Current transportation asset assessment methods often rely on manual or heuristic approaches, which limit scalability and efficiency. Although AI has been proposed as a potential solution, many AI systems produce black-box predictions that lack transparency, interpretability, and uncertainty disclosure, making them difficult to trust in safety- and liability-sensitive settings. This project develops trustworthy and explainable AI methods for digital twin-enabled transportation asset management by generating human-readable explanations grounded in visual evidence and standardized asset guidelines, such as the Manual on Uniform Traffic Control Devices for Streets and Highways (MUTCD). The proposed approach explicitly quantifies uncertainty and improves robustness through ensemble modeling, while also evaluating fine-tuning strategies to support industry-ready deployment of the AI-powered pipeline in practice.
