While a real wind tunnel test of a car or a general vehicle that is considered as a bluff body, is vortex shedding present in the tests?

If so, from my point of view, many simulations and papers that shows many vehicle manufacturers are not correct, as they do (3D) CFD simulations that are steady and the are not... is this statement correct?

Thanks to all!

If vortex shedding is present or not depends on the shape of the vehicle as well as on the Reynolds number (i.e. on the speed of the car, and on the air temperature). You are probably right that steady-state solutions will not give you very accurate results, expecially on the drag. Vortex shedding will lead to increased drag as compared o the steady state drag. So, your statement is correct, under the assumption that the Reynolds numbers in actual tests are high enough to result in shedding. Although I don't know any actual results, I would strongly expect that. Especially on very bluff designs like SUV's or trucks! However, I would think that the manufacturers provide the experimentally obtained drag (which naturally includes all flow features) rather than relying solely the CFD result.

Many thanks for your contribution Mani!

Your reply helps me to introduce a set of observations I've seen in steady and unsteady simulations I've been developing these days. While running steady simulations of a 2D side profile of a real car, with a inlet velocity of 10m/s, solver (Fluent; k-e Realizble) threw me oscillatory results when switching to 2nd order schemes, in Cd, Cl, Mass-Weighted averaged quantities and residuals, so I concluded that my simulation, although it was with a low velocity, flow showed vortex shedding instability and I must run an unsteady simulation. In 1st order schemes, simulation gave my converged non-oscillatory results and a Cd=0.303 (very good result though!)

In the unsteady results, Cd was in the range 0.33-0.36, quite far from Cd=0.31 that manufacturer says. Now, new questions come to me:

1. When you said "Vortex shedding will lead to increased drag as compared o the steady state drag", you mean that the MEAN Cd computed from the unsteady results will be higher than steady simulations (like in my case)?

2. Can I assert that results made with 1st order discretizations are correct if my objective is only the computation of the Cd and not seeing the real behaviour of the flow?

3. I still cannot understand how steady RANS models can show me vortex shedding when I make contour plots in different iterarions, as simulation has been unsteady... I expected to see the same plot in different iterations; steady RANS seems to be a pseudo-unsteady solvers to me. Can anybody explain me this behaviour?

The manufacturer's results, if measured, were on a 3-d car. Can your question be answered completely without doing the 3-d calculation?

Jim Park's remark is correct. You may expect the drag to be smaller in the 3D case, so it's really hard to say how much of the discrepancy is due to vortex shedding, and how much is due to using a 2D versus 3D model.

But that left aside...

>the MEAN Cd computed from the unsteady results will be higher than steady simulations (like in my case)?

Exactly. The time-averaged Cd, processed from unsteady computations, will be higher than the Cd obtained from the steady-state equations. Reason: the unsteady result will include power input to the flow (due to unsteady vortex shedding) that is not accounted for in the steady computation. This effect may or may not be small, depending on the contribution of vortex shedding to the total drag.

>Can I assert that results made with 1st order discretizations are correct if my objective is only the computation of the Cd and not seeing the real behaviour of the flow?

First of all, I am not surprised your lower order model does not show any oscillations. Most likely the 1st order model is much more diffusive than the higher order models, smoothing out any flow instabilities. You are essentially increasing the effective viscosity by introducing numerical diffusion (not real diffusion). This will not give you the most accurate Cd, but you can surely use it as a first order approximation. If unsteady effects are negligible, and you do this in 3D, then it should be sufficient.

>I still cannot understand how steady RANS models can show me vortex shedding

The RANS model will not show you any small scale vortical structures. However, there could be large vortices shedding from the vehicle (think of a Karman Vortex Street behind a cylinder). Even if vortices are not actually shedding, the recirculation regions behind the vehicle could be pulsating (changing size and shape over time). If none of this is apparent in your unsteady result, even using a high temporal and spatial resolution, then you can say that there are no significant unsteady effects. In that case there won't be much effect on the Cd, and I would definitely proceed to the steady 3D case.

Thank you very much, Mani and Jim_Park.

Well, that's true: all manufacturers CFD results I've seen are in 3D and mine are in 2D. They are not comparable, but two questions arise after reading your replies: my 2D Steady, k-e Realizable, 2nd order scheme simulations gave my oscillatory results that showed me the unsteadyness of the case.

Q1. I don't know if I've understood well one thing about the 3rd answer you gave me, Mani: althoungh I make plots between 2 iterations made with an "oscillating converged" STEADY RANS solution, they are not showing me nothing REAL, as the oscillation indicates me that I will get real results running an unsteady sim. Is this correct? I wish I could show you those plots, they are really disturbing to me...

