Automation and Sustainability: Developing coating thickness detection algorithm for construction H-shaped steel components

Abstract

To enhance the longevity and sustainability of structural materials, H-shaped steel coated with specialized protective layers has emerged as a superior solution due to its high resistance to rust and fire. However, ensuring the optimal performance and lifecycle efficiency of such materials relies heavily on the precise, automated assessment of coating thickness—an area where traditional detection methods remain error-prone and inefficient. This study addresses the need for a sustainable, automated solution by developing an intelligent detection algorithm based on ResNet, a deep convolutional neural network architecture known for its high performance in visual recognition tasks. A comprehensive dataset of 23,900 annotated images was constructed under a statistically robust sampling strategy, ensuring a 95% confidence level with a 5% margin of error in a balanced 50–50 distribution. Of these, 19,100 images were used to train the CNN model, while 4,800 were reserved for rigorous validation and testing. The dataset was systematically categorized into three distinct coating conditions: uncoated, partially coated, and fully coated. Using a 5-fold cross-validation approach, the proposed model achieved a high accuracy rate of 93%, demonstrating its effectiveness in classifying coating levels. This advancement represents a critical step toward automated, scalable quality control in steel production and infrastructure maintenance. By reducing reliance on manual inspection and minimizing material waste or under-coating, the system contributes to sustainable construction practices and extends the service life of structural steel components.