Hi.

Would someone, especially from an academic institution, please enlighten me as to why grid validation stdies are necessary? Why is it that in industry such exercises are never carried out? If experience, past on from one cfd-practitioner to another, why is it necessary to spend hours on grid validation studies?

In CFD grid is the most important thing which determines the success of the solver.No matter how good your scheme may be,if you dont have the right grid you wont get the solution.

Grid effects: Resolution(i.e)damping , aliasing , stability, dispersion error and may be there are many others that I dont know.

bye

"Why is it that in industry such exercises are never carried out?"

Rather too sweeping a statement I think. The cost of garbage in - garbage out (due to insufficient resolution) is higher in industry than in academia, and the studies are done, at least where I worked.

Of course I can't speak for all of industry.

1) The CFD industry consists of both CFD code developers and application users.

2) One of the most time consuming tasks for any serious CFD code developer in the "industry" is "validation".

3) CFD software makers do not sell "un-validated" codes. Grid independence is one of the very important things considered during the validation process.

4) The CFD application users most times assume that the code they are using is validated for the problem they are trying to solve. If this is not the case, then they do perform a lot of validation cases. Nobody will want to buy untested/non-validated software.

5) Grid independence is a very important issue especially for turbulence modeling and is definitely given a lot of time in the CFD development industry.

>why grid validation stdies are necessary?

A CFD grid needs to be designed appropriately to resolve all significant flow features. The flow, of course, is not known a priori, so a first CFD computation can only be regarded as a first guess. How are you going to tell how accurate this first guess is? With just a single shot you have no idea what the uncertainties are.

Grid resolution studies are performed in order to get an estimate for the error induced by insufficient spatial resolution. Without the study you really can't tell if your solution is reliable or not. Even in the rare case that you get a perfect match with experimental data, it may just be a coincidental match and not a reliable solution. There is no way around grid resolution studies, and any serious CFD engineer/scientist performs them routinely on any given problem, regardless if in industry or academia.

Experience helps in divising a good first guess but does not supersede grid resolution studies. Only unexperienced CFD users think they can do without.

>Why is it that in industry such exercises are never carried out?

Where do you get such idea? Sure, the pressure to get fast results may be higher in industry than in academia. The temptation is to save time and believe that the grid which was validated for a similar problem will be sufficient for the new problem. However, as I said: Only unexperienced or shallow users of CFD fall for that misguided assertion.

>If experience, past on from one cfd-practitioner to another, why is it necessary to spend hours on grid validation studies?

In order to know accurately what resolution you need at every location in the flow field, you would need to know the flow. If you were experienced enough to tell in advance what the flow field looks like, why would you have to do CFD? Experience helps in performing efficient grid resolution studies, not in doing without.

It is a fact that in hardly some industrial investigation, whether with own code or with commercial code, grid validation is done. Often the excuse forwarded is "time limit" or "job pressure". The result is obvious. G-O. I feel it should be the job of the code vendors to educate their users on aspects and issues like convergence, stability, grid independence, etc. Why with flows only? Even in so-called simple heat conduction problems, grid independence is not attempted. I have seen "results" (surprisigly highly appreciated by customers of such results who are unaware of what they are receiving and only aware of and impressed by colored vector/contour plots) for flow situations where the nearest-to-wall grid velocity vector is just in the opposite direction to the immediate adjacent grid velocity - denying all the education of basic physics. Such results are termed "results for engineering application" by those result producers (and their recipients too!), who forget the fact that engineering is based on basic principles of science only.

Dhania,

In order for you to appreciate the essential role of grid dependency study, consider the following demonstrative simple example.

Take a simple ODE with appropriate BC (e.g., on the [0,1] x interval, y'+y=0, y(0)=1) for which analytic solution is available.

Solve this case numerically (e.g., using central differences on a uniform grid). You will notice the numerical solution deviation from the exact solution approcahes 0 as the grid is refined, provided your discretization is consistent and stable. You may also notice that non-uniform grid with smaller cells where y' is high (i.e., a manually-adapted grid), is more efficient (i.e., a lower error is achieved for less computation vs the uniform grid with the same No. of interbvals).

