Efficient and accurate structural health monitoring (SHM) of bridges is a growing necessity, especially for aging infrastructure under constrained budgets and limited inspection windows. While high-fidelity three-dimensional (3D) numerical models enable robust vibration-based SHM, their computational intensity often limits practical application. This presentation introduces a novel, artificial intelligence (AI)-driven framework that integrates reduced-order numerical modeling with transfer learning to deliver fast, accurate, and scalable SHM. The proposed method employs reduced-order models, which are simplified finite element representations that keep critical dynamic behavior, to generate large datasets of simulated vibration responses under various damage scenarios. These data are used to pre-train a deep learning model, which is then fine-tuned using a limited dataset from a high-fidelity 3D model of the same structure. This hybrid training strategy bridges the gap between speed and accuracy, allowing for efficient SHM without sacrificing reliability. A historic steel truss bridge serves as the case study, representative of many aging structures across the U.S. Preliminary results demonstrate that the framework effectively localizes and classifies damage, achieving accuracy comparable to high-fidelity models with significantly reduced computational demands. This approach supports integration into existing inspection and monitoring workflows, offering a practical, cost-effective enhancement to nondestructive evaluation (NDE) programs and enabling more frequent condition assessments. By leveraging deep learning and vibration-based sensing in a computationally efficient way, this research advances the real-world viability of AI-assisted NDE for infrastructure systems.