Turbulent Flow Simulations with Elmer - A Request

Numerical methods and mathematical models of Elmer
kishpishar
Posts: 54
Joined: 17 Jun 2015, 10:04
Antispam: Yes

Turbulent Flow Simulations with Elmer - A Request

Post by kishpishar »

Hi Peter,

The difficulty of simulating turbulent flows with Elmer has been discussed in these forums many times in the past. I am also aware that most of Elmer development happens through funded projects and there has been little interest in this area. However I would still want to rake up this issue and make a passionate plea to the developers. The reasons:

I've been using OpenFOAM quite regularly for some industrial turbulent flows. However, in the recent past I've also tried my hands on with Elmer - using Navier-Stokes and k-epsilon together for real-life flows in complex geometries - with varying degrees of success. What makes Elmer attractive is the inherent multi-physics capability (Fluid flow + heat transfer + turbulence + electromagnetic effects or conjugate heat transfer with stress analysis on solid part) and the relative ease of case-set up for complex multi-physics, which is not the case with OpenFOAM.

In my view, there is no competitor to Elmer in multi physics when it comes to open source. This problem with turbulence is the only glaring gap in Elmer - a software with such vast scope, multitude of features and models and designed at the very outset for complex multi-physics applications.

If Elmer perfects just one turbulence model (k-eps is enough) good enough for high-reynolds industrial flows, I'm convinced that its use will see a quantum jump. At least we could do a lot more testing with these modifications:

(1) Activating stabilization for the k-epsilon equation (AFAIK, k-eps model has only bubble stabilization which in the current form cannot be used for many real-life meshes with prisms or pyramid elements - comes in the way of extensive testing)

(2) Implementing heat transfer in the k-epsilon wall laws and make the formulation more robust.

As Elmer is otherwise so attractive from a variety of standpoints, I think the issue of turbulence is long overdue and needs urgent attention. I'm willing to help in whatever way I can except coding - for e.g, take part in testing, benchmarking etc. I can also provide real-life industrial test cases and meshes etc. to Peter and the Elmer team.

It's my sincere wish that the Elmer team takes a closer look at this soon ..

-Kumar
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Turbulent Flow Simulations with Elmer - A Request

Post by raback »

Hi Kumar,

We tend to be opportunistic and guide the development to areas with most promise. Having people supporting the development is one important factor.

I looked a little to the k-e model. Now it is solved as a (2,2) fully implicit system. Going to two independent system where the production terms are explicit (on r.h.s.) could be beneficial. The also other stabilization methods would be easier to take into use. I had a look at some more recent FEM implementations and there could be some unused tricks that are fairly easy to implement.

We probably could use some 1-2 weeks on the topic next few months. Having some suitable test cases (with mesh and reference solution) would certainly be of great help. Maybe there would be additional people wanting to join?

-Peter
kishpishar
Posts: 54
Joined: 17 Jun 2015, 10:04
Antispam: Yes

Re: Turbulent Flow Simulations with Elmer - A Request

Post by kishpishar »

Hi Peter,

Many thanks for your kind reply. Fascinating and indeed heartening to know that you could allocate some time and resources to tackle this vexed issue.

Looks like the k-epsilon formulation in Elmer is quite old. Perhaps you have seen this paper already - here are the details of the numerical implementation of the k-epsilon model in the FEM code FEATFLOW. They have used a positivity preserving discretization of the convection terms by recasting the PDEs in a slightly different form along with algebraic flux correction. Perhaps something similar could be tried without much efforts?

http://num.math.uni-goettingen.de/bail/ ... mierka.pdf

I hope that more people would want to join us in this.

-Kumar
dmitry
Posts: 53
Joined: 09 Dec 2014, 16:52
Antispam: Yes
Location: Russia

Re: Turbulent Flow Simulations with Elmer - A Request

Post by dmitry »

Hi!
Kumar, thanks for invitation.
As programmer I have some basic skills but not enough to dive into Elmer FRORTRAN code. As engineer, I think, I am able to do something.
I think, first of all, we should prepare and supply solution algorithm, based on some resent works, preferably in pseudo code. It will be nice if Elmer team will point the certain paper that will be possible to implement that we should work on.

