Automated dimensional inspection of automotive components using computer vision through YOLO

This paper presents the development and deployment of an automated visual inspection system using artificial intelligence and computer vision based on the YOLOv8 algorithm, designed to replace conventional aluminum jigs for the dimensional verification of extruded tubes in a Brazilian automotive parts factory. The research followed the Design Science Research (DSR) methodology, including six stages: problem identification, objective definition, artifact development, demonstration, evaluation, and dissemination. A dataset of 800 labeled images—500 compliant and 300 non-compliant—was captured under varying conditions and annotated using Roboflow. The model was trained in a Python environment using Ultralytics’ YOLOv8 with image resolution of 640×640 pixels, over multiple configurations up to 376 epochs. The best configuration (150 epochs) achieved precision of 0.97, recall of 0.969, [email protected] of 0.9917, and [email protected]:0.95 of 0.9397, with real-time inference speed under 25 milliseconds per image. The system proved effective in detecting missing, deformed, or misaligned connectors, and accurately classifying geometric compliance. Validation by the industrial director and production manager confirmed its readiness for deployment, emphasizing gains in agility, digital traceability, and cost savings. The replacement of physical templates by AI-driven inspection systems represents a significant step toward Industry 4.0 integration, enabling increased flexibility, standardization, and sustainability in production processes. Future work will focus on augmenting dataset diversity to enhance the model’s robustness against geometric and lighting variations, improving its generalization for broader industrial applications.