Visual inspection of perceived quality using machine learning

The article "Contrôle visuel de la qualité perçue par apprentissage automatique" proposes an automated system for quality control in industrial settings to address the subjectivity and variability of human inspections. It presents a machine learning approach using YOLOv5 for defect detection and XGBoost for severity estimation, utilizing a new dataset of PVC plates with various defects. The results show high accuracy in detecting defects and estimating their severity, with a mean Average Precision (mAP) of 0.99. Practical tests with human operators confirm the model's robustness and effectiveness. The study highlights the potential of machine learning to improve the efficiency and reliability of quality control processes in industrial environments.

In order to assure the quality of industrial products, it is necessary to implement perceived quality checks in the manufactured parts. However, this task is complex because of its time-consuming and hazardous aspects, due to the subjectivity in inspection of different operators. We propose an approach aiming to detect, classify and evaluate the defects in perceived quality to help the operators in this task. To evaluate our methodology, we have created a new use case. Our approch is evaluated based on two criteria, the automatic detection of defects and the estimation of their criticality.