open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi all,

I am running a TELEMAC-2D simulation in both SAINT-VENANT FE and SAINT-VENANT FV (finite elements vs finite volumes), due to FE velocities being much lower than predicted. FV provides more realistic velocities and a more stable stolution (FE had temporal oscillations), at additional computational expense.

One of the necessary outputs for this study is shear stress, which involves the use of friction velocity u* (USTAR) as a proxy (in postprocessing, using USTAR^2 * 1000). When I run the simulation in FE, the friction velocities are reasonable, around 0.10-0.13 m/s, which gives shear stresse up to 50 N/m2 (high, but this is a high velocity channel). However, when I run the simulation in FV, the friction velocities (not the shear stresses!) in the same areas reach 1100 m/s. This is obviously unrealistic, but I have noticed that these values are 10^4 times greater than values from FE modelling. Dividing u* by 10000 before the operation thus yields realistic shear stresses.

It could be a coincidence, but I suspect that between FE and FV, there are different units being furnished to the calculation of u* that have this impact on the results. Unfortunately, I do not have the time to look into this right now, but I thought I would bring it to the attention of the community. Either something is wrong in the source code for SAINT-VENANT, or there is something else going on regarding the change of numerical scheme I'm simply not aware of, in which case I'm awfully curious to know what's going on.

Kind regards,
André Renault

Some more information... the spatial distribution of shear stress is also very different between FE and FV.

The attached figures are the 2D distribution in FE and FV respectively, as well as a graph showing the a cross-section of the lateral distribution. The channel is trapezoidal.

Finite element:


Finite volume:


Both, cross-section:


The exact values do not matter so much as the relative distribution, in this cas. I am not sure which to trust. SRH-2D gives a relative distribution similar to the FV (higher at the edges), which could be because SRH-2D is also a FV solver.

It should be noted that I didn't take much time to change the parameters for finite volume scheme. Nevertheless, the spatial distribution of bed shear stress should not be overly dependent on model parameters, in my opinion (which could well be wrong!).

Kind regards,
André Renault

Apologies for the triple post... I was unable to edit my previous post after a few minutes.

I should add that I am on v8p1r2. I am well aware that this version is no longer supported, but due to my company's IT policies (which I am currently pushing to change), I am not able to upgrade to v8p5. It may well be that the bug has been fixed since, but I have no real way of knowing. Since this is a model for a client, I am not able to share it, either.

Kind regards (again ),
André Renault

Hello André,

2 colleagues of mine did great job to update Finite Volumes schemes in releases 8.2 and 8.3. Hard to says if your issue was solved from 8.2 or not, but it will be impossible to assist you with such old release.

Cannot you install your own TELEMAC installation in a temporary folder?

Chi-Tuan

Hello Chi-Tuan,

This is good to know, thank you for the hat tip regarding updates since v8p1.

The local installation is an excellent idea, frankly I did not think of it. I will try this when I can. Thank you!

André

Hello,

Interesting topic. What's the last update? Curious to know about the differences between FE and FV here. How about the velocities?

When you compute USTAR^2*1000, you compute the total shear stress, which is highly dependent on the friction coefficient especially if you use a Strickler < 30. If you want to analyze the shear stress as part of an erosion risk assessment (for example), it might be more appropriate to look at the skin friction instead of the total shear. For normal bed material sizes (say sandy or finer), the skin friction would be significantly smaller than the total shear. More importantly, it should not depict such a drastic different distribution across the river (unless even the velocities have similar patterns, but this is not so likely I would say).

Kind regards
PL