A Tale of Two Years: 2013 and 2014 in Simulation and CFD

In case you missed them, two very good articles were published recently on the topic of CFD and simulation.

[The year is still “new,” right? I started writing this article on 03 January and am only getting back to it now.]

Tech Briefs magazine shared the results of their annual discussion with leading simulation software vendors in Industry Roundtable: Analysis & Simulation Software 2013.

And with a forward leaning slant, CAE Watch gave us their annual look at the coming year in computational fluid dynamics, CFD in 2014.

What We Learned from 2013

The eight industry executives interviewed by Tech Briefs provide a pretty good perspective on the state of analysis and simulation software based on what transpired in the previous 12 months.

Simulation is Good

They all agree that simulation is a valuable if not necessary tool for many industries, especially ones built around complex products. SIMULIA’s Dale Berry summed it up “Simulation is not only a necessity, it is a do-or-die requirement for survival in many industries today.”

The mantra is simple: simulate early and simulate often. By performing simulation early in the design process you’ll be more effective in influencing the design before it becomes too rigid (as happens as the design matures). By simulating often you’ll be able to more fully explore the performance envelope and parametric variations on the geometry.

[This is where a cynical reader usually feigns shock that simulation software vendors advocate more simulation.]

An interesting corollary to simulating early in the design process is that the CAD models are usually less mature – in other words, sloppier. While this won’t necessarily effect some types of simulation, it can play havoc with mesh generation for CFD. Of course, the need to defeature might not be as great because detailed design hasn’t started.

Testing Isn’t Going Away

As strong as the advocacy of simulation is, no one has drunk enough of their own bath water to believe the need for physical testing has disappeared. Simulation and testing have always worked hand in hand.

I can still remember back in the late 1980s when a CFD workstation – loaded with completed computations and ready to compute more – was taken to a wind tunnel test for a special aircraft with the purpose of verifying that things were going as expected or to figure out why they weren’t. If I really stretch my memory I can recall my first exposure to CFD while working one summer at the 10×10 supersonic wind tunnel at NASA Glenn (then NASA Lewis). The simulations were performed pre-test to guide the test schedule and post-test to validate the simulations. [And we actually discovered something about the numerics of the solution that had a bigger than expected effect on the results.]

You might be tempted to think that testing is real while simulation is not. But testing (and I’m thinking primarily of wind tunnel testing) is a simulation also. Putting a tiny airplane in slightly larger box and blowing wind over it isn’t really the same as flight test. Scale effects, interference effects, measurement accuracy, and the pesky detail about just how closely the model matches the CAD all lead to interesting engineering.

You have all heard this joke: No one believes the CFD except for the engineer who ran the simulation. Everyone believes the wind tunnel test except for the engineer who ran the test. But never forget that all good humor is based on a kernel of truth.

One recent example flies in the face of this advice against cessation physical testing. As cited in Desktop Engineering, Oracle Team USA used only CFD for all their fluid dynamic design in a successful run at the America’s Cup.

More Computing Power

Remember that advocacy of more simulation? It comes at a cost. Faster CPU, more RAM, bigger disk. And then there’s software license cost.

Fortunately, server clusters – both local and remote (aka cloud) – are available to fill that need. Plus, the software providers are targeting those platforms with their product development.

MSC Software’s Dominic Gallello hit the nail on the head when he said “Yesterday the computer was the bottleneck. Today, it is the engineer who has to interpret all the data.”

Noteworthy

  • The article begins “In today’s environment of doing more with less…” We should stop saying that because I’ve heard it for 30 years in this business. In this regard, there’s nothing different about 2014 from 1984. As I overheard one senior executive say shortly after starting my first job, “When things are bad here they are really, really bad. And when they’re good they’re not a helluva lot better.” [Or “not hella better” as the kids say.]
  • ANSYS’ Barry Christenson and Altair’s Mike Kidder both deserve credit for pushing for CAE (i.e. simulation, performance) to be the language that bridges designers and analysts. This gets back to the core issue of designing for performance instead of geometry. [I will not hold it against Christenson for using the term democratization.]
  • The award for best quote goes to CD-adapco’s David Vaughn for “2014 is the year that design optimization grows a beard.”
  • I found it odd that two of the industry execs (Vaughn and Kidder) cited software licensing issues in the context of high performance computing. Kidder goes into detail by saying that software costs have not mirrored the commoditization of hardware.

What to Expect in 2014

Perhaps more interesting than looking back with 20-20 hindsight is looking forward and making predictions. On the CAE Watch Blog, Shengwei Ma is “ranting from the perspective of end users” about what’s to come in CFD during 2014.

