External Vehicle Aerodynamics
 

Hi All,

This is a relatively new field for me coming from internal rotating machinery.

To get myself started I did the Ahmed body and had no problem getting convergence and good results.

Then I got this geometry... http://imgur.com/XucSViN http://imgur.com/XucSViN

I've tried literally every form of mesh manipulation, time step, algorithm, solver choice, steady state, transient, domain extension, shrinking, free slip, turbulence modelling that I could think of (Several times over to be embarassingly honest). And yet no convergence. Every item from the convergence checklist posted here. Study of how this has been done before. And yet always there is a massive zone of high residual cells along side the car and trailing it, even with the most perfect 20 layer prism inflation 1.05 growth rate in the fluid zone, super refined wake region (up to 40 million cells!).

What additional considerations can I make because this is driving me insane and seems like it should be a relatively easy problem.

It is easy until you start looking at the details....

I can suggest two things:
1) The tyres can form nasty cusps with the ground and generate very poor mesh in this area. It is useful to fill in the worst of the cusp to avoid the worst elements. This is sort-of realistic anyway as the tyres sag a little under the vehicle weight anyway.
2) Bluff bodies like this are always going to shed large transient vortex structures. You are going to have a hard time getting any steady state run to converge. I would be looking at SAS or DES so you can resolve the larger turbulent structures in the wake.

As soon as you say SAS or DES then the simulation is no longer "easy"...

Thanks for the response. The tires are not truly tangential close up, but I may flatten them a bit more to help out a bit the RMS residuals of the system as a whole.

And I think you are right that I will have to go transient with more advanced turbulence models as it makes sense that it is transient, and also, the back end shape is not a 2D type surface like the ahmed body.

But a question if you happen to know about the applications in this field. For those who do blunt body/vehicle external dynamics consistently do they really have to run transient? It appears to be the issue in this case, but would be maybe worse in something like a truck because it is even bigger and blunter.

This just surprises me because some blunt body simple simulations I could get to run steady state, and if I were doing external blunt body flows regularly it seems awfully expensive to have to run transient special turbulence cases regularly like SAS modification or LES or DES. I assumed they ran steady state and I was just somehow, despite all the research and thought, missing some piece.

No, my understanding is that most serious bluff body simulations for vehicles are done transient with LES or LES-derivative simulations.

Is a sign of transience when a steady state solution has residual oscillation that is independent of time step size? I believe noticing that as a clue that I needed to switch. As opposed to time step making a marked difference in residual oscillation frequency/amplitude being a sign that mesh quality and or boundary conditions not being appropriate.

If so I definitely don't want to forget that and will be writing it in my personal best practices notes.

Yes, it is. And yes, it is a big warning sign that a transient simulation is required.

Are you monitoring the coefficient values you are interested in, as you are running the simulation? Are they changing that much, or are they steady?

You might try comparing an uncoverged SS solution coefficeints, with a converging transient solution to see if, for your case, do you really need to worry about the SS flucutions as much as you are.

I have been monitoring Cd, and you are right, despite the lack of convergence it is very precisely and consistently oscillating about 0.405+/-0.015. I don't have a converged transient analysis to compare to, but see what you are getting at. I think it's quite possible that this value I am finding (minus the oscillations) could be quite accurate, but I'm unwilling to say as such since the numerical side is still so iffy.