Detection of glass defects with computer vision and artificial intelligence

In today's world, the increasing global population and corresponding rise in consumption demands have made it essential not only to ensure product quality during manufacturing but also to maintain reliability throughout the product's lifecycle, including its use by end consumers. Detecting potential defects before or during consumer use is crucial for enhancing customer satisfaction and preventing possible safety risks. In this context, the integration of artificial intelligence and computer vision-based systems into quality assurance processes significantly reduces human error, while improving both product safety and user experience. This study aims to detect surface defects such as cracks, scratches, and fractures on glass surfaces after the production stage, focusing on the products already in the hands of end users. Glass is a widely used material in various industries including automotive, construction, home appliances, cosmetics, and packaging, where visual and structural integrity is of critical importance. Therefore, accurately identifying such defects at the consumer level plays a vital role in maintaining product reliability and brand reputation. A dataset consisting of 4,975 images (4,065 for training, 463 for validation, and 447 for testing) was used in this study. The dataset was employed to train deep learning-based object detection models, specifically YOLOv5, YOLOv8, and YOLOv11, for identifying defects on glass surfaces. The models were evaluated using accuracy, precision, F1 score, and mAP50 metrics. The results demonstrated that the YOLOv11 model outperformed the others in terms of accuracy, F1 score, and mAP50, while YOLOv8 showed better performance in terms of precision. Overall, YOLOv5 exhibited lower performance compared to the other two models. These findings indicate that YOLO architectures can be effectively utilized for detecting glass defects at the consumer level and can offer significant contributions toward automating post-sale quality assurance processes.