PSI - Issue 64

Shirley J. Dyke et al. / Procedia Structural Integrity 64 (2024) 21–28 Author name / Structural Integrity Procedia 00 (2019) 000–000

27 7

material properties, damage status, and inspection history, to decide on the most suitable inspection action. The reward function in the RL model aims to develop an inspection policy that minimizes both the economic costs and risks associated with bridge inspections. After training with simulated data or historical data, the RL agent can make data driven recommendations regarding inspection timing and methodology to improve inspection efficiency and effectiveness.

Fig. 4. Results for different sample cost scenarios: expected cost vs variance of the cost (based on Zhang, et al., 2023a).

6. Summary In the rapidly evolving landscape of applications of AI to structural engineering and monitoring, achieving perfect results remains an elusive goal. However, the integration of AI into the research and practice of structural engineering is inevitable; it must be approached carefully and with a clear understanding of the capabilities of the methods. The capabilities developed in the IISL to facilitate the collection, organization, and predictive modeling of structural data address several of the many challenges provides several pioneering capabilities. These methods have transformative potential, yet there is work to be done. For the foreseeable future, they are excellent tools to aid the human engineer or inspector, and can augment procedures typically used and can help prioritize where we should place our attention. The capabilities developed are adept at processing extensive visual datasets for structural health monitoring, enabling rapid and reliable assessment of infrastructure after natural disasters. They are particularly innovative in their ability to deal with large sets containing unstructured images, to extract and organize contextual data, and to synthesize information needed to support robust decision-making regarding infrastructure health and resilience. For the effective utilization of AI in structural engineering, it would be beneficial to shift toward the collection of more high-quality, structured data. Collecting and sharing more data will be critical for improving AI-based methods overall. High-resolution images and comprehensive datasets are needed that delineate the condition of infrastructure both before and after hazard events, for reconnaissance, and throughout the lifecycle of a bridge, for inspection. Guidelines are needed, and should be followed, for collecting data that can be better analyzed by machines and methods. And the use of automated tools to provide metadata, such as GPS information, will provide greater context and support the development of new methods that can exploit these data. Metadata such as the location, time, and specific conditions under which the images were taken are essential for providing context. Engineers will need to be meticulous in establishing and curating meaningful datasets that are representative of various structural types and conditions to ensure robust AI model training. Such detailed and well-maintained datasets will significantly enhance the precision and reliability of AI tools in assessing infrastructure, thus aiding in the advancement of a proactive and data-informed approach to structural engineering. Acknowledgements We acknowledge the support from the US National Science Foundation under Grant Nos. NSF 1608762 and NSF 1835473. The authors would also like to acknowledge collaborators mentioned in this work, Prof. Chul Min Yeum (U. Waterloo), Prof. Jongseong (Brad) Choi (SUNY-Korea), and Prof. Alana Lund (U. Waterloo).