Technology Is Not As Important As It Used To Be

About a year ago, Antony Jameson was quoted as saying that CFD has been on a plateau for about 25 years. Ma echoes this by pointing out the lack of exciting developments in CFD and the relative sameness of competing products. Given this environment, the cycle goes like this:

  1. Software vendor needs to make sales.
  2. Software vendor has no exciting new developments.
  3. Software vendor adds minor features.
  4. Software vendor brands this upgrade as new technology.
  5. Customer upgrades.
  6. Software vendor starts over at Step 1.

I think Ma is being a bit cynical. The overnight sensation business model wouldn’t apply to CFD even if it were real. (You’d be surprised how many years of preparation it takes to become an overnight sensation.) Plus, I’ve always believed in evolution over revolution when it comes to software development.

But I agree 100% with how Ma concludes this section of his article. Let me quote it here. “Building trust with potential users is more and more important for CFD software vendors.” Social media tools like blogs, Twitter, YouTube allow that trust to be built through a dialog.

More Is More but Less Can Be Better

Probably as a result of the software development cycle Ma cites above, we find ourselves with software that keeps getting bigger and bigger and you can state this in many ways. Ma says 99% of users will only use 1% of the software’s features. Or is it 90-10? Or 80-20? It doesn’t matter because the nugget of truth is there. Here at Pointwise we call them 2% commands because that’s how often they get used.

However, when you need it, a 2% command comes in very, very handy.

I agree with Ma 100% when he advocates vertical applications that target specific classes of geometry. In my opinion, the best opportunity for automating mesh generation is by targeting vertical applications like turbomachinery, aneurysms, wing-body, etc.

And this is why we architected our product with a scripting language that lets it be customized with macros and templates and why we freely share those scripts online. It’s why we added a plugin capability so users can export their grid and boundary condition to their own custom file format.

Back to Ma’s point about feature-bloat, interested readers should consider Clayton Christensen’s The Innovator’s Dilemma.

Meshing, Meshing, Meshing

I don’t know whether to feel belittled or buoyant that Ma says the top 3 problems in CFD are meshing, meshing, and meshing.

He cites the Catch-22: you need to generate a mesh to compute the physics but you need to know the physics to generate a mesh that will compute the physics accurately.

This is why my meshing friends advocate solution-adaptive meshing. But even that’s not a panacea.

I also don’t know of a single CFD code that will automatically – and I mean that literally – converge to an accurate answer given only a mesh and boundary conditions.

No one said CFD was easy. But Ma’s point is that as CFD begins to be applied by non-experts we really need to remove meshing as a potential stumbling block.

A Wealth of Open Source

Let’s begin where Ma ends: “open source CFD software will probably still only make the rich richer.”

If you equate open source software with free software it becomes somewhat paradoxical. How do you become wealthy with free software? Volume!

It seems that Ma’s point is that open source software’s drawbacks (including lack of strong 3rd party support) will limit adoption thereby continuing everyone’s reliance on commercial software – thereby making the rich richer. [Because everyone who works at a commercial CFD software company is filthy, stinkin’ Robin Leach rich.]

Clouds, GPUs, Webinars

  1. Ma believes everyone will eventually migrate to the cloud for its “pay per use” license model. But I’m still not convinced that there is a consensus opinion of what exactly the cloud is good for.
  2. Ma seems intrigued by the potential benefits of GPU-accelerated CFD codes but then points out that the coding issues involved are non-trivial [very true] and the graphics cards themselves aren’t cheap.
  3. Ma thinks most webinars are a waste of bandwidth because too many are simply self-promotional instead of being educational. Like any other medium, webinars can be abused.

In Conclusion

Like I wrote in the opening, two very interesting articles that I recommend you read. And even though I may sound somewhat critical of Ma’s forecasts, I applaud his courage in making predictions.

Let me know what you think in the Comments below.

And if you have suggestions for a Pointwise webinar or want to learn more about our Glyph Script Exchange, click the button below to send me an email.

action-learn-more-200x50

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

14 Responses to A Tale of Two Years: 2013 and 2014 in Simulation and CFD

  1. jdrch says:

    Ma believes everyone will eventually migrate to the cloud for its “pay per use” license model.

    This could be disastrous. Clouds tend to be one-size-fits-all solutions that aren’t necessarily optimized for the task at hand or type of computation the user specializes in. You can easily wind up with lower performance at the least.

  2. John Chawner says:

    Hi Judah:

    First, I hope I’m accurately representing what Mr. Ma wrote. Second, in my opinion Mr. Ma is very concerned about software license costs because he mentions it frequently in his writing. So from that standpoint, a pay-per-use cloud would be a nice thing.

