Simulations using the OLCF supercomputer could lead to better, faster turbines for jet engines and more
Simulations performed on Oak Ridge National Laboratory’s (ORNL’s) Titan supercomputer could clear the runway for more efficient jet-engine turbines and help set a new benchmark for turbine design.
The study by an international team of scientists modeled air flow over a 3D turbine blade using the computational power of Titan, the now decommissioned 27-petaflop Cray XK7 at the Oak Ridge Leadership Computing Facility (OLCF), a US Department of Energy Office of Science user facility at ORNL. The simulations tracked billions of potential values to yield more detailed and accurate results than traditional approaches. Those results demonstrate the potential for computational simulations to one day help replace expensive real-world wind tunnel experiments, said Peter Vincent and Freddie Witherden, two authors of the study recently published in Computers and Fluids.
“Reducing or eliminating the need for wind-tunnel testing would be a transformational development, and this study demonstrates how it could be possible,” said Vincent, a professor of computational fluid dynamics in the Department of Aeronautics at Imperial College London. “The technology we are developing is very accurate and captures all the flow physics, including the turbulence. So, it could one day start to replace wind tunnels as the gold standard, helping make flight cheaper, cleaner, and more economical.”
The spinning rows of turbines in a jet engine capture energy from the exhaust gases created by the combustion chamber. The turbine blades recycle that energy upstream to power compressors, fans, and other auxiliary components. The unsteady flow of hot, churning gases over the blades can lead to turbulence and other complications that slow down or otherwise degrade performance.
“When designing turbines, there is a balance to be struck between maximizing aerodynamic efficiency, which can reduce with increased blade spacing, and minimizing overall weight, which increases with reduced blade spacing,” Vincent said. “Designers want to keep the blades close together to avoid unsteadiness in the flow and extract as much energy as possible, but they also want as few blades as possible in order to keep the engine light. To help strike the right balance, they need computational tools that can accurately capture the unsteady, turbulent flow physics in the vicinity of the complex blade geometries. That capability so far has been missing from the industry’s standard set of tools.”
Current designs rely on a combination of physical testing in wind tunnels and on approximations such as the Reynolds-Averaged Navier Stokes (RANS) simulations, which are time-averaged equations that offer a kind of shortcut to predicting turbulent flow. But wind-tunnel experiments can be expensive and time consuming, and RANS models too often fail to account for all relevant physics.
“Because you’re averaging everything out, you’re losing a lot of the unsteady physics,” Vincent said. “So, it’s certainly cheap but not necessarily trustworthy.”
Vincent and a team of fellow scientists from the United States, United Kingdom, Canada, Germany, and Japan received an allocation of computing time on Titan to model 3D flow over a turbine to better understand the flow physics. The study required the computational power of more than 180 million core hours on Titan—a supercomputer that was capable of 27 quadrillion calculations per second before it was decommissioned in 2019—and used 5,760 GPUs to undertake a simulation with 55 billion potential values using PyFR, a Python-based framework.
The computing time was supported by an allocation grant from the Innovative and Novel Computational Impact on Theory and Experiment, or INCITE, program.
“The OLCF was completely enabling for us,” said Witherden, an assistant professor at Texas A&M University. “Before Titan, this study wouldn’t have been possible. These were some of the largest and most accurate computational fluid dynamics calculations ever undertaken, capturing all the physics at very high fidelity. The scale of the problem was so big it simply would not have been possible to run it on any other machine. It took 2 years to set up and run the simulations, and then we analyzed the data.”
The increased level of detail revealed by the simulations could enable engineers to build better turbines and transform the design process, Vincent said. Succeeding generations of supercomputers offer top speeds that dwarf Titan’s and open possibilities for further simulations. Examples include Summit, the OLCF’s current flagship machine at 200 petaflops (200 quadrillion calculations per second), and Frontier, the nation’s first exascale system (more than 1 quintillion calculations per second) set for delivery to the OLCF in late 2021.
“The state of the art has moved on since we began, and we hope to ultimately deploy these tools as almost a commodity,” Vincent said. “In a wind tunnel, there are a limited number of instruments that can be set up to capture a limited amount of data, but with our computational simulations, we have access to the entire flow field. We could access anything, at least in theory. This rich data from these high-fidelity simulations could be used, for example, to train a new generation of improved turbulence models via machine learning approaches that improve the accuracy and reliability of RANS methods. This study demonstrates how it could be possible.
“These particular findings apply directly to increased fuel efficiency for aircraft, making aviation cheaper and more environmentally sustainable, but the technology in PyFR can be applied to a wide range of problems—wind turbine design, automotive design, ship and submarine design, and building design. They all exhibit unsteady flow physics, so they can all benefit from the technology we are developing.”
Related Publication: Iyer, A. S., Y. Abe, B. C. Vermeire, P. Bechlars, R. D. Baier, A. Jameson, F. D. Witherden, and P. E. Vincent, “High-Order Accurate Direct Numerical Simulation of Flow Over a MTU-T161 Low Pressure Turbine Blade.” Computers and Fluids 226 (2021). https://doi.org/10.1016/j.compfluid.2021.104989.
The research was supported by the DOE’s Office of Science. UT-Battelle LLC manages Oak Ridge National Laboratory for the DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, visit https://energy.gov/science.