CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   How good is CFD? (

kai February 24, 2000 22:27

How good is CFD?
I am using PHOENICS to compute the characteristics of an airfoil.

The results are compared with the standard results get from wind tunnel.

The CFD results are so "terrible".

I had try to correct my model and setting in various method.

However, the problem still remain the same.

So, assuming that the model and setting are correct, then, is it possible that in certain circumstances like my case that there is absolutely no "good" or " logic" results produced by CFD at all?

Mahesh Prakash February 24, 2000 23:26

Re: How good is CFD?
The answer to your question is. CFD is as good as the person who is using it.

Ben February 25, 2000 00:36

Re: How good is CFD?
hi, kai,

You look like a newer in CFD. I had some experience in helping some students to use some CFD software, say Fluent. It usually happens for a beginner to obtain a "ugly" result. Things you need to do is that 1. read the software manual or handbook carefully. 2. find some CFD book to read in order to have a initial knowledge regarding CFD.

By the way, although PHOENICS is not best one, I think that it is not bad. If you use correctly, it should give a resonable result.

good luck!


kai February 25, 2000 07:29

Re: How good is CFD?
I have to make my statement clear, first what I mean for bad results are the solution given by CFD is roughly 5% error from the solution given by the wind tunnel.

What I mean is that "If you use different CFD software like PHOENICS, FLUENT and etc, they will give you a result that is may be much more accurate or terrible".

Therefore, in this circumstances, How CFD can be evaluted?

May be in the future CFD will be replaced by something else that are more persistent in giving accurate answer.

kai February 25, 2000 07:35

Re: How good is CFD?
Please read my message to Ben.

I don't agree with you.

That's mean CFD will no longer give a "good" solution.

If you use CFD, you absolutely be confined with certain rules or formulas.

Assuming that you are perfect, but as long as you are confined by these rules or formulas that are true or valid untill now but not in the future, we still can't make a standard to evaluate the results given.

yangang bao February 25, 2000 11:47

Re: How good is CFD?
In my opinion, 5% error is accurate enough! Why are you still complaining the numerical results?

John C. Chien February 25, 2000 12:26

Re: How good is CFD?
(1). Let me try to understand your question in the first place. (2). You are using a code called PHOENICS to calculate characteristics of an airfoil. That is great, becuase someone has written the code and it is available to you. (3). In order to use the code, you also must have done some homework exercises that this is the right code for you to study the airfoil problem. (There are many situations where people are forced to accept the system's decision. In that case, my assumption here would not be applicable.) (4). So, the preliminary home work exercise is very critical to determine whether the decision will match the requirement. (5). For someone in the similar situation, it is likely that a multi-million dollar CFD research laboratory will be created just to do flow over an airfoil. (6). Now, you have been using the code, and you have learned how to obtain solutions to flow over an airfoil. At least, the code works. And you also have learned something. So, someone should present to you a certificate of accomplishment. ( companies do present certificates to students when they finish the training.) (7). Now, back to your "terrible" CFD results. Have you shown the result to your instructor or the support engineers? If not, then you need to do so. (I would not like the user to get real good results, if he was using my free demo code. But that's something else.) (8). Have you shopped around to see whether there are better codes for flow over airfoil? This is important, it applies to CFD code selection,.... to even the political system selection. (9).My answer to your question is: we are all looking for the right answer here, not just you. you will find the right answer, if you stay here long enough and if you are willing to experiment with some of the suggestions presented here. (10). By the way, the CFD froum is still here, because readers are still busy looking for the right answers. And even if I had the right answer for the flow over an airfoil, it would turn the forum into one-way system if I told you the right answer. (we want you to get the right answer. and we believe that you can do it.)

clifford bradford February 25, 2000 14:03

Re: How good is CFD?
i have been reading your posts. it seems like you are very impatient. 5% error is not bad. it is strange that people will take results they calculate by hand as being accurate when thee formulas often have so many assumptions embedded in them. do you know that stress calculations done with handbook formulas can be up to 20% off. so your 5% error is not bad. as john said CFD is difficult and if you continue to read here you'll see that many people who post here are people with years of experience in CFD yet they ask questions just like everybody else. have patience and you can improve your knowledge.