The same is true for ANY discrete numerical solution, including CFD. Therefore, grid independence should be proved for any case studied. Industrial calcualtions are no exception.

I think a reason you may not see formal grid validation studies performed routinely in industry is not just the time pressures, but also the philosophy that the CFD codes are simply other tools to be used to provide guidance, to confirm things the designers expect to see. The CFD results themselves are not the be-all-end-all of the problem.

Rather, the CFD analysis is merely providing a sanity check. Used in that manner, numerical accuracy to within 5% of analytic/experimental results is not deemed necessary. Instead, the questions being checked are "is the pressure drop within the range we expect?", "is there heating in this area or not?" for example. In these cases there may already be a crude prediction from other analysis tools, and these are rather global questions that a first pass grid should be able to answer.

The CFD purists out there will certainly disagree, of course.

Just my $0.02 ....

I do appreciate all your responses and I was in now way trying to ruffle feathers. I am new to this field and I am merely questioning what I see happening in the industry where I work in.

There is a philosophy question at stake here. Are you doing the CFD for the sake of the CFD or are you doing it to support a design? If CFD is being done to support design, the quality metric is not the quality of the CFD, it is the quality of the resultant design. Which can sometimes be enhanced by accepting "lower" CFD quality in order to evaluate more design variants within the constraints of the available respources (which are always limited). You need to know what you are doing to make this work, and you certainly need to be using other tools than just CFD to validate your design, or else you're really sitting out on a limb! To summarise, a validated design is essential, but it may not be essential to validate the CFD on its own. The reality of the design process is that there is often an experimental validation already built in to the process.

It would all sound good, if your assumptions were correct in every case: that without grid resolution studies you may end up at a slightly inaccurate but still qualitatively correct result.

That may be the case (especially if the case you are running is just slightly different from previous validated cases), and then I agree that other validation methods (experiment) may be more important, when CFD is not the primary design tool.

Unfortunately, there is no guarantee that saves you from getting a completely (even qualitatively!) wrong result because your grid is so inadequate that you miss out on even essential flow features. This is not at all unusual, and there has to be a check against that possibility if you want your CFD result to be of even the slightest engineering use.

How often do people realize that their CFD results are not only inaccurate, but do not even predict the qualitative trends observed in the experiments? Then what do you do? Randomly play around with parameters because you have no idea what might be wrong? Scrap the CFD, defining the time you spent on it "lost time"? How is that saving you time?

Some people may have the strange impression that certain practices that are more common in academia than in industry are philosophical ceremonies of some secret spiritual club, meant to look important but really superfluous. Grid resolution studies are not performed for the fun of it. Yes, they do take time, and even when they're done, you will not end up with a perfect solution. It's just a very basic and necessary evil without which your CFD solution has hardly any meaning at all, and is really a waste of time. In the end you work more efficiently by spending a little more time in doing things right, than trying to save time by neglecting common sense. This is not a philosophy but a down-to-earth engineering fact. Simple as that.

No matter how accurate a result you need, there is always the need to get a feeling for the uncertainties. You can't just throw numbers at people, having no answer for the question they are going to ask you: "How accurate are those numbers?" How you approximate that uncertainty is up to you, but in CFD spatial resolution studies are part of the unavoidable deal.

It's a very good question. Very old, but still controversial, as you can see from the response.

I personally think your question touches a very sensitive spot in numerical analysis. (1) How much time you spend on validation of your results (not just by grid resolution) will depend on how important the numerical method is in your design process.

(2) How much importance you are willing to assign to your numerical analysis will depend on how trustworthy you think it is.

(3) How trustworthy you think the results are, may either depend on the results of your validation study, and on your comparison with experiments, (which will be hard to interpret without validation), or on your best unguided "guesstimate", or on the advice you got from your favorite psychic. (4) Now, your best practice will determine how trustworthy >you< are, and how trustworthy your results really are, from an objective point of view. And that, in turn, may be a factor for your boss to decide how much money and time to spend on CFD, where we return to point (1).

That's why I think your question is quite important.