Application Of Machine Learning For Crack Detection On Concrete Structures Using Cnn Architecture

ABSTRACT Concrete structures and the associated cracking represent a persistent challenge within the fields of construction and civil engineering, a dilemma that has persisted for decades. Generally, cracks arise from the expansion and contraction of materials, which can lead to various forms of damage within buildings. Engineers typically address these damages and irregularities through manual inspections or by employing predictive models developed using the machine learning techniques, thereby facilitating a comprehensive assessment of the structural health of the edifices. This research aims to implement tools such as the VGG 16 network model alongside other machine learning methodologies to identify the cracks in concrete structures. We propose a model that integrates Convolution Neural Networks (CNN) with a VGG-based architecture. For image processing and segmentation, we have employed the gradient boosting algorithm. The datasets utilized in this study were sourced from the Kaggle platform, and the Hugging Face Transformers library was leveraged for implementation. To assess the performance of the developed models, metrics including precision, accuracy, recall, and F1 score were employed. The accuracy achieved by our model can be benchmarked against the accuracy rates of Vision Transformers (ViT). The proposed method demonstrates significant validation accuracy rates, positioning it as a highly innovative and novel machine learning model capable of effectively predicting cracks in concrete structures.