[manar]

Hello everyone
I am modeling an air free burning dc arc , I got some results from simulations but I am not 100% sure that I can rely on since the continuity residual is relatively high (attached picture2), I tried to monitor the total temperature (volume weighted average ) which is stable after multi thousands of iterations (attached picture 1) .
my question : can I consider these results , Or I should insist to reduce the continuity residual (URf already reduced - mesh refined ) ? I am suffering because the arc modeling is very touchy since no inlet and outlets ?

[grapheeene]

N-S Residuals are very very high.

If you had already reduced URF for continuity I suggest to converge momentum (and, maybe, energy) only.. Maybe your viscous model is not the right one for your situation. And then add Species or other phenomena.

[manar]

Hi Grapheeene
I cant solve flow equation alone because originally a virtual temperature would ignite the arc which is due to a UDS applying electric current in on of the electrodes ,after the arc is ignited , an electric current would conduct causing some forces to the momentum equation , so all the equations are linked simultaneously . can you explain more about the species ?
or if there any way way to decrease the continuity residual? what about the integral volume , I see it is stable ? is it not enough ?

[LuckyTran]

I always ignore the continuity residual because it's rather meaningless. It's normalized by the worst value during the first 5 iterations. If your initial guess is perfect, continuity is stuck at 1. If your initial guess is terrible, continuity residual goes up even though the solution gets better. You can test the first case rather easily by simply clearing the stored residuals, if your simulation is already converged the continuity residual will be stuck at 1.

[manar]

Dear LuckyTran
that's what happening to me right now , the continuity residual is stuck at 7.6e-1 though other residuals are below 1e-3 , and volume average (temperature
) is stable .

[neha03sharma]

Currently I am doing a steady state simulation in ANSYS-CFX. The geometry is a human face with respirator at the face, so it's a little complicated geometry. I am having issues with convergence. Please see the attached image, can I consider it as a converged solution.

In the second image, I am tracking velocity at a random point at outlet.

PLEASE REPLY
RESIDUAL_CONVERGENCE.PNG

VELOCITY@outlet.PNG

[Aohnia]

Originally Posted by LuckyTran

I always ignore the continuity residual because it's rather meaningless. It's normalized by the worst value during the first 5 iterations. If your initial guess is perfect, continuity is stuck at 1. If your initial guess is terrible, continuity residual goes up even though the solution gets better. You can test the first case rather easily by simply clearing the stored residuals, if your simulation is already converged the continuity residual will be stuck at 1.

Hi LuckyTran,

Indeed, I don't know well why it is recommended to reach a value of (e-3) for normalized residuals to consider that our calculations are converged.

I'm simulating an open channel steady flow and I rarely obtain values lower than (e-3) for my calculations. However, residuals are constant (I do more than 100.000 iterations) and variables like mass-balance or static pressure at specific points have also constant values.... so I suppose my simulation is converged even if residuals are high. My reasoning is that maybe my initial conditions were far from the "correct" answer and, as residuals are normalized by the first iteration values, that's the reason residuals are so high.

But maybe I'm wrong? What do you think? Should I pay attention to residual values or just to their "forme" (constant)?

Here you have an example of my residuals and other variables plots...

[LuckyTran]

Yes I would say the results are clearly converged after 60 000 iterations (or 80 000) if you want to be safer. Notice that I say results are converged (and not residuals), results being these monitors you have given here.

Okay and now you can go back and observe the residuals and see what happened there. Once you build up enough experience (if ever) then you can start to make some judgements on what the residuals normally look like for your problem.