Multi-fidelity modelling for structural identification

Asset-management decision-making is often improved by obtaining a better understanding structural behaviour through monitoring, which can then help avoid unnecessary repair, retrofit and replacement of existing infrastructure. Interpretation of monitoring data in the presence of biased and systematic uncertainties may require computationally time-consuming numerical models to approximate real structural behaviour. These models could be replaced by less time-consuming machine learning-based surrogate models. When and how this should be done is the subject of current research. In this paper, the use of surrogate models in a multi-fidelity framework for structural identification of a full-scale bridge is presented. The effects of varying degrees of fidelity are studied in a transparent manner within a structural-identification framework. The use of models with multiple fidelities helps obtain accurate model updating results in less time compared with using only one high-fidelity model class for simulations.