Minisymposia


Minisymposium I: Dummy models retired? Active digital human models for automotive safety research

Jörg Fehr, Institute of Engineering and Computational Mechanics, University of Stuttgart,
Syn Schmitt, Human Movement Simulation Lab and Stuttgart Research Centre for Simulation Technology, University of Stuttgart
joerg.fehr@itm.uni-stuttgart.de

In the last decades, real and virtual dummy models have been used in the safety developments of cars. However, there were persistent doubts if the biofidelity of human beings is approximated well enough by crash test dummies.

In order to overcome limitations imposed by the use of mechanical surrogates, the active and interdisciplinary field of human body modelling endeavours to stop simulating crash test dummies in the assessment of vehicle safety concepts. Instead, the use of virtual human models is suggested. Additionally, human body models can also be used to investigate ergonomics, comfort and other areas in automotive engineering.

The minisymposium shall give a scientific platform for different aspects of scientific computing of the human body especially in automotive engineering. Different families of modelling approaches like multibody dynamics or finite element methods will be considered. Furthermore, topics such as active vs. passive human models, application of muscle forces to actively change the body posture in pre-crash phase or control algorithms to mimic the human behaviour will be discussed.

Topics include engineering, mathematical and computer science aspects like model validation, run-time comparison or multi-scale approaches. Various fields of application will be presented, e.g. like occupant simulation, passenger simulation, comfort aspects and ergonomics concepts.


Minisymposium II: Get in shape - algorithms and data structures for complex, changing domains on adaptive Cartesian meshes

Tobias Weinzierl, Tomasz Koziara, School of Engineering and Computing Sciences, Durham University,
Martin Ruess, Aerospace Engineering, TU Delft
tobias.weinzierl@durham.ac.uk

Regular Cartesian meshes or block-structured adaptive Cartesian meshes are among the most popular data structures to solve partial differential equations (PDEs) on supercomputers. They also play an important role in other application fields such as particle in cell methods or molecular dynamics where they act as helper data structure for the actual computation - cf. the classical linked list algorithm, e.g. Their structuredness and simplicity facilitate algorithms that exploit state-of-the-art architectures. The regular data access pattern fits to vectorisation and allows the tailoring of the operations to memory hierarchies, and the regularity in space fits to data decomposition approaches predominant for shared and distributed memory parallelisation. They however face one fundamental disadvantages: the world is not aligned in cubes!

Several approaches have been proposed to overcome this drawback: Linked-list algorithms use the regular grid as index data structure to arbitrary point clouds. Immersed boundary methods embed geometry information into the mathematical formulations on top of the regular grids. h-adaptive methods to some extent give up the spatial regularity and make the Cartesian grid's resolution locally follow the geometric requirements. Other methods introduce mappings from the regular topology onto the real geometric world - they buckle the grid. The focus of this symposium is to bring together researchers from a broad range of application fields to discuss algorithms handling non-regular spatial information within the Cartesian world. These algorithms are the backbone of many applications and their properties codetermine the efficiency and capabilities of the evolving simulation codes. The talks hence will overview the algorithms' flexibility, capability, performance, and simplicity on a methodological level.


Minisymposium III: Towards Exascale Simulations and Applications

Philipp Neumann, Alfredo Parra Hinojosa, Informatik, Technische Universität München,
neumanph@in.tum.de

The implementation of highly scalable software on upcoming exascale sys- tems represents a big challenge. The priority program SPPEXA of the German Research Foundation (DFG) has been established to bring together experts to address and solve respective issues. In this minisymposium, speakers from various SPPEXA projects present their latest work with particular regard to numerical simulation software for application on peta- and exascale supercom- puters. Common issues arising for each particular application are discussed and strategies to resolve them are pointed out.


