[Tapeworm]

Hi,

New on the forum and the first thread, so give me a heads up if I missed something important.

Have been looking around on the forum for a while and have gotten some nice, useful information. But some questions remains...

To the case,
Me and a colleague are now doing our master thesis project where the objective is to create and validate a numerical model of an experimental setup of a tube-in-tank heat exchanger with a phase change material (PCM). We are working with ANSYS Fluent 17.0.
I don't really have a straight forward question, I am more looking for general tips/hints and inputs. We can't really get the simulation stable and are often getting crashes/errors after a couple of time steps (often different errors of the "Divergence detected in AMG solver"- kind, most often due to the temperature).

Here is the geometry of the model:
it is approximately 1m long, 30cm in diameter and it has 104 tubes. The inlet is on the bottom to the left and the outlet is on the top to the left. The heat transfer fluid (HTF) is coming in from the inlet at 0.15m/s and are spread out in 52 tubes. The HTF is then mixed on the right hand side and are further going in 52 tubes on the top towards the outlet. The HTF is one domain and the other is the PCM.
The temperature BC are ~120 C for the PCM and HTF inlet temperature is 35 C. (The whole domain is patched to 120 C).

We have tried different kinds of meshes,
this is one of them. We have also tried to fine tune the mesh with an increased number of cells and different types (tetrahedral or hexagon)
.

We have tried mesh sizes around 1 million cells up to 4 million cells. To much? To few?

Basic information about the model: 3d, pressure-based, RNG k-epsilon, transient case read into 3d, double precision, pressure-based, RNG k-epsilon, transient solver.

Some questions:
- How fine do you think the mesh needs to be? And what type of mesh would be most appropriate?
- What time step size should be sufficient? Could we do a fixed or is the adaptive a better choice?
- Any hints in general that should be good to consider?

I bet I missed a lot of information needed to give advice, so let me know and I will fill in all that's needed.

Would really appreciate some input!

Thanks in advance!

[dmirel]

Hey,

Regarding the mesh, I would do a tethraedral mesh, second order,for accurate results, a minimum of 0.2mm and a maximum of 60mm. For example, on the pipes I would mesh them with a 5-10mm target length, minimun 0.2mm and for the outer shell with 50mm target length maybe because you don't need small mesh on big parts that you don't study so you could use the power for the areas of interest. I'm a proeficient user with ANSA, wich is excellent for pre-processing so I don't know if this could be done in ANSYS Meshing, it was just a hint. I'm using ANSYS only as a solver.


Hope this helped even a little bit

[Tapeworm]

Originally Posted by dmirel

Hey,

Regarding the mesh, I would do a tethraedral mesh, second order,for accurate results, a minimum of 0.2mm and a maximum of 60mm. For example, on the pipes I would mesh them with a 5-10mm target length, minimun 0.2mm and for the outer shell with 50mm target length maybe because you don't need small mesh on big parts that you don't study so you could use the power for the areas of interest. I'm a proeficient user with ANSA, wich is excellent for pre-processing so I don't know if this could be done in ANSYS Meshing, it was just a hint. I'm using ANSYS only as a solver. Could you have a guess on approximately how many cells we need?


Hope this helped even a little bit

Thank you for the input!
Yeah, we are looking at the tetrahedral mesh as well, think that would be more accurate for our case. We have recently also tried out some "contact sizing"-mesh options. Haven't been able to run a simulation using the contact sizing yet though.

We are both beginners and haven't really worked with ANSYS or larger CFD simulations before. How much time (approximately) is usually needed in order to solve a problem like this you think? Of course it depends on the PC performance, but do you have a guess? Are we talking hours, days or weeks? We do have access to "super computers" at our School/department, which have massive PC performance (we are connected to 1 node which has 24 cores and 512GB of RAM with the CPU Intel Xeon Processor E5-2690 v3). Though we need to run some simulation at our private PC which are avarage to high performance personal use computers.

[Tapeworm]

Couldn't edit my last post, sorry for double post.
Approximately how many cells do you think is required for a setup like this?

[oozcan]

Hi,

You have asked some questions successively maybe we can proceed step by step.

What boundary conditions do you have in order to validate to an experiment you are doing? in other words,you need to focus on what exactly you are working on. Exact Objective?

