Deep learning-based structural crack detection framework for structural health monitoring of reinforced concrete buildings

Abstract

Structural health monitoring (SHM) is an important activity in the safety and long life of reinforced concrete buildings and bridges. Structural cracks should be detected early to avoid further degeneration and reduce the expenses incurred on maintenance. Conventional approaches to crack inspection are mainly based on the manual visual evaluation that may turn out to be time-consuming, labor intensive, and prone to subjective error. New technologies in artificial intelligence and computer vision give hope of automating the process of structural inspection. This paper proposes a deep learning-based model of automated crack detection on reinforced concrete surfaces through convolutional neural network (CNN). The Concrete Crack Images Dataset (40,000 labeled images divided equally into crack and non-crack categories) is proposed to train and evaluate the offered model. Before training, images have been processed by resizing, normalization, and dividing into a set of data to guarantee uniformity of inputs in the model and credible performance assessment. CNN architecture includes several layers of convolutional and pooling which extract features hierarchically and dense which classify features in binary. The experimental results indicate the proposed framework has the high detection performance with the overall accuracy of about 99, and large precision, recall, and F1-score indicators. The findings reveal that the model demonstrates strong performance in detecting crack patterns under controlled experimental conditions. The suggested solution can aid the automated structural inspection and smart infrastructure monitoring systems, which will help in making structural health monitoring more efficient and scalable in the modern civil engineering practice.