CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   Mercedes-Benz External Flow Solution using STAR-CD (

Ed Frulla June 9, 1999 11:28

Mercedes-Benz External Flow Solution using STAR-CD
Mercedes-Benz performed an evaluation of CFD results for the external flow over an E-Series sedan verses the results obtained from a Wind Tunnel.

They used STAR-HPC and a 10,000,000 point mesh.

I would like to obtain a white paper of this experiment for evaluation.

Does anyone know where I can obtain this information?

Was it ever published?

John C. Chien June 9, 1999 13:16

Re: Mercedes-Benz External Flow Solution using STAR-CD
(1). Do you expect to see something new? (2). A real car geometry is highly complex. Whether the result is realistic or not depends mainly on the turbulence model used in the computation. (3). But so far, the turbulence modeling for 3-D complex geometry with flow separations is still in research stage. The drag prediction will depend on the accurate prediction of the flow separation locations. It is easier to send e-mail to the car company to locate the information if you think it exists.

Dr. Hrvoje Jasak June 10, 1999 09:59

Re: Mercedes-Benz External Flow Solution using STAR-CD
I'm sorry, but you're wrong: this calculation has been performed on 2 meshes: one with 1,2 million cells and another with 9 million. The drag coefficient for the fine mesh is within 0.5 % of the measured value, and that is with the standard k-epsilon with wall functions (not bad!). Also, the difference between the two results was considerable: you really need this sort of resolution.

The reference for the paper is:

Reister, Leidig, Bauer, Robertson: Flow around a passanger car comparison of numerical results with measurements, Proceedings of the 29th ISATA Conference, Florence, 1996

(If you contact Computational Dynamics, you might be able to get it there.)

John C. Chien June 10, 1999 10:44

Re: Mercedes-Benz External Flow Solution using STAR-CD
(1). Thank you for the information. (2). 0.5% of the measured value using the standard k-epsilon with wall functions is very good. I agree with you that a fine mesh is essential to get a good result.

andy June 10, 1999 12:16

Re: Mercedes-Benz External Flow Solution using STAR-CD
Given the known limitations of the k-e turbulence model for this class of flow, I would tend to view an "accurate" prediction with some suspicion and ask the obvious question: how accurate was the lift coefficient?

John C. Chien June 10, 1999 13:32

Re: Mercedes-Benz External Flow Solution using STAR-CD
(1). With considerable difference between the result of 1.2 million points mesh and that of 9 million points mesh, one really need another data point to establish the convergence trend. (2). I don't know whether the solution at 15 million points will over shoot the test data or not. Anyway, this is not something one can do easily everyday. (3). I think, it is possible to have good result when the design is good. Formula-1 would be the ultimate candidate to test the turbulence model.

Peter Dittrich July 3, 1999 16:11

Re: Mercedes-Benz External Flow Solution using STAR-CD
The k-e turbulence model is a crude approximation for closing the numerical solution of the coupled transport equations. To claim that with pure grid refinement an accurate solution of the turbulent flow problem can be achieved is a gread oversimplification of the fundamental turbulence problem.

For example, consider stagnation flow in front of the car. Since the k-e turbulence model scales the fluctuations proportianal to the mean velocity gradients, for stagnation flows the turbulent kinetic energy is greatly underestimated.

As long as there is only a law of the wall model and no near wall turbulent boundary layer model is used, all these solutions must be considered "accidental".

