[Getdp] Electrokinetic - GetDP vs. Analytical Solution

michael.asam at infineon.com michael.asam at infineon.com
Tue Aug 21 10:38:27 CEST 2012


Hi Jochen,

you could try to use higher order basis functions. Preferably you should use
a P adaption instead of using globally the same order.
Have a look in the Wiki for an example:
https://geuz.org/trac/getdp/wiki/DualFormulations
The best is to combine H and P adaption.

One additional note:
Of course there is always an error by the discretisation of the round shape by plane
face elements. Unfortunately GetDP is not able to handle geometrical higher order
elements...

Have a nice day!
Michael


From: mail at jchn.de [mailto:mail at jchn.de]
Sent: Thursday, August 16, 2012 11:15 AM
To: Asam Michael (IFAG ATV BP PD 1 M1); getdp at geuz.org
Subject: Re: Electrokinetic - GetDP vs. Analytical Solution


michael.asam at infineon.com<mailto:michael.asam at infineon.com> hat am 6. August 2012 um 11:55 geschrieben:

> Hi Jochen,
>
> sorry for the late response.
> The deviation is due to an inadequate mesh-size because the characteristic
> length varies linearly between the two electrodes, whereas the E field is
> proportional to r^-2.
> So the best is to use a field in Gmsh to specify the mesh size by
> an equation, or in case of more complex geometries do adaptive meshing.
>
> Please find attached a modification of your example for adaptive meshing.
> As your geometry is axisymmetrical I've reduced it to 2D.
> Just generate a mesh with Gmsh as usual and run GetDP. Do the post-
> processing "GenerateAdaptionFile" and plot it in Gmsh.
> There do a right click in the main window on the according view number
> and choose "Apply As Background Mesh" and then mesh the geometry
> once again.
> With the adapted mesh size the result should be better.
> For 3D the procedure should be similar.
>
> I hope this is of some help.
>
> Kind regards,
> Michael
>



Dear Michael,



thanks for your help!

Using your adaptive meshing example (2D), deviation from the analytical result can be decreased from 0,05% with linear variation of mesh size to 0,006% with applied adaptive meshing.

But when I transferred this procedure to my 3D problem, the deviation stayed at approx. 5%. What I had to do, was decreasing the outer radius from 200 m to 50 m and modeling just one quarter of the hemisphere to get more elements per volume (see attached files). with approx. 600000 elements, I get a deviation of 1,03%, which can be decreased by adaptive meshing to 0,94%.

Now my RAM is on its limit (32 Bit Win7). Are there any other ways to increase the accuracy without increasing the element size?



Greetings,



Jochen
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.geuz.org/pipermail/getdp/attachments/20120821/987ed640/attachment.html>