Mobile Deep Learning Framework for Automated Concrete Surface Crack Detection and Assessment on Construction Sites

Abstract

Nowadays, there are an increasing number of historic high-rise civil buildings worldwide, most of which are made of concrete. Concrete can lose strength due to environmental factors and constant loading. Consequently, the external surface of the structure may be damaged (crack and spall). If these anomalies are not looked into and fixed, the structural integrity could be compromised. Therefore, crack detection is essential when inspecting building structures to evaluate their safety. To ensure the longevity and reliability of structures, it is crucial to have professionals do building inspections on a regular basis. Traditionally, building inspections have been carried out using both human-based visual inspection techniques and deep learning (DL) techniques, which have recently demonstrated remarkable success in mobile edge integration. By integrating IoT-enabled sensors and cameras via a mobile application that acts as a real-time data gateway, this system enables the automatic assessment and detection of concrete surface cracks on construction sites. To achieve this, the study presents a mobile deep learning framework for automated concrete surface crack detection and assessment on construction sites (MDLACS-CDACS). The MDLACS-CDACS model undergoes data preprocessing, feature extraction, detection, and classification. A set of experiments was conducted to validate the MDLACS-CDACS techniques for inspectors, offering an effective way to measure the degree of damage of cracks found in image-based assessments. The study obtained an accuracy of 97.85% in damage classification without overfitting. The suggested MDLACS-CDACS outperforms state-of-the-art methods in balancing efficiency and accuracy, and real-time on-site inspection is made possible by its inference speed on edge devices.