Parametric Study of Inspecting Surface Defects in Investment Casting

Abstract

Metal defects detection has always been an essential task for the majority of various industries, moreover, it is the core element in the metal inspection too. This research paper explores the effectiveness of different deep learning algorithms for surface-defect detection in investment casting using the Inspection 4.0 approach. The study compared the performance of four popular deep learning algorithms, Fast R-CNN, Faster R-CNN, ResNet, and YOLO, using the accuracy metric as a performance evaluation measure. The results show that ResNet achieved the highest accuracy rate of 95.89%, followed by Faster R-CNN with 90.23%, Fast R-CNN with 89.21%, and YOLO with 86.43%. The findings of this research demonstrate that ResNet and Faster R-CNN are effective deep-learning algorithms for automated surface-defect detection in investment casting. On the other hand, Fast R-CNN and YOLO exhibited lower accuracy rates.The outcomes of this study provide valuable insights into the effectiveness of deep learning algorithms for surface-defect detection in investment casting. The high accuracy rate achieved by ResNet and Faster R-CNN can guide the development of automated inspection systems for investment casting in various industries such as aerospace, automotive, and medical.