Hallo everybody

Could anyone give me some idea of how precise CFD-programs can predict wall shear stress in complox geometries. I know it is difficult, but If anyone has experiences with it I would like to know??

Regards

Bo

(1). Most turbulence models were calibrated for the flow over a flat plate (Law of Wall type) data. (2). So, the model has built-in solution for such standard flows. (3). For flow over smooth surface with favorable pressure gradient, there was no big problem there. But for flow over smooth surface with adverse pressure gradient, most turbulence models will give differenct results. (4). For flow over complex geometry, with or without flow separations, the predicted wall shear stress is "not useful" at all. (But the flow field results might provide parametric information which can be very valuable for design and analysis purposes.) (5). Predicting accurately the wall shear stress and heat transfer in complex flows is a research field. We are not there yet. (can you predict accurately the wall shear stress and the wall heat transfer for flow over a simple backward-facing step? Not really. It's possible, if you have the test data. Well, it can be fine tuned to match the data of a particular type of problem. )

Thank you

But how far away from the correct values are you?? And can the use of the two-layer models give improvements to the predictions of the wall shear stress?

Regards

Bo

(1). If you have only the 3-D cfd results and no previous data or experience, then the wall shear stress is totally useless (the absolute value). (2). So far, we are using its trend only. I mean the behavior between two calculations. (3). But this can be very dangerous, because the flow field can also change in the parametric study. (4). IT IS ESSENTIAL THAT YOU HAVE SOME TEST DATA, in addition to the 3-D cfd predictions. (5). For certain types of problem, which have been studied for many years, it is a different story. There, test data have been used to guide the modeling, thus the results will be more repeatable. (6). So, the major part of cfd analysis is to fine tune the turbulence model. If you are using a fixed model with fixed set of parameters, then you are not going to get accurate results for the wall shear stress and wall heat transfer at all. (7). It is not just difficult, it is because we know very little about the real flow problem.

Alright then. So if I do not have measurements of wall shear stress in the 3D geometry that I am working on I have no chance at all to predict the wall shear stresses?? Then I am in really trouble because what I have seen until now is that no good method is available to measure the wall shear stress in complex geometries with liqued flow. Alright in descrete points it might be possible, but thats not worth that much to me. Do you have any idea how much CFD programs are of the correct values of the wall shear stress??

Regards

Bo

(1). Only on special occasions, when I was doing validation, (I mean developing a turbulence model), I had to run some test cases against measurement in wall skin friction coefficient, for simple flow benchmark problem. (2). That is why I have said many times, that 99% of times, the cfd results obtained are wrong. (the results are still very valuable, except that the absolute values are all wrong.) (3). The correct answer is simply not in the formulation and the code. But since we are solving the conservation equations in some ways, 3-D results will reveal more detail information about the flow field. (4). If no one is doing research in industries or academic institutions on that particular problem, then, it is unlikely that you will get any correct answer at all for that particular problem. Remember that a code is simply a library of subroutines and models. These alone do not guarantee a solution , not even an accurate solution. (5). Each time I run a 3-D commercial cfd code, I have to do it several times in order to get it started, otherwise, it will diverge. It is a function of many things. All I can say is, these subroutines and models do not have the intelligent to find a solution. (6). But since the testing sometimes is not possible, cfd is one alternative. But it is not a sure one yet.

WSS depends on the flow situation inside the domain you are modeling. I got quite good results using FIDAP for the calculation of WSS compared to experimental WSS estimates from Laser Doppler discrete point measurements, the flow is laminar, though. But it was directly validated for both steady and unsteady flow conditions and the geometry was complicated too.

For turbulence flows, it is very hard to match to the experiments because to me, I think those turbulence models are developed like a curve-fit (ie. law of the wall) and there are so many parameters involved which in commercial codes were optimized for certain geometries. It would better you get your own code and input with your experimental data for a validation of WSS, otherwise it is quite hard to get good results for the details...

John is right in saying that, in general, the wall shear stress CFD predicts is not very reliable. This statement is obviously not true for particular turbulence models calibrated using test data on specific types of geometry.

So, when you say you have a complex geometry, you have to be more specific. For example, the flows over a backward facing step are quite comples (involving separation), but lot of Reynolds averaged (K-e, second order closures) have been tested (or "calibrated) for these flows. You have to find turbulence models that were developed using data from flow fields which have the same "complexity" as yours.

LES is probably more reliable if you are simulating a flow field for the first time and you have no test data or apriori knowledge (from past research) about the flow. The problem is that LES of flows in compex geometries is not quite mature enough. LES so far are done mostly on structured meshes and the numerical artifacts such as artificial dissipation and/or dispersion for schemes on such grids are well characterized. Numerical dissipation especially has a significant effect on the predictions in LES. Codes for LES on unstructed and structured multi-block meshes are still under development. No word is out on their performance in really complex fields.

For LES using complex meshes, you can visit the these websites

http://www.scorec.rpi.edu/~kjansen/

http://www.csar.uiuc.edu/F_info/Res_CFD_97-98.htm

http://vortex.mech.surrey.ac.uk/LESCF/db182.html

http://www.public.iastate.edu/~tsimons/

(1). Well, that's a good idea. (2). When you are having trouble in getting the wall shear stresses or heat transfer, it is the time to run some low Reynolds number laminar flow calculations. (3). At least in this way, we can simply say that it is the problem with the turbulence model. (4). I think, when you are solving problems with cfd approach, everything should be included. I mean, you are suppose to create the best mesh, run the test, validate the turbulence model, and plot the results. Otherwise, a person doing the turbulence modeling research is not going to find a job at all. (What I am trying to say is, if in industries people only run a code, then it is useless to train a person to do turbulence modeling research, because after graduation, he can only find the number crunching jobs) (5). So, it takes the whole society's effort to make a small progress.

