The Organising Committee is is happy to announce the list of the accepted mini symposia.
MS1: Convergence of Artificial Intelligence and High-Performance Computing for Computational Fluid Dynamics
- Organisers1:
- Andreas Lintermann, Jülich Supercomputing Centre, Forschungszentrum Jülich GmbH, Germany
- Guillaume Houzeaux, Barcelona Supercomputing Center, Spain
- Corentin Lapeyre, Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique, France
ParCFD Minisymposium ProposalArtificial Intelligence (AI) technologies are penetrating into all sectors of research and industry, where they automate and accelerate processes, and uncover new unseen relations in huge amounts of datasets. Among many other disciplines, method development in Computational Fluid Dynamics (CFD) continuously benefits from these technologies. AI methods, and notably deep learning techniques, are used to develop new models for CFD, e.g., reduced-order models, surrogates and closure models (e.g. for turbulence and chemistry) aiming at efficiently modeling complex physics that are otherwise expensive to compute. The quality of such methods is often a function of both the quantity and the accuracy of the underlying simulation data used for training and generated by CFD methods, as well as the physical constraints imposed on the training. The generation and processing of high fidelity simulation data necessitates the application of High-Performance Computing (HPC) systems. Modular and heterogeneous systems with accelerator and/or specialized AI-components as blueprints for upcoming Exascale systems have the potential to deal with the demands of complex and intertwined simulations and AI-data processing workflows. This minisymposium aims at gathering experts in the fields of development and application of parallel CFD methods incorporating novel AI methods, and pure AI method developers contributing to the fields of CFD and HPC alike. It will build a platform for discussion and exchange in the context of the convergence of AI and HPC with respect to parallel CFD methods that could benefit from the power of next-generation Exascale computing systems.
A list of potential contribution may include, but it is not limited to:
- Turbulence and other subgrid processes modeling with AI
- Modeling of combustion processes with AI
- Surrogate modeling for CFD
- AI-incorporated full-loop implementations of CFD workflows
- CFD-centric AI training on modular and heterogeneous HPC systems
- Physics-informed learning methods, e.g., with deep neural networks, graph neural networks, or generative networks
- Application of AI methods to various fields of CFD: aerospace, automotive, energy, coating, bio, etc.
- Feature extraction from complex flow fields using AI methods
1The organizers are all members of the European Center of Excellence in Exascale Computing “Research on AI- and Simulation-Based Engineering at Exascale” – CoE RAISE (https://www.coe-raise.eu)
MS2: Advanced HPC algorithms for pre-exascale CFD simulations on heterogeneous supercomputers
- Organisers:
- Xavier Álvarez-Farré, Heat and Mass Transfer Technological Center, Technical University of Catalonia, Spain
- F. Xavier Trias, Heat and Mass Transfer Technological Center, Technical University of Catalonia, Spain
- Andrey Gorobets
- Charles Moulinec, Daresbury Laboratory, Science and Technology Facilities Council, UK
- Guillaume Houzeaux, Computer Applications in Science and Engineering, Barcelona Supercomputing Center, Spain
- Takayuki Aoki, Global Scientific Information and Computing Center, Tokyo Institute of Technology, Japan
The evolution in hardware technologies enables scientific computing and especially CFD, which is very much HPC-demanding, to reach new levels. Nowadays, heterogeneous HPC systems are rather common in the solution of both industrial and academic scale problems. To take advantage of such systems, the computing subroutines that form the algorithms, the so-called kernels, must be compatible with distributed- and shared-memory multiple instruction, multiple data parallelism, and more importantly, stream processing, which is a very restrictive parallel programming paradigm. Therefore, complex hierarchical parallel implementations are required for combining the different parallel paradigms and the corresponding computing frameworks. In this context, this Minisymposium aims to bring together CFD-driven people working on advanced and cutting-edge numerical methods for solving large-scale CFD simulations with a special focus on efficiency, portability and sustainability.
Contributions can cover, but are not limited to, with the general aim to demonstrate their benefit to large-scale CFD simulations:
- Hybrid approaches combining MPI with OpenMP, OpenACC, OpenCL, or CUDA on heterogeneous HPC systems
- Architecture-oriented solutions for parallel linear solvers
- Sparse algebra kernels
- Eigenvalue problems
- Load balancing, Adaptive mesh refinement
- Domain-specific languages
- Parallel-in-time methods.
