"Tying Two Forces" Feature Article, May 2007

tying two forces

Software upgrades have made analyzing for fluid and structural interaction easier and more common than in the past but, alas, not much quicker.

by Jean Thilmany, Associate Editor

Only a decade ago, analyzing the myriad forces and flows that came together to inflate an airbag could be a lengthy and complicated undertaking.

Flash forward 10 years and mechanical engineers can now commonly combine fluid and structural analyses to best find how these forces, acting in tandem, affect an object. In fact, so common is fluid-structure analysis today that many finite element analysis packages offer ways for an engineer to model both the structural and fluid forces that affect a design, said Bob Williams, product manager at Algor Inc. of Pittsburgh.

But there's still work to be done in this field, including speeding up solving times and analyzing fluid-structure interaction in one fell swoop, rather than passing it back and forth between FEA and computational fluid dynamics programs joined by an interface, said one consultant. Some industry vendors agree.

In perhaps five years, when those problems have been resolved, most mechanical engineers will be able to solve for FSI without giving it a second thought, according to Williams.

Fluid-structure interaction, or FSI, analysis lets mechanical engineers study how fluid flow around or through a part or assembly can affect performance. Say an engineer is designing the fan that circulates air through a computer to cool electronic parts. To make sure the fan effectively cools electronic parts, the engineer must look at how fan design affects airflow through the computer.

When Predictive Engineering was asked to certify the safety of a private 10-passenger submarine (above), chief technologist George Laird ran FEA, CFD, and fluid-structure interaction analyses (below) to ensure that ocean pressure would in no way deform the vessel's hull.

Analyzing the way an airbag inflates is a classic FSI example. Engineers need to study exactly how the bag will inflate, as its position in a crash could save—or take—lives. Here, airflow affects how the bag, a structure, will inflate. The aerospace industry also does a lot of FSI. Analyzing how air flows over the wing of a plane is another popularly cited example of a common FSI problem, Williams said.

Although FSI problems have existed from time immemorial, the past 10 to 15 years have seen engineering software vendors offering methods to help solve those problems more easily.

"It used to be that you would do heat transfer, do fluid flow, do structural, and look at each result separately," Williams said.

So, in the past, most companies solved FSI problems by running them first in their FEA program, then manually passing the finding to the computational fluid dynamics package—and the engineers who dealt with CFD—so the fluid aspect could be solved next. Then, the CFD finding would have to be sent back to the FEA application to be updated with the new information. The updated problem would need to be repeated in FEA, passed to CFD, and so on until the analysis was complete, Williams said.

To help end the hassle, vendors began creating applications that closely marry FEA and CFD, so results don't have to be laboriously transferred between programs.

Today, for example, Algor's FEA software offers the capability to solve for FSI, Williams said.

Earlier this year, Abaqus Inc. of Providence, R.I., and cd-Adapco of London announced that they'd joined their finite element analysis and computational fluid dynamics products—Abaqus and Star-CD, respectively—to allow engineers to solve FSI problems. Abaqus already partners with Fluent Inc. of Lebanon, N.H., to offer the same FSI capability between FEA and CFD packages. Ansys Inc. of Canonsburg, Pa., the maker of multiphysics analysis software, owns Fluent, so its software is designed to work with Fluent's as well.

The interface allows FSI problems to be solved by one group of users working with the software application they know best, said Dale Berry, the Abaqus director of industry solutions. Companies no longer need both their FEA and CFD analysts to study FSI problems. One is enough.

"Some of our users had worked with Abaqus for things like powertrain simulation for years now and they also used Star-CD within their fluids group to look at things like cooling," Berry said. "So these are two different user groups with different expertise within different domains. We've found these types of companies don't want to necessarily bring groups together to solve FSI. They just want to solve FSI problems where structures guys can focus on structure and fluids guys can focus on fluids without learning another code."

In other words, both of these groups can solve FSI problems using the software—either Abaqus or Star-CD—they're most familiar with. Although the results will actually be passed between both packages for solving, the user doesn't notice this aspect, which happens seamlessly—not manually. Analysis results are returned in their home application. Designs can be changed within that package, if necessary, and analyzed again.

Berry pointed to an Abaqus customer, Vernay Laboratory Inc. of Yellow Springs, Ohio. The company makes slow-control valves that include a rubber nozzle inside each valve. As the fluid pressure rises inside the valve, the nozzle expands to allow the valve to maintain a constant flow rate regardless of inlet pressure. Vernay's engineers use FEA software to determine how the nozzle will operate within the valve under certain conditions. They use Star-CD to solve for fluid flow.

The companies' cooperation now lets mechanical engineers solve those types of problems without needing to hand over FEA results to the fluids side, waiting for fluids to solve it in Star-CD and pass it back to the engineers to be analyzed again.

The Handoff

The problem is, applications still solve for fluid flow and structural deformation separately. An interface allows for quick handoff from one package to the other, but the passing back and forth and the separate solving, while not manual, still take considerable time, according to George Laird, chief technologist at Predictive Engineering, an FEA consulting service in Portland, Ore. He encounters FSI problems on a near-daily basis.

While vendors have made great gains in easing an engineer's FSI headaches, the industry still has room to improve, Laird said.

"FSI is moving along, but it's still not at the holy grail," Laird said. "While programs like Ansys and Fluent are trying to make it one-stop shopping, you're still passing it back and forth, Ansys to Fluent or Abaqus to Fluent. The holy grail is: Flip the switch, and it solves everything in one gigantic chunk."

