The project, led by Arline Osorio Moreno under the supervision of Professor Thomas Kjeldsen, focuses on the critical problem of cross-site generalization. While an AI model might excel within a controlled training environment, its accuracy often falters when deployed on a network with different construction standards or varying camera quality. To address this, the research compares four distinct methodologies: vision-language models, vision foundation models, supervised deep learning, and classical machine-learning techniques.

Early findings suggest that larger, generalized models do not necessarily equate to superior performance in specialized engineering contexts. Arik Voronov, AI R&D Lead at Dynamic Infrastructure, noted that more targeted approaches often provide greater consistency when exposed to site-specific variations. This research aims to move beyond theoretical capabilities, establishing the rigorous testing protocols necessary for operators to trust automated systems in the field. The partners plan to develop these findings into a formal peer-reviewed publication to further bridge the gap between academic research and practical infrastructure management.