Simulation and Meshing takes too long (6days).

[Anonymous]

Hi.

I have a 3D model need to be simulated and details on the heat sink is important to be studied. I run this simulation at first using this autosize mesh setting and it takes only 2hours for volume meshing and another 13hours for iteration run.

The customer requires us to run the simulation with fluid gap elements of 3 because they are comparing this product with Solidwork Flow Simulation. So, they asked us to use a computer with our specs instead of using the cloud to run this simulation. Using automatic mesh sizing and refine fluid gap element to 3, when I run it after 2days the simulation was still at the process of volume meshing. Then, I change the mesh settings as follows:

Automatic Meshing Settings

Surface refinement	1
Gap refinement	1
Resolution factor	3.0
Edge growth rate	1.1
Minimum points on edge	2
Points on longest edge	10
Surface limiting aspect ratio	20
Surface growth rate	1.2
Enhancement growth rate	1.1
Refinement length	0.422143605322
Fluid gap elements	2.0
Thin solid elements	0.2

I increase the resolution factor to reduce approximate element count. When I run it still takes too long for the volume meshing process. Since the computer was not hang or freeze, I left it to run and the total time taken for meshing and simulation to run is 6 days (I am not sure how long is the meshing and how long is simulation because I was not in office). So, before I conclude this to the customer and rerun the simulation with fluid gap element of 3, is there any ways I can make this simulation faster especially on the volume meshing process?

Your help on this is highly appreciated.

If it helps, the computer specs I use to run the simulation is as follows:

Attached is the CFZ file and report generated from previous simulation.

[Anonymous]

Hi

If I did not miss some details, the model appear to be symmetric along YZ plane. Split it and you could reduce your element count by 50%.

[Jon.Wilde]

I see that you lowered the refinement length - what is the gap size between fins here?

As an aside, be sure that you have one clearly defined inlet and one outlet to the larger air domain. You need to avoid air entering and exiting over the same suface if you can help it.

As you have fans, cutting this in two might not work, as the flow is likely to swirl and not be symmetrical - shame though 🙂

[Anonymous]

Another thought, do you have enough memory for about 18M elements? Perhaps you're using the page file, which slows things down considerably. Check CFDMesher.exe in the task manager

[Jon.Wilde]

Maybe not quite, I have it meshed with 19m elements 😉

(I used a machine with 128GB RAM).

I would guess this utilised 40-50GB of RAM.

I only made one change to the default mesh sizing - two gap elements.

The overall mesh actually looks like it could be a little finer:

Gap mesh:

[Anonymous]

---

@Anonymous wrote:

Another thought, do you have enough memory for about 18M elements? Perhaps you're using the page file, which slows things down considerably. Check CFDMesher.exe in the task manager

---

Hi erichul, thanks for your reply.

How do you check that there are 18M elements? Because what I can see from the approximation is only 920k (yeah, I know it should be more).

The computer that I used to run the simulation has 32GB RAM. Therefore, if according to Autodesk 2GB RAM for 1Million elements, then yeah this would not be enough because I would need to have 36GB RAM for this simulation. 

I attached is the task manager screenshot. So, is the paged pool or page faults signifies that I am using the page file?

[Anonymous]

---

@Jon.Wilde wrote:

Maybe not quite, I have it meshed with 19m elements 😉

(I used a machine with 128GB RAM).

I would guess this utilised 40-50GB of RAM.

I only made one change to the default mesh sizing - two gap elements.

The overall mesh actually looks like it could be a little finer:

Gap mesh:

---

Hi Jon, Thanks for the reply.

"As an aside, be sure that you have one clearly defined inlet and one outlet to the larger air domain. You need to avoid air entering and exiting over the same suface if you can help it."

Regarding this, the customer already conducted a lab test on this model and they have the result to compare with this software accuracy. It was done on a bench with external natural convection. So, I suppose the boundary condition set is correct according to this external natural convection except the fluid volume is not following Autodesk suggestion because it was particularly defined by the customer. 

1. How long did you take to run the simulation?

2. Did you split it into half as suggested by erichul?

3. Is there any changes necessary for the model geometry and boundary condition compared to what I set?

4. So, when you said the mesh could be a little finer means that 3 fluid gap element would be better right?

If no, so apparently I would need to tell the customer that we need to use a machine with bigger RAM to simulate this according to their specification right?

Will cloud be able to simulate this, and will they be able to open the file after cloud simulate it?

Because I have read here that if there is insufficient RAM to solve the analysis,when trying to open a study that had run in the cloud which is too large for your computer to open, it is expected the UI will crash.

Thanks in advance for your assistance.

[Anonymous]

Hello

The element count is approximate and sometimes it can significantly overpredict or underpredict. I prefer to keep my task manager open to check the memory usage. To avoid even starting a job with a oversized mesh I've set a default flag, mesh_size_limit = 250. This basically abort jobs containing a mesh with more than 25M elements. Additionally, you'll see the mesh count in the output dialog and in the report generated by Autodesk CFD. For very large cases, one case follow the mesh generation a bit more closely by looking at the mesh.log file. Even better would be if there is was flag to terminate the mesher when it exceeds a specified memory usage, however, this isn't available.

[Anonymous]

---

@Anonymous wrote:

Hello

The element count is approximate and sometimes it can significantly overpredict or underpredict. I prefer to keep my task manager open to check the memory usage. To avoid even starting a job with a oversized mesh I've set a default flag, mesh_size_limit = 250. This basically abort jobs containing a mesh with more than 25M elements. Additionally, you'll see the mesh count in the output dialog and in the report generated by Autodesk CFD. For very large cases, one case follow the mesh generation a bit more closely by looking at the mesh.log file. Even better would be if there is was flag to terminate the mesher when it exceeds a specified memory usage, however, this isn't available.

---

Well, thanks for your explanation and advice erichul. I will take note and apply this the next time I run simulation.

By the way, this is a great idea and maybe you could suggest it to the development team. It will be helpful in the cases like mine.

"For very large cases, one case follow the mesh generation a bit more closely by looking at the mesh.log file. Even better would be if there is was flag to terminate the mesher when it exceeds a specified memory usage, however, this isn't available."

[Jon.Wilde]

To answer the questions directed at me:

No, I said I would not cut in half as you have fan swirl right?

I meshed using a cloud machine

The only change I made to the mesh was to switch to 2 fluid gap elements and 'solve' 🙂

Does that help?

Thanks,

Jon