John C. Chien February 25, 2000 14:21

Re: How good is CFD?
(1). I think you are right about nature of the solutions from the commercial CFD codes. Different codes will give different answers.(same code also will give different answers by different users at different time) (2). Therefore, it is important for the users and the developers to do more validation work on their codes. The same is true for any products on the market. The commercial software does not give warranty on the use of the code, not to mention the warranty on the accuracy of the solution. (3). The reason why CFD is attractive is not because commercial CFD codes will produce very accurate results. It is because the use of CFD in the product design can improve the quality of the product, which is sometimes very hard to achieve by using wind tunnel testing. This potential benefit in the improved product design is the main source of attraction. (4). In a one-way system or society, the so-called right answer is always handed down in one pre-determined direction. On the other hand, in an interactive system or society, it is the responsibility of the user to find the right answer. It may not be very cost effective. But it will stimulate the over-all economic activities, thus provides higher probability of getting the right answers. (5). Up to now, CFD or any computer-aided engineering approach, is the by-product of the computer systems. So, the future of CFD is not controlled by the CFD commercial codes or its solution accuracy, but by the development of computer systems, both hardware and software. (6). The near-term suggestion is: more validation effort is required by the users, since there is no warranty on the solution accuracy by the vendors of the code. It is hard to get a product warranty in a flea market or from a swap meet. (7). So, a success story of applying the CFD technology in the product design by a company will have a long term impact on the value of the company because it reflects the dynamic side of the management and engineering team, which far out-weight the actual investment in the direct CFD related activities.

Aspens February 25, 2000 15:00

Re: How good is CFD?
Comparing with wind tunnel result is a good method. But how are you sure that that result is CORRECT? One is very difficult to say that this or that result is correct, we live in a real world, error should be taken rationally: first believe there SHOULD be error for everything, both CFD or experiment. second, one should try to tackle this error, and improve it. In several cases, CFD workers discover new things before experiment workers. Also, how are you sure that the calculation condition is exactly same with that of experiment? (this can induce significant error in turbulence modelling).

It is difficult to evaluate whether 5% error is good or not without seeing your detailed application (CFD is not for fun basically, it is application oriented). In some cases, 1% error is unacceptable (say, you can not spell the exact name of your advisor tomorrow), while some times 10% error is well accepted (say, I can not spell exactly your name tomorrow). It depends on your specific situation.

However, I still think 5% error is a good one. I remember, in boiling heat transfer, even the classic formula has very big error (say, sometimes 30~100%).


Jurek February 25, 2000 17:21

Re: How good is CFD?
5% is very good. This are tolerances you also have at FEM with different codes, and FEM is linear and has no Turbulence.


Alton J. Reich, P.E. February 26, 2000 06:49

Re: How good is CFD?
In my experience, most practitioners of engineering analysis (flow, structural, heat transfer, etc.) will tell you that they'd be happy to get answers that are within 5% of "reality". In many tryes of analysis, an answer within 5% is more than good enough. In structural analysis, for example, factors of safety built into allowable stresses leave ample margin for errors well in excess of 5%.

The question in CFD has long been "How good does the answer have to be?" The answer is often that it depends on what you are going to do with it. If you're using the answer to compare several design alternatives, then the computed solutions will all have a similar amount of error. In a sense, the errors "cancel" each other out, and allow the engineer to select the best design. In this scenario, 5% error (or more) may be perfectly acceptable.

If your goal is to prove that the algorithm you've just developed matched reality for a particular physical flow, then 5% error may look like a huge amount. It all depends on what you intend to do with the answer.

You may also find that a finer grid in critical regoins will improve the accuracy of your solution. The cost is computation time. You may find that it is possible to reduce your error to 1%, but it may cause the calculation to take 5 times longer. In a "production" CFD environment where CFD is used to support other processes, that time increase may be unacceptable, and therefore, 5% error represents the lesser of two evils. In a different environment without time constraints, you may want to try to minimize error regardless of computation time.