So,if time-dependent melting process is one step ahead of heat transfer then you could give priority to PCM or you need to pay attention mesh you will do.(like mesh depending on turbulent flow regime,so, I do NOT know what regimes have fluid phase(s) you are working on, that is why you should need to calculate some boundary conditions first.)

[Tapeworm]

Originally Posted by oozcan

Hi,

You have asked some questions successively maybe we can proceed step by step.

What boundary conditions do you have in order to validate to an experiment you are doing? in other words,you need to focus on what exactly you are working on. Exact Objective?

So,if time-dependent melting process is one step ahead of heat transfer then you could give priority to PCM or you need to pay attention mesh you will do.(like mesh depending on turbulent flow regime,so, I do NOT know what regimes have fluid phase(s) you are working on, that is why you should need to calculate some boundary conditions first.)

Hi,

The objective is to both simulate the melting process and the solidification process. Both the HTF and the PCM material are user specified with mostly piecweis-linnear relation for the temperatur range of interest. More or less the only difference between the two cases are the initial temperature of the PCM and HTF.
We do feel quite confident with the boundary conditions.
Now in the beginning we are assuming adiabatic conditions around the model and do only have heat transfer (coupled wall) between the contact regions.

We keep getting crashes due to temperature limitation in XX number of cells in zone Y. The really strange thing with this is that when we refine the mesh (increase number of cells) the crashes comes earlier compared to if we have a slightly more simple mesh (decreased number of cells).

[oozcan]

Originally Posted by Tapeworm

Hi,

The objective is to both simulate the melting process and the solidification process. Both the HTF and the PCM material are user specified with mostly piecweis-linnear relation for the temperatur range of interest. More or less the only difference between the two cases are the initial temperature of the PCM and HTF.
We do feel quite confident with the boundary conditions.
Now in the beginning we are assuming adiabatic conditions around the model and do only have heat transfer (coupled wall) between the contact regions.

Well done!

We keep getting crashes due to temperature limitation in XX number of cells in zone Y. The really strange thing with this is that when we refine the mesh (increase number of cells) the crashes comes earlier compared to if we have a slightly more simple mesh (decreased number of cells).

Could you give me more detail of mesh you do? it seems like answer could be based on mesh.

[Tapeworm]

Originally Posted by oozcan

Could you give me more detail of mesh you do? it seems like answer could be based on mesh.

Thank you for your time and effort! Really appreciate it!

Here are som screen shots from the mesh, including also information about the mesh quality. I also included a log file from the last simulation we made.

One thing we are not sure about regarding the BC are the turbulence-settings under the Velocity inlet. Which model should be more suitable and what setings? As for now we are using the standard "Intensity and Viscosity Ratio" With Turbulence intensity of 5% and Turbulent viscosity ratio of 10.

Thanks for the help!

Edit: For the log file i just included the beginning and ending of the simulation due to large file size.

[oozcan]

it should be compliance between minimum size and maximum tetra size in ANSYS-Meshing e.g. maximum tetra size should be two-times more than min. (up to me) to provide for better transition of computational nodes and I could not see what values skewness and max. aspect ratio are.

So cylindrical solid bodies mesh generally are recommended as sweepable but not tetra as it decreases mesh cell and improving quality for CFD.

As for turbulence, could you post secreenshot then we shall check it out

[Tapeworm]

Originally Posted by oozcan

it should be compliance between minimum size and maximum tetra size in ANSYS-Meshing e.g. maximum tetra size should be two-times more than min. (up to me) to provide for better transition of computational nodes and I could not see what values skewness and max. aspect ratio are.

So cylindrical solid bodies mesh generally are recommended as sweepable but not tetra as it decreases mesh cell and improving quality for CFD.

As for turbulence, could you post secreenshot then we shall check it out

Thank you!

Here are the values for Skewness and Asp. ratio:
Skewness:
min 1,23e-003
max 0,867
Av 0,328
Std Dev 0,114

Aspect Ratio
min 1,1078
max 10,335
Av 3,4944
Std Dev 1,7858

For the mesh with the cylindrical body, do you refer to the whole model or only the PCM or HTF tubes?