By the way, I found another paper on K-e implementation (see attachment).
After solution algorithm is ready, we will ask Elmer team to check it on errors and perform the source code development.
Also I can do some humble donation (I'm not a rich guy) for those who will do coding.
Also I can prepare, perform and supply results some of cases from my area of interests.
Finally I can write xml for ElmerGUI.
Attachments
Implementing a k-epsilon Turbulence Model in the FEniCS Finite Element.pdf
(374.99 KiB) Downloaded 356 times
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Turbulent Flow Simulations with Elmer - A Request

Post by raback »

Hi

Funnily I had found exactly the two papers you referred to and found useful stuff in both.

Elmer actually has a serial implementation of FCT as presented by some articles by Kuzmin et al. I actually quickly tried to apply them to k-e solver but the implementation assumes single component. Therefore splitting at least the inner logic to k and e equations would allow utilization of FCT. Parallelization of FCT is also planned as it is also needed elsewhere.

Both the formulations solve (k,e) in time. Even if we are looking for steady-state solution it may be advantageous for convergence. In fact, FCT is only applicable to transient cases.

It is great if we find an active group to push this forward! What would be ideal communication media for such collaborative effort?

-Peter
kishpishar
Posts: 54
Joined: 17 Jun 2015, 10:04
Antispam: Yes

Re: Turbulent Flow Simulations with Elmer - A Request

Post by kishpishar »

Hi Peter, Dmitry,

Thanks Peter for actively looking into this, really appreciate!

If Elmer already had an implementation of FCT, I think the task would be even easier. In that case, we could seperate k and epsilon parts into two solvers (as opposed to a single KE solver with 2 degrees of freedom) and use the solvers in the order - Navier-Stokes, Epsilon and K, respectively in the main time loop.

Absolutely agree with you about both the formulations (k,e) in time. Steady state can be anyway reached by marching in time and it will certainly aid convergence - that has been my experience in flow simulations with Elmer.

I am also hoping that a few more people come on board. I am open to any form of communication media, what would be the best for you?

-Kumar
kevinarden
Posts: 2237
Joined: 25 Jan 2019, 01:28
Antispam: Yes

Re: Turbulent Flow Simulations with Elmer - A Request

Post by kevinarden »

Hello,

I can certainly help with bench marking and testing. Improving the turbulence solving capability would be a significant improvement for Elmer.

Kevin
kishpishar
Posts: 54
Joined: 17 Jun 2015, 10:04
Antispam: Yes

Re: Turbulent Flow Simulations with Elmer - A Request

Post by kishpishar »

Hi,

I have found out that Comsol uses the same implementation and tricks as presented in this article by Kuzmin et al. - the article under the link:

https://www.researchgate.net/publicatio ... retisation

This is discussed in detail in Comsol CFD Module Users Guide pp. 178-179. Here is the link:

https://doc.comsol.com/5.4/doc/com.coms ... sGuide.pdf
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Turbulent Flow Simulations with Elmer - A Request

Post by raback »

Hi

How we use a GitHub repo for small turbulence cases. It could include reference, mesh, and sif files which would eventually be refined to solve the case more optimally, and results in compact form. People who have existing cases and meshes could start and others could refine this when the solvers are improved.

I think the GitHub limit is 500 Mb. Would that suffice to any real cases? We could use "Mesh Levels" to scale up the sizes.

We could perhaps target that reformulated k-e + parallelized FCT would be usable till end of June. Initially only very light tests would be needed (step+diffuser among elmer test could suffice) to ensure that no errors are introduced in the process. To limit the amount of work only k-e model would be dealt with.

-Peter
kishpishar
Posts: 54
Joined: 17 Jun 2015, 10:04
Antispam: Yes

Re: Turbulent Flow Simulations with Elmer - A Request

Post by kishpishar »

Hi Peter,

Great suggestions, perfectly agree with you on all points.

I think the GitHub limit of 500 Mb would be enough for initial testing and joint activities. Once we perfect the reformulated k-e + FCT, the users can test more complicated cases individually at their end and upload to your servers as well.

I can also provide OpenFOAM reference solutions for any given mesh and boundary conditions.

The time frames are entirely up to you - at the end of the day, we will have a longstanding issue resolved!

-Kumar
Post Reply