In the 1950s, James Arthur Nicholls proposed a novel engine design for rockets. Instead of propelling rockets via conflagration (burning fuel), he contended, one could propel rockets via detonation: combustion that creates a shockwave, building its own pressure, nullifying the need for the compressors used in traditional combustion engines and substantially increasing efficiency. Calling this design a rotating detonation engine (RDE), Nicholls prototyped it until it stalled due to difficulties in studying the volatile engine concept. Now, researchers at the University of Michigan have helped to resurrect the concept using powerful simulations run on Summit, the most powerful supercomputer in the country.
“So many of the techniques we have for conventional jet engines and gas turbines don’t work in these kinds of extreme environments. So simulations are the only way to go. There’s no way around it,” said Venkat Raman, the professor of aerospace engineering at the University of Michigan who is leading the project, in an interview with Coury Turczyn at Oak Ridge National Laboratory (ORNL). “However, simulating the complicated physics in RDEs is very challenging. These simulations have as much physics as the most complicated problem you can think of—fluid mechanics, shockwaves, chemical reactions, a heat transfer to the wall.”
In collaboration with several other organizations, Raman and his team began to run these complex simulations – at first, only putting around 30 percent of the simulations on Summit, which, despite its 148.6 petaflops, didn’t deliver a substantial speedup. Researchers at Oak Ridge National Laboratory, which hosts Summit, then helped the researchers completely rewrite the simulation code they were using for optimization on Summit. With 95 percent of the code now running on the massive supercomputer, the team further boosted the simulations using neural networks trained on Summit’s GPUs.
“To me, Summit is a step change. Its compute power is something I did not expect. Even the biggest calculations that we thought we were going to run would fit on 20 nodes of Summit – and Summit is over 4,000 nodes. So we really had to increase our ambition once we got access to the machine,” Raman said. “I think that sort of step change requires you to rethink how you do your simulations and what simulations can actually be done. We changed our algorithmic frameworks, we changed our codes, we even changed how we answer the questions.”
This new scope has allowed Raman and his team to approach the full second of simulated time considered necessary to meaningfully represent the functions of RDEs. “Such calculations would take months on a CPU-only machine. With Summit, you’ll finish the simulations in about a day,” Raman said.
RDEs still face a steep uphill climb to commercial viability, with many more simulations and experiments necessary before the engines could see practical use in the field. Still, Venkat Tangirala, RDE program manager at GE Research, cites Raman’s simulations as a powerful step forward for the experimental design.
“Venkat Raman’s advanced code for Summit stands head and shoulders above anybody else’s. He’s delivering stunning results,” Tangirala said. “We have the rotating detonation technology and test rigs here for experiments. And now we also have his computations going hand in hand to help us understand what our tests are telling us. I really don’t know how many research institutions are set up to do that, but that’s how we’re making tremendous progress.”
To read Coury Turczyn’s article discussing this research, click here.