An Optimal Modelling Strategy for Vehicle–Bridge Interaction and Frequency-Based Damage Assessment

Bridge structures subjected to moving vehicle loads display highly coupled dynamic behaviour arising from structural flexibility, interaction forces, and surface or rail irregularities. Accurately representing these vehicle–bridge interaction (VBI) effects is essential for predicting vibration responses and enabling reliable vibration-based Structural Health Monitoring (SHM). This work proposes a comprehensive modelling and learning framework that integrates high-fidelity numerical modelling, reduced-order dynamic representation, and supervised classification to support robust bridge health assessment. A detailed finite element (FE) model of a steel truss bridge is first constructed and rigorously updated through experimentally obtained modal parameters, ensuring that the numerical model closely reproduces the measured dynamic characteristics of the physical structure. The updated FE model is subsequently condensed using Component Mode Synthesis (CMS), producing an efficient reduced-order VBI formulation that preserves the dominant modal content while enabling large-scale simulations across a wide range of operating and damage conditions. This reduced model is employed to generate an extensive synthetic dataset reflecting multiple structural states, vehicle speeds, track irregularity patterns, and load configurations. Frequency-domain features extracted from these simulations form the input to a one-dimensional Convolutional Neural Network (1D-CNN), trained to discriminate between health scenarios with high accuracy. Validation against independent vibration measurements obtained from a controlled laboratory bridge–vehicle setup demonstrates that the classifier reliably identifies the correct structural condition, even under varying operational conditions.