open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello everyone,
here is my problem :
We first did a modelisation of the fishway (vertical slot for those who know this type of fishway) with TELEMAC 2D and we use the K-Epsilon turbulence model. It fit right with the measurement we took on the scale model.

The geometry of the probem is composed by two areas. Near the slots, the flow is more disturbing and vertical velocity component seems to appear. thus we need to modelise the flow with TELEMAC3D. There are also areas with horizontal re-circulation and slow velocity.

We succeed to modelise the problem whit TELEMAC3D but the problem is that the flow does not influence enough the all part of the problem : there are some large areas where the velocity component are close to zero. Thus the results does not correspond with the 2D results and with the measurement. The vertical velocity profile in the agitated areas does not correspond ether, but they have a mean close to the mesurement.

We think that the turbulence model is not enough disperive and that it soulhd more influence the problem. We try with K-Epsilon model first, then with Smagorinski (with the coefficient put up to 0.2). The results with Smagorinski seems beter, but not good enough.

Can someone guide us in the choice of the turbulence model ?

(sorry for my English) and thank you for riding me.

have a nice day

Hello,

The k-epsilon model should be the best but it is difficult to use. Depending on your geometry the turbulence that you get in your model could be only what you give at the entrance, if the turbulence has no time to develop between the entrance and where you have measurements. So upstream boundary conditions are very important (see subroutine kepcl3.f, very small values are given at entrances but other tentative formulas are proposed but commented out). You have also in subroutine cstkep.f a number of hardcoded options, like deciding whether we want Dirichlet or Neumann conditions for k and epsilon at bottom and free surface. For example we put Neumann for k at the free surface because it must work with wind, but in your case Dirichlet on k would be better, as k should be 0 at the free surface.

With best regards,

Jean-Michel Hervouet

Good evening,

Thank you for your quick answer.

The geometry of the entire problem is a succession of pools that have the same measurement and communicate with each other. Because we don't want to have to much compute time, we decided to modelize only one pool with the 3D model. Do you think that if we modelize a channel with small value of velocity before, the turbulence will have the space needed to develop ?
I will seek to change the boundary condition of the k-e model as well.

Thank you for considering my problem.

Hello,

The results seems better with the modeling of the entire problem. I think it's due to the more rough/bigger mesh because I don't change much in the turbulence model.
Do you think it will improve the results to use the turbulence solver ? It could be a lead ... I try to use the simplest way to achieve this before entering in the details of the code and the boundary condition.

Have a nice day

Hello,

As the turbulence develops mainly from the friction on the bottom, if you put no turbulence at the entrance, from there the k-epsilon model will start again developping turbulence from the bottom, but it may take some distance to do so, so in your model you may see only a partly developed turbulence. Other models do not have this drawback, as they depend only on the velocity gradients.

Regards,

JMH

Hello,

Thank you for your response, it's very helpfull.