Hi everyone,

I am (still...!) working on the aerodynamics of a car with strong wake(s), actually a single seater car (kind of a small formula one...).

Despite all of my work on the surface and volume mesh, I am still not satisfied with my drag and lift results. Not that they are not realistic, but mostly that they still oscillate too much for the accuracy that I need (more than 1% compare to the max value in SCx and more than 2% in SCz).

My number of prism layer is 15 (I cannot more due to computer capacity) with a first cell at Y+=1, and a geometric expand ratio of 1.3 (actually I tested from 1.15 to 1.3 but not so much difference on the convergence).

My volume mesh is quite well refined, but because of strong wakes, I want to use a mesh adaption (I have, I think, computer memory space for more than 2 millions cells more). But I am not sure of wich criteria would be the best? Maybe another than the proposed ones?

Does anyone has experienced a better drag and lift convergence with a good mesh adaption?

One more thing, maybe some phenomena are not steady, but I really want a steady study.

Thanks a lot in advance, 'cause I really get crazy after so much work on this!!!

PS: I am running a sst turbulence model, and I was thinking of using a bigger physical timescale. I now have a timescale close to a second, maybe should I try something like 100s? more? (because 20s didnt change anything...)

Hi,

I think you will find the lift and drag are in fact not steady so your failure to find a steady state solution is in fact correct. Have you tried a transient simulation and looked at the lift and drag over time?

Glenn Horrocks

Hello Glenn,

No I haven't tried this, because I'ld like to compare different configuration on this car, so I'ld like to compare kind of average values...

I think transient simulation would take too much time, and I don't want to spend more than a week to calculate solutions! (Currently it's about 200 iterations a day).

Don't you have any clue of an appropriate mesh adaption criteria? I'm sure you are right when you say lift and drag are not constant (as the wind tunnel data that I have), but so far I found the percentage of oscillations I told you a bit too big. I understand the oscillations of my values in the steady state study are not the ones that I would find in a transient simulation, but I really have to stick to a steady simulation I think. And I really would like to test a mesh refinement (I just tried a very big timescale, around 100s, but it was not the trick obviously!)

Thanks!

Hi,

You need to establish mesh size independence for you simulation, but you should also establish whether it is significantly transient or not. If the transient nature is significant and you try to do a steady state simulation then your answer will be wrong.

Appropriate mesh adaption criteria depend on the simulation. For external aerodynamics something like pressure or velocity gradient sound a good place to start.

"I'ld like to compare different configuration on this car, so I'ld like to compare kind of average values..." - in a transient simulation you get a time history of lift and drag and then average that against time. You can easily still reduce it to a single number.

"I think transient simulation would take too much time" - do you want the correct answer or not? The correct answer pays no regard to how much time you have available. Get more computers and run parallel.

Glenn Horrocks

Hi Glenn,

First thank you for your fast answer (I think you are in Australia, so that's why there is a big delay with me in France...!).

You are defenitly right about the mesh size independence, and I haven't thought about that with a mesh adaption...

I can have another computer so I will follow your advice about the transient simulation, you definitly set my mind to this idea.

One last thing. I know how are measure these values (of lift and drag) in the wind tunnel (from where I have datas), in terms of duration and steps (to speak as a CFX language). Do you think I should use the same parameters in CFX or should I try first a long duration with small steps to have an idea of what should be the optimum time duration and steps (to then reduce computer time)? Maybe if there is a kind of periodicity...

Pascal

Hi,

In a transient simulation you also need to establish that you have timestep independence. In CFX a good way of getting an idea of the timesteps you need is to use adaptive timestepping, and get it to home in on 3-5 timesteps per iteration. In most cases this will give you a timestep size which is a good compromise between computational speed and accuracy. Start off with a really small timestep size and let the software find it from there.

Glenn Horrocks

Ok!

One (very!) last thing. I still have some high RMS MAX (10^-2). This appends always close to my boundary layer in different places, although the quality of my surface mesh is very good (even the quality of my prism layers). I think this is due to the transition from prism to tetra, where there is a big 'scale' difference between cells.

So far, I think i's a bad idea to increase to much the geometric ratio when building prism layers. To use SST model, I need first cell at Y+=1 (even a bit below), so it's difficult to then have a good transition from prism to tetra... (I cannot use something else but prism, because of the complexity of the geometry)

Do you think I should use the scalable wall functions and use a k-e model (for example RNG)? Like that I could increase the size of my first cell (let's say at Y+=20) and have a better transition from prism to tetra (while keeping a low geometric ratio). What do you think about that?

Sorry to ask you this kind of questions, but throughout all the courses that I have, theory is easy to apply with small studies, when you are not limited with computer capacities!!

Thanks again

Hi,

If you are looking for accurate boundary layer modelling you need to make sure the entire boundary layer is contained inside the prism layers. You may need 20, 50 or more layers to do this. Make the transition from prisms to tets such that the volume of adjacent elements is as close to 1 as possible.

