optimum grid sizes
 

hi all, how would one determine the best grid sizes for a computational model. For example, in my case i am working on external flow over a train in 2D. My train sizes L=150m and height=3m and the whole domain is a rectangular size of L=300m x h=50m. Currently, i have a Quad meshing interval size of 0.8 using Gambit 2.0. I also know that the smaller the grid the better the accuracy but requires a huge computational time. Is there any guidelines in choosing an optimum grid size? Please help. Thanks.

Re: optimum grid sizes
 

Mesh which solution is mesh-independent is the best mesh.

If you do not have resources to produce such grid you can create initial (coarse) mesh. Obtaining solution on this coarse mesh and using gradient-adaptation tools you can fine tunning your solution. Final mesh will be fine in regions of high gradients and coarse enough in other regions.

Hope this will help,

Re: optimum grid sizes
 

to my knowledge....

optimum grid size or grid type has to be an approximation and depends on the complexity of the geometry and the regions of interest in the geometry.

best way is to start with a coarse mesh from gambit and adapt the whole grid or part of it in fluent. But adapt function can only help to a certain limit. If you intend to refine the grid further, build the a finer grid in gambit again. Trial and Error method.

if there is any other way, I would like to learn it as well.

Thank you

mp

Re: optimum grid sizes
 

thanks ff and mp for your advice. I had seen some tutorial regarding adapt function in Fluent. I would like to try that tomorrow. Now, what if i am doing a unsteady simulation and also to require the train to move some discrete time steps. Currently, my plan is to break the overall movement of the train into 10 steps for example to travel a short distance. That is to say, after the first step/solution is converged, i will used this solution as an intepolated file to be used for the second step. This will carry on until all 10 steps is completed. The question now is after the first step is converged, and the adapt function had performed to fine tuned on this step, will it have any implications later on? Meaning, the second and all subsequent steps.

optimum grid sizes (unsteady sim.)
 

In the past, i came across a similar situation. I think it's not proper to use gradient based adaptation in this case since gradients change after each time step, and so on what set of gradients one would refine the mesh? Instead try the boundary adaption, or adding boundary layer in gambit itself. That will keep the same extent of refinement in all time steps.

R there any other approaches?