Published Paper Details:

This paper presents the development and assessment of an automated optical Quality Control (QC) system to optimize defect detection and dimension measurement in metal casting operations, merging high-resolution imaging, motorized positioning mechanisms, and machine learning algorithms to economize time and resources traditionally required in QC processes. Utilizing an optical method for coordinate measurement and a fine-tuned DETR model for defect detection, the system demonstrates a promising reproducibility error of 0.005mm in dimension measurement and a moderate mean average precision metric (mAP-50) score of 20 in surface defect detection, pinpointing a need for further model optimizations. The paper suggests enlarging the training dataset and exploring alternative computational models or pre-processing techniques to enhance system performance, thereby contributing a foundational framework for advancing automated QC systems in the metal casting industry towards a more accurate and resource-efficient paradigm.