[dli]

Hi,guys. I have problems on the arrangement of different types of unstructured mesh,such as tetrahedron,pyramid,prism,hexahedron and polyhdral. With the cell-centered scheme and face address,we must build the connectivity of those data,for example ,face to element,element to nodes(for gradient calculation based on nodes and display in the tecplot). so I have some questions about them, can you guys give me some advice?
1. a computation domain may compose different types of mesh,just mention above, usually the mesh file can give use some information about this, but with the cobalt data format file,there are no direct information about which types of the elements are, only gives the number of face,but the pyramid and prism both have 5 faces, we can also judge them by the number of nodes,but here comes a question, how can I know the number of the node of an element?
2.the second question is still the connectivity of nodes and elements, in Tecplot, it has its own data format,for example,in requires a special order of the eight nodes of a hexahedron to avoid a wrong line connection. so can there are some algorithm to solve those problem?

thanks.

[mprinkey]

I haven't used tecplot in a very long time or cobalt at all, so I can't answer specific questions about them, but parsing and organizing unstructured CFD meshes is something I know a bit about, so perhaps I can offer some guidance.

(1) A pyramid will have two tri and three quad faces. A prism will have four tri and one quad face. If Cobalt mesh file says cell 4003 is composed of faces 100,101,102,201,202, say...you will need to inspect each of the faces so see how many nodes each have and count tri/quad faces. Nodes per cell time are just simple geometry: Tets will always have 4 nodes, hexs will always have 8 nodes, pyramids will have 5 nodes, and prisms will have 6 nodes.

(2) I don't know the Tecplot requirements that you are meeting, but ordering nodes is often associated face orientation. Once you order the nodes to get all of the faces (say) pointing away from the cell centroid, then you can make sense of the node order relative to the template that TecPlot requires.

So, first, you have to order the nodes on the faces. Take the three nodes position vectors (A,B,C) and build two edge vectors (A-B and A-C). Then form the cross product of the two CROSS(A-B,A-C) and dot with (A-CellCentroid). That will tell you whether the face order (A,B,C) points into the cell or out of the cell. If it is the wrong way, use a face order of (A,C,B) instead.

Quads (and general polygon faces) are a little harder. First you have to make sure to avoid "bowtie" faces where the edges intersect. The general algorithm is to find the ordering of the nodes the minimizes the perimeter. So, if you have (A,B,C,D) as the nodes on the quad, you can compute the perimeter as |A-B| + |B-C| + |C-D| + |D-A| and compare it to the perimeter of (A,C,B,D) as |A-C| + |C-B| + |B-D| + |D-A| and choose the ordering with the shorter length. That will be the non-bowtie ordering. The treatment of poly faces is the same idea, but there are more than two permutations that you need to test, but the minimum perimeter ordering is still the "right" answer.

For quads and poly meshes, you should really verify that all of the nodes are coplanar. That is usually an implicit assumption in any linear finite element or finite volume formulation. This isn't an issue for tris because three (non-colinear) nodes are always coplanar. But, now you can analyze the face direction by using any three nodes as I outlined above with tris. If the face is pointing the wrong direction REVERSE THE ENTIRE ORDER of the node list. You must be careful permuting quad/poly face node lists that you do pick a bow-tie configuration. By reversing the entire node list, you keep the same edge list and the same perimeter, but invert the face orientation.

With your faces oriented relative to the inside of the cell, it should then be easy to build connections and node orderings.

[FMDenaro]

I don't know if you have already the guide for tecplot format datafile

http://www.hgs.k12.va.us/tecplot/doc...rmat_guide.pdf


you should go in the section for FE format (i,j,k-ordered data is for structureed grid). If I remember correctly, you have headers for the first identification of elements, then the list of nodes for each elements, then the topology...Usually using non-structured grids you should have a pointer that for each node gives you the linked nodes and the structure of the element (for example by its faces).

[dli]

Originally Posted by mprinkey

I haven't used tecplot in a very long time or cobalt at all, so I can't answer specific questions about them, but parsing and organizing unstructured CFD meshes is something I know a bit about, so perhaps I can offer some guidance.

(1) A pyramid will have two tri and three quad faces. A prism will have four tri and one quad face. If Cobalt mesh file says cell 4003 is composed of faces 100,101,102,201,202, say...you will need to inspect each of the faces so see how many nodes each have and count tri/quad faces. Nodes per cell time are just simple geometry: Tets will always have 4 nodes, hexs will always have 8 nodes, pyramids will have 5 nodes, and prisms will have 6 nodes.

(2) I don't know the Tecplot requirements that you are meeting, but ordering nodes is often associated face orientation. Once you order the nodes to get all of the faces (say) pointing away from the cell centroid, then you can make sense of the node order relative to the template that TecPlot requires.

So, first, you have to order the nodes on the faces. Take the three nodes position vectors (A,B,C) and build two edge vectors (A-B and A-C). Then form the cross product of the two CROSS(A-B,A-C) and dot with (A-CellCentroid). That will tell you whether the face order (A,B,C) points into the cell or out of the cell. If it is the wrong way, use a face order of (A,C,B) instead.

Quads (and general polygon faces) are a little harder. First you have to make sure to avoid "bowtie" faces where the edges intersect. The general algorithm is to find the ordering of the nodes the minimizes the perimeter. So, if you have (A,B,C,D) as the nodes on the quad, you can compute the perimeter as |A-B| + |B-C| + |C-D| + |D-A| and compare it to the perimeter of (A,C,B,D) as |A-C| + |C-B| + |B-D| + |D-A| and choose the ordering with the shorter length. That will be the non-bowtie ordering. The treatment of poly faces is the same idea, but there are more than two permutations that you need to test, but the minimum perimeter ordering is still the "right" answer.

For quads and poly meshes, you should really verify that all of the nodes are coplanar. That is usually an implicit assumption in any linear finite element or finite volume formulation. This isn't an issue for tris because three (non-colinear) nodes are always coplanar. But, now you can analyze the face direction by using any three nodes as I outlined above with tris. If the face is pointing the wrong direction REVERSE THE ENTIRE ORDER of the node list. You must be careful permuting quad/poly face node lists that you do pick a bow-tie configuration. By reversing the entire node list, you keep the same edge list and the same perimeter, but invert the face orientation.

With your faces oriented relative to the inside of the cell, it should then be easy to build connections and node orderings.

Thanks a lot ， I have solve this problem，your idea help me gain more insight of.this