Challenging Meshing...
 

I continue to pave my way to a working yacht simulation...

(For reference, see my post "Convergence speed in free surface simulations" further down in the forum)

Thank you all who provided input so far, the solution has improved quite a lot.

I now have a nice mesh with sensible y+ (5-8) values and a much larger domain. Together with a decreased Timestep for the Volume Fraction equations I now converge quite rapidly without the oscillations I posted about earlier.

However I am not totally happy with the accuracy yet, and I suspect one part of that is due to the water surface being rater undefined, even at Interface Compression = 2. The interface need about 5 element widths for volume fraction to go from 0 - 1, so I would like element heights to be at most 50mm high on the surface. Obviously I can't mesh my entire 40m x 80m x 12m domain with 50mm elements.

(I sould say that I am using ICEM 10.0.1 for meshing, with a mixed quad/tri surface mesh, hexa boundary layer and tri volumes)

I can put a Density on the surface. However this makes the elements small in all dimensions, not only their height, thus causing a radical increase in element count. What I really would like is to have 6 layers of prisms at the water surface, with equal heights across the domain but with length/widths larger at the far field and smaller nearer the hull.

Can anyone think of a way to create this?

It's harder than it sounds at first glance, because I want the prisms in the middle of the fluid, not attaced to a wall.

If I create a box geometry to grow the prisms from, CFX sees that as a wall which I need to provide a boundary condition for, and there is no "Treat it like it's not there" boundary condition ;-)

If I place a single surface geometry at the water surface, I can create a beautiful surface mesh on it, but the prism mesher refuses to run since my blocking is not enclosing a proper volume. ("shells xxx and yyy not consistently oriented").

Also running the Tetra+Prism mesher on this design (with very detailed boundary layers) is fruitless, it either crashes or comes out with a mesh full of negative volumes and uncovered faces, so I must go the Blocking->Surface->Prism->Volumes path.

As you can see I'm kind of stuck, and I spend waaay too much time on this project already, so any creative ideas are much appreciated!

/C

Re: Challenging Meshing...
 

Hi,

Define a "air" body and a "water" body in ICEM. The interface between the two bodies can be meshed using prism layers to give you a nice interface.

Glenn Horrocks

Re: Challenging Meshing...
 

Oh I wish it was as simple as it sounds. I thought I had all the quirks in IEM worked out by now, but it seems not.

No matter what I do the prism mesher tells me "Shells XXX and YYY not consistently oriented. Application finished."

I tried with the interface surface curves separated and joined, with structured blocks and unstructured blocks, with the block edges as separate edges and with the block edges joined. With "Interior Wall" swithced on and off. With "Body" objects in the volumes and without. With the side blocks covering the whole side and with them split at the interface.

If I remove the blocking on one of the volumes, I can do prisms into the other one, but then when I do the surfaces on the other volume the shells don´t connect...

I'm thinking since you made the comment it must be possible, but HOW???

Re: Challenging Meshing...
 

Hi,

Have you tried doing it with just tets and prisms and no hexas? Prism in ICEM does not like hexas in the grid (at least last time I looked).

Glenn Horrocks

Re: Challenging Meshing...
 

Chebeba: sounds like you are having tet & prism & hex elements in the domain. Don't you think this only makes thing worse? If you really need to do that, try to create two different meshes and join them up using interface in Pre. -

Re: Challenging Meshing...
 

Yes use hexas because I like to get as fine a mesh as possible into the 3.5M elements that is my budget to fit in a 4GB RAM computer...

The prism mesher is fine with hexas, I have a beautiful boundary layer mesh on the hull using mostly hexas. Mind you I had to work A LOT with the geometry and the blocking to get it to do 30 layers without crashing. (Also, with hexas you need to turn off (set passes=0) the pre-smoothing the prism mesher does, otherwise it stops, maybe that's what got you last time you looked:)

The issue I'm hitting obviously has to do with the "T" topology between the interface surface and the sides of the domain. I keep thinking there must be a way to do the blocking that results in a shell topology the mesher likes, but at the moment I can't think of more things to try...

/C

Re: Challenging Meshing...
 

Hi Chebeba,

You can 3.5 million elements on a 4GB RAM computer? Are you running on Linux or on Windows?

I am using CFX at work and we HAVE to use Windows, I don't get anywhere near 3.5 million elements onto 4GB. Windows takes up 2GB just for the operating system.

Just curious what you are running. Good luck on your meshing issues.

Re: Challenging Meshing...
 

So why do you install 4GB if you're running windows? Waste of silicon, 32 bit editions of windows will only adress 2GB no matter how much hardware you have. (Or actually maybe 2.6 GB with various tricks, search other forums for this). It's not the OS taking 2GB, it's 2GB RAM wasted to no use.

You need 64 bit OS for >2GB. I run Linux. /C

Re: Challenging Meshing...
 

Hi Chebeba,

By definition, a 32-bit processor uses 32 bits to refer to the location of each byte of memory. 2^32 = 4.2 billion, which means a memory address that's 32 bits can refer to 4.2 billion unique locations, i.e. 4 GB, NOT 2GB.

Windows Professional 2000 takes up 2GB of memory and the remaining 2GB can be used for your application (i.e. CFX).

Window XP and some server versions have a /3GB switch that puts Windows into 1GB and allows your application to access 3GB.

I am pretty sure, although not positive, that some server versions of Windows can go much higher in terms of the amount of RAM they can address, I think that this is using the /PAE switch, but this is one of the tricks that you refer too and may not give as great an advantage as it seems.

There is no question that Linux is superior for doing CFD, but sometimes when you work for a big industrial company with business controls and corporate directions, you have to fit into the system ... there is NO possibility of using Linux at my company. I have 10 years of experience in UNIX and LINUX operating systems, so they are my preference.

Re: Challenging Meshing...
 

Well then switch to a 64 bit windows version and you'll suddenly find the OS taking a lot less than 2GB :)

...and go bang your calculations in Steve Ballmers head, not mine :) !

(But watch out for flying chairs...)

/C