Is CFD Stuck in an Ivory Tower?

A reader asked whether CFD is an “ivory tower” tool and, if so, what would it take to make CFD more reliable.

The simple answers are “no” and “validation and verification.”

But nothing is that simple.

CFD Is One Big Tangled Ball of String

Here’s why this reader is concerned.

“From my perspective, it seems there is a lot of uncertainty in applying CFD because the complexity of flow features and how they interact which each other and a geometry.  One has Mach number and Reynolds number, which translates to compressibility, shocks, boundary layers, laminar to turbulent transition, separation regions, small and large eddies, and steady and unsteady.  Then there are all the different methods to model such flow, laminar, LES, DES, RANS, DNS, incompressible, compressible, low order, high order, boundary conditions, etc.  Then there is implicit and explicit and all the algorithm mumbo jumbo which goes along with it.  Sometimes, when listening to non academics who use CFD, it seems like it is one big tangled ball of string.  And academics, in general, seem to focus on their region of interest and are forgetful of the existence of other regions.  I have not seen a handbook (such as Hoerner’s book on drag) for CFD.”

Is CFD an Ivory Tower Tool?

An ivory tower is a world in which intellectuals engage in pursuits that are disconnected from the practical concerns of everyday life. While I bet each of us knows a CFDer or two who fit that description, CFD is used every day to obtain engineering information – that is, information that can make a product better, faster, or cheaper. Therefore, I have to say that CFD is not an ivory tower tool.

In fact, I think we might be asking the wrong question. It’s not the tool that’s in the ivory tower – it’s the people. The keys to successful CFD are not in the tools – they’re in you. For successful CFD you need:

  • a firm foundation in fluids,
  • an awareness of your CFD solver’s range of applicability,
  • a familiarity with the device you’re simulating and
  • an understanding of how to properly use the CFD software.

With all this knowledge, you’ll know how to get the best answers out of your CFD solver and how to best interpret them. As Harry Callahan said, “A man’s got to know his limitations.”  The same is true for CFD.

Where Does CFD Uncertainty Come From?

The level of uncertainty boils down to whether verification and validation (V&V) has been performed. Verification is the responsibility of the code developer and ensures that the software is coded properly and produces accurate results for what I’ll call generic benchmarks.

On the other hand, validation is your responsibility and quantifies how accurate the software is for your particular application.

Let’s say your area of responsibility is the exhaust nozzle on a transonic fighter aircraft (which, coincidentally, is where my CFD career started). When you’re given a CFD code to use the first action should be to obtain documentation from the code’s developer about all the verification testing. This will include things like flat plate boundary layers and the NACA 0012 airfoil and backward facing steps – generic benchmarks. The second step will be for you to apply the solver to see if you can reproduce the verification results because then you’ll know you’re applying the software correctly. Next you’ll apply the software to a nozzle for which you have data – perhaps something proprietary or maybe an open benchmark. With everything you’ve learned, you’re ready to begin applying the software to new configurations.

Online Resources

Rather than repeat them all here, let me Google a few things for you.

Workshops, Meshing, and Uncertainty

The pursuit of ever increasing reliability in CFD is an ongoing activity exemplified by the Drag Prediction Workshop, High Lift Prediction Workshop, Propulsion Aerodynamics Workshop, and many others. They provide benchmark cases for comparison and through their multi-year efforts take a long-term look at how accuracy is improving over time. They are even pulling us mesh generation folks into the fray and asking us to quantify how good any particular mesh is. But that’s a topic for another post.

*Note: Harry Callahan was quoted directly, hence the use of the masculine. However, as we all know, women are CFDers too.

This entry was posted in Applications, Events, Software and tagged , , , , . Bookmark the permalink.

7 Responses to Is CFD Stuck in an Ivory Tower?

  1. Jeff Waters says:

    Good stuff, John. It’s also in what you are expecting. I spent about 9 years promoting an upfront CFD tool meant for non-specialists. Saw plenty of cases where we set the expectation to “trend” vs “dead nuts accuracy” and the customers were ultimately very happy. The alternative for these customers was continuing to build and fail in the lab 30 times. Why not trim that down to 10 times by picking better directions based on design variables?

    The purists always hated this approach because, “The world will come to an end and your company will go under if you use those kind of CFD results and get it wrong.” Well, that’s a strawman assuming that you are a dumbass will blindly accept such results and do no final testing or validation.

    I believe there is very much a need for both ends of the CFD spectrum… even in the same company. Here’s a little video expressing my thoughts on the matter:
    http://lifeupfront.com/2010/09/01/expert-upfront-cae-roi/

  2. John Chawner says:

    Thanks, Jeff. You are correct – setting expectations is also important. You can use a CFD tool that’s always 10% low on its numbers if you use it to determine trends. However, you should also be wary of what might be causing that 10% difference because sooner or later it’ll bite you.

  3. Nice article, John. Your point on the every-day use of CFD to disqualify it as an ivory tower tool is well taken. Perhaps we need to develop more turnkey apps (like prescription medicine) for specific problems will make the use and application of CFD more affordable and usable.

  4. John Chawner says:

    Hi Pat. Turnkey (aka vertical) apps are a great way to package CFD to make it more useful for the non-expert. From a meshing standpoint, vertical apps let you automate the meshing and remove it as a barrier to entry. Similar benefits are available on the solver side as well because the solver options can be limited to those that make sense for the particular app. You mention affordability and that’s an interesting aspect. Do you make a vertical app less expensive because it’s limited in applicability? Or do you make it more expensive because of the extra value it delivers?

    • John, one might think “less expensive” to attract a wide audience but when was the last time you heard a discussion about aerodynamics at your local supermarket? On the other hand, a big company might be interested in an expensive vertical app but companies hire Ph.Ds. to make their own “in-house” codes. One way is to demolish (or renovate) the ivory tower might be to “hide” CFD in consumer products such as games, engines and augmented reality apps.

  5. John Chawner says:

    Pat, a less expensive app would definitely appeal to a wider audience but I don’t think we need to worry about consumer level interest just yet. The best hope for mass appeal is to create a CFD app for iOS or Android and that turns into the next Angry Birds. Angry Fluids anyone? But if you could afford to outfit every designer in your company with a CFD app tuned to the types of widgets you build, that would be great.

    A more expensive app would appeal to a narrower audience who recognized by the value of having an app that does exactly what they want right out of the box.

    (shameless plug follows)
    That’s why we’ve written our Pointwise meshing software so it can be customized with our Glyph scripting language and extended using our CAE plugins. A company’s Ph.Ds don’t have to reinvent an entire meshing app from scratch – they just build on ours. And they also have the full meshing app to apply to configurations that are outside the scope of the verticals they’ve built.
    (end plug)

    As for CFD in consumer products, a lot of the fluids simulations in games are getting very sophisticated. As an example, here’s an article, The Science of Fluid Sims, that I’ve blogged about before: http://www.fxguide.com/featured/the-science-of-fluid-sims/

  6. Thanks for your reply and the Fluid Sims link. The Glyph scripting language is a good solution to extending the use and application of your products (congratulation on the coding that must be behind the scenes.)

Leave a Reply