Minisymposium IV: Optimal Control based on Reduced Order Models

Stefan Ulbrich, Jane Ghiglieri Technische Universität Darmstadt Graduate School of Computational Engineering
ghiglieri@gsc.tu-darmstadt.de, ulbrich@mathematik.tu-darmstadt.de

Reduced Order Models (ROMs) aim to reduce the dimension of ODE- or PDE-problems significantly while maintaining the main features of the original complex model. ROM techniques serve in many cases as promising model reduction tool for simulation purposes. In an optimal control or optimization setting, the success of the ROM highly depends on the extraction of the data from the original model that is needed to actually generate the reduced order model. The proposed minisymposium will consider a range of model reduction strategies in an optimization context ranging from optimal control in real-time systems to parameter estimation. An interesting issue is also the characterization of the quality of the model and the development of error estimators.

The proposed minisymposium brings together researchers working actively in the field of reduced order modeling. The aim is to discuss successful developments, to address current challenges, and to identify new directions.


Minisymposium V: High Order Methods for Unsteady Flows

Andrea Beck, Thomas Bolemann Numerics Research Group Institute for Aerodynamics and Gasdynamics University of Stuttgart,
beck/bolemann@iag.uni-stuttgart.de

In the recent CFD Vision 2030 Study by NASA, the ability to predict viscous turbulent flows with possible boundary layer transition and separation has been identified as the single most critical area in Computational Fluid Dynamics development. For these types of flow, various challenges arise due to the complex interaction of the separation and transition. The spatial and temporal schemes are not only required to be accurate over a wide range of scales, but must also lend themselves to efficient parallelization. In addition, their multiscale character often makes a full resolution of all scales prohibitively expensive, which introduces the combined need for a physics-based closure of the unresolved scales and robust numerical schemes. For these types of problems, high order methods have gained considerable interest in the research community, flanked by recent efforts towards their industrialization. Due to their inherent low dissipation, high order schemes are generally less forgiving of numerical instabilities introduced by misrepresentation of subgrid terms, including aliasing errors, which arise in underresolved simulations. The stabilization of high order methods through de- aliasing, filtering or energy-conservative operators is currently an active research topic. Linked to this issue is the inclusion of explicit or implicit subgrid scale modelling strategies in a high order setting. While robustness without loss of accuracy is a key ingredient for simulation of underresolved non-linear multiscale flows, efficiency both in terms of numerical and implementation aspects plays a decisive role in the attractiveness of high order methods. Parallelization strategies in space and time, h/p-adaptivity and efficient time integration schemes are among the issues that are of particular interest in this respect. This mini symposium is open to contributors with either theoretical or/and practical interest in the development of high order discretizations and their application to challenging flow problems. The topics to be discussed at the symposium include, but are not necessarily limited to:

  • Stabilization techniques for high order methods in underresolved situations;
  • Explicit and implicit closure of unresolved scales for multiscale problems in the high order context;
  • Application of high order methods to non-linear multiscale problems;
  • Efficient numerical building blocks for high order methods, e.g. time integration, h/p- adaptivity
  • Implementation and parallelization strategies for high order methods;

  • Minisymposium VI: Simulation-based identification

    Bastian von Harrach, Department of Mathematics - IMNG University of Stuttgart
    harrach@math.uni-stuttgart.de

    When all necessary parameters are known, the outcome of a physical experiment can be predicted by numerical simulations. Often the goal of the simulations is to identify the parameters from a comparison of the simulated outcome with real measurements. Such inverse simulation-based identification problems are of enormous importance for novel tomography techniques and for controlling industrial processes. The aim of this minisymposium is to bring together experts from different groups to discuss the state of the art in simulation-based identification.

    3rd International Workshop on Computational Engineering

    Registration Deadline

    September 1, 2014

    Date

    October 6-10, 2014

    Venue

    Universität Stuttgart
    Computer Science Building
    Universitätsstraße 38
    Campus Map
    70569 Stuttgart
    Germany

    Contact

    Miriam Mehl, miriam.mehl@ipvs.uni-stuttgart.de