Fluid Flow Simulations with Two-Phase Effects on Next Generation Supercomputers
Malte Hoffmann, Claus-Dieter Munz
Institute of Aerodynamics and Gas Dynamics, University of Stuttgart (Germany)
Local Project ID:
HPC Platform used:
Hazel Hen of HLRS
Flow simulations play an important role in many areas of industrial product development, as the properties of components can be predicted before the first prototypes are produced. Accurate simulations of large systems such as an entire aircraft, however, demand the computing capacity of the largest supercomputers and many problems are too complex even for these powerful machines.
Until a few years ago, computer performance could be raised by increasing the clock speed of their processors. This development has now come to a standstill. To make large supercomputers, as well as ordinary desktop PCs faster never-the-less, they are equipped with more and more processors that work in parallel. Modern supercomputers already come with hundreds of thousands to millions of calculation cores, and this number will increase significantly in the future.
Software that is to run efficiently on such systems must be carefully adapted for this purpose. The overall problem, which is to be solved in the simulation, has to be decomposed into many small sub-problems and distributed to the cores. At the end of the simulation, the partial solutions have to be reassembled into a total solution.
A team of scientists from the Institute for Aerodynamics and Gas Dynamics as well as the Visualization Research Center of the University of Stuttgart and the High-Performance Computing Center Stuttgart has worked on such a code in a project supported by the Federal Ministry of Education and Research. With this software, gas and liquid flows can be simulated very precisely and very efficiently on supercomputers.
The group concentrated on the particularly challenging case of flows in which liquids and gases occur together, for example, because parts of a liquid spontaneously vaporize by a pressure drop in a process called cavitation.
Cavitation is a widespread phenomenon, it occurs in the normal coffee machine for private households as well as on ship propellers of the largest freighters. It can lead to damage on many components because the cavitation bubbles recollapse, resulting in very high pressure. It is therefore vital to gain a deep understanding of this phenomenon through simulations.
In addition to the actual simulation, the scientists also dealt with the question of how the resulting enormous data sets can be made manageable and users of the software can gain as much information from their simulation as possible.
At the end of the project, the developed software was released as open source code and is now available to other scientists and product developers.
The group has performed several simulations on HPC system Hazel Hen, the Cray XC40 supercomputer of the High Performance Computing Center Stuttgart, which is one of the largest and fastest supercomputers in the world featuring more than 185,000 processors. The research team already gained valuable insights into the physical processes that occur during cavitation. Figure 1 shows water flowing through a throttle. Due to the acceleration of the fluid the pressure drops and cavitation occurs. These cavitation areas are producing high pressure waves during collapse. These waves can lead to damage on the throttle itself.
To reduce the analysis time of such big simulations, an in-situ algorithm was developed at the Visualization Research Center which detects the collapses.
Figure 2 shows such a detector image which gives information where most of the collapse occur. With this information the investigated component can be redesigned to reduce the effect of cavitation.
HONK project webpage:
Selection of publications:
A Robust High-Order Discontinuous Galerkin Solver for Fluid Flow with Cavitation (http://iccfd9.itu.edu.tr/assets/pdf/papers/ICCFD9-2016-196.pdf)
Simulation of real gas effects in supersonic methane jets using a tabulated equation of state with a discontinuous Galerkin spectral element method (http://www.sciencedirect.com/science/article/pii/S0045793016304078)
Toward a Discontinuous Galerkin Fluid Dynamics Framework for Industrial Applications
Dipl.-Ing. Malte Hoffmann, Prof. Dr. Claus-Dieter Munz
Institute of Aerodynamics and Gas Dynamics
University of Stuttgart
Pfaffenwaldring 21, D-70569 Stuttgart (Germany)
email: munz [at] iag.uni-stuttgart.de