Structural Model Updating and Model Selection: Bayesian Inference Approach Based on Simulation

Structural model updating, and model selection have been in the focus of intensive research for many decades and still represent a major challenge. In this paper, the Approximate Bayesian Computation (ABC) using a Nested Sampling (NS) technique is employed to deal with model updating and model selection issues. The proposed framework is based on simulations, it can update a single model but also to find the most likely model from a set of competing models. Moreover, instead of learning a single point estimates, the ABC-NS scheme learns a distribution over the unknown model parameters allowing us to quantify predictive uncertainty. The ABC framework offers the possibility to use different discrepancy metrics measuring the similarity between the measured modal data and the ones obtained from simulations. The performance and the robustness of the simulation-based inference procedure in structural dynamics are demonstrated through two numerical studies using modal data.