Automated and sustainable coating assessment for H-shaped steel using DNN

Abstract

This study presents the development of a fully automated recognition system leveraging Deep Neural Networks (DNN) to identify and evaluate coating patterns and structural attributes of H-shaped steel components. The system is designed not only to improve the efficiency and accuracy of inspection processes but also to contribute to sustainable construction practices by enabling early defect detection, reducing material waste, and optimizing lifecycle maintenance. The computational pipeline consists of eight key stages: (1) data input, (2) detection area computation, (3) categorization of residual regions, (4) calculation of potential parallel distances, (5) optimal pattern recognition, (6) model-object comparison, (7) output generation for Matplotlib visualization, and (8) 3D plot construction from output coordinates. This pipeline allows for precise, non-destructive evaluation and facilitates real-time integration into automated fabrication or quality control workflows. The system was trained and validated using a robust dataset comprising 115 standard-type and 99 special-type H-shaped steel samples, representing a broad spectrum of commercially used profiles. The proposed model achieved 100% accuracy in identifying critical geometric features such as width, base plate thickness, and wing plate thickness. The overall recognition accuracy exceeded 99.12%, with an average real-world application accuracy of 99.73%, indicating excellent performance and reliability for industrial deployment. By automating the recognition process and enhancing the reliability of coating and structural assessments, this research supports the development of smart, resource-efficient infrastructure, aligning with key goals in sustainable manufacturing and resilient civil engineering.