CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   Residual convergence... (http://www.cfd-online.com/Forums/fluent/41228-residual-convergence.html)

Cyril June 2, 2006 05:49

Residual convergence...
 
Hello !

I'm simulating a supersonic 30mm shell (Mach2) in the air. The residals are converging, but no more than 5e-3...

Is it important ? can I trust this result ?

And what are exactly the residuals ? (I understand them as the difference beetween 2 cells... Am I wrong ??)

Thank you very much !!

Cyril

Emanuele June 2, 2006 06:17

Re: Residual convergence...
 
Residuals are the differences in the value of a quantity (for example x-velocity or y velocity or k,ecc.) between two iterations. This means that the more the residual is low the less the results will change if you will continue to iterate. Residuals depend from the models that you use (for example with natural convention normally they don't decrease a lot because there isn't a strong fluid dynamics component) and from the initialization because if you use an initialization that is very close to the final solution residuals will not decrease a lot as you use an initailisation that is far from the final solution. Often you can find a period during which residuals seem not to decrease but if you continue to iterate they restart to decrease. Furthermore residuals means mathematical convergence; you have to monitorate a quantity that is fundamental for your case (for example the lift or the drag, ecc.) and you have to decide if your case is converged to look both (often to the residuals and to the fundamental quantity (if your fundamental quantity doesn't change even if you continue to iterate your solution will not change more). I hope this helps Regards Emanuele

Fabrice June 8, 2006 01:08

Re: Residual convergence...
 
Hi,

I agree with Emanuele's answer, what you can do to decrease the residuals is to do the calculation in double precision (2ddp for 2D case and 3ddp for 3D case).

Cyril June 8, 2006 04:33

Re: Residual convergence...
 
The problem, with double precision, will be the time. I'm working with some hudge meshes (about 3-4 1.e6 cells !)

Here is what I've got : http://img105.imageshack.us/img105/8...nstitre2xk.png.

My geometry fits the shockwave. Here is what I study... http://img149.imageshack.us/img149/6...nstitre2xv.png

What do you think...?

Fabrice June 8, 2006 06:30

Re: Residual convergence...
 
Hi,

First of all 3-4 e6 cells, it's really really heavy!! Can't you input a mesh which corsens when you move out from your object of study (use size function for this). You will decrease the number of node elements and therefore earn a lot of time for the calculation.

Furthermore, for shockwave study, you need at least a second order discretization of the pressure and velocity, the first order will give you wrong results.

Check in the Fluent manual if you can't increase the convergence speed with another coupling between pressure and velocity (SIMPLE, SIMPLEC, PISO).

I hope it will help you.


All times are GMT -4. The time now is 21:42.