Automated Bounding Dimension Retrieval: A Machine Learning-Driven Framework for Engineering Drawing Interpretation

Accurately retrieving bounding dimensions from engineering drawings is critical for manufacturing processes, influencing machining feasibility, material selection, and cost estimation. This paper presents an innovative Automated Bounding Dimension Retrieval (ABDR) framework, integrating state-of-the-art object detection and Optical Character Recognition (OCR) technologies with a novel ratio-based analysis methodology. The proposed multi-stage pipeline systematically identifies primary, secondary, and tertiary dimensions across multiple views, addressing challenges such as view separation, bounding box detection, and text classification. The ABDR framework was evaluated on a dataset of engineering drawings adhering to ISO standards. The view detection module, implemented using YOLOv10n, achieved 98% mAP50, while the bounding box detection module using YOLOv11n reached 98.7% mAP50 and 97.5% mAP75. The text extraction module, employing EasyOCR, attained 83.55% accuracy and 95.87% normalized edit distance. The overall pipeline achieved 81% accuracy in dimension retrieval. By automating this traditionally manual and error-prone process, the ABDR framework provides a robust and scalable solution, offering significant advancements for technical drawing interpretation and manufacturing workflow automation.