Our research highlights serve as a collection of feature articles detailing recent scientific achievements on GCS HPC resources.
At its core, combustion is one of humanity’s oldest technologies. While this burning process helped the earliest humans cook food and defend themselves from predators, today’s combustion processes move cars down the road, help aircraft take flight, and govern countless industrial processes for making everything from tires to the white pigment used in furniture and paints. Ever since the industrial revolution, humanity has continued to harness the power of controlled burning to incorporate into new technologies.
However, as researchers and the public at large begin to reckon with issues posed by climate change and seek to limit the amount of fossil fuels being used in our daily lives, researchers have been studying combustion on a molecular level in hopes of making the process safer, cleaner, and more efficient.
To that end, a research group at the University of Duisburg-Essen (UDE) has been using Gauss Centre for Supercomputing (GCS) resources in order to study the critical milliseconds to microseconds when fuel is injected into an engine, observing how the flame ignites and changes over time. Ultimately, it wants to understand the chemical interactions taking place in molecular, nanosecond detail.
“When we look at combustion, we have several different aims,” said Prof. Dr. Andreas Kempf, head of UDE’s Fluid Dynamics Chair and principal investigator on the project. “We want to minimize the amount of fuel needed, ensure that there are no unburned hydrocarbons in the combustion reaction, and then, of course, minimize the amount of nitrous oxide and carbon monoxide emissions.”
In order to develop combustion processes that do all three of these things, Kempf and his collaborators are using supercomputing resources at the High-Performance Computing Center Stuttgart (HLRS) to run high-resolution, four-dimensional simulations of ignition processes.
The full picture
By its nature, the small, controlled explosions happening during combustion in an engine are hard to observe—they happen very quickly in an extremely hot, sooty, volatile environment, making it difficult to record or photograph the process.
However, researchers at the German Aerospace Agency (DLR) who collaborate with the UDE team use a method called laser-induced fluorescence to get a sharper image. Essentially, researchers expand a laser beam into a laser “sheet” or “light blade.” At ultraviolet frequencies, these light blades can slice through a combustion reaction, illuminating the many individual particles making up the chaotic, turbulent ignition process. For the most recent collaborative work, the UDE and DLR teams have been studying methane, a relatively well-understood fuel that has relatively simple reactions in comparison to diesel or other more complex fuels.
Even when taking snapshots at microsecond intervals, though, experimentalists are not able to see the whole picture of the ignition reaction. Think of it like a photographer trying to capture how a building looks from every side, but only getting a snapshot of each side at different points in the construction and demolition processes.
Simulation, however, can recreate ignition conditions in 3D and follow the many individual particles at nanosecond intervals, allowing researchers to put their model in motion and observe the many different particles simultaneously. In order to truly optimize combustion processes, researchers must not only see ignition process playing out in high resolution; they have to be able to chart how individual fluid particles and chemical “species” reactions influence the combustion reaction as a whole. In fact, when running their most recent simulations, the researchers observed how formaldehyde, a byproduct of the methane ignition reaction, has significant influence on the process.
“With simulation, we have access to every major and minor chemical species that play a role in the combustion reaction, which is very important for understanding the reaction” said Eray Inanc, doctoral candidate at UDE and research leader on the team’s most recent paper. “We can correlate velocities, strain, heat transfer, reactions, and species transfer of major quantities of particular materials in the reaction with minor quantities, such as formaldehyde, which plays a significant role in the overall reaction. Rather than just following one chemical species at a time, simulation allows us to see everything.”
The team ran two sets of simulation to compare the accuracy and computational costs of two different modelling approaches. The first relied on “tabulated” chemistry, meaning that the researchers generate a table describing the different thermochemical states at a given point in the ignition reaction, such as the amount of fuel or the amount of oxidizer (particles capable of taking new electrons in a chemical reaction). While this approach is computationally cheap, researchers introduce assumptions about the physics in the reaction, making the simulation less accurate.
The second approach, direct chemistry, tracks the many individual reactions occurring at each point in time. While this requires the computationally demanding task of solving transport equations for the chemical species in the simulation, it results in a much more accurate picture of the process. The team found that the difference in accuracy was worth the additional computational cost.
Kempf noted that without access to leading HPC resources such as those in the GCS centres, his team would be unable to make the same kind of advancements in its field. “In the field of turbulent combustion, you have a big transition happening toward new topics, and it is also a field where most researchers are coming from the world’s top universities,” he said. “To compete with our international contemporaries and competitors, we need access to truly high-end supercomputing power, and we are lucky this is possible in Germany due to the GCS initiative.”
With the team’s methane research completed, it looks to next-generation supercomputers to take their concept and apply it to more complex fuels. Kempf and Inanc indicated the current simulations provide the resolution necessary to get an accurate model of fuel ignition in simple fuels such as methane, but with next-generation computers, the researchers could study things like biofuels and diesel fuels. For methane, the team needed to run about 60 transport equations throughout the course of its simulations, but more complex fuels would require hundreds of transport equations due to the additional chemical complexity.
The team was able to effectively scale its code to take full advantage of the Jülich Supercomputing Centre’s (JSC’s) old JUQUEEN supercomputer, and effectively scaled on Hazel Hen. As HLRS prepares for its next-generation supercomputer, Hawk, to come fully online in the first half of 2020, Kempf is confident the team will once again be able to take advantage of the additional computational muscle.
“Our code has been ported to several different architectures, and we feel good about our abilities to port it to new architectures,” he said. “Our impression of Hawk is that it is a well-rounded, balanced system design. Much like how the GCS centres upgrade their systems in a ‘round robin’ fashion, we port our code to all three centres to ensure it remains portable, and we are optimizing it all the time.”