Welcome to the Professorship for Continuum Mechanics

We strongly believe that the rapid advances in data sciences and machine learning will not and should not leave physical modeling untouched. The group promotes a data-driven perspective for carrying out engineering analysis and design tasks, in calibrating and validating computational models, but also in discovering new ones that are predictive of multiscale behavior. We employ a probabilistic-Bayesian mindset  to achieve the fusion of data and models and attempt to address fundamental challenges relating to dimensionality & model reduction and the discovery of salient patterns in the presence of small data.




Graphical illustration of a deep probabilistic generative model to coarse-graining (CG) dynamical systems. (Top) Particle density obtained by simulating 250,000 interacting particles. In the limit of infinite scale separation, the density can be modeled by the inviscid Burgers' equation. (Middle) Predicted density from the coarse-grained model trained on data from the first two time-steps (more details in paper). (Bottom) Particle density at different time instants into future. Shaded areas indicated predictive uncertainty. Observe the appearance of shock fronts which can be captured by the CG model.