Absolutely! I was talking about theese values (see screen shots).
Do you think we even need a turbulent flow or could we assume it as laminar? The flow in the small tubes that passes through the PCM is for sure laminar, but the question is how the fluid behaves at the inlet and the "mixing" side on the opposite side of the inlet.

[oozcan]

fluid flow behaves as laminar if Re<2300 in closed system or turbulent flow regime will be occurred technically or in theory.But some forget turbulent flow model in FLUENT accept fully-turbulent flow,that is, if model you have designed has not wider or more length (dimensionally) to achieve to be ''FULLY TURBULENT FLOW'', you can select ''laminar'' for sure.

if all hypothesis still show that fluid flow regime is accepted as turbulent, you need to calculate ''turbulent intensity and viscosity ratio numerically.you can easily find basic equations of two and write them down regarding boxes in FLUENT. I think you should get lower values of two if you calculates them because as far as I see, boundary conditions are lower.

[kaazto28]

Heat exchanger with PCM.

[kaazto28]

I think, Time step must be adaptive. Because I also try to solve the PCM, but İt is not solve. I do not know to use adaptive time-step.

[Tapeworm]

Originally Posted by kaazto28

Heat exchanger with PCM.

Thank you! We will keep you informed if we manage do make some progress, please do the same!

Originally Posted by kaazto28

I think, Time step must be adaptive. Because I also try to solve the PCM, but İt is not solve. I do not know to use adaptive time-step.

Yes, we also think you need to use the adaptive time step. Those settings are pretty straight forwards I think.

We did some calculations on the flow and we will assume laminar flow in the whole domain as well.

[kaazto28]

http://imageresim.com/images/41627614081587625706.png last 3695 iteration

And then, http://imageresim.com/images/28439154858810415250.png

I suppose I will give up this job.

[Tapeworm]

Would really appreciate some input in what might be happening here:

I was running the simulation and it seemed stable. The process is expected to take approximately 7200 seconds to finish. The melting process starts after ~200-300 seconds. The pictures included is from the FLUENT log at the flow time of 1363s - 1365s. During these few seconds, the simulation totally diverges and it can't seem to be recovered. Any suggestions on what can cause this or how to avoid this situation?

I mean, there is no new critical process happening in the model. I would understand if it was during these few seconds that the PCM starts to melt or something, then the calculations may be more complex and the simulation diverges. But looking at the results files during the seconds before this divergence, it behaves just the same as previously (from the start).

Any hints?

Thanks!

[Tapeworm]

Edit: Forgot to include the pictures of the log-file. Here they are.

[Tapeworm]

What I did manage to do was to decrease the number of iterations from 30 to 10, but still keeping all other settings as before. This resulted in the simulation going on until the following state (see screen shot of log-file).

Here is also a picture of the temperature and the mass fraction of the model. For the mass fraction, the blue represents solid state and red is liquid.

[kaazto28]

""During these few seconds, the simulation totally diverges and it can't seem to be recovered. Any suggestions on what can cause this or how to avoid this situation?""

I think, ANSYS notes that "time step" should be reduced in training notes. Iteration is not to be increased.

"The whole domain is patched to 120 C."
Why? Did not it have to be a lower temperature? And did not HTF have to be a higher temperature? for example htf inlet 120 C, pcm domain 35 C. I do not understand.

[Tapeworm]

Originally Posted by kaazto28

""During these few seconds, the simulation totally diverges and it can't seem to be recovered. Any suggestions on what can cause this or how to avoid this situation?""

I think, ANSYS notes that "time step" should be reduced in training notes. Iteration is not to be increased.

I use the adaptive time step, so it can decrease it to 0.0001s (if I remember correctly). Do not realy understand what you are trying to say here...

"The whole domain is patched to 120 C."
Why? Did not it have to be a lower temperature? And did not HTF have to be a higher temperature? for example htf inlet 120 C, pcm domain 35 C. I do not understand.

Cause I am simulating the discharge process. So the PCM is "fully loaded" and has stored all the energy possible. At this state the PCM and HTF should be assumed at a uniform temperature. The liquid/solid temperature is in a range between 105-110 C.

So for this process I want to extract the energy in the PCM. So the inlet temperature of the HTF is at 45 C and the outlet temperature will be higher.

The situation you are mentioning is for the charge process, when we want to store the energy in the PCM.