which is more important? Getting the answer right to 1.0E-9 or getting a workable solution to a problem in a workable time?

Hey, well said!

"All models are wrong, some are useful" G.E.Box

Hi, I am always looking for the 80% solution in 40% of the time required for the 100% solution (if one exists!).

regards, Tom

How about a response from the people who dont like messages from 'mouse-clicking cfd people'?

I am trying to open a debate on the subject which seems to produce large differences of opinion.

Yeh, come on you lot, I've never written a line of code in my life and I use CFD nearly daily, always productively.

It seems CFD is still a lot nerdy. With computer albino geeks watching "The Matrix" ten times or colgate advertisers as annoying as austin powders.

Lomi

I was about to comment with something like, "Hey! Why don't you post a comment or question about something useful?!" until I looked at the number of people who have read the threads. Your original post has already been read 180 times and I'm sure that number will continue to climb. For further prooof, look at the post entitled, "One year without John C. Chien" which has been read something like 450 times. No other posts come close to these numbers. And there's a good reason: CFD is boring. Talking about it makes it tolerable, but having conflicts (minor ones only, please) makes it downright fun! Boy do I miss J.C. Chien...

Hey monkey man, if it wasn't for the pretty pictures CFD generates then the entire CFD community would be comprised of mathsheads and underpaid engineers looking for explanations

Bill's original post is key. There's 3 things that are required in industry/business: accuracy (enough), time (little) and cost (less). Academia excels in the former but at the expense of the second. Getting 2 satsified is not impossible. Balancing all 3 is a gruesome. CFD is still dripping from its acadmic origins, drenched in an absolute accuracy requirement that was essential for its initial applications such as nuclear, aerospace etc. Now people have woken up to the potential of using it for design, during product design, outside of that little R&D office populated by the PhDs. I say people; managers really who understand (or have been sold) the cost advantage of using CFD to replace costly make and test methods.

It shouldn't be surprising that there is now somewhat of a backlash that has resulted in a number of 'trust in CFD' initiatives. A realisation that most of the available CFD methods still require those PhDs to achieve the desired usefulness in the application of CFD.

If all the academic effort being expended to derive marginally better turbulence models or numerical schemes was spent on automating and burying the CFDness then tools that utilise CFD would be as wide spread as CAD (well, maybe not but you get the idea).

So, come on you clever CFD people, give us 'gridless' CFD, hard wire or autoselect the best numerical solvers, give us easily digestable usefull results instead of reams and reams of bloody vector plots.

please?

Rock on Harry Fulmer. You speak the language I wanted to hear, because your statement is validation for the validation work I'm doing for my PhD. Rather than focusing on discretizations, I'm using an optimization scheme (I guess that part is like old-fashioned PhD stuff) to decide which are the best combinations of already-available CFD tools and models based on controlled windtunnel and exposure chamber experiments.

Where I work, we have a few CFD engineers, but most of those who could be using these tools in the field barely even have a clue what a PDE is. They need a Black Box that works to get them in the ballpark. After that they can use real-time monitors to navigate to the pitcher's mound and engineer the changes that need to be made immediately.

On the flip side, without the CFD industry as a whole though, the commercial codes wouldn't work as well as it does because those who do specific work in one area (DNS, RNG-ke, discrete phase, imaging) get really good at it and eventually the commercial codes pick up on the best of their products making the CFD more and more like a black box all the time.

I am assuming you're using a commercial code too? Or an in-house code that works? ;-)

Just a thought, we still do not really understand (mathmatically) turbulence. So, we don't know the exact solution to all the practical problems. If we don't know the exact solution, how do we know the approximate solution is 80% accurate ? In other word, how can we garantee the black box always gives the 80% accracy ?

I think the only good test is with reality, rather than with analytical solutions from other derivations (however, they are great for general comparisons, especially in the code verification process). Compare the solution, whether it be from fuzzy logic, neural network, statistical model, discretization, or whatever, to reality. Estimate uncertainty of the numerical method (i.e. the model). Estimate (hopefully "calculate") uncertainty of the measurement of reality. Compare the two with a simple measure of their differences. Use your own safety margin that applies to your situation, but definitely not an arbritary one. Just because Joe CFD uses 80% doesn't mean it is a hard and fast rule for everyone. If you need to get into the ballpark, then maybe you can have 60% agreement. If you need to see a speck of dust on home plate, then 100% agreement may not be enough because the level of uncertainty will be relatively enormous and you won't be able to discern for "certain" that you have seen the speck of dust.

