Uncertainty quantification in a mechanical submodel driven by a Wasserstein-GAN

The analysis of parametric and non-parametric uncertainties very large dynamical systems requires the construction a stochastic model said system. Linear approaches relying on random matrix theory Soize (2000) principal component can be used when undergo low-frequency vibrations. In case fast dynamics wave propagation, we investigate generator boundary conditions for submodels by using machine learning. We show that use non-linear techniques in learning data-driven methods is highly relevant. Physics-informed neural networks Raissi et al. (2017) are possible choice method to replace linear modal analysis. An architecture supports necessary physical system uncertainties, since goal learn underlying probabilistic distribution uncertainty data. Generative Adversarial Networks (GANs) suited such applications, where Wasserstein-GAN with gradient penalty variant Gulrajani offers improved convergence results our problem. objective approach train GAN data from finite element code (Fenics) so as extract faster predictions submodel. submodel training have both same geometrical support. It zone interest quantification relevant engineering purposes. exploitation phase, framework viewed randomized parametrized simulation submodel, which Monte Carlo estimator.