Why It's Important

Though we may not think of it on a daily basis, turbulence is a big deal. It's the key to transferring heat in a power plant so that it generates electricity more efficiently. It can allow us to keep ever-smaller electronics cool and working properly. It can also let engineers better predict how wave motions can damage marine-based platforms like offshore facilities. In general, controlling turbulence and using it productively can be the difference between efficiency and safety versus our systems getting torn apart.

An important concept in this work is the boundary layer. Whenever a fluid – whether a liquid like water or liquid helium, or a gas like air – passes through a solid channel, there's a thin layer of fluid that sticks to the walls of that channel and facilitates the transfer of heat and momentum between the channel and the fluid.

"… By understanding the full spectrum of instabilities that occur within the boundary layer, then we can come up with ideas to excite those structures that are conducive to either dynamic or heat-transfer improvements… And by moving them, [we] can then do cool things with them; we can abate the negative impact or… improve or enhance [performance] without the major penalty in terms of shear stress or skin friction." – Iman Rahbari, Purdue University

Research Scientist Iman Rahbari, working in the PETAL team of Guillermo Paniagua at Purdue University, wanted to use acoustic streaming to control the boundary layer and skin friction in a fluid traveling through a narrow channel. The idea was to create sound waves in the fluid to disrupt and change the specific near-wall flow patterns. To begin this work, he built a simulation using a unique one-two computational punch, leveraging the huge memory of PSC's Bridges and SDSC's Comet platforms and the huge number of processors on the Stampede2 supercomputer at TACC.

How XSEDE Helped

While Purdue offers powerful campus supercomputing "clusters" to its scientists, Rahbari realized that his simulation would require even more power. Initially setting up the simulation would require the computer to store large amounts of data ready for its processors to work on it – moving massive data between storage, as in the hard disk of a personal computer, would create bottlenecks that choke the calculations. Instead, the data needed to be stored in the RAM memory contained in each processing node. Bridges, with its 12-terabyte-RAM extreme memory nodes that offered nearly a thousand times more memory per node than found in a typical PC, was ideal for this step.

In the first stage, the scientists used Bridges to create the sound waves and complete the model. The next simulation splits into thousands of different pieces, all of them being computed at the same time to make the computations finish in a reasonable amount of time. The scientists ramped up their calculations by first running at lower numbers of cores on Comet and then expanding to tens of thousands of processors on Stampede2. Thanks to an XSEDE allocation, Rahbari was able to get computing time on all three systems.

"[The initial simulation] is very memory-intensive rather than CPU-intensive ... Our matrices are sparse and large, while solving the eigenvalue becomes larger and larger; our memory cost grows exponentially … Bridges was a massive help for us, we were able to [simulate] one of the largest stability problems ever solved at that time … This is an asset that is very unique to XSEDE."—Iman Rahbari, Purdue University

Rahbari set up his simulations to test different speeds of flow and different fluid resistance, or viscosity. He simulated a fluid moving through the channel at Mach 0.75 and 1.5, three-quarters of, or one-and-a-half times, the speed of sound in the fluid. By testing fluid flows with a Reynolds number – the ratio of inertial to viscous forces – of either 3,000 or the less-viscous 6,000, he sampled two interesting viscosity ranges. At the higher Reynolds number, the fluid would flow with high turbulence, with a lot of mixing. At the lower number, the flow would be unstable, with moments and areas of low-friction or laminar flow, and moments and areas of turbulence. By vibrating the walls of the channel to create sound waves, he was able to test different sound frequencies and volumes to see the effects on heat transfer and friction with the channel walls.

The simulations showed that he could tune acoustic streaming to maximize heat transfer between the fluid and the walls of the channel while minimizing friction and stress on the container. In the first peer-reviewed study that reported this new technique, Rahbari and Paniagua reported their results in the Journal of Fluid Mechanics in 2020. Today they are repeating their experiments in the laboratory, verifying that the computer predictions match real-world performance.

This work used the Extreme Science and Engineering Discovery Environment (XSEDE), via allocation TG-CTS170041, which is supported by National Science Foundation grant number ACI-1548562. Specifically, it used the Bridges system, which was supported by NSF award number ACI-1445606 at PSC; Comet, NSF award number ACI-1341698 at SDSC; and Stampede2, NSF award number ACI-1540931, at TACC.