The point is that CFD is not perfect, so just use it to the level of confidence that you are comfortable with... provided you do all the right stuff to get your CFD answer: numerical convergence, grid convergence, realistic boundary conditions, proper application of solvers, turbulence models, etc. then use the answer with as much confidence as you are able based on the CFD analysis itself (did you achieve a grid independent solution? did you use realistic BCs? was the code verified before you used it?) and how well it compares to some measurement of reality. CFD is a good tool. However, it is not the absolute perfect tool, maybe someday it will be, but not today.

Don't forget to use a little common sense and critical thinking. Remember the famous saying, "Don't lose sight of the forest for the trees."

Totally agree. That is why it is impossible to build a black box to let anybody use it unless the black box has the universal solution. So, academic has to continue to try to find the universal solution or something close to it (if it does not exist), and for industry, the commerial CFD software has to be used by people who know how to do the right stuff (have CFD common sense).

Maybe our definitions of Black Box are different... My Black Box is a tool that gives reliable answers and is really easy to operate, not necessarily the universal solution -- but don't get me wrong, I'll be happy to use the universal solution when its ready!

What goes on inside the Black Box is really of no concern to the user. The only thing that matters to the user is the output based on certain inputs. The user of such a device (or code, to which I am alluding) would pick up the calibrated unit in the morning, go out and use it all day collecting information, and then bring it back at the end of the day for analysis and recalibration.

The accuracy of the output is compared against reality to determine how well the tool worked over the course of the day. If someone from R&D or the local University sends over something to make it run better, then that's great, then it would be a Better Black Box rather than just a Black Box.

CFD is fast becoming this, and its O.K. for this to be happening as long as there is a gateway early in the process so that the Black Box is not used without calibration and so that the right stuff goes in and reasonable stuff comes out. Black Box isn't a bad thing if its controlled. Kind of like driving a car, right? Are you a car mechanic? No. Most of us are not. Can we drive? Yes, most of us can. So we can operate the unit without knowing exactly how it works. Cars are Black Boxes.

I get a feeling that CFD and it's accuracy is often put in the line of fire by the world CFD-concience often sprung from academia. The bottom line is that CFD used in industry should be used as a TOOL to predict problems and performance. Not to give the final answer to all problems - at least not yet. When some people say: "Uuuuuh, CFD...is it really reliable to use that?", I often have to say: "It's either that or it's pure and wild guessing". And hit me with a frying pan if I don't make better judgements after a (even quick'n'dirty) simulation, compared with just wetting my finger and put it in the air for inspiration.

A while ago, there is a company selling the so-called 21 century black box type of easy to use tool which is based on a theory even some of your 'clever" CFD PhDs found difficult to understand.

The code is called powerflow. What happen to this company now ? Searching the whole CFD discussion forum, last time someone would care to mention powerflow was in 1998 ! It doesn't seem that many people use it or if some did get a chance to use, maybe they are not happy to talk about it - or maybe there is nothing good to say about it.

So the concept of black box is good, but when you get non sense from a black box tool, you're screwed really ! Let's hope one day, there will be a black box that works and will conque the CFD world.

To stretch monkey's car analalogy a little further. You have to be qualified to drive the thing. You don't expect to get in the car, release the brakes and just roll off to your destination without touching the steering wheel. It's easy to get down to solutions at 10e-9 that are completely wrong because of incorrect setup. The quality and accuracy of the solution are judged by the operator, not the code.

I guess I'm saying I don't mind 'CFD mouse clickers' as long as they have a degree in aerodynamics.

Degree in aerodynamics? A Mechanical Engineering degree should be all that is required.

Cars, one day they'll drive themselves for you. Then all you'll need to know is where you want to go.

Cars, CFD codes; all have some way to go before they truly satisfy our needs.

OK, I'll accept mech eng at a pinch ;-)

We can drive cars does not mean everybody drives well (otherwise there would be no car accidents). Years of experience gives us the knowledge (or skill) to drive well and safe. That's exactly for CFD, you either learn it from graduate school or learn it by using it for a long time, then you'll have your reliable answers.

> I guess I'm saying I don't mind 'CFD mouse clickers' as long as they have a degree in aerodynamics.

I would like to take issue with the "degree in aerodynamics" - I would use the expression "a good understanding of fluid dynamics or aerodynamics", something which you may not necessarily get from any degree, or necessarily need a degree to have. Fluid flow is far too complex for dummies, if you do not understand the physics you are looking for trouble. CFD is a wonderful tool to help one develop that understanding.