Failed Volume Mesh + Meshing takes long times
 

Hi All,

I'm an Engineering Student working on a research project that investigates airflow within aircraft cabins. I'm new to ANSYS Meshing, and I've been running into many problems trying to get a fluid region of the cabin air meshed to feed into Fluent.

The geometry was made from scratch using AutoDesk Inventor, and was then imported into ANSYS SpaceClaim for cleanup. I'm meshing using generally the default values for fluids, and I'm trying to start with coarse global settings.

ANSYS Meshing is saying that there is some issue with edges of the geometry, but I couldn't identify any problems using the error checking tools in SpaceClaim. When I do a section view, the highlighted problem geometry seems to be a seat armrest in the cabin. However, isn't that very odd, as I had created the seats uniformly? I've also been able to preview a surface mesh without any problems, but the volume mesh seems to be getting it stuck.

Meshing also seems to be complaining about high NURBS counts, and every attempt I make to generate a mesh takes about 1-5 hours before stalling and failing. I'm on a 3.0GHz machine with 32GB of RAM, if that makes a difference.

This has been bugging me for several weeks already. Any help would be vastly appreciated, as these problems are pushing me uncomfortably close to the project deadline for school :( . If it helps, I can also attach the geometry in question in whatever file format is desired.

Thanks!

Pictures
 

Sorry, I forgot to attach pictures. Here they are.

just little advices:
1)I Think that the mesh is far too complex. This is a job for Aircraft builder department with a huge computational power.
2)If there are simmetry plane, let's take advantage of it and cut the computational domain by a half
3)You need to simplify the geometry around the seats. Ask a professional CAD designer for an help.

Let me know if you need further help.

Geometry Simplification
 

Thanks for your reply! Really appreciate it. I'm fairly new to the meshing process, but I've seen some published papers with very simple geometry doing similar work with global airflow in aircraft cabins on a 2.0GHz CPU with 1GB of RAM.

As I am using a workstation with an 3.5GHz i7 and 32GB of RAM, I am hoping that I would be able to push complexity a little further. I've thought about simplifying things using symmetries, (e.g. mirroring down the center aisle) however I am a bit concerned that phenomena like the Coanda effect would not be properly detected and simulated.

I do think that I should simplify the seats, but I don't have any real CAD designer contacts around. Would it be more preferrable to do simplifications in CAD like SpaceClaim, or could I do defeaturing in Meshing? Do you have any suggestions for resources that I can look at/general steps I should consider taking?

Many many thanks!

Honestly I didn't think about Coanda effect. I'm really sorry. I spend most of my time with high pressure internal industrial flow simulation. I'm not an aeronautic engineer.

Regarding the computational power:
I use a portable computer (DELL M3800 with i7 processor and 16 Gbyte Ram and in my opinion it is better to stay below 3 Million elemets (meshing within the workbench tools).
Unfortunately I don't have any experience about "outside workbench mesher" (like ICEM), therefore I don't know if a different mesher (inside or outside the ANSYS "family") will do the job more efficently...
I know for sure that there are special softwares that solve very large meshes breaking them in parts, but I've never had the chance to be trained for using them.

answering your questions:
My opinion is that it will be better to simplify all the difficult geometries inside the cabin, by replacing it with very simple geometries (seat = two squared boxs etc...), without making any fillet or chamfer radius. In other words you have to re-design everything with very simple geometry.
Infact once the mesh will be done for the first time, it will be necessary to look very carefully at the mesh quality. (this process is very time-consuming inside ANSYS workbench workflow).
Finally, if you succeed, it will be time to start FLUENT or CFX. Generally speaking CFX is much more tolerant regarding provisional mesh (I mean mesh that aren't perfect).
Regarding the CAD: I use CATIA, wich is the top for this kind of work, due to the huge set of surface shaping functions. It produces very clean geometry with controlled tangency contact and curvature... and it is very rigorous when it comes to "reversing the geometry" to extract the "negative" in order to extract the fluid out from the void between the solid.

Let me know if you need help looking at your geometry. I don't have much time but I can take a look.

The screenshots you have attached show us that you have basically no mesh controls applied to your model. Are you just importing and pressing "generate mesh"?

Try with just one row of seats first (cut your geometry off at some location" in order to figure out what sizing you'll need. Then when that is working you can start with the entire domain. This will save you time in the end rather than waiting for meshing to maybe finish and not knowing what happens.

Simplifying Geometry/Refining Mesh
 

Thanks so much MB72 and jrunsten for all your help so far! Really appreciate it.

MB72, I will try simplifying the geometry and the mesh to something more managable. Do you have any specific experience regarding controlling the amount of high order NURBS? It does seem however, that when I check for orthogonal quality and skewness, some contact issues lie in regions between the human models & chairs as well as between adjacent chairs, where skewness is as high as 0.99 in a couple of cells. The geometry used to be quite dirty when it was made in Autodesk Inventor, which I have tried my best to clean up using SpaceClaim, closing in gaps and tangency issues. I also generate the fluid region using SpaceClaim. However, when I go to try and detect contact issues in SpaceClaim Direct Modeller, it does not turn up any problems. I'll see if I can ask around about CATiA, and whether my institution has a license for possibly using that.

jrunsten, you are right about there being only global mesh settings applied at the moment. Essentially I've been trying different global inflation and relevance settings, and the mesh fluctuates quite wildly from about 1.5 million to as much as 22 million nodes. I've decided on building up from a coarse mesh now, and I will follow up shortly about how local sizing goes, when I rework some things later today. Do you have any suggestions for going about refining the mesh for my type of situation? I realize that there is an option to highlight poor mesh quality areas through statistics in ANSYS Meshing, but how do I go about fixing the detected areas from there? I've tried looking at ANSYS Academic Resources, but they are generally all pretty broad instructions.

To All: I've attached a copy of my cabin in STEP format, if it helps to visualize my situation. Once again, I really appreciate all of your time! Any help would be very highly valued.

https://www.dropbox.com/sh/nubihoiji...HxQY6_nMa?dl=0

About your question: I don't have any experience with NURBS. I can't open the file now but I will.
However pay attention: The Workbench mesher (without defeaturing options) has always had problems with the tangency of curved bodies (or curved body with flat body). The reason is clear: in the fluid domain there is a very sharp edge next to the bodies contact. Always in this condition you are going to obtain skewness around 0.99! Therefore when you place two bodies in contact, it is better to have a very "flat & broad" contact surface & around

First inspection
 

Dear StarCFD

I've revised your file.
The problem is mainly the contact between the human body and the seat.
I think it is possible to heal the geometry.
With few modification I was able to shift from 0.99 to 0.96 skewness, but I'm managing to bring it below 0.90.
Meshing this geometry took less than 10 minutes on my computer.
Tomorrow I will post back the healed geometry.