The work group led by Prof. Claus Dieter Munz has developed the FLEXI flow solver. It is regarded internationally as one of fastest, most efficient programs for simulating turbulent flows at high resolutions. FLEXI has been deployed successfully in industry-related research projects and now is available as open source.
Outside rear view mirrors not only affect a vehicle’s air resistance but also it road noise emission. While driving, very complex flow patterns that depend on the wind and vehicle speeds develop around a mirror of this type. In a joint project with Audi AG, the work group around Prof. Munz has investigated a particularly undesirable phenomenon, the so-called mirror whistle. Munz is the acting head of the Institute of Aerodynamics and Gasdynamics at the University of Stuttgart, holds the professorship for numerical flow mechanics, and is a Principal Investigator in the SimTech Cluster of Excellence. The acoustic waves that the mirror generates by interacting with the flow at specific frequencies create resonances that produce the mirror’s whistling noise. “Audi had developed a mirror for which the computer simulation showed no mirror whistling while the finished prototypes when they were tested in fact very much did,” is how Munz described the reason for the joint project.
The contradictory results originate in the numerical simulation: the software used by industry is not up to handling, spatially and temporally, at a sufficient resolution the complex patterns that develop in turbulent flows. On the contrary, simulation programs work with averaged values and by modeling the turbulence. Explains Munz, “FLEXI, on the other hand, is capable of carrying out large-eddy simulations, meaning that it can locally resolve a flow spatially and temporally.” While this method of simulating turbulent flows on the computer costs much more, it is also much more precise. It gives insights into substantially more complex flow phenomena than standard simulation software – in fact, even the mystery of what made that mirror whistle.
The project participants simulated the problem at the Stuttgart High Performance Computing Center (HLRS). The nearly 3,300 kernels of the Cray Systems XC40 Hornet supercomputer with 44 million degrees of freedom per variable calculated 12.5 million time steps. “That means it took a night of computing, which is commonly still acceptable for industry,” said Munz. The simulation’s result: an edge on the mirror geometry was the culprit causing the mirror to whistle. “Even more important than the individual finding is that we proved that large-eddy simulations produce results for flows from which industry can directly profit.”
This also emerged in the BMBF (Federal Ministry of Education and Research) project HONK (www.honk-projekt.de), in which Robert Bosch GmbH and the University of Stuttgart Visualization Research Center (VISUS) participated with IAG and HLRS. In this recently completed project, the partners studied how to make it feasible for computing power commonly used in-house by industry to handle large-eddy simulations for industrial problems. Serving as application examples were diesel flows in injection piping, injecting of natural gas into an engine, and the established manufacturing technique of deburring of components with water jets. “For this,” Munz says, “we had to extend FLEXI to general state equations, so that we could also handle fluids in addition to gases.” That allowed the project partners, for example, to simulate cavitating turbulent flows in real-time, which let them also capture the pressure waves generated by the imploding cavitation bubbles, something that is not yet feasible with commercially available programs. In cavitation, steam-filled bubbles form in a fluid whose implosions near a wall can lead to erosion of the material. With high-resolution simulations like HONK, it is possible to study techniques for keeping such bubbles away from components by manipulating flows and thus minimizing the damage from cavitation.
“It’s not just the computations that matter with such simulations but also the visualization of the data,” Munz emphasizes. Picture this, for example: in a simulation comprising 50,000 time steps and that computes 17 million values for each of five states, the output consists of more than four trillion numbers! “Given this mass of data, it is very important that we adapt the visualization already in parallel with the computation.”
Bosch, by the way, was very much sold on the results of the HONK project. Not least at the company’s urging has FLEXI in the meantime become available as open source. Bosch hopes that a broader base for the simulation program will result in industry, so that FLEXI’s development continues not just at the University of Stuttgart. FLEXI is already being implemented by research groups at the universities in Cologne, Munich, Bochum, and Milan. In the Munz work group in the IAG, at present ten projects based on FLEXI under way. The researchers want to run simulations with it on the Stuttgart super computer for turbulent flows, which to date have not been simulated in satisfactory fashion. In addition, they are expanding its use to other, more complex physical models, such as multiphase and multicomponent flows, that is, flows in which several aggregate states (gaseous/fluid, condensation/evaporation) exist simultaneously or which, for instance, consist of differing gases or fluids.
Source code, documentation, and application examples: www.flexi-project.org
By Michael Vogel