A novel method of computing and modelling turbulent reacting flows developed by University at Buffalo researchers produces results equivalent to those generated by expensive supercomputers and is anywhere from 30 to 100 times less expensive to use.
The research by engineers in UB’s Computational Fluid Dynamics Laboratory is expected to have a major impact on how engineers conduct computational simulations of chemically reacting turbulent flows, such as those involved in hydrocarbon combustion and propulsion.
While knowledge of how turbulent flows affect internal combustion could greatly improve the efficiency and environmental impact of all kinds of engines, their complexity has remained an overwhelming hurdle to those trying to simulate them.
‘This methodology will revolutionise the way people compute turbulent combustion,’ said Peyman Givi, Ph.D., UB professor of mechanical and aerospace engineering.
The research is said to overturn the conventional wisdom in the field that held that a computational technique called Large Eddy Simulations (LES) will not describe complex reacting flows with the same accuracy as attained by Direct Numerical Simulation (DNS), which requires supercomputers.
‘You can use this LES approach and get DNS-type results,’ said Givi. ‘People have known about LES for a long time, but our results demonstrate for the first time that we are able to implement it for chemically reacting flows and get reliable results.’
Givi said that until now, two main approaches have been followed for computation of turbulent combustion: DNS and Reynolds averaging.
‘DNS provides incredibly detailed results on phenomena that occur over very small time and length scales,’ Givi explained, ‘but you need days and sometimes months of supercomputer time to obtain results and they are basic research-type results, providing information on every detailed aspect of the flow.’
Most engineers and engine designers cannot access supercomputers because of the time and expense involved, so they generally use a much older technique, Reynolds averaging, that originated in the late 1800s. Givi said the Reynolds averaging approach can only provide averaged or greatly ‘smeared’ results, and in many cases, they cannot be trusted.
The new approach taken by Givi and his colleagues is somewhere in between DNS and Reynolds averaging, he said. According to Givi, LES provides researchers with a way of obtaining the ‘smooth’ solution for a very ‘wiggly’ phenomenon, somewhat analogous to the way that a professional photographer can use air-brushing and other techniques to smooth out the sharp contrasts in an image. ‘LES is the solution of filtered values of the equations that describe the phenomenon,’ he said.
For example, he explained if you wanted to create a weather map for Buffalo, you would not need to know the temperature and barometer and other measurements for every single centimetre in the city. Instead, the readings for each square mile would be sufficient.
‘If you do it right, the filtered values will be the same as those obtained from the filtered DNS data, assuming such data are available,’ said Givi. ‘The advantage is you can perform LES on your workstation with a fast turnaround time and obtain reliable results.’
A number of laboratories, including Sandia National Laboratories and the Rolls-Royce Engine Company, have expressed interest in using the method.