Vitaly Fishman February 28, 2000 09:53

Re: How good is CFD?
I am working in the field of Electronics Cooling. I tried to use PHOENICS to perform Thermal Analysis of Electronic Devices. It is really terrible!!! May be the man that wrote this code could force it to give appropriate results, but no one else. My opinion is that Good CFD is the software, which gives you the possibility to understand why your results doesn't match the experiment results. And gives good results after you fixed your "mistakes". In my field the softwares like Flotherm and Coolit give very good results, once you understood their logic. I think that comparison of CFD results with experiment results is the only way to evaluate it. At you can find plenty of different CFD codes. I believe that CFD is very good, but you have to find one that is good for you.

John C. Chien February 28, 2000 13:22

Re: How good is CFD?
(1). I am glad that there are satisfied CFD code users out there. (2). What is missing here is the story about the "terrible" results and the story about how the "good" results were discovered. (3). Which part of the process is responsible for the "terrible" results, and which part is responsible for the "good" results? The code( the mesh?, the turbulence models? the boundary conditions? or the solver?), The user(knowledge about the CFD? about the code? about the problem itself?). (4). I think, this is an interesting area.

John C. Chien February 28, 2000 14:40

Re: How good is CFD?
(1). I think, we really need to address the accuracy of CFD results in a more scientific way. (2). And this can be approached in a systematic way, namely, to evaluate the accuracy in terms of the various CFD steps involved. (3). We can evaluate the problem side and the process side. (4). For example, on the problem side, is it 2-D or 3-D? What would be considered as the limit in mesh size? Is the turbulence modelling critical to the accuracy of the solution? What level of turbulence modelling is best suitable for this problem? (5). On the process side, Is the first-order solution algorithm acceptable in this case? Is the solution sensitive to the mesh arrangement? Is the automatic mesh generation acceptable? Is the local mesh refinement required? What are the convergency criteria? ...... (6). I think, it is possible to design a table or chart to eveluate these parameters, before and after the solution is obatined to aid the evaluation of the solution accuracy. (7). For example, in the mesh density issue, one can plot a curve of solution VS number of mesh points. At the same time, the number of mesh points can be related to the computing time or the cost. In this way, high quality solution is going to be related to the high mesh density and also the higher computing cost. This sort of issues must be evaluated first, before one can put a cost on the solution accuracy. (8). Now, back to the basic issue of solution accuracy. If one can not obtain accurate solution regardless of the mesh density used, then apparently, the code must be returned to the factory for adjustment or repair. In this way, the user responsibility and the code vendor responsibility can be clearly defined. (9). What I am saying is: if the code vendor can supply a chart of solution VS number of mesh points, then he has demonstrated the convergency and the accuracy of the solution. And the accuracy of the user obtained solution must then be evaluated against this established chart, in terms of the mesh density. (10). I think, in order to develop a healthy industry in CFD, this type of guidelines must be followed. If a user is working on a brand new problem, then he needs to establish this chart first. (11). This same approach can be extended to include other parameters of the CFD process.

Fred Uckfield February 29, 2000 03:04

Re: How good is CFD?
In my experience 'bad' results are due to the following things (in order of importance):

Innaccurate geometry

-too much simplification

-critical geometries not accounted for

Errornous flow/heat BCs

-e.g. power dissipation known only to +-10%

-fan curves not appropriate with local downwind geoemtric obstruction

Not all physics captured

-wrong turbulence model

-e.g. no radiation accounted for

-coarse grid

Like Vitaly, I work in the electronics cooling field so the above may be rather specific. Interesting to note that grid is last, there are many more important aspects to get right before worrying unduely about grid!

