Comparative Analysis of 2D-CAD Comparison with Custom-Trained Computer Vision Models

Abstract

In manufacturing environments, the ability to efficiently compare and analyze 2D Computer-Aided Design (CAD) drawings is critical for ensuring product quality, minimizing errors, and streamlining the design iteration process. Traditional manual comparison methods are time-intensive and prone to human errors, particularly when analyzing annotations, dimensions, and complex structural details. To address these challenges, this study presents a comparative analysis of state-of-the-art deep learning models—YOLOv8m, YOLOv8x, YOLO-NAS, and Faster R-CNN—for automated CAD design evaluation. The models were trained on a dataset of 5,000 CAD images, encompassing diverse mechanical components with varying complexities. The proposed system leverages object detection and Optical Character Recognition (OCR) techniques to extract and compare dimensions and notes with high precision. Experimental results demonstrate that YOLOv8m outperforms other models in terms of detection accuracy. The findings highlight the effectiveness of deep learning-based CAD comparison systems in reducing verification time and improving design evaluation accuracy. This study provides insights into the strengths and limitations of different computer vision models for CAD analysis, contributing to the advancement of automated design validation in manufacturing industries. This approach enhances the recognition accuracy of deep learning models, making dimension extraction and recognition more practical. The system achieves 90-95% accuracy in detecting dimensions and notes in the CAD designs, and 85-90% accuracy in text recognition. These findings highlight the effectiveness of AI-driven CAD comparison systems in reducing verification time and improving design evaluation accuracy, contributing to the advancement of automated design validation in manufacturing industries.