Mesh Conversion : structured to unstructured
In our CFD analyses, Structured hex meshes are what we mostly work with, set up using ICEM. However, when we export to CFX for setting up the boundary conditions and solving, the mesh has to converted to an Unstructured mesh.
I'd like to know : Why? and How exactly does this take place, considering the mesh doesn't look unstructured once the conversion is done!?
Please feel free to point me to literature that provide explanations.
Wikipedia and the Ansys website mention that CFX utilises a Control Volume based Finite Element discretisation scheme. I'm not very clear what that implies either.
Thanks for your help!
The first thing is that even .uns format can have structured mesh but structured mesh cannot be called as unstructured. Structured meshes have uniform nodes positioned to give a fair look to geometry thus establishing a relation between the neighboring nodes. But can be called unstructured since it obviously follows a regular pattern of structure called grids.
CFX Solvers uses FINITE VOLUME METHOD that means that it accounts and computes on 3D derivatives. So when you convert to .uns format, the volume mesh is also formed. This helps the CFX solver program to understand the nodes and conduct convergence computation. You can check the volume mesh by clicking the volume option at the right hand side below premesh, i guess and confirm in ICEM-CFD.
The mesh still looks structured because it is structured only but the volume has been created now after conversion. Thats all.
You said control volume based finite element descritization scheme, which means to say FINITE VOLUME METHOD.
Thanks for your reply.
I understand your answer to CFX being a Finite Volume solver. The question struck me because I hadn't come across any other software being described with that kind of terminology and I was under the (wrong) impression that CFX is FEM based.
I'd like more details regarding the reason and mechanics behind the conversion and the characteristics of the .uns format. I've posted the same question on Linked in sans the part about CFX being a FVM based solver. (see URL)
Clarification regarding CFX
CFX isn't a Finite Volume solver.
It's a Control Volume based Finite Element Solver.
Have a look at the attached picture.
FEM can be done using either a structured or an unstructured grid. Converting from structured to unstructured doesn't change the grid or the basic numerical methodology, it just changes the way the grid is described and the amount of grid information that has to be stored. A structured grid can be described very simply: x0,y0,z0, deltax, deltay, deltaz, nx, ny, nz. The numerics in a solver that uses a structured grid can be written quite simply. If a code is written so that it can utilize a more complex unstructured grid, then the numerics doesn't rely on simple formulas: the location of each node and it's connectivity to and distance from its neighbors all have to be stored in the grid file.
Thanks for your post. It does make sense to me.
I'd still like to try understand better/practically - how an existing structured mesh is converted to an unstructured format.
By the way >
What exactly does 'Control Volume' based Finite Element method then mean ?
The attached picture in my previous post also describes that, in CFX, the nodes are at the vertexes of the cells, like in the Finite Element method. Where does the 'Control Volume' fit in ?
Structured to Unstructured
When you create a structured grid, as in a grid generator like Pointwise, it only takes a few lines to describe the grid. The example I gave above was the simplest: rectangular with even spacing in all directions. Codes that rely exclusively on this kind of grid may not even read a grid file; the grid description can be part of the input file for the code. They can use any numerical methodology. The locations of the nodes and/or elements can be calculated simply "on the fly": a node is described by its indicies: (i,j,k), and calculating it's location is simple: something like (x0+i*deltax,y0+j*deltay,z0+k*deltaz). The calculation of derivatives is straightforward.
Of course there are limitations: you can't let the mesh deform during the simulation.
In an unstructured grid, nodes can be located anywhere and numbering of the nodes doesn't indicate connectivity; each node is defined based on its location and its connectivity to other nodes has to be stored in the grid file. Cells can have different shapes. A large grid file is required, which contains the locations and connectivities of the nodes, the definitions of the faces and cells, a directionality to determine whether the gradient is positive or negative across a cell face, etc. I said one thing wrong in my first response though: distances don't need to be stored in the grid file; those can be calculated "on the fly".
The numerical methods are the same: for example, you can calculate a laplacian on a structured grid or on an unstructured grid, but it's simple on a structured grid and more complex on an unstructured grid. If you're using an unstructured grid you generally define a local coordinate system centered at the cell centroid, calculate the laplacian there, and then map it to x-y-z space (assuming the equations are being solved in x-y-z space.)
When you create a structured grid in Pointwise and then export it as an unstructured grid for CFX, Pointwise knows that CFX needs a large grid file that contains the full description of the grid, as if it were unstructured. Pointwise calculates the location of each node and all of the connectivities and writes a grid file that contains all the information needed by a code that uses unstructured grids.
There's a pretty good description of the difference between FEM and FVM here: http://www.cfdesign.com/OnlineHelp/2...ite_Volume.htm
can anyone help me know how to convert .uns to .nas format in icem cfd.
It's very basic I know, but I haven't been able to do it. :p
|All times are GMT -4. The time now is 09:40.|