No need to go to k-e for y+>20. Stick with SST and use the automatic wall functions and let the software sort it out. Do a sensitivity check to gauge your level of accuracy, and check against good experimental data if possible.

Glenn Horrocks

Hi Glenn,

Thank you again for your response. This was something I was wondering about the boundary layer. I finally end with an Y+=1, 16 layers and a growth ratio of 1.1 (it appears that a bigger one gives an earlier (and maybe wrong but not sure yet)) transition point in some parts of my geometry. So far it correlates a point that I've seen in the manual. Even if my last prism layer stays below Y+=45, I think it's the best choice so far. It's too bad, because I know I can go really further in Y+, after so much time spend on the surfacique mesh!!

I'll try to see if it's possible for me to use bigger computer capacity do define my mesh (and thus to solve it). At least for a while, and then I will try to see what is the optimum and what could be the smallest size of my mesh to have decent results.

Wahou! So big stuff!

So if you don't hear from me for a while, that means I'm getting closer to the right thing!

Hi again,

'Make the transition from prisms to tets such that the volume of adjacent elements is as close to 1 as possible' when you say 1 you mean the volume ratio? But how is it possible to check this? (I'm using Tgrid)

Ok,

I think I got it, in quality measure I check size change, and then I just do mesh -> check, and I've got my 'volume ratio'. Am I right? But it's really difficult to see where, close to boundaries (and especially prism cap), this ratio is too high in Tgrid.

So I see some max value at 4, but average at 1.1 (which is good). I guess I won't be able to avoid these max values, due to skewness problem then at intersection of boundary zones. (I have checked that CFX tells me that the max residuals are always at these intersection, for example symmetry plan, boxes, etc...). But do you know a way to improve this ratio with Tgrid?

Do you think I will have to deal with these max values of size change (leading to high max residuals), and stay focus on the rms residuals? Believe me, I can't do a better job on the surfacique mesh after so many weeks...

Sorry for all these posts today! I know it doesnt deal with cfx but thank you again if you are of help.

Pascal

Hi,

I am not familiar with TGrid so can't help you there.

A max volume ratio of 4 and average of 1.1 should be OK for most simulations. If you are looking for very accurate boundary layer resolution you need to go better but for most general cases that should be OK. However you say the max residuals always occurs at the boundary so you may well have to improve that area.

Glenn Horrocks

Hi,

I see you are using the turbulence transition model. This model is very sensitive to mesh quality. A growth ratio of 1.1 is probably too big. Try 1.05 or less. Also 16 layers is probably not enough. With the small growth ratio putting more layers in should be possible.

Can you describe your simulation? Why are you using the turbulence transition model in a simulation with a large wake region? The turbulence transition model is tuned for aerofoils so no guarantees about its accuracy on other things.

Glenn Horrocks

Hi,

So you would recommend me to use an simpler model like RNG k-e instead of SST? But then I use the scalable wall functions... So then first cell at something like Y+=20 is enough right?

As I told you my model is a single-seater, close to something like a formula 3 (but I'm sure you see what is it!). Airfoils, tyre ramp, gurney, underbody, diffuser, the complete car! But wheels are closed (to simplify the mesh and thus reduce it).

I was using SST because it was one of the way to have an automatic wall function.

It's difficult, because I'm sure SST is a right thing to describe behaviours of my airfoils, but I agree it can be a problem in a large wake zone... What would be your recommendation?

Thanks,

Pascal

Hi again Glenn,

I see that in the SST model you can check or uncheck the transitional turbulence. SO would you say it's correct if I leave it unchecked? I don't see many explaination about that in the manuel...

Ok I see that we just misunderstound... I have always used the SST model without any transitional turbulence model. When I was refering to an article in the manual, it was just because between a growth ratio of 1.3 and 1.1, I was seeing a difference looking at the wall shear in different places, and thus in my SCx and SCz. But I wasn't doing the same approach as describe in the manual with very low Y+ and a transitional turbulence model. So to leave it unchecked is correct right?

Thanks again Glenn, and very sorry if I overload you email box, it's not my intention

Pascal

Hi,

If you are looking at race car aerodynamics then forget turbulence transition. A far more important feature is resolving separations and wake effects. Therefore I recommend you look start your modelling using a standard SST turbulence model, but once you are comfortable with the SST model look at the DES and SAS turbulence models in CFX. These are extensions of the SST model to allow you to capture the transient wake and therefore its effect on lift and drag.

Glenn Horrocks

Hi Glenn,

Thanks again, I think everything is clear with your recommendations and also with all the first results that I have.

As I already said I will follow your advice with the transient simulation.

Many thanks again for all the time you have considered each to answer me.

Best Regards,

Pascal