Using Regularized Linear-Regression Surrogate Models for Accurate Probabilistic Structural Identification

Model-based data interpretation has the potential to increase knowledge of structural behavior and support asset management. Models are usually conservative and contain many parameters and sources of systematic uncertainty, which need to be taken into account for accurate model updating. However, interpreting measurements using physics-based models is computationally expensive. Supplementing physics-based models with inexpensive surrogate models might facilitate practical implementation of data interpretation. In this paper, development of regularized linear-regression surrogate models for simulating structural behavior of a full-scale bridge and their use in error-domain model falsification for structural identification is presented. In this methodology, uncertainties from systematic sources and surrogate model error are considered explicitly during model updating. Results are verified with knowledge of parameters used to simulate measurements on a full-scale bridge. Use of simple regularized linear-regression models helps achieve accurate knowledge of updated structural behavior, which can then be used for making better asset management decisions.