@Article{cmes.2026.077356,
AUTHOR = {Toa Saito, Kantawatchr Chaiprabha, Kosuke Takano, Gridsada Phanomchoeng, Ratchatin Chancharoen},
TITLE = {A Deep Learning- and AI-Enhanced Telecentric Vision Framework for Automated Imaging-to-CAD Reconstruction},
JOURNAL = {Computer Modeling in Engineering \& Sciences},
VOLUME = {146},
YEAR = {2026},
NUMBER = {3},
PAGES = {--},
URL = {http://www.techscience.com/CMES/v146n3/66803},
ISSN = {1526-1506},
ABSTRACT = {This paper presents an automated imaging-to-CAD reconstruction system that combines telecentric vision and deep learning for high-accuracy digital reconstruction of printed circuit boards (PCBs). The framework integrates a telecentric camera with a Cartesian scanning platform to capture distortion-free, high-resolution PCB images, which are stitched into a single orthographic composite. A YOLO-based detection model, trained on a dataset of 270 PCB images across 23 component classes with data augmentation, identifies and localizes electronic components with a mean average precision of 0.932. Detected components are automatically matched to corresponding 3D CAD models from a part library and assembled within a Fusion 360 environment, producing a 3D digital replica. Experimental results show a similarity score of 0.894 and dimensional deviations below 2%, outperforming both SensoPart image measurement and manual vernier methods. The proposed approach bridges optical metrology and CAD automation, providing a scalable solution for AI-assisted reverse engineering, digital archiving, and intelligent manufacturing.},
DOI = {10.32604/cmes.2026.077356}
}