GOODWIN: Grey-box approaches for digital twins of complex mechanical structures and systems under uncertainty.
This training network is aimed at training a next generation of PhD students in the development, application and exploitation of state-of-the-art techniques in multi-fidelity grey-box modelling for high-dimensional mechanical analysis under uncertainty. Specific focus is dedicated to highly transient dynamic simulations such as crash, nonlinear structural dynamics and complex systems.
In engineering simulation, mathematical models of a structure or system are usually classified as either being “white box” or “black box”. The former approach uses first-principle equations to model the structural behavior and hence requires deep insight into the nature of the problem. The latter starts from observational data on the structure or system to construct a mathematical model based on e.g., machine learning approaches. Grey box structures are aimed at combining both approaches to relax the need to exactly model the underlying physics, while requiring considerably less data than a pure black box approach.
This training network specifically will jointly focus on the development of procedures and modelling approaches combining costly, but high-fidelity first principle based numerical codes with mid-fidelity surrogate models and low-fidelity data driven approaches based on limited sets of experimental data, including measures of impreciseness and uncertainty. The dedicated tasks of the consortium are divided as follows:
- KU Leuven (David Moens, Matthias Faes): coordination and inverse techniques for uncertainty quantification
- Leibniz University Hannover (Michael Beer): imprecise probabilistic methods
- University of Thessaloniki (George Stefanou): probabilistic approaches
- ETH Zurich (Bruno Sudret, Stefano Marelli): surrogate modelling approaches for high-dimensional models
- TU Munich (Fabian Duddeck): crash simulation and reduced order modelling
- Université de Compiegne (Pascal Lardeur): non-linear dynamics
- Politecnico di Milano (Enrico Zio): data-driven black-box approaches