PSI - Issue 64

Konrad Bergmeister et al. / Procedia Structural Integrity 64 (2024) 14–20 Konrad Bergmeister / Structural Integrity Procedia 00 (2019) 000 – 000

16

3

inspections is not consistent even when the same standards/procedures are to be followed. The discrepancies are attributed to human perception, differences in training and experience, and fear and uncertainty while performing inspections (Abdallah et al. 2022, Moore et al. 2001, Graybeal et al. 2001). Finally, the dissemination of the data is shown to increase the level of uncertainty, as the process is not always consistent. While these studies refer to specific visual inspection practices in the United States, the contributing factors that lower the quality of visual inspections are also applicable to European standards that follow similar principles, as discussed in the work of Turksezer et al. (2021). Moreover, the discrepancies in the procedures and standards within the European countries are expected to augment the concerns mentioned above on data dissemination quality. The limitations, hardships, and high cost of visual in-person inspections are well understood and therefore, significant research efforts have proposed alternatives. Promising solutions rely on detailed photos obtained using Unmanned Aerial Vehicles (UAV), commonly known as drones. The studies of Koch et al. (2014), Zink and Lovelace (2015), Morgenthal et al. (2019), Liang (2019), Lei et al. (2018), Seo et al. (2018), Kim et al. (2018), Ellenberg et al. (2016) and Mandirola et al. (2022) among many others, is a short list of relevant research efforts. The common goal among these studies is to employ UAV-obtained images and, through different methodologies, construct 3D digital models of bridges with or without employing image recognition to detect defects such as cracks and corrosion of steel reinforcement induced defects automatically. In this way, the goal is to transform the bridge inspection process into a digital and automated streamline, allowing for increased efficiency in disseminating data and improving consistency. Moreover, including AI-enhanced image recognition and damage detection further accelerates the process. The current paper discusses these technological developments that pave the way towards a data-driven, automated inspection of bridges to promote safety, early warning, and intervention. In particular, the major contributing elements to be discussed are UAV-obtained images, AI algorithms that allow for image recognition and machine vision, and a thorough risk-oriented approach to assess visually accessible defects. Their importance, contribution, and limitations are discussed, while omitting the algorithmic details of their implementation, effectively delivering a high-level Using UAVs to access and inspect infrastructures and bridges is widely proposed and pursued. Similarly to other studies, Mandirola et al. (2022) point out several advantages compared to traditional methods involving in-person, conventional survey techniques. These advantages include (i) the capacity for vertical hovering and landing, requiring minimal operational space, (ii) the flexibility to operate at low altitudes, (iii) the ability to conduct detailed inspections of inaccessible areas and capture high-resolution images from various perspectives, (iv) lower operational and acquisition expenses, (v) decreased risk for operators, and (vi) faster surveying and data processing speeds. The resolution of the images collected with drones, together with the accessibility of associated metadata (such as GPS/RTK position, time, camera parameters, etc.), is of paramount importance. Therefore, it is fundamental to capture images leveraging on professional equipment. For example, Fig. 1 shows the equipment adopted in this work in order to capture high-resolution images of a viaduct. The configuration consists of the DJI Matrice 350RTK equipped with the ZENMUSE H20 camera. This technology ensures high resolution images, up to a resolution of 0.1mm/pixel Ground Sampling Distance (GSD). The benefits of utilizing UAVs to access and inspect bridges are straightforward. However, specific regulations and safety measures exist to promote safety. Such measures limit or prohibit the use of UAVs in certain instances. At the same time, in other cases, the resulting quality of UAV inspections is low due to the nature of the inspection. UAV flights are prohibited above traffic, either regarding the traffic above or below the bridge. In cases with no traffic under the bridge, the Global Positioning System (GPS) is oftentimes inaccurate when the UAV is under a bridge of low height (e.g., up to 7-8 meters). Regarding the quality of the UAV-obtained images, vegetation near the bridge and low illumination result in low quality, while specific areas and components of a bridge are frequently inaccessible (e.g., in between parallel beam segments very close to each other). Finally, weather conditions such as heavy rainfall/snowfall or intense wind gusts affect the UAV's maneuverability, stability, and the resulting quality of the obtained images. description of the proposed scheme. 2. Data acquisition and AI tools 2.1. Drone photography