A Computer Vision Framework for Structural Analysis of Hand-Drawn Engineering Sketches.

Area of Science:

• Structural Engineering
• Computer Vision
• Computational Mechanics

Background:

• Manual conversion of engineering sketches to digital models is time-consuming and error-prone.
• Hand calculations for structural analysis are slow and prone to mistakes.
• Need for efficient tools to bridge the gap between conceptual design and analysis.

Purpose of the Study:

• To develop an autonomous framework for converting hand-drawn engineering sketches into analyzed structural models.
• To leverage computer vision for automated feature extraction and interpretation of structural diagrams.
• To integrate sketch analysis with structural analysis software for a complete workflow.

Main Methods:

• Utilized computer vision for object detection to identify raw features in sketches.
• Implemented a number-reading model to transcribe handwritten numerals.
• Employed feature association models to establish relationships between detected elements.
• Integrated the generated structural model with OpenSees for analysis.

Main Results:

• Object detection achieved 99.1% mean average precision.
• Number-reading model accuracy reached 99.0%.
• Feature association models demonstrated accuracies between 95.1% and 99.5%.
• The framework correctly analyzed 45.0% of images entirely and 55.0% partially.

Conclusions:

• The proposed framework successfully automates the conversion of engineering sketches to analyzed structural models.
• This technology offers significant potential for improving efficiency, safety, and sustainability in structural engineering.
• The framework shows promise for application to various structural types beyond beams, including trusses and frames.