MS3: Solutions, ideas and perspectives for CFD, HPC and Exascale computing
- Organisers:
- Giorgio Amati, Information Technology Department, CINECA, Italy
- Neil Ashton, Computational Engineering Product/Tech Strategy and HPC, Amazon Web Services, UK
Probably in this year, and in the next years, a few Exascale class supercomputers will be delivered. To achieve a 1018 flop per second a huge number of computing nodes and accelerators like GPGPU will be used trying to reduce as much as possible total power consumption.In this mini-symposium, focusing on CFD, a description of main ideas for Exascale, and as a direct consequence for HPC, will be presented.
Topics of interest of this mini-symposium include, but are not limited to:
- System co-design
- accelerators (GPGPU, FPGA, RISC-V,…)
- Processor evolution
MS4: High-performance lattice Boltzmann computing and its application
- Organisers:
- Amirul I. Khan, School of Civil Engineering, University of Leeds, UK
- Jianping Meng, Daresbury Laboratory, Science and Technology Facilities Council, UK
The lattice Boltzmann method has been developed as an emerging modelling tool for fluid flows and beyond. The method is rooted from the kinetic theory and is well-suited for high-performance computing due to its linearity and locality. In this mini symposium, we will invite contributions to reflect the rapid developments of the lattice Boltzmann method. Meanwhile, we also welcome contributions from other kinetic theory-based methods, e.g., discrete velocity (Boltzmann) method gas-kinetic scheme and so on.
Contribution can cover, but are not limited to, the development of LB codes and their applications
- GPU porting of LBM software
- LBM application in gas-dynamics
- Multiphase flows with LBM (gas-liquid, solid-liquid and phase changes)
- Porous media
- Free surface flows
MS5: CFD for supersonic and hypersonic aerodynamics
- Organisers:
- Davide Modesti, Aerodynamics Group, Faculty of Aerospace Engineering, Delft University of Technology, The Netherlands
- Jian Fang, Daresbury Laboratory, Science and Technology Facilities Council, UK
Recent years have witnessed a renewed interest in hypersonic flight. For instance, commercial aircraft manufacturers are heavily investing in hypersonic technology that can drastically cut down intercontinental passenger flight time making same-day global round trips a reality. Also access to space has become more attractive for private companies which are largely investing in reusable launchers to lower the costs. These engineering applications call for a more in depth understanding of the flow physics and modelling of high-speed flows. The progress of computational fluid dynamics (CFD) and High-Performance Computing in the past three decades has boosted large-scale high-fidelity simulations of high-speed flows. However, many aspects of high-speed flows are still not well understood, especially those related to boundary layer transition, shock-wave/boundary layer interaction, non-equilibrium thermodynamics and fluid-structure interaction. In terms of engineering applications, classical turbulence models for compressible flows often rely on simple compressibility corrections whose validity is not always clear. This mini-symposium will target and communicate on the progress in CFD for supersonic and hypersonic aerodynamics.
Topics of interest of this mini-symposium include, but are not limited to:
- High-performance computing of high-speed flows
- Numerical schemes, models and algorithms
- Shock-wave/boundary layer interaction
- Turbulence modelling for high-speed flows
- Fluid-structure interaction for high-speed flows
- Laminar-to-turbulent transition for high-speed flows
- Application of CFD to supersonic/hypersonic engineering
MS6: High order Finite Difference schemes for large scale CFD simulations
- Organisers:
- Sylvain Laizet, Turbulence Simulation Group, Department of Aeronautics, Imperial College London, UK
- Andrew P. S. Wheeler, Whittle Laboratory, Department of Engineering, Cambridge University. UK
For fundamental turbulent flows in academic configurations, the usefulness of highly accurate numerical schemes for high-fidelity simulations (Direct and Large Eddy Simulations DNS/LES) is now fully recognised. For very simplified geometries, in terms of accuracy and computational efficiency, the most spectacular gain is obtained using spectral methods based on Fourier or Chebyshev representation. Unfortunately, for fundamental problems with complex geometries, the full spectral approach is no longer viable, especially for large-scale CFD simulations. It is well-established that high-order finite-difference schemes on structured meshes is now a credible option to spectral methods for large scale CFD simulations. This is mainly due to their ability in resolving small turbulent scales close to the mesh size, their low truncation error, their easiness of implementation, their simplicity and their low cost. This mini-symposium will present several tools based on high-order finite-difference schemes to perform large-scale high-fidelity simulations of turbulent flows with practical applications.
Topics of interest of this mini-symposium include, but are not limited to:
- Implicit high order finite difference schemes
- Penalty methods for simulations of complex geometries
- Porting finite difference codes to modern hybrid architectures
- Applications to engineering size problems