[bstern]

Hey folks,

I was hoping that someone could offer me some direction on an issue I've been asked to investigate. We’re trying to determine whether our overall approach to CFD analysis is sound. It’s less a question of validation than of strategy and expectations.

We have several seats of CFDesign that we use for natural convection thermal analysis of electronic products. We follow the guidelines that Blue Ridge Numerics provides us, and frequently work with them on technical issues. While the results of the analyses are generally in the ballpark, and are good for comparing the relative performance of designs, they rarely match up with reality closely enough to rely solely on the CFD models to predict the performance of our products. So, we end up building bench mock-ups with power resistors in place of the real circuitry to predict more accurately what will occur. Senior management has asked the question of “if we’re building these bench mockups, why are we spending all this time and money on the CFD analysis tool?”.
Some folks have suggested switching to a higher-end CFD that would give us good enough results to stop messing with bench mockups. Others have suggested abandoning CFD analysis and just building more mockups. And others feel that we’re ok where we are, using CFD for evaluating conceptual designs, but looking to the bench mockups for the best prediction of real performance.

So, what I’m looking for is a source of information on recommended strategies for the use of CFD in a design organization (particularly for natural convection heat transfer). Any thoughts?

Thank you for reading my lengthy post.

[phsieh2005]

Hi, bstern,

What is the cost of building the bench mock-ups? This cost includes the cost of making the prototype, time required of making them (lead time). If the cost of building the bench mock-ups is less than (or about the same) running the CFD simulation, then, just go with the bench mock-ups. It is the real thing!

Even if the bench mock-up is more expensive than running CFD simulation, you still want to at least make one so that you can validate your CFD results.

There are a couple of reasons that we use CFD simulation:
1. cost several thousand dollars to build the prototypes and several days/weeks to run the tests. then CFD makes sense because you can narrow down to few concepts that are mostly to work.
2. some measurements are difficult to make. For example, mixing of two liquids. Or need to measure temperature at tight spots. Then, CFD can give you more details at any locations easily.

phsieh

[julien.decharentenay]

Hi,

Your problematic seems very interesting. At present you are CFD to compare designs and validate the design using mock-up models.

I would personally see your problem from the following angles:

1) Is CFD a cost-efficient process compared to mock-up models? In my opinion, such cost analysis should also include time component.

2) Is there a CFD approach/model/tool that would give you a "better" answer? That is a very open ended question (what is "better"?). A way to approach this would be to ask a few CFD software vendors to model an appropriate benchmark (blindly) and see what comes out of it.

Also I have no experience in electronic cooling, I do know that best practice guidelines, use of automated process, sensitivity studies are part of the answer to be able to get repeatable and accurate answers.

Good luck.
Julien de Charentenay

[lbossi]

Hello Ben,
interesting question.

I would second phsieh2005 and Julien de Charentenay answers and check weather it is cost-effective and if another simulation tool would deliver better results.

I am confident that with an opportunity for several seats, CFD vendors would happily demonstrate their capabilities with a free benchmark. Then you can use this to compare with your CFD results and actual tests.

In addition, you could contact other companies for similar products / problems and ask their engineers how they correlate CFD to test results. (During industry events for example). If this falls into competitive information, try some universities, there must be some doing research in your domain.


Hope this helps
-Lorenzo

[jchawner]

Ben:

First, I think eliminating mockups completely and replacing them with CFD is not necessarily a good approach. Maybe a better goal would be to use CFD to minimize the number of mockups you have to test. This of course assumes that the overall CFD process is cost effective relative to tests.

Of course, getting each individual CFD solution to be as accurate as possible is what everyone strives for. But also what matters when you do your validation studies is to determine whether a series of solutions gives you the correct trends. For example, if all your solutions are 20% too high relative to test data, but the increments (deltas) from configuration A to configuration B match data closely, that's good.

Best Regards

[jurich]

Hello Ben,

Interesting question. I agree with the discussion of relative cost others have contributed, and add the following:

You are developing a data set of (a) CFD model cases, (b) bench model case, and (c) real world installations. Compare the correlations of (a) vs. (b) vs. (c) over time to get your answer about approach.

You've basically said bench model (b) correlates well enough with real world (c) to run the business in the past. But so far, CFD (a) doesn't correlate well enough with (c) to run the business. Has it been getting closer over time? Are you continually getting better at it, or have you plateau-ed?

Also, your comparisons put you in a good position to figure out what might be limiting your CFD accuracy. Just guesses, but perhaps your configuration of boundary conditions, or what turbulence modeling you use? From that information you could make a judgement about improving your process or a possible switch in software.

Good luck,

Joe Urich