    Second, I think “the cloud” is so immature at this point everyone talks about it while presuming one specific benefit when in fact it will be just another platform that will appeal in different ways to a different segments of the market.

    Third, pay-per-use has nothing to do with the cloud. You can put software in the cloud that requires an annual license just like desktop software. You can license desktop software with a pay-per-use model.

    Finally, I agree with you that everyone will not migrate to the cloud. Some will, some won’t. It’s just another platform.

    Thanks for commenting.

  3. Martin says:

    I’m still confused about the term “accuracy”. As far as I know, we still are not able to predict drag when transition is natural, i.e. not tripped. This is also true to some extent for max L/D. In regards to the Oracle team, I wouldn’t be surprised if they used past WT data to validate and enhance their results and still be able to say they did not do any testing. Frequently we add historically determined offsets to CFD results for better predictions.

  4. John Chawner says:

    Martin, I’m not certain of your context for the term accuracy. Everything you wrote is true. But it doesn’t mean that CFD isn’t sufficiently accurate for design work.

    • Martin Hegedus says:

      Oh, my context is that I get the impression when reading various blogs, polls, articles, etc. that “meshing, meshing, and meshing” is the #1 issue involved with achieving accurate answers. And by accurate, I assume we mean quantitatively accurate, and not qualitatively accurate. (Which, may be bigger assumption that I should make.) In general, vendors and some individuals don’t say “sufficiently accurate” or “accurate enough” they say “accurate”. And I’m going to assume “accurate” means within a couple of percentage points. I don’t know of that many people who would say +/-5% and greater is accurate.

      Take the challenge Dr. Wu made to Pointwise. The better grid will be determined by the grid which gets closer to experiment. I’m curious how many CFD professionals think this is true? Am I one of the few that thinks this is false?

      Or take my comment about turbulence modeling being the #1 issue for quantitative results. I get the impression that turbulence modeling is an issue for only a small group of individuals. In other words it is “my” problem. I’m not sure why it is only “my” problem. I, and others that I know, are frequently making choices between design A and B based on CFD results which are not that far apart. Truthfully, many times I have very little idea about how accurate that delta actually is. There is very little guarantee that the delta even goes in the correct direction. I’m not doing necessarily anything that I consider “extreme” or cutting edge in regards to fluid dynamics. Personally, I do find it challenging to make previously “good” designs “better” with CFD alone. And, sure, gridding is a pain. But it does not end there.

  5. John Chawner says:

    I see your point. Perhaps the way to look at it is that CFD is accurate enough – the problem is throughput. And the problem with throughput is the time required to preprocess (CAD to mesh to solver). That’s not saying we still can’t improve on accuracy, but for some folks it’s not the main issue. I’d be willing to be that for others, the main issue with CFD is how to compare and contrast a thousand solutions for parametric variations of a design. The list of issues could go on and on.

    And when I say accurate enough, I don’t mean that it shouldn’t be improved. I mean that someone who has done their V&V work can apply their CFD code to get meaningful engineering data for design. (As goes the quote, all models are wrong but some are useful.)

    • Martin Hegedus says:

      Yes, I do agree. But, it’s not just about improving accuracy (we know it’s a though nut to crack and probably won’t be cracked) it’s about minimizing the arm waving and the “my schwartz is bigger than yours”. Selling CAE software, or cloud services, shouldn’t be approached as if one is selling a used car. It’s an engineering tool. I must admit that sometimes I truly wonder if some CFD software and cloud vendors actually do engineering design for the sake of understanding their customers or maybe even just test their programs on real engineering problems? Or, do the just rely on their customers to do the dirty work?

      I also agree with this statement:

      “Building trust with potential users is more and more important for CFD software vendors.”

      And that leads to another discussion, which will be left to a later date.

  6. vivin j geevarghese says:

    Interesting read. I have a query,though. Do the major companies like GE who uses CFD based wind turbines develop their own CFD software or do they use the open source CFD software available??

  7. John Chawner says:

    Vivin, that’s a very good question and I don’t know the answer. I know that the “build vs. buy” debate is ongoing within many organizations. For “build” to win that debate it must show strong competitive advantage.

  8. vivin j geevarghese says:

    Ok. If I may ask,what do you think is the global CFD market size and where can I find data regarding the same?

    • John Chawner says:

      Hi Vivin:

      I don’t have that number at my fingertips. A lot of people publish numbers on the estimate growth rates of the various CAE market segments that can be found via Google. If you find anything interesting, let me know.

  9. John Chawner says:

    Vivin, you can email your questions to jrc@pointwise.com.

Leave a Reply