My solution ratio doesn't converge, some advices to improve it?

Marco Alfonso Diaz Neciosup

Hi, I'm doing my degree final project about car aerodynamics and I'm using Hypermesh CFD, I'm trying to simulate a Tesla model S, at the moment, I'm doing simulations to see a first impression of the results, if they're coherent. Here comes the issue, when I see the plot manager, my solution ratio doesnt converge, specially the "eddy_frequency", also the other equations doesn't converge at all. In the other side, the residual ratio it is said to reduce 10^-5 units, in my case, it's around 10^-2. Here I leave an image:

I would like you to give me some advices on how to improve these results or where should I look to improve. Also I don't have a lot of power computing so I'm limited in terms of meshing. Thanks everyone who's willing to help me.

acupro

Here's an interesting item on convergence.

https://community.altair.com/community/en/how-does-one-judge-the-convergence-of-an-acusolve-cfd-solution?id=kb_article_view&sysparm_article=KB0122435

Based on your plots showing decent residuals, but pretty high solution ratios, I would guess this is actually a transient flow - as is the case with most automotive external aero cases. In other words, you won't get a nicely converged steady state solution, because the flow is actually transient.

As discussed in that article, you may want to review your solutions of interest, and see if you're happy with some sort of averaging of the last several time steps of the steady approach.  Sometimes the transient results - time averaged - may be very far from what you see in the steady approach, so you may want to run a transient, too.  Maybe try a simulation with a time increment like 0.001 sec, and do enough time steps to get at least 5 or 6 flow passes of the car length, to see if a repeating pattern develops.

Marco Alfonso Diaz Neciosup

acupro_21778 said:

Here's an interesting item on convergence.

https://community.altair.com/community/en/how-does-one-judge-the-convergence-of-an-acusolve-cfd-solution?id=kb_article_view&sysparm_article=KB0122435

Based on your plots showing decent residuals, but pretty high solution ratios, I would guess this is actually a transient flow - as is the case with most automotive external aero cases. In other words, you won't get a nicely converged steady state solution, because the flow is actually transient.

As discussed in that article, you may want to review your solutions of interest, and see if you're happy with some sort of averaging of the last several time steps of the steady approach.  Sometimes the transient results - time averaged - may be very far from what you see in the steady approach, so you may want to run a transient, too.  Maybe try a simulation with a time increment like 0.001 sec, and do enough time steps to get at least 5 or 6 flow passes of the car length, to see if a repeating pattern develops.

First of all, thanks for the post given, it has some interesting information. The simulation is steady flow and incompressible, also I used SST model turbulence, maybe I should try K-epsilon which is also used in this type of simulations. 

After reading the post, the residual ratio seems to be pretty decent, but the kinetic_energy, pressure and velocity are very high. The results obtained are decent so actually I'm happy with that, but also I have to justify that the Residual and Solution Ratio are coherent and converge. Anyways, thanks for the reply!

acupro

Marco Alfonso Diaz Neciosup said:

First of all, thanks for the post given, it has some interesting information. The simulation is steady flow and incompressible, also I used SST model turbulence, maybe I should try K-epsilon which is also used in this type of simulations. 

After reading the post, the residual ratio seems to be pretty decent, but the kinetic_energy, pressure and velocity are very high. The results obtained are decent so actually I'm happy with that, but also I have to justify that the Residual and Solution Ratio are coherent and converge. Anyways, thanks for the reply!

Selecting a different turbulence model will not change much - as the flow behavior is actually transient, not steady-state.  In other words, at different points in time (different instantaneous snapshots) the actual flow will show different results - mostly coming from vortex shedding but there can be other instabilities also.  This is shown by the good residual ratios - but the too-high solution ratios.