Simulation-based spot welding inspection on automotive chassis using YOLO-powered image processing

Authors : Adem Dilbaz, İlker Ali Ozkan

Pages : 170-180

Doi:10.18245/ijaet.1729908

View : 64 | Download : 99

Publication Date : 2025-09-30

Article Type : Research Paper

Abstract :This study examines a simulation-based testing platform designed to enhance the quality control processes of Resistance Spot Welding (RSW), a technology widely used in the automotive industry. A virtual testing environment was developed to eliminate the need for physical prototypes. The platform was assembled by placing ESP32-CAM-based virtual cameras on a vehicle chassis obtained from the RoboDK library within the simulation environment. A dataset of approximately 1,000 real RSW images from Kaggle was labeled using Roboflow and converted into a format compatible with YOLO(You Only Look Once) architecture. During image processing and object recognition, YOLOv3-s and YOLOv5-m models were utilized. The models’ classification performance was evaluated using metrics such as F1 score, precision, recall, mean average precision (mAP), and Confidence Score (CS). Both models required low hardware requirements; however, YOLOv5-m displayed overall superior performance. Notably, the YOLOv5-m model achieved higher confidence scores in detecting critical welding defects classified as Class 2 (explosion weld); an approximate increase of 8–9% was observed in experimental results, reaching a CS of around 0.58. In addition, the F1 score for Class 2 (explosion weld) improved by approximately 5–6%, reaching a value of around 0.85. This simulation-based method has made RSW quality control faster, more cost-effective, and reliable. Consequently, robotic welding systems can be thoroughly tested for accuracy and safety in a virtual environment before being integrated into the production line.
Keywords : otomotiv endüstrisi, derin öğrenme, direnç nokta kaynağı, görüntü işleme, RoboDK, YOLO

ORIGINAL ARTICLE URL