Q2. Two manufacturer's papers I have show that the 3D simulation finished with an STEADY Realizable and 2nd order schemes (it began with 1st order schemes to get an initial solution), but any paper shows any oscillation in their results (neither CD, Cl nor residuals) so I conlude there is not any vortex shedding present. The question is, it could be possible that because there are "more physics" implicated in a 3D CFD simualtion, it is easier to converge them than 2D ones? I mean, 2D simulations tends to instabilize the flow much more than 3D ones?

Thanks a lot. You're really enlighten me!

This is not a criticism, just an observation that you've probably made as well.

When driving behind a semi-trailer (a 'great bloody lorry' in Great Britain I think) at highway speed, the vortices shedding from the trailer really shake my little Accord. It's definitely unsteady flow behind those vehicles and speeds. Point being that there are flow conditions in the real world where the shedding is real and significant. Not all speeds, but some.

If you're observing that some simulations (yours or the transporations folks') show no shedding (unsteady behavior), either the simulation boundary conditions (flow speed, etc) are not in the region where shedding occurs - OR the simulations are not realistic.

The lack of realism may be 2D instead of 3D, or the code(s) is/are too diffusive, or the mesh is wrong, or ... .

OK Jim_Park, clear as water!

Now I will extend my research to 3D simulations. But imagine 1 minute the following example I propose you to think about: now I go to 3D simulations with the same conditions as I introduced into my 2D simulations (same vel. inlet, turb.int, etc.) and, after converging a 1st order scheme simulation, then I switch to 2nd order and it also converges (as all manufacturer's papers I've seen). If I had this situation, how would it be explained? How a steady RANS in 3D can get a converged result and 2D sim. with the same conditions don't? With your last replies I guess that it could be explained as 2D sim. has (over)predicted a flow instability (vortex shedding) that in 3D -and also with many probabilities (because it is a 3D sim) in the reality- does not exist; and this is because 2D sims. are not as "real" as 3D. Do you agree with the last conclusions? If you want and the last statement was correct, you could give a more formal explanation , it would be realy enlightening to me!

Thank you very much Jim_park for your replies. They are helping me a lot with the insights of my sims.

>the vortices shedding from the trailer really shake my little Accord. It's definitely unsteady flow behind those vehicles and speeds.

How do you know that your Accord does not shake in the steady bound vortices coming off the sides of the truck? The unsteadiness might actually originate as a reaction to the wakes, because of the presence of your car. Like you, I would expect that most likely the wake is unsteady, but if that is from actual shedding of vortices or just from swaying of bound vortices or from other sources is far from "definite"... which brings me to the point: Unsteadiness is not necessarily an indicator of vortex shedding. There could be something else going on that needs to be examined.

to Q1:

Oscillations in the steady computation "might" indicate a flow instability, but that's not certain. There could be other reasons for such convergence problems. What I meant is that you should examine the result from a time-accurate unsteady computation. Print out the flow field at every time step in the unsteady computation and then look at them (you could create an animation in your favorite visualizer). You could also examine the real-time history of the aerodynamic forces (drag, lift) on the vehicle.

to Q2:

Again, oscillation of the steady computation does not prove unsteadiness or even vortex shedding. Vice versa, even if in reality there should be vortex shedding, the manufacturer's CFD code might still produce a converged steady state. That very much depends on the method (scheme, solution method, numerical parameters...). Either way, unfortunately you can not draw a conclusion based on the steady convergence history. You will need to run a time-accurate unsteady computation to find out if the flow is unsteady.

About 2D versus 3D:

Yes, it is possible (actually likely) that your 2D simulation tends to shed more easily. The prime example for this is vortex shedding over a cylinder (for Re about 50 and above) versus shedding over a sphere (for Re about 200 and above). 2D flow is more constraint and likely to behave more "violently" than 3D flow. If your steady-state convergence really suffers because of flow instability, I wouldn't be surprised to see that your 3D computation converges better. However, again you can not draw any conclusions until you have actually run it and examined the unsteady result.

"How do you know that your Accord does not shake in the steady bound vortices coming off the sides of the truck? The unsteadiness might actually originate as a reaction to the wakes, because of the presence of your car."

Caught again. I overstated the conclusion. Although I've felt the shake from a good ways behind the truck and I've observed straps or short pieces of rope flaying around, it's not conclusive. Guilty as charged.

The trouble is that I don't know what the manufacturers have published, how good a job they did with errors, resolution, and on and on. I also don't know if the referees were thorough. From what you're seeing, it appears that there are still some puzzles to be resolved.

You're in the best position of anyone to work this thing out. There's surely a good paper there someplace. And of course, some valuable knowledge. Good luck.

Ok guys, last posts were really clear! Now I've got a wider knowledge about these fenomena that occurs in my 2D sims. and I'm ready to go and see what will happen in 3D ones. I'm really curious to get 3D results and compare them with 2D to assimilate/verify all discussed in this great thread you've done (at least it was great to me

Many thanks Mani and Jim_Park!