Contact
 
IRTG 2657 News
Presentation on Distribution-free Bayesian Updating with Hybrid Uncertainties - 21 March

Presentation on Distribution-free Bayesian Updating with Hybrid Uncertainties - 21 March

Symbolic image: presentation on distribution-free Bayesian updating with hybrid uncertainties Symbolic image: presentation on distribution-free Bayesian updating with hybrid uncertainties Symbolic image: presentation on distribution-free Bayesian updating with hybrid uncertainties

On Tuesday, 21 March, Dr. techn. Matteo Broggi will give a presentation about distribution-free Bayesian updating with hybrid uncertainties.


The complexity of model updating depends on the presence of different level of aleatory and epistemic uncertainties. Deterministic model updating generally accounts for the presence of only epistemic uncertainty, where parameters are fixed but unknown constants. On the other hand, the presence of hybrid uncertainties is considered by Bayesian model updating, where the aim is not only a single set of parameters, but a reduced space of epistemic variables which can represent model predictions in agreement to a multiple sets of observations. In the presence of hybrid uncertainties, parameters are modeled as random variables with only vaguely determined uncertainty characteristics, e.g., by means of p-boxes. However, analysts need to assume a distribution family for the epistemic parameter. This is a very significant assumption that greatly constrains the possibility to accurately capture the real variation affecting the physical system. Thus, staircase random variables are employed in this presentation to characterize epistemic uncertainties without any pre-hypotheses of the distributional characteristics. Additionally, a two-step approximate Bayesian computation is employed, where the Euclidian and Bhattacharyya distances are utilized as uncertainty quantification metric. The performance of the proposed procedure is demonstrated with an illustrative application using a shear building model and the 2020 NASA challenge.


Join the presentation with this link!