Using Pattern Recognition to Identify Design Errors on Pipe System Drawings

Manual quality checks of isometric pipe system drawings in the Engineering, Procurement and Construction (EPC) industry are time consuming, prone to human error, and require significant expertise. The high cost of design errors, combined with the complexity of these congested drawings, makes automation a critical challenge. To address this, we propose a deep learning-based approach to automate the detection of design errors in isometric drawings. Our solution utilizes an object detection framework trained on a diverse dataset from various EPC projects, incorporating expert engineering knowledge from McDermott. By identifying design mistakes represented by specific object assemblies, the model assists engineers in document verification, freeing them to focus on higher value tasks. The model is validated on real-world datasets and overcomes challenges such as small object detection, class imbalance, symbol variability, and high-resolution image processing. Techniques like tiling and data augmentation significantly improve detection accuracy. Our results demonstrate high precision in identifying design errors, showcasing the adaptability of the model for similar engineering diagram applications. This research advances automated drawing analysis and provides a scalable AI-driven solution to capture engineering expertise and integrate it into quality control workflows.