Real-time Bayesian data assimilation with data selection, correction of model bias, and on-the-fly uncertainty propagation

Abstract

The work introduces new advanced numerical tools for data assimilation in structural mechanics. Considering the general Bayesian inference context, the proposed approach performs real-time and robust sequential updating of selected parameters of a numerical model from noisy measurements, so that accurate predictions on outputs of interest can be made from the numerical simulator. The approach leans on the joint use of Transport Map sampling and PGD model reduction into the Bayesian framework. In addition, a procedure for the dynamical and data-based correction of model bias during the sequential Bayesian inference is set up, and a procedure based on sensitivity analysis is proposed for the selection of the most relevant data among a large set of data, as encountered for instance with full-field measurements coming from digital image/volume correlation (DIC/DVC) technologies. The performance of the overall numerical strategy is illustrated on a specific example addressing structural integrity on damageable concrete structures, and dealing with the prediction of crack propagation from a damage model and DIC experimental data.