PSI - Issue 64

Mercedes Solla et al. / Procedia Structural Integrity 64 (2024) 293–300 M. Solla et al. / Structural Integrity Procedia 00 (2019) 000–000

294

2

of the corrosion process cannot be detected by traditional visual inspections). A critical aspect of efficient repair activities is detailed damage diagnosis and, therefore, innovative approaches for RC-TI maintenance and safety are urgently needed. Advancements in sensing technologies and structural health monitoring (SHM) systems enable continuous monitoring, allowing for early detection of deterioration and timely interventions, thus fostering the prolonged use of the TI (Tešić et al. 2021). The gained knowledge leads to an improved assessment of the actual safety level and improves the rationalization of decision-making regarding maintenance and intervention strategies. The use of Non-Destructive Testing (NDT) techniques remains as a suitable solution for TI monitoring while preserving the integrity of structures (Elseicy et al. 2022). Among these techniques, Ground-Penetrating Radar (GPR) technology has emerged as a widely used and effective tool for TI monitoring (Rasol et al. 2022a). GPR provides comprehensive subsurface imaging, offering valuable insights into TI condition (Diamanti et al. 2017, Solla and Fernández 2022). However, it also presents challenges particularly in processing and data analysis, which often involves manual interpretations that can be subjective and time-consuming, especially for large amounts of collected data. The introduction of Artificial Intelligence (AI) and cloud-based systems in condition monitoring, particularly in SHM, has become a top priority for practicing engineers and researchers, especially for real-time analysis. AI technologies offer notable advancements in processing vast amounts of data collected by GPR (Kuchipudi et al. 2022) thereby contributing to eliminate analysis subjectivity, saving processing time, and reducing overall operation costs compared to traditional methods. AI algorithms can automatically analyze GPR data, detect anomalies, and provide valuable insights into the condition of TI assets (Novo and Kaufmann 2023). Recently, Deep Learning (DL) techniques have emerged as state-of-the-art (SoA) methods for signal processing and object detection, offering enhanced detection speed and accuracy compared to conventional image processing techniques. Several architectures can be applied to GPR data in various dimensions, including 1D, 2D and 3D. DL applications help to eliminate analysis subjectivity by detecting hidden patterns between target features, while maximize the quality and accuracy of the results obtained. Within rebar detection, the You Only Look Once (YOLO) method can detect and localize steel bars and combine them with 1D Convolutional Neural Networks (CNN) to estimate rebar diameters (Li et al. 2021). However, the effectiveness of DL models heavily relies on the availability of large and high-quality datasets for training. In the context of GPR data analysis, objects with a circular section, such as rebar, often exhibit a characteristic hyperbolic reflection pattern. Nevertheless, the reflected signatures in GPR data are highly variable due to several factors, including the type, size and orientation of the targets, but also influenced by various external and signal-related sensitivity factors such as ringing noise, energy couples, airwaves, signal attenuation and soil condition. Moreover, operational factors can significantly influence the quality, resolution, and reliability of the collected data, mainly: (i) the selection of the antenna frequency (influencing image resolution and detection accuracy), and (ii) the orientation of the data collection direction, thus compromising the reflected signature (perpendicular direction provides hyperbolic reflections). Given these considerations, the objective of this paper is to compare two different GPR systems, SFCW (frequency domain) and pulsed radar (time domain) operating in different frequency ranges and resolutions, in terms of rebar detectability and imaging capabilities.

Nomenclature GPR

ground-penetrating radar SFCW stepped frequency continuous wave TI transport infrastructures RC reinforced concrete NDT non-destructive testing SHM structural health monitoring AI artificial intelligence DL deep learning SoA state-of-the-art YOLO you only look once CNN convolutional neural networks