Deep Learning-based Machine Vision for Casting Defect Detection: Data-Driven Quality Assurance in Casting Manufacturing

The incorporation of transfer learning utilizing the Mask-RCNN architecture for defect identification has significantly advanced the pursuit of improved quality assurance in casting manufacturing. This paper offers a thorough investigation of a deep learning-powered machine vision system developed to transform quality control in the casting industry. The method we used has resulted in an impressive mean Average Precision (mAP) score of 99.4%, indicating the outstanding accuracy and efficiency of the model in detecting and classifying defects in casting. After conducting thorough testing, the model continuously demonstrated high precision and recall rates, accurately differentiating between occurrences of “cast_defect” and “cast_normal”. Furthermore, throughout individual testing situations, the model demonstrated a strong detection rate of 98%, emphasizing its consistent ability to accurately identify both normal and defective casts within single images. This study highlights the crucial importance of transfer learning in the Mask-RCNN architecture, providing a novel and data-oriented approach for identifying defects. The achieved high mAP value and consistent, high-performance levels confirm the ability of our model to greatly enhance quality assurance procedures in casting manufacturing, ensuring superior product quality and operational efficiency. This work signifies a significant advancement in utilizing cutting-edge technology to guarantee the accurate identification of defects and improvement of quality in the complex field of casting manufacture.