HYDAMO - Hybrid data-driven and model-based simulation of complex flow problems in the automotive industry


Researchers: Derick Nganyu Tanyu, Lennart Abels
Project funding: Bundesministerium für Bildung und Forschung (BMBF), Förderschwerpunkt Mathematik für Innovationen
Project sponsor: DESY
Partners: Axel Klar, Technische Universität Kaiserslautern; Jörg Kuhnert, Fraunhofer ITWM, Kaiserslautern; Lars Aschenbrenner, Volkswagen AG, Wolfsburg; Matthias Schäfer, ESI Software Germany GmbH, Neu-Isenburg
Duration: 01.04.2020 - 31.03.2023

There are essentially two different paradigmatic approaches for mapping complex physical processes: classical physical modeling with associated numerical simulation (model-based) and prognostic methods based on the analysis of large amounts of data (data-driven). In recent years, the efficient combination of both approaches has become a research topic in its own right. However, research is far from an interlocking, problem-adapted application of the principles.

The goal of HYDMAO is to integrate data-driven and model-based approaches into an overall solution based on a continuum mechanics problem from the automotive industry that has been insufficiently understood to date. This is intended to decisively improve the computer-aided mapping of the associated process. A prototypical example with great industrial and societal importance is considered: The interaction of a vehicle with complex materials such as sand, mud or snow. On such surfaces, vehicle stability is not always given: Collisions or vehicle rollover may be unavoidable. These situations must be handled appropriately in terms of occupant safety. In particular, the question arises as to whether, when and which airbags should be deployed. This problem can only be solved efficiently by a suitable computer-based mapping of the process. Our application partners Volkswagen AG and ESI Software Germany GmbH underline the general relevance of the project, which extends far beyond the prototypical example.

The subproject Parameter Identification of Complex Nonlinear Dependencies of the WG Technomathematics aims at reducing the high-dimensional parameters in a generic model to their inherently nonlinear but low-dimensional structure by Deep Learning approaches and to identify them for the subsequent numerical simulation. The focus here is on stability analysis in addition to machine learning (ML) approaches.