Some packages, notably LS-DYNA from Livermore Software Technology Corp. in Livermore, Calif., does solve for both fluid flow and structural deformation simultaneously, said Laird, who often turns to this application when solving his own FSI problems. Depending on his needs, however, he also uses programs like Femap or NX Nastran from UGS of Plano, Texas, and CFdesign from Blue Ridge Numerics Inc. of Charlottesville, Va. With these applications, he'll frequently pass results back and forth to be updated and solved, for a complete FSI solution.

For instance, Laird recently was asked to certify the safety of a 10-passenger submarine for a private customer. The 40-foot-long vessel was designed to operate in depths to 1,200 feet. The structure would include bigger-than-normal viewing portholes, and it had to be light enough to be carried on the yacht that would take it to its launch site. The unique design deviated from the standard American Bureau of Shipping code for hull thickness, frame stiffness, and porthole and hatch design.

Laird needed to ensure that the ocean's pressure wouldn't adversely affect the hull, the frame, or the porthole, which made this an FSI problem, he said.

Fluid-structure simulations rely on both FEA and CFD results to demonstrate how a fluid and a structure would affect design. FSI is used in industry, to model how liquid moving through a pipe or valve, for example, might interact with the structure of that vessel, as indicated above.

For another FSI example, Laird points to a company that asked him to find the best speed and design to move a conveyor belt under an ultraviolet laser system. The twist in the case that put it firmly in FSI territory? The belt was studded with nubs that affected the conveyor's movement.

"So as the belt moved, it created turbulence under the leading and trailing edge of the bumps," Laird said.

The tolerance between the laser system and the top of the bumps was tight, and any reduction in airflow that could be made would aid conveyor movement, Laird said. For this project, he used CFdesign to give motion to the structure and look at the airflow around the nubs, and NX Nastran, the FEA package, to study how the shape of the nubs affected flow. He used Femap and CFdesign for the submarine project.

FSI is also helping Bechtel Corp. in an environmental cleanup project. The San Francisco-based engineering company is building a nuclear waste treatment plant for the U.S. Department of Energy in Hanford, Wash., site of the first and most extensive U.S. nuclear defense production program, according to the DOE.

The Hanford project's goal is to stabilize nearly 53 million gallons of radioactive and chemical waste for more than 10,000 years. The area is the former site of the Hanford Engineering Works, which in 1943 began producing plutonium for atomic weapons. By 1964, nine plutonium production reactors were operating at the site, all of them on the banks of the Columbia River, according to the Bechtel Web site.

Hanford Engineering stopped plutonium production in 1989, after the fall of the Berlin Wall. Now, these millions of gallons of radioactive and chemical wastes are stored in 177 underground tanks built from the 1940s to the 1980s. The tanks were designed to last 20 years, according to the Web site. Radioactive waste has leaked from many of the tanks, contaminating the groundwater and potentially threatening the Columbia River, which could affect people downstream in Portland, Ore., and other cities. It is this tank waste that the new waste treatment plant will process and immobilize for safe long-term storage.

Seismic Design Standards

Laird said the effort is analogous to building a big chemical treatment plant, albeit one that must be stabilized against the earthquakes that sometimes rock southeastern Washington. That's where FSI comes in: Ensuring that tanks won't crack or leak when liquid sloshes in them is an FSI analysis problem, Laird said. In other words, the liquid can't affect tank structure and tanks can't crack, even if both tank and liquid are jostled with great energy.

On April 30, 2006, the CBS program 60 Minutes claimed that Bechtel ignored warnings about the need to upgrade seismic design standards, according to the show's Web site archives.

Bechtel responded on its own Web site that the impact of changing seismic criteria on the project has long been a matter of public record. DOE provided Bechtel with original ground motion criteria for the design. The DOE had commissioned a study in early 2004, which resulted in the department's decision to increase the seismic requirements by nearly 40 percent.

Because its original plant design met those seismic requirements, Bechtel didn't remove or redo any construction work, although it did review tens of thousands of design documents to ensure that they meet the new standard, according to the site. FSI studies are done as part of that review, Laird said.

Although the past decade has seen a flurry in FSI technology, there is still room for improvement. Declining hardware costs should help speed FSI solver times.

Although the past decade has seen a flurry in combined fluid and structural analysis technology, it can take a long time to solve these types of problems, particularly for the applications that analyze these problems by handing them back and forth between structural and fluid packages rather than solving jointly, Laird said.

"The iterative approach can be insanely long," Laird said. "It can take weeks."

Of course, if your model is simple, you may need to make only one run. But the problem to be solved is almost never simple.

FSI models are complex by their nature, Laird said.

"But complexity also depends on the number of assumptions you're making about how things are flexing and bending," he said. "So it could take 24 hours on a supercomputer to get a good run and you might need 20 runs. It's the solution times that kill you.

"You solve it and that's a week, then you go back and tweak something and that's another week," Laird said. "In a design environment where you want to look at several design iterations, it could be several months by the time you're done."

But both Laird and Williams say that declining hardware costs should help speed FSI solver times in the near future. Splitting up the problem to be solved across multiple computer processors—parallel processing—can also cut solution time, the two said.

Because FSI solutions are so computer intensive, hardware costs play a role in making FSI more manageable. And hardware costs have been coming down in the past five years. A computer that now costs $1,000 would have run about $10,000 only five years ago, Williams said.

That pricing structure gives even small companies a crack at FSI, Laird said. It's those declining costs in addition to ramped up software that will allow nearly all mechanical engineers to routinely solve for FSI in the not-so-distant future, Williams added.

"In another five to 10 years I don't think designers or engineers will even think, 'I'm doing a fluid-structure interaction problem,' " Williams said. "They'll just think, 'This is my product; this is what the environment it's used in looks like,' and they'll let the software go do whatever behind-the-scenes stuff it needs to do.

"We've gotten closer to that type of system already," he added, "and there's no reason to believe we won't be there in five to 10 years."