Dr. Hrvoje Jasak July 6, 1999 04:55

Re: Mercedes-Benz External Flow Solution using STAR-CD
In the interest of wider audience, I feel obliged to reply. The "fundamental turbulence problem" has absolutely nothing to do with the issue here. What we are talking about is whether one engineering method (CFD) may be used to replace an older, traditional and more expensive method (prototype testing) for the purpose of ESTIMATING the characteristics of a road car under realistic working conditions. Let us first consider the experiment: in this case, apart from the Cd and Cl number, several velocity profiles and pressure distribution were also available. In real industrial environment, this is rarely the case. The two numbers were then used to "optimise" the shape of the car body to keep the fuel consumption under control. Now consider this: 1) the flow around the car is unsteady and hence cannot be described by a single number. 2) The prototype in the wind tunnel was simlified: although the wheels were turning, the brake disks were replaced with flat plates (in reality they've got cooling passages), the radiator grill was blocked, the engine was switched off, the pasanger compartment was sealed etc. etc. 3) The tests were done in a wind tunnel with specific characteristics of inlet turbulence (not in equilibrium), not to mention that the velocity profile near the belt wasn't quite right. Also, very few roads are as smooth as the wind tunnel belt and I have never seen a commuter route which keeps the same wind-to-car-axis angle all the way. The point I'm trying to make is that people do not usually drive their cars in wind tunnels and in any case, the one in there wasn't fit for driving. The Cd and Cl parameters are engineering quantities, produced (very well!) for a certain purpose. Now, let us consider the calculation. I could again keep going on and on about the simplifications that were being made, but instead I would just like to concontrate on the turbulence model. We, of course, may decide to exactly resolve the turbulence (the equations are well-known), but right now this is a bit outside of everybody's price range. Therefore, turbulence modeling is required. I had a nice discussin with Prof Bradshaw (Stanford) about this problem, and he said something that really nailed me to the ground: turbulence modelling is NOT a science, but a bunch of engineering correlations. Therefore, we cannot talk about them as "right" or "wrong" but only as "appropriate" and "not appropriate". Many people (including myself about 5 years ago) have been saying a lot of rubbish about the k-e model and wall functions, some of us have been declaring allegiance to Reynolds Stress, low-Re, LES etc., but the point is that (as an engineering correlation) we cannot discount the k-e model without first establishing its limitations. In my mind, this aerodynamics calculation (still on a relatively coarse grid) shows that the "old" k-e model may have a LOT of mileage in it if used with care! Of, course, you may claim there are accidents and accidents, but if you look around you, you may find that most (all?) of our engineering environment is based on correlations which have more or less basis in the physics, all of which have serious limitations and in spite of that prove to be up to the job! BTW, if you switch on the air-conditioning in your car, the fuel consumption will rise by about 10% and I'm sure nobody told you that when you were buying a car.

John C. Chien July 6, 1999 09:46

Re: Mercedes-Benz External Flow Solution using STAR-CD
(1). I have expected this to happen. So, I waited and waited. Finally, I am seeing something which is very common to most people in carrying out testing, comparison, and conclusions. (2). The statements are consistent with what I had expected. So, there is no surprise. And Peter's comment was important to bring out this somewhat detailed description. (3). I think, one could say that under very limited conditions, the calculated results looked fairly good when compared with test results using simplified car models. At this level, I think it was acceptable because it was not designed to test the Turbulence model and the wall function. (4). Unfortunately, the extended statements to cover the validity of the two-equation k-epsilon equation model with wall function was misleading, regardless of whether it was used mainly for the engineering purposes or not. ( In early 90's, I developed a 2-D code on PC to calculate flow over a car center plane geometry. I was able to copy the geometry of a car directly from magazine photos and performed the equivalent constant viscosity flow calculations. The result was close to the stated drag coefficient mentioned in the magazine. The computing time for each case on PC took only about 10 minuites in those days. But I could not say that my laminar flow calculation had predicted the right drag coefficient of a car. It was used to demonstrate the use of a 19 point piece-wise quartic curve to represent a realistc car shape.) (5). The validity and the limitation of the two equation k-epsilon models and the use of the wall functions are well-known. I don't think it will be changed by this car simulation results. (6).Unfortunately, the names mentioned in the posting will probably receive somewhat negative impression. (7). I am not going to comment on the personal love and hate feeling about the particular turulence model. ( I don't think that it is the right approach ( with strong personal feeling) to do the engineering or the scientific study. Next time, when a paper is presented, it is probably important to present results using two turbulence models and two numerical methods in additions to the two meshes required, so that a more consistent conclusion can be made.) (8). As I said before, the survival of CFD depends on unsolved problems. The sure way to die ( for a field or a company ) is to pretend that the problem has been solved. I am afraid that many large companies without in-house CFD staffs and projects are heading in the direction of no return.

All times are GMT -4. The time now is 05:03.