I would like to get a definition of the term "structured grid". How does this differ from unstructured grid. I am working with a program cfd2000 which has a structured grid, and developing meshes for models is fairly difficult. What are some easier options ?

C Shores HVAC R and D Engineer

Hello,

My understanding is that a structured grid is organized in rows and columns of cells (for two-dimensions) so that a program sweeping over the entire mesh can directly address the neighboring cells to evaluate differencing expressions. There is a direct relationship between a cell's location in its row and column and the location in the CFD program's arrays used to store physical variables associated with the cell.

Unstructured grids have no obvious relationship between neighboring cells and the location of their data in the data structure. So the location of the cell and the location of the data must be connected using indirect addressing.

With structured grids, your program is less complex because you address your data (values for the variables in each cell) directly with subscripts that exploit the fact that the cells are in rows and columns. The down side is that, in order to bend the structured mesh to the physical space being modeled, cells can take on an odd shape or undesirable size.

If you substitute "node" for "cell" in the above paragraphs, the paragraphs still apply.

This extends to three dimensions, but someone else will need to provide the word that goes with row and column for the third dimension.

That's a start. I'm sure a lot of folks will take a stab a clarifying. Hope they succeed.

(1). I think Craig's problem is not really about the difference between the 'structured' and the 'unstructured' meshes. (2). The problem is he is having difficult time in generating a mesh using the code, which is using structured mesh method. (3). Let me use an example, the five dollar bicycle I bought recently in a flea market (you can say that it is a structured mesh). This is because I was able to replace the old seat and tires with the new ones, and now it is working nicely when I ride it in the park. (4). The reason why it is a structured mesh, is because there is a rule you can follow. So, once you have learned the rule, you can follow it and make it to work. (5). The cost of the structured mesh is the learning of the rule. The user of the structured mesh must know the geometry and the topology of his problem first. (6). Has anyone try to create a complete mesh around a bicycle using a structured mesh approach? I don't know. But I think it is quit possible. But it's going to take a lot of time to study the construction of the bicycle before one can create a complete mesh around a bicycle. The bicycle has been around over a hundred years, and we are still having problems in creating a structured mesh? (7). What is "unstructured mesh"? It is a cut-and-paste approach. You first create meshes here and there, in anyway you like (the code has its own way of doing thing I guess). And then cover the gap and empty space with smaller meshes, until the computational domain is filled with small cells of meshes. The point is: you don't have to follow the pre-determined rule. (8). So, the "unstructured mesh" is a custom-made bicycle. You don't have to follow the industry-wide rule in using this approach. (9). If you are after the "mesh" itself, then I guess the custom-made mesh will take less time to create. (10). But if you are after something else, like the convergence of the solution, the accuracy of the solution, etc.. then, it is hard to say whether the custom-made bicycle is more reliable, or faster. (11). "Engineering" is a sound "compromise". My suggestion is: multi-block structured mesh will give you better insight and control, while the unstructured mesh will give you a faster mesh only. (don't try to generate an unstructured mesh by hand, you must have a code to do so. In that case, the code takes over the task of mesh generation) In the end, the quality of the solution depends only on the quality of the mesh. And the quality of the mesh depends on your understanding of the problem and how you arrange the mesh. So, you still need to know a lot about your problem.

Thanks to both you and Mr. Park. I have an idea now of the issue. From my standpoint, I would like an automatic mesher, but I have also had convergence problems which apparently are more likely with an unstructured mesh.