The turbulence created when diesel motors, internal combustion engines or gas turbines fire results in all manner of instabilities. Being able to model and predict those dynamic flows requires a complicated array of physics calculations.
The multi-physics presented by these kinds of turbulent reactive flows makes their behavior one of the world’s biggest simulation and modeling challenges, according to Joe Oefelein, a professor of aerospace engineering at Georgia Tech University.
“We’re not just talking about fluid dynamics or thermodynamics or heat transfer,” he said. “We have all those things going on at once.”
This translates to a high degree of mathematical complexity — and mammoth computational needs. Fortunately, Oefelein and his team have access to the most powerful tool ever created for just such an endeavor: the Summit supercomputer at the U.S. Department of Energy’s Oak Ridge National Laboratory in Tennessee.
Being able to develop models and run simulations on Summit is opening the door to designing more efficient and better performing propulsion systems, Oefelein said. Based on early experiments, he estimates that Summit can run his team’s code 25x faster than its immediate predecessor.
“If we can run two times faster, for any given problem, we’re happy,” said Oefelein. “Twenty-five times is extremely relevant to us.”
Supporting Complex Simulations
While continuing advances in computing hardware get some of the credit for that increase in performance, it’s really Summit’s architecture, which features more than 27,000 NVIDIA V100 Tensor Core GPUs, that packs the biggest punch.
Summit’s vast capabilities are making it possible for Oefelein’s team to more effectively simulate the processes that occur when fuel is injected and mixed with an oxidizer, resulting in combustion.
The objective, Oefelein said, is to be able to enhance or prevent certain phenomena with a goal of maximizing performance while minimizing emissions. Summit’s computing power lets the team improve the resolution of important features of the flow, enabling calculations that previously weren’t possible.
Eventually, makers of propulsion systems could tap into the resulting knowledge to inform their design processes.
“If we can improve the predictive capabilities of simulations, we can use those to understand how to make engineering models that industry can run very quickly and reliably,” Ofelein said. “As we move toward more predictive models, you’re going to see innovative design. We’re going to get improved fuel economy, reduced emissions and high performance.”
The team’s first simulations on Summit have been aimed at producing a dataset that captures the impact of turbulence on the mixing and combustion processes. Next, Oefelein will move on to combining that knowledge with other experiments to validate models with more powerful predictive capabilities.
Reconsidering Research Strategies
Working on Summit has enabled Oefelein’s team to completely rethink its workflow and its overall approach to research. Many limitations have been lifted, opening doors to potential advances.
Case in point: Tensor Cores, which are designed to speed training and improve deep learning performance. Oefelein said his team is planning to take full advantage of them as it improves its algorithms.
“That’s an example of a very desirable feature that exists now that didn’t before,” he said. “It helps us not only change our way of thinking, but it becomes like a new opportunity to take advantage of the hardware in a slightly different way.”
And if all goes as planned, it won’t be long before Oefelein’s team’s work is enabling propulsion system makers to take their technology in new directions as well, leading to cars, planes, boats and even rockets that produce fewer emissions.