A final word about 'good/bad' results. If you're using CFD to get something USEFUL out of an analysis then accuracy is one of many USEFUL things, certainly NOT the ONLY USEFUL thing. The majority of my work is involved with qualitative 'what if' analyses early in the design stage when the design hasn't even been finalised yet, let alone the geometry known! I get use out of the models by comparing trends. I have confidence in these trends, I look at % differences, I do not look at absolute degC, I am happy, I am more productive using this brand of CFD than not using it at all. If CFD is cars them I'm driving a VW Golf, you lot can go on driving Lotuses, TVRs, Caterhams etc... :)

John C. Chien February 29, 2000 10:37

Re: How good is CFD?
(1). It is really nice to hear your comments on the electronic cooling requirement. (2). I think, your type of design analysis is very different from the traditional CFD analysis of a well-defined problem, such as flow over an airfoil, flow over a cylinder, flow through a pipe,...etc, where the geometry is simple and well-defined. (3). In those simple cases, one is interested in the detailed flow field and the exact wall properties. So, without the fine mesh near the wall, the physics will be changed and the results will be "bad". (4). So, we can say that there is a category of CFD problems where the geometry is simple and well-defined, but the detailed flow field is important because it is used in the design of some other things, such as an airplane,...etc. (5). On the other hand, there is also another category of CFD problems, where the exact geometry is not well-defined, but the flow field is more sensitive to the arrangement of the geometry, which is not yet defined. In this type of CFD problems, the configuration change will determine the trend of the design. The detailed flow solution of a fixed geometry is no longer useful in this type of problems. (6). This is a rather interesting point of view, related to the "good" and "bad" CFD results. At least, for these two groups of CFD problems, the "good" and the "bad" CFD results are not defiend in the same way. (7). I am sure that I can say, one is more related to the search of the detailed physics (so that it can be used in the design correctly), and the other is more coupled through the optimization of the configuration (so that new arrangement of components can be derived to satisfy the heat transfer requirement). These seem to be very important criteria behind the general "solution accuracy" of CFD results.

low March 1, 2000 01:22

Re: How good is CFD?
So, do you think that CFD can only perform itself greatly in "fluid" and not other fields? In other words, is CFD not as good as FEM? Because FEM is capable to handle more "things".

John C. Chien March 1, 2000 11:01

Re: How good is CFD?
(1). CFD for complex geometry alone is 20 years behind the structure problems using FEM. (2). The commercially available general CFD codes capable of handling complex geometry is only rencent event. In late 80's, complex geometry problems must be treated separately as a special case, and it usually took several months to a year to get the geometry and mesh generated. In those days, the structure FEM were already in common use. (3). In electronic cooling design, it happens that the configuration does have a strong influence on the local heat transfer. As a result, parametric CFD analysis can produce the optimum configuration. On the other hand, it is hard to optimize the airfoil configuration because the shape is the only thing one can easily change. And one needs to know the precise answer to the shape change before he can use it in the final design. (if one is allowed to add and substract several wings at the same time, then it would be easier to design the wings by optimization, similar to the electronic cooling) (4). The convection term and the turbulence mixing are two major difficulties in handling fluid flow problems. Normally, these are missing from the FEM structure problems. (5). The convection term can be handled by the new stable solvers and larger mesh sizes, but the turbulence modeling has to be modified for each problem. So far, there is no ideal general turbulence model which will fit every application. (6). In the electronic cooling design, if the configuration is fixed and is not allowed to change, then accurate predictions of the flow field and heat transfer will become very critical, because a small change in geometry will have big effect on heat transfer. In this case, the optimization method will not be effective. One must still depend on the accurate CFD solution to improve the design.

COBOK March 1, 2000 13:35

Re: How good is CFD?
CFD stands for computational fluid dynamics. It says nothing particular about the numerical method used to solve the fluid mechanics problem. So, the door is wide open for everything: finite differences (actually, pioneering CFD works were based upon FDMs), finite/control volumes, finite elements, spectral methods, boundary elements and etc. When someone opposes CFD to FEM, he/she shows nothing but the lack of understanding of basic things in computational mechanics.

All times are GMT -4. The time now is 00:49.