Comparison of Deep Learning Algorithms on Defect Detection on Metal Laptop Cases

Defect detection is crucial in industrial quality control. There are innumerable methods available for manual inspection, but these methods are time and resource consuming. We proposed an automatic defect detection method on metal laptop cases using deep learning algorithms. In this study, we used a 6-DOF robotic arm to collect the defect dataset and these defects are learned by three different methods. In this paper, we compared renowned detection methods YOLOv3, Faster R-CNN, and SSD over traditional defect detection methods. These methods outperformed incompetent conventional methods. The accuracy of Faster R-CNN had the best result compared to SSD and YOLOv3 on the defect dataset. Faster R-CNN had a recall rate of 80% and precision of 80%, whereas YOLOv3 had a recall rate of 73% and a mean average precision of 83%.