open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi everyone,

as part of my master thesis I am trying to reproduce bedforms (mainly dunes) within a 0.4m wide rectangular channel. I am using Sisyphe together with Telemac3D (version v8) in the non-hydrostatic version in order to have a better representation of the velocities and stresses near the moving bed.

In some test runs I found out that using SUPG for the advection of velocities in combination with the MURD N scheme for the advection of k-epsilon gets me the most promising results in terms of dune structures. However, SUPG seems to be prone to stability issues (abrupt model breakdown with NaN values) especially when I apply higher discharges to the model. At the moment my average velocities are around 0.7m/s (max. up to 0.85 m/s) while my smallest mesh element size is around 0.03m (with a time-step of 0.05s this means the CFL number could go up to around 1.4). As my computational power is limited, unfortunately I cannot go much further down than 0.05s. Further, having decreased the time-step by half still caused the model to collapse later on. Only with 0.01s I managed to run the model through, however with this time step I got some instabilities at the bottom near the outflow boundary.

My question is now if there is any possibility to make the SUPG scheme more stable, e.g. by using different solvers or preconditionings for certain parameters later on? Or if finding the optimal time step for each discharge scenario is the only thing I can do?

Attached you find a picture showing the listing printout for the described NaN problem as well as both steering files for Telemac3D and Sisyphe.

Looking forward to your help!
Cheers,

Samuel

Hi Samuel,

at BAW we had a PhD about dune modelling and telemac3d/sispyhe (Annalena Goll) some years ago. Do you know the PhD? In the shear stress she took additional the turbulence into account. I continue her work.
Contrary to her simulations I use the MURD scheme (4/14) also for the velocities. I think it is more safe to use the same advection scheme for velocities and k,epsilon...
Furthermore I found out that the k-value becomes incredibly high if the water depths decrease to zero. Probably it is not the case in your flume, but nevertheless it is good to check the values.

The vertical distribution of the planes is also very important and where to take the shear stress (first or second layer or something in between).

Annalena used the slope effect of Apsley and Stansby (but probably not helpful to stabilise the simulation).

MASS-LUMPING = 1 and IMPLICITATION = 1.0 will increase your stability.
It would be helpful to investigate in which step the NAN occurs first. If you do a restart just before getting NANs and then with a listing period of 1 should show at which step.


Best regards,
Rebekka

ps: your name sounds German... if yes we can have a talk by phone, just write me an email: This email address is being protected from spambots. You need JavaScript enabled to view it.

Hi Rebekka,

thank you very much for your reply. Yes, I am aware of the work from Annalena. Many of her findings I already take into account for my simulations (like the vertical distribution of the planes and where to extract the shear stress etc.). With respect to using Mass-Lumping and a full Implicitation this unfortunately does not help in the case of the SUPG scheme. I switched to the MURD scheme (4) for now, however my results are just not as good as before.

Like you assumed yes I am German. Your offer of having a talk by phone sounds great. I will get back to you by Email tomorrow.

Best regards and thank you in advance,
Samuel

Hello Samuel,

Currenty, the recommended advection scheme is LIPS, i.e.
SCHEME FOR ADVECTION OF... = 5
together with
SCHEME OPTION FOR ADVECTION OF... = 4

as reported in the appendix E of recommendations in the TELEMAC-3D user manual only available since release 8.1,
see e.g.: wiki.opentelemac.org/doku.php?id=documentation_v8p1r1

Hope this helps,

Chi-Tuan

PS : sorry, I cannot speak German, ich habe fast alles vergessen. :)

Hello Chi-Tuan,

thank you. I was unaware of the new recommendations. I will try them instantly and tell you how my model performs.

Cheers,
Samuel

PS: keine Sorge, mein Englisch ist ja gut genug