open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello Everyone,

This is my first post, so please forgive if I make some beginner mistakes.

I have been trying to simulate flow on a artificial (concrete) floodplain and channel setup(Images are attached below). It is approx 20mX20m in size and water depths range at around 20cm, just to give an idea of the scale of the setup.
I have tried all viable numerical schemes to simulate the setup. Wave+Characteristics overprdicts the water depths by almost 10%. However only the scheme 14 for advection of velocities gives the best results.

However what I have observed when using the scheme 14, water seems to jump onto the floodplain area at depths much lower than the bank elevation. Also the velocity vectors at the channel walls are very random in direction and magnitude with respect to middle of the channel. I searched through the forums and the closest I found was the issue regarding number of elements used to define steep levees. the problem is that the channels in my setup are essentially rectangular and the mesh refinement required at that point will be extreme when compared to the rest of the setup. So my question is: Are the vertical walls of the channel really the culprit here, and if so is there a way to deal with them effectively without increasing the computation time to an insane amount?

Hello

vertical walls could be the culprit in a direct and in secondary way.
When the slop is very steep, the wetting and rying algorithm of finite element kernel could fail especially for small water depths. It can then give unphysical results.
To overcome such issue, you need to refine the mesh locally. This will increase the model size but if you optimize the refinement locally it will not be prohibitive.
You can also try to use finite volumes which behave much better thant finite elements in such kind of applications.

I hope this helps

with my best regards

Riadh ATA

Thank you for the suggestion.

What I did was to resample the outline files defining the channel separately in bluekenue to Equal Distance of 4 mm so that the mesh is forced to have elements of size 2 mm at the outlines (which correspond to the channel banks).
The general element size on the channel is 3 cm, but to capture the steep slope of the channel walls, a element size of even 4 mm is not sufficient. Hence you can see the scale to which the size of elements differ (and how much it will reduce the time step). Computation cost increases by more than 8 times when using a mesh refined in this way.
As per your second suggestion, I will try Finite volume scheme (HLLC 1st order) now. I will report back if there is any success or failure in the process.

I tried using HLLC scheme to simulate the flow. But what I observe is that bulk of the water stagnates near the inflow and water depth raises to more than 0.8 m, instead of moving downstream towards the floodplains. Water depth in the channel downstream is less than 10 cm.
Can you suggest me the error that might be causing this? Is there a recommended FV scheme?

Hello aditlk,

additionally to what Riadh suggested, you can try to use THRESHOLD DEPTH FOR RECEDING PROCEDURE. You can find all the information you need in the user manual.

Cheers,
Sebastian

Thank You everyone for the suggestions.

I tried to use FVM but could not solve the problem I mentioned before. I have also tried the "Threshold depth for reverse procedure", and it actually prevents those random floodings to appear (especially at the beginning). However the velocity vectors at the boundaries of the channel (at the vertical walls) is still erratic in direction and magnitude.

So I have taken the first suggestion by Riadh and have now refined the mesh at the channel banks. Now lets see what happens with the simulation. I will report here if something new happens.