Computer Vision Quality Control
Client: Automotive Manufacturer
The project
Deployment of an automated computer vision quality control system on an automotive production line. Real-time detection of 12 types of defects (scratches, dents, faulty welds) at 60 parts per minute.
Architecture & implementation
Training of an object detection model on a proprietary dataset of 120,000 annotated images, using data augmentation and transfer learning from YOLOv8. Precision achieved: 98.4% (mAP@0.5).
Monitoring & adoption
Edge computing deployment on NVIDIA Jetson GPUs installed directly on the production line. Inference latency under 30 ms, and no data ever leaves the plant (sovereignty and confidentiality).
Results & production rollout
Integration with the existing MES (Manufacturing Execution System): automatic defect alerts, line stoppage when needed and real-time reporting. Result: 42% fewer defects and 15% higher productivity.
Screenshots
