Natural gas-fired power plants were the largest source of electricity generation in the United States in 2020, constituting about 40 percent of the country’s power supply — and beyond electricity, gas-fired turbines power hospitals, factories and more. Unfortunately, difficult-to-predict oscillations can damage turbines over time. Now, supercomputer-powered research at Oak Ridge National Laboratory is helping to predict – and prevent – those oscillations.
The research team was composed of software developers from Cascade Technologies and engineers from Caterpillar subsidiary Solar Turbines – which, contrary to what one might expect from the name, designs and manufactures industrial gas turbines. The team modeled a 14-burner gas turbine combustion chamber, capturing the turbulent flow inside in order to predict the thermoacoustic oscillations that can vibrate turbines into dysfunction.
“At the extreme, these instabilities can shake a system apart, and they often don’t appear until late in the design cycle when all the components are built and assembled, and a physical test of the full engine is performed,” explained Yonduck Sung, the senior principal engineer for Solar Turbines who led the study, in an interview with Matt Lakin of Oak Ridge National Laboratory (ORNL). “At that point, we may have to start all over.”
But there was a problem — Solar Turbines’ computational resources were only enough to model one burner. “We need to be able to spot the problems before they start and before we cut metal,” Sung said. “But understanding and predicting these combustion instabilities using high-fidelity numerical models requires very large computational resources that far exceeded our in-house computing systems. We needed more computing power.”
So, to run these simulations, the researchers turned to the most powerful publicly ranked supercomputer in the United States: Summit, ORNL’s IBM-built, 148.6-Linpack petaflops system, which has also ranked second in the world on the Top500 list for two years running. Using the Cascade Technologies-developed charLES computational fluid dynamics code, they were able to model 14 operating burners on Summit.
“This kind of problem couldn’t be run anywhere but on Summit,” said Sanjeeb Bose, CTO for Cascade Technologies. “These large-eddy simulations would have taken months to run on a conventional computer. Summit’s world-class speeds shrank that time to a matter of days.”
The simulations successfully predicted combustion-induced oscillation for the first time.
“These insights hold paramount importance for our efforts to continue delivering reliable gas turbines without compromising emission standards and performance,” Sung said. “Simulations like these can speed up important discoveries by replacing expensive physical tests and reducing time in the design cycle.”
To learn more about this research, read the reporting from ORNL’s Matt Lakin here.