Defect Inspection For Casting Products Using Convolutional Neural Networks

Abstract

During and after completing conventional manufacturing processes, quality control is conducted with a view to confirming the integrity of the products. Vision-based inspection systems have been extensively adopted in numerous industries associated with the smart factory concept due to improvements in precise and fast inspection. Computer vision is also known for its drastic reduction of the human inspection cost. This paper focuses primarily on the development of four models for the inspection of casting products that is supported by convolutional neural networks and transfer learning techniques in deep learning, thereby enabling the classification of products with or without defects. The performance of the proposed algorithm for inspecting casting products has been validated using more than 700 images of casting products, resulting in more than 98% prediction accuracy and instant prediction time.