John, you are straying away from the topic, but that's another good point: the way of CFD approach should include what you have said, best mesh, test, validation, reports. However, that mostly exist in academic only, some industry use this approach but I bet most do not. In my opinion, industries choosing CFD as an analysis tool is because it is 1) Cheap or getting cheap; 2) Resonability to convicence; 3) the way the commercial codes are designed.

They are different from academic as so pick on the solutions. They want a quick solution, most of the time, good or bad, that's it. They do not look at the solution so close as in academia (otherwise I wouldn't be spending so much time in school), let alone WSS, imagine somebody discussing wall shear stress gradient in a project, are you crazy, nobody's gonna to understand you...

The way the commercial CFD codes were developed to be a blackbox helps in this way, too. I bet a high-school graduate knowing about autoCAD can now or soon generate a fairly complex geometry/mesh in gambit and following predures in fluent to get a solution.

(1). Well, we now have two issues, one is the state of the wall shear stress prediction and the other is the source of this problem. (2). Back to the wall shear stress prediction, I think we all agree that a low Reynolds number model type must be used to cover the whole flow field in order to compute the near wall region. This require fine mesh near the wall. In addition, we need a reliable turbulence model. (3). In order to make the model reliable, one needs to have good test data under the controlled environment, which can only be found by doing careful research project. (not in industries though) (4). So, it seems to me that the reliability of the wall shear stress prediction by cfd is actually linked to the controlled test data. (5). The industries issue can be changed by the public. For example, the environment issues, the health issues related to the tobaco usage, the safety issue of automobiles, etc... So, the public does have the influence on the industries in terms of the quality of products and services. (6). And one of the most important goal of this forum is to educate the public and the industries about the usefulness of cfd and also how to provide the proper environment to encourage the cfd development. (7). The other way to check the state of the art of wall shear stress prediction is to read the technical journal papers and see how many papers have included the wall skin friction coefficient as part of the cfd results. Most include the pressure distributions, but very few include the wall skin friction distributions. (8). The wall shear stress is related to the loss of the flow. In turbomachinery, the efficiency is related to the power output, and the understanding of the loss can lead to the more efficient design. So, the accuracy of the better turbulence model can have great impact on the power output. I think, in gereral, the industries understand that relationship. But like the tobaco industries, still, the future is in the hand of the people. (9). So, the wall shear stress prediction, though it looks like a part of turbulence modeling effort, it has great impact on the energy usage in daily life. The better turbulence model developed can save a lot of energy and money. (through the more efficient products designed)

First of all Tahnks for all the good answers.

Well with regards to 2):

You say that a low Re model should be used near the walls and a fine mesh sjhould be made.

Now If I do that could I then get results for the WSS that hardcore CFD people (like you) would not regard as usefull??

Bo

(1). If you pick a low Reynolds number turbulence model(or the like) and use it consistently with the required fine mesh near the wall, then, for the simple benchmark problem like the flow over a flat plate or in a pipe, you should get fairly accurate wall shear stress result. (2). This is simply because the model you pick was validated for such type of flows. (3). For other types of flows, you are not likely to get the same kind of accuracy. (4). But, then, if you have good test data of that problem, you could go ahead and fine tune the model again until the results match the data. Presumely with this exercise, the future prediction of the same type of problem, using this updated model, should give you the same degree of accurcy. (5). It is more or less like fine tuning the car. Each one is designed differently, but each one can be adjusted to near perfect condition again. (6). This is the message I was trying to say. The engineer should have the capability to adjust the model for the particular problem at hand, and the working environment also should allow the engineer to perfect the model and perform the task. (6). If you say that, once a model is validated for the flat plate case, it is safe to use it everywhere. Then, you will end up with the current state of the art of the wall shear stress prediction. If it is not in the option list of the code, you are out of the luck. This is not the healthy environment, right? (7). So, there is always this "extra step" which makes a professional different from the ordinary person. And that is how we can all move ahead and make some progress. (There is no short cut, if you are interested in becoming a professional.)

Hallo John

Well I woul dvery much like to have WSS data to compare my simulations with. However I dont have that. You keep saying that WSS predictions are not that accurat and so on. But do you have any idea of the percentage that the predicted WSS are wrong (generally)? Does it over- or under predict?

Secondly you talk a lot about fine tuning the model, which I easily can relate to. But with a geometry with different flow structures I do not think I can fine tune the model as it might be good in some areas and bad in tohers. Is that correct??

And finally. Are you aware of any work going on for solving this problem thatwe have discussed here??

Regards

Bo

(1). In the last few years, I have been checking out some commercial codes. (2). For some very simple cases, it is possible to get numbers within a factor of two. (3). In other more complicated flow problems, still fairly standard, the number is easily over a factor of two. (4). Since these are commercial codes, it is hard to know the source of errors. I mean, it could be the turbulence model, or it could be the numerical algorithm, or both. I run the commercial codes daily, so these are fairly common to any users. (you can easily run a case and make your own conclusion) And when the results are not accurate, you simply look at other features of the results. (5). Well, for the persons who write their own codes, the adjustment to the turbulence model is not a big issue. (6). I am not trying to say that cfd codes are not accurate when predicting the wall shear stress. I am trying to say that if the turbulence model is not validated for the specific complex flow problem being solved, it is likely that the wall shear stress prediction is not going to be accurate. (100% sure. and it will be several times off the right answer. That's life.) And if you are not following the rule such as fine mesh near the wall, you can get almost any number you want. You can easily show that even the physics of the flow will change because of the mesh size. (7). CFD is a very difficult subject by itself.