How to speed up the simulation of a large amount of similar steady state cases
 

Hi,

I need to simulate many (5000+) similar steady state cases with almost exact the same geometry/boundary conditions/meshing. The only thing different among these cases is the inlet and outlet velocity. Is it possible to speed up the simulation?

Thanks.

In our tests, even the slightest speed change (e.g., change from 0.6m/s to 0.7m/s) will result very different solutions so it is not possible to use one case's result as the initial iteration values of other cases.

To my understanding, CFD will turn every case into a large linear system (Ax=b) using finite volume method and use CG method to solve the x. Will these cases have the same (or similar) A so I can solve A^-1 for one time and use it for the 5000+ cases?

anyone has an idea?

Can you write some scripts?

I am familiar with a project working with scripts to run steady state pump analysis using ansys cfx. One case is used to set up the problem. After that a script is used to generate all the solver input files with different mass flow rates. The script also takes care of solving all the files. The scripts cannot change the mesh..

This sounds like an suitable approach for you when you are talking about changing inlet and outlet velocities.

Hi, Thanks for the reply.

However, this is not what I want. automation is not a problem here --- What I really need is "Speed Up" the simulation as the title mentioned since 5000+ simulations will take forever. My question is, if the cases are pretty similar, what are the similarities underneath the CFD simulation engine (at an algorithm level) so I can utilize some results instead of calculating them over and over.

To be precise, when we apply the Finite Volume Method and turn the cases into linear equations, are there similarities among the matrixes?

If the discretized equaiton is linear in the discrete solution,
then the matrix A in Ax=b depends only on the geometry of the mesh.
If the mesh is identical for each case, you can invert A just once, and
re-use it for all cases. Each problem now becomes just a matrix-vector
multiplication.

If the discretized equation is non-linear (non-linear governing equations or
non-linear schemes for linear governing equations), then the above procedure is
not applicable.

bjohn