A general framework for uncertainty quantification under non-Gaussian input dependencies
share
Uncertainty quantification (UQ) deals with the estimation of statistics of the system response, given a computational model of the system and a probabilistic model of the uncertainty in its input parameters. In engineering applications it is common to assume that the inputs are mutually independent or coupled by a Gaussian dependence structure (copula). In addition, many advanced UQ techniques rely on a transformation of the inputs into independent variables (through, e.g., the Rosenblatt transform). The transform is unknown or difficult to compute in most cases, with the notable exception of the Gaussian copula, which admits a simple closed-form solution (also known as Nataf transform). This makes the Gaussian assumption widely used in UQ applications, despite being often inaccurate. In this paper we overcome such limitations by modelling the dependence structure of multivariate inputs as vine copulas. Vine copulas are models of multivariate dependence built from simpler pair-copulas. The vine representation is flexible enough to capture complex dependencies. It also allows for statistical inference of the model parameters and it provides numerical algorithms to compute the associated Rosenblatt transform. We formalise the framework needed to build vine copula models of multivariate inputs and to combine them with UQ methods. We exemplify the procedure on two realistic engineering structures, both subject to inputs with non-Gaussian dependence structures. For each case, we analyse the moments of the model response (using polynomial chaos expansions), and perform a structural reliability analysis to calculate the probability of failure of the system (using the first order reliability method). We demonstrate that the Gaussian assumption yields biased statistics, whereas the vine copula representation achieves significantly more precise estimates.
READ FULL TEXT