Timing is everything | National Center for Supercomputing Applications at the University of Illinois
Timing is everything
05.02.12 - Permalink
by Barbara Jewett
NCSA's GPU machines allow researchers to improve the fluid dynamics codes used for rotorcraft design.
There's an old joke that has a flight instructor telling his student takeoff is easy, it's landing that's hard. Helicopter and tiltrotor pilots know that landing is even more difficult if you're battling the wakes generated by the rotors.
Unfortunately, rotor wake is a major problem in rotorcraft. That's why Earl Duque, manager of applied research at Intelligent Light, a New Jersey-based software company specializing in computational fluid dynamics (CFD) workflows, and Christopher Stone, founder of Computational Science & Engineering LLC, based in Chicago, used NCSA's Forge and now-retired Lincoln to develop and test a simulation software that would more accurately predict rotor wake interactions. The pair participated in a five-year project working on CFD research for rotorcraft as part of the Army's Vertical Lift Research Center of Excellence based out of Penn State.
Rotorcraft refers to anything that has vertical lift from a rotorunmanned air vehicles (UAV), helicopters, or the newer tiltrotors like the V-22 Osprey. The Osprey has a three-bladed proprotor, a turboprop engine, and a transmission nacelle mounted on each wingtip. For takeoff and landing, it typically operates as a helicopter with the nacelles vertical and rotors in a horizontal position. Once airborne, the nacelles rotate forward 90 degrees in as little as 12 seconds, converting the V-22 to a more fuel-efficient, higher-speed turboprop airplane. The U.S. Marine Corps began using Ospreys in 2007 and the Air Force in 2009. They have been deployed in Iraq, Afghanistan, and Libya in both combat and rescue operations.
"The big problem with helicopters is the interaction between the vortex wake coming off the rotor blades and how those vortex structures affect the lift behavior and drag behavior of the rotor blades themselves," explains Duque.
He notes that the disturbed air may further change as it passes over the fuselage of the vehicle, creating new wake structures. This can impact the vehicle's tail, causing control problems in the tail area.
The traditional way of developing new helicopters is to design one, build it, and then fly it to test it. It's not uncommon during this demonstration and validation phase to discover significant wake problems.
"One of the big problems whenever there is a new design for a helicopter," Duque says "is that inevitably there is a tail rotor shake that happens. That means the wakes come off the rotor hub or the rotor blades themselves, hit the tail, and cause problems with control of the vehicle. Problems like the pilot can't land."
To alleviate tail shake requires extensive wind tunnel testing and flight tests, which are costly. Thus the Army is sponsoring research to develop the software capability to predict tail shake problems in helicopters and UAV during the design stage. Being able to accurately model the wakes and predict tail shake will dramatically reduce the number of wind tunnel or flight tests required.
For their project, says Stone, they took standard grid-based finite-difference (Eulerian) CFD software and fused that with a particle-based method that tracks vorticity, the vortex particle method (VPM). Vortex particles are chunks of fluid that are free to move around to represent the motion of the fluid, as opposed to the grid-based system where one tries to solve the equation at each static grid point.
"We came up with the idea to combine these two disparate simulation methods that we theorized would enable us to better predict the wake interactions. 'Better' meaning more cost effectively and also more accurately," Stone says. "The grid-based methods have been in use for a very long time because the software was already there and the numerical methods are pretty good for modeling the flow pretty close to the fuselage body or the rotor wing. But farther away from the body the grid-based methods are not very accurate."
If the CFD used is not accurate enough, Stone explains, the vortex that's striking the tail rotor won't have the right physics thus causing the tail rotor shake. With the particle method, he and Duque were hoping to more accurately predict how a vortex would flow back and strike the tail rotor, providing improved predictions of the rotorcraft's handling.
One of the biggest challenges with developing the new code became apparent early in the process: the particle-based method turned out to be extremely expensive. It was not how the program was written, says Stone, "it was just mathematically a very expensive way of doing the calculations." But the GPU computing revolution was beginning, and the technology helped him quickly overcome the cost issues.
"The year we started this project, GPUs came into use. So we were able to make use of GPU computing to really get this project going. We started using Lincoln right when it was deployed. Now we use Forge," says Stone. "We were able to redesign our particle codes to use GPUs and honestly, without those GPU capabilities, we would not be able to do the simulations that we're doing."
Stone notes performance was not just a couple factors faster, there was a magnitude difference. But more importantly, he says, is that with access to Lincoln and Forge he and Duque were able to attempt simulations that were not previously possible.
Once the codes were modified for GPUs, they focused their attention on fundamental research, looking at the algorithms and how to couple the grid-based code with the particle code. They looked at fluid flow against a static wing, trying to predict some simple flows in order to validate the research. They then progressed to looking at loads on the rotor and fuselage in real-world flight conditions.
The past nine months they've been working on this with the ROtor-Body INteraction (ROBIN) fuselage shape, which is used by Army and NASA rotorcraft groups for code and measurement validation. Their original theory that the particle method is superior to predict the rotor wake far down stream was upheld with this project, notes Stone. But they are still working on the validation.
Another important issue when it comes to interaction of the vortices and the rotors, says Duque, is the audible noise generated from the helicopters themselves. A blade hitting a vortex, as well as the vortex generated by the blade ahead of itself, can cause an undesirable noise character the Army would like to eliminate. Currently there is no software capable of modeling those interactions.
"It's research," says Duque. "We've shown the methodology works, but there is still work to be done."
Although the team's work was focused on rotor-wake interaction on the same vehicle, Duque says they briefly looked at predicting wakes multiple rotor lengths downstream. They also looked at other potential applications for the code, such as wind energy projects where the wake interaction between turbines is really important. But, Duque says, "we haven't gone very far with that."
One of the interesting tangential outcomes from this research, notes Duque, is that while there are other people doing similar types of CFD work, in the rotorcraft world Stone was the first to utilize GPUs in the coupled Eulerian and vortex particle method codes. So in addition to being cutting edge on the method, they are also leading the way with the technology employed. Duque says he thinks that the knowledge they gained of how to use GPUs was one of the most beneficial aspects of the project. In fact, he says, he has made use of GPUs on other projects.
"We turned out a pretty open-ended research project," Stone says. "We came up with some good methods but it is not yet 100 percent. The main problem is that the grid-based codes tend to use a compressible flow, and the particle methods use an incompressible. So how to join those two methods turned out be extremely complicated to get the physics right. We've come up with a good way of doing that, which gives decent results. But it's still an open question of the exact mathematical way of doing it right. With access to GPU resources, though, we'll get there."
Project at a glance
U.S. Department of the Army