Enhancing structural health monitoring: Conditional GAN-based crack detection in concrete and asphalt surfaces

Abstract

Structural Health Monitoring (SHM) techniques are essential for identifying structural defects and damages to optimize maintenance and repair schedules. While automated vision-based damage detection systems have been explored in the past, their reliability and deployment challenges in real-world scenarios have limited widespread adoption in structural inspections. This research presents an innovative approach to SHM by leveraging advanced Deep Learning (DL) techniques to overcome persistent challenges in automated vision-based damage detection systems. This paper introduces a novel application of Pix2Pix, a Conditional Generative Adversarial Network (CGAN), designed for efficient and precise crack detection and localization in concrete bridges. The approach is centered on a robust DL model trained on an extensive and diverse dataset of cracked concrete and asphalt pavement surfaces. Streamlined workflow of the proposed system enables rapid processing and generates highly accurate representations of detected cracks in under 0.7 seconds, significantly enhancing real-time monitoring capabilities. To demonstrate practical applicability, a user-friendly prototype application was developed for detecting deterioration in concrete and asphalt pavement surfaces. This tool not only identifies existing damage but also supports longitudinal studies by enabling comparisons with previous assessments for crack propagation behavior analysis. The proposed system’s ease of use and automation capabilities make it a promising solution for large-scale structural inspections, offering a significant step forward in the field of SHM.