Data-informed finite element modeling of thermal effects in bridges exploiting frequency and temperature monitoring data

Temperature changes can strongly affect the dynamic behavior of bridges, making it difficult to distinguish between environmental effects and actual structural damage. To develop reliable damage detection methods, it is essential to generate realistic data that include both thermal variability and damage scenarios. However, real data from damaged structures are rare, so numerical simulations are often used. This study proposes a data-informed finite element (FE) modeling approach that reproduces temperature effects directly from measurements. Instead of relying on traditional thermo-mechanical models, the method learns how temperature influences the effective mechanical properties of the structure using long-term monitoring data (modal frequencies and temperature). A set of temperature-sensitive parameters is first generated within realistic ranges, and the FE model is used to compute the corresponding modal frequencies. These data are used to train a surrogate model linking dynamic response to material properties. The surrogate is then applied to field monitoring data to infer how these parameters vary in practice. A regression model is finally built to relate temperature to the identified parameters. Once trained, the model enables realistic simulation of thermal effects for any temperature history. Importantly, it also provides a consistent framework to introduce and simulate damage scenarios, allowing the generation of synthetic datasets that capture both environmental variability and structural degradation. This is particularly valuable for testing and validating damage detection algorithms under realistic operating conditions. The approach is demonstrated on the KW51 railway bridge, accurately reproducing observed thermal trends and enabling the generation of realistic data for both healthy and damaged states.