As a CFD analysis consultant I have been subjected to the question "Is CFD Exact?" many times. How exact is exact,exactly? Confusing isn't it? The answer really depends on CFD analysis objectives...

I've been asked to carry out results data to 3 or 4 decimal places!
Really? Do results showing 4.015 psi drive product development in a
different direction than 4 psi? If so, you don't need CFD simulation,
you need a professor and enough capital to embark on a 6 month research
project. Don't get me wrong, CFD analysis can be
exact! However it requires a high degree of scrutiny and more time to
capture every minuscule detail. I come across many engineers who feel
the need to include every detail is required to derive any value from
simulation. This cannot be further from the truth. CFD can be exact
but CFD does not have to be exact to drive product development
decisions.

CFD results should always be verified with hand calculations and
various sanity checks to ensure their validity; however, CFD being a
mathematical model constrained by user inputs is inherently different
than the physical world so differences in results should be expected.
If an initial baseline analysis shows a 10% discrepancy with physical
testing does that mean there is no value in the analysis results? What
about the power in visualizing the flow fields, temperatures gradients,
and other trends not accessible through physical testing or hand
calculations? This insight, and it’s interpretation,
will provide the information required to make informed design
modifications. As long as analysis assumptions are maintained across
design configurations and sound engineering judgment
is used, the result yielded will be a qualitative comparison of which
design is better. Given enough time and resources CFD can be spot on;
but it does not have to be if developing a better product (performance,
cost, weight, ...etc.) with less time and testing is your goal!

Right on. I'll pile on with this question: "Exactly as accurate as what?"

What's the actual accuracy of your pressure gauge in the lab? If you run the physical test 100 times, what is the spread of results even if you had an impossibly accurate pressure gauge?

Posted by: Jeff Waters | 04/09/2013 at 10:58 AM