Best practices for CFD simulations

Senior Editor

Designers increasingly run their own flow and heat-transfer simulations rather than sending them out to analysts. Formal classes are the best place to learn a CFD program. But even after the classes, users need help getting the most from the simulation software. That help, in the parlance of management gurus, is called best practices.

Ed Williams, CEO of Blue Ridge Numerics and developer of Cfdesign software in Charlottesville, Va., offers several best practices for CFD users. He cautions though, "Giving a person software does not replace the need for engineering training and experience. You have to know when temperatures are too high and flow rates too low. CFdesign is a tool that requires practical engineering knowledge for its best use.

Product-development engineers usually have the necessary practical knowledge. What they tend not to have is lots of experience and familiarity with CFD," he says. To make best use of the software, Williams suggests the following.

Use simulation software at the project's front end. Many companies still use CFD software late in the design game. Their argument is that there is no time to apply it early on. "Simulation works best in design trade-off studies with physical testing being left for final verification. If the design team uses simulation tools correctly, it need test only one physical prototype and avoids major rework. What's more, finding the most innovative and best design is more likely to happen in a digital environment. It lets teams do what-if brainstorming. Qualification or validation is still best done in the physical realm on a flow bench or test rig. It's important to apply the right tool at the right time," says Williams.

Describe the simulation's goal. "Engineers often tell us what they want to simulate by showing us their CAD model. Actually, we must first understand the simulations' goal. Without this it is difficult to define a reasonable approach to the problem. When the goal is uncertain, engineers need to step back a bit and call on their engineering knowledge," says Williams. The main question is: What do we want to learn? "With a little prodding, engineers get specific and say they want to find, for example, if stuff will deposit near the valve seat and block its operation. Simulations need focus," he says.

Build an appropriate model. Great detail is usually unnecessary early on, but the detail should be sufficient. "The biggest payback comes by using simulation early on when all details have not been worked out. This lets engineers verify that their conceptual approach to dealing with the flow or thermal issues seems reasonable. If it is not, then even radial changes to the design approach can be explored because many of the details, tooling, and PCBs have not been developed," says Williams.

Of course, there are cases when details also cannot be left out. "An engineering client opened an avionic box we analyzed in CFD to reveal the top portion was just a nest of cables with stuff everywhere," he says. "Then the guy admonished us because none of the cabling was in the digital model and the simulation showed air freely circulating at the enclosure top. But this was the first time we had seen the actual enclosure, and the blocking cables and components were not indicated in the CAD assembly that was used in the simulation. 'Was the cabling important to the simulation?' the engineer asked? Yes, the detail was important."

Building a good model takes some thinking. "You wouldn't model that wire nest, but you can apply a free-area ratio to its volume," says Williams. So if the cabling fills about 70% of the space, apply a 0.30 free-air ratio in an appropriate input field to simulate the flow blockage. This capability is also sometimes referred to as a distributed resistance.

Solve for airflow first and the thermal problem later. Just see where the air is going. "You know it has to flow around important components such as processors and hard drives, so see if it's doing that," says Williams.

He tells of another company that simulated computer equipment enclosures with a sheet-metal case and punched holes in the side. They'd add thermocouples and heaters and then tape some holes closed to see how and where internal temperatures changed. The exercise did not take long with an in-house sheet-metal shop, and results were adequate.

Eventually customers wanted plastic enclosures, so the team had to move towards molding them. Because the cost of building the mold is high, it was now critical that they got the thermal design right on the first pass. "Fortunately, the engineers could do in our CFD software just what they did in sheet metal: Model the enclosure and vents and assign different free-area ratios on both sides to simulate vents. And then just as they did with tape, vary the free-area ratio and see where the air goes. These simple models run quickly," he says.

Build a concept model first. "Solve the simple things first and then go on to more complex ones. Find actionable bite-sized places to apply CFD and build on success," he says. A concept model might be just a plane for the PCBs, or lumped parameters for hot devices. One initial goal is to see where air goes and determine initial temperatures of the most important components.

A variation on this practice focuses on the things you can change, especially for late-in-the-game simulations. Identify the adjustable parameters and experiment with them.

Allow sufficient time for a simulation. On one occasion, Williams says a customer showed him a 700-part model and asked for an analysis right then and there. Their perception after reading several magazine articles on the topic and looking at the result pictures is that they should be able to knock out a simulation in 30 min before lunch. "On a model of that complexity, it just won't happen in 30 min. We are certainly working to that goal but that is not the reality today. A simulation with such detail might take 4 to 8 hr," says Williams.

Expect to make a few assumptions. "Many companies switching from hand calculations, spreadsheets, and physical testing to simulations think they will be giving up assumptions and engineering estimates. That's because they mistakenly believe results will be exact or perfect. The software is a tool, not magic, and it's easy to get the right answer to the wrong problem. If you model the inside of an electronic enclo-sure and don't add any thermal boundary conditions on the out-side of the box, the mathematical model will ensure that the enclo-sure does not loose heat to the outside. In reality, heat even escapes through bottom surfaces that the enclosure sits on as well as through other surfaces. Pick up your laptop after it has been sitting on a desk and it will be clear that some heat has transferred to the desktop. A simulation model is a mathematical model and need not obey natural laws. It will do what you ask it to, such as making a surface a perfect insulator. If the model does not mirror its physical prototype, results from each will differ," he says.

Another misperception is that the software will tell engineering teams what to do next. "That feature is not yet in the planning stage," adds Williams.

MAKE CONTACT
Blue Ridge Numerics Inc., www.cfdesign.com