنفرانس ملی ساخت افزایشی20 بهمن 1404 · تهران