Spalling Damage Detection in Irrigation Structures using Convolutional Neural Networks

Abstract

Irrigation structures play a crucial role in water resource management but often face deterioration of their concrete components, particularly concrete spalling. This type of damage frequently occurs in hard-to-access areas and significantly affects the strength and service life of structures that are subjected to water pressure and continuously changing environmental conditions. This study proposes a method for detecting concrete spalling in irrigation structures by developing an artificial intelligence system using a Convolutional Neural Network (CNN). The CNN model was trained with a dataset combining spalling damage data collected from multiple sources and a newly developed spalling dataset. The approach involves capturing aerial images of irrigation structures using an Unmanned Aerial Vehicle (UAV) and creating three-dimensional models (3D models) models of the structures. Simultaneously, the spalling dataset was developed and used for CNN model training. In the final stage, the trained CNN model was applied to detect spalling damage on the concrete surfaces of the 3D models. The results demonstrate that the CNN model accurately identified and localized instances of concrete spalling, enabling rapid damage detection and structural assessment of irrigation structures. Four CNN architectures—VGG16, VGG19, ResNet101, and ResNet152—were evaluated and compared. Among these, VGG16 achieved the highest accuracy of 86.40%, making it the most suitable model for practical application in irrigation structure inspections.