Faculty of Engineering and Physical Sciences, University of Southampton (U. K.)
Local Project ID:
HPC Platform used:
Hazel Hen of HLRS
The flow of air over aircraft wings is an important application for scale-resolved numerical simulations that is only just becoming feasible with improved algorithms and computer hardware. An improved understanding of the complex physical behaviour of turbulence, including transition to turbulence and the interaction of turbulence with shock waves, will pave the way to improved
engineering design. This includes reducing air resistance (and thus lowering fuel consumption), increasing the robustness of wings, and improving wing characteristics.
For passenger transport aircraft, the flow over the wing becomes locally supersonic, terminating in a shock wave. The changed flow physics relative to low speed flight leads to an additional source of drag, known as wave drag, which must be included in the design optimisation of complete wings. Additionally, these shock waves can strongly interact with the boundary layer, leading to local flow separation and eventually to an undesirable phenomenon involving low-frequency oscillations known as transonic buffet, which can affect the structural life of wings.
Professor Neil Sandham and his colleagues from the University of Southampton have been working on a PRACE-supported project that has been investigating shock-related buffet. Their simulations have gone beyond previous research on the topic, increasing the complexity of the flow and moving from low speed flow over the aerofoil to higher speeds where shock waves are seen. “Previous studies of buffet have been limited to the averaged equations of motion where what you have to do is add empirical turbulence models, which are unreliable when you have flow separation as you see in buffeting,” explains Sandham.
Using high-performance computing (HPC) system Hazel Hen hosted at HLRS, the researchers have run simulations where all of the scales of the turbulent flow are resolved (excluding the internal structure of the shock waves). Although they have not been able to simulate at the Reynolds numbers (the governing parameter of turbulent flow) seen in full flight mode, they have been able to reach figures that provide similar physics seen in flight. This provides useful insight into the physics of buffets.
The simulations carried out ran on around 30 000 cores, and were able to successfully capture the buffet process. “We saw variations in the boundary layer separation and the interactions with shock waves,” says Sandham. “One novelty we found is that these shock wave motions do not occur at the same frequency as the buffet. In the simulations we used large computational domains, so we are pretty confident that this is not the cause.”
The team was also able to run cases with modified geometries to investigate how this might affect the buffet process. In particular, the team was able to add a splitter plate to the trailing edge of the wing to see how the buffet responded. As well as this, the researchers investigated the use of simpler methods called large eddy simulations, where the large and medium scales of flow are resolved but the smaller scales are modelled. “The idea is to create a similarly accurate method that uses less computational resources,” explains PhD student Markus Zauner. “We were successful in creating a large eddy method that reproduced the buffet and used a factor of 16 times less computational resources than direct numerical simulations, although we did find that these were sensitive to resolution.”
One difficulty of the project was that it required a lot of effort to generate the grids upon which the calculations were performed. This labour-intensive process was part of Markus Zauner’s subproject, in which he was able to apply some error indicators and set up a grid generation system that allowed them to optimise the grid much more effectively than in the past. This software has now been released as open source and so is available to other researchers working within the same field.
“My work involved setting up the simulations and ensuring that we had a suitable grid to work with,” Zauner explains. “Then we had to set up the production runs. This sounds straightforward, but because we were producing quite a lot of data, we had to be very careful to establish a process to avoid loss of data.”
The amount of data produced during the project has been astounding, even compared to other High-Performance Computing projects. “In our simulations, we produced huge amounts of data, but of course you do not actually need all of it to reconstruct the three-dimensional flow dynamics,” explains Zauner. “You do not know at the beginning which parts of the simulations you are going to be interested in, so after you have done them you go through post-processing and find interesting examples that show particular flow behaviour. Even after all of this, though, we were still left with around 30 terabytes of selected data.”
Sandham and his team will continue to exploit the data produced by the simulations long after the project has ended, and smaller versions of their database will be released to the public in association with their publications. This will ensure that the results of the project get used as much as possible and will hopefully yield some more findings in the near future.
Zauner, M, Sandham, N.D. 2019 Modal analysis of a laminarflow airfoil under buffet conditions at Re=500,000. Accepted for publication in Flow Turbulence and Combustion.
Zauner, M., De Tullio, N, Sandham, N.D. 2019 Direct Numerical Simulations of Transonic Flow Around an Airfoil at Moderate Reynolds Numbers. AIAA Journal 57(2), 597-607.
Jacobs, C.T., Zauner, M., De Tullio, N., Jammy, S., Lusher, D and Sandham, N.D. 2018 An error indicator for finite difference methods using spectral techniques with application to aerofoil simulation. Computers & Fluids 168, 67-72.
Prof. Dr. Neil D. Sandham
Faculty of Engineering and Physical Sciences
University of Southampton
Southampton, SO17 1BJ (United Kingdom)
e-mail: n.sandham [@] soton.ac.uk
NOTE: This is a reprint of the article published in PRACE Digest 2019, p.18-19. The simulation project was made possible by PRACE (Partnership for Advanced Computing in Europe) allocating a computing time grant on GCS HPC system Hazel Hen of the High Performancee Computing Center Stuttgart (HLRS).
HLRS Project ID: PP17174149