Floating Point Exception error in car cfd


#1

Hi
I am trying to make a cfd simulation of car in a wind tunnel. I have problems with setting rotation to wheels, and I think it might be the reason of “Floating Point Exception” error I am getting. I would appreciate help in setting rotation of wheels and solving problem with simulation run, as in future I would like to investigate more complex model of car and I wouldn’t like to move on if I can’t get the simple model first.

(I have tried different initial conditions for turbulent kinetic energy and omega but it did not help)
(I know there is Student Formula Workshop. I have been trying to follow their step by step guide, but still failed)

Here is the link:
https://www.simscale.com/workbench/?pid=8115297672374577766#tab_0-0


#2

Hi @pkciuk!

This problem has already been addressed by my colleague @jprobst in the past: Most Popular Errors of Simulation. Tell me if that helped you - if not either one of the PowerUsers or me will jump in to help you out!

All the best!

Jousef


#3

Hi @pkciuk,

Looking at your geometry, there are some complex features that you may be able to remove in order to reduce your problems in meshing that is likely the primary cause of your simulation error. The inner area of the wheel containing the star shaped wheel rim can be removed for example. Simplifying any other such features will certainly help in meshing.

Your mesh on the other hand has significant problems as the no of illegal cells is above 500 for Mesh 1 and above 2000 for Mesh 2. This should be the primary reason for the “floating point errors”. As mentioned, fixing of the cells lie primarily in further ensuring your geometry is as simple and defect free as possible.

Hope this helps!

Cheers.

Regards,
Barry


#4

Thanks @Get_Barried and @jousefm I really appreciate your tips. I will work on them now and see if that helps. Btw, did you look at the wheel rotation setup as well @Get_Barried?


#5

Hi @pkciuk,

Are you referring to the boundary condition setup? If so what about it?

Cheers.

Regards,
Barry


#6

Hi @pkciuk!

For the rotation setup just make sure that the rotation points are exactly in the middle of your wheel. To check that you can create a geometry primitive Point and see if your coordinates are correct. If this is not the case this might indeed cause your simulation to crash at some point - some users had this problem several times.

Cheers,

Jousef


#7

Thanks @Get_Barried and @jousefm for help. Currently I am working on reducing illegal cells (preferably to 0), cause it might cause the biggest problems indeed. Also, do creation of MRF cell zones significantly improve accuracy of solution? Cause I would like to try making rotation of wheels without that mesh setup. @Get_Barried what I meant was whether I have properly created a dummy for MRF cell zones.


#8

Hi @pkciuk!

A post that might help you understanding the different approaches: Viscous/Inviscid CFD Flow. It depends on what configuration you are investigating. The setup for the MRF zones looks good I would say and you may continue with your simulation.

Cheers,

Jousef


#9

Hi,
You have both helped me much so far, @jousefm and @Get_Barried and I got 1 more question. If you could see the latest simulation of that particular project (“Incompressible 5”) and tell me, is there any way to achieve better convergence on Fz force? Is it because of p residual?


#10

Hi @pkciuk,

By convergence do you mean closer results? The simulation is already converged (albeit an oscillatory convergence) and your residuals are not ideal but are converged.

I’m assuming you are comparing the force on the z direction to some experimental data. How far off is the results?

Cheers.

Regards,
Barry


#11

I meant just the residuals, whether there is a way to set numerics so they are lower values.


#12

Hi @pkciuk,

You can try to set the relaxation factor for U , K and omega to 0.5 to see if converges better. You may need to increase simulation time however. Additionally you can also further lower the relaxation factors for P, U, K and omega even further provided the first lowering of values does help with the residuals. Do note again that every time you lower the relaxation factor you may need a longer simulation run time so keep that in mind check for steady-state convergence where the residuals do not fluctuate before deciding to increase the end time.

For example if the residuals have not reached a steady-state and are still converging and the simulation ends, then clearly you need a longer end time.

If that does not work, you may need to use higher order schemes or a different gradient scheme. Do try the relaxation factors first to see if they help.

Cheers.

Regards,
Barry