Simulation failed

[gang fang]

Hi, Dave
Very Thankful for your patience and support.
The project file link is:
https://drive.google.com/file/d/1hHX2fl ... sp=sharing

Regards,
Gang

[David Parker]

Hi Gang,
I was able to run the simulation no problem. I used a time step of 0.001

Code: Select all

     1 3.000E+00 8.779E-02  (   0)   4.581E+01   1.351E+28  <  6137- 3|  21> [   0 -   0]
     1 5.000E+00 2.145E-02  (  -6)   5.571E-01   7.629E-01  <  6137- 3|  19> [   0 -   0]
     2 7.000E+00 8.782E-01  (  10)   1.294E+00   6.629E-01  <  6137- 3|  12> [   0 -   0]
     2 9.000E+00 8.280E-02  (   0)   1.285E-01   7.575E-02  < 13325- 3|  16> [   0 -   0]
     3 1.200E+01 5.529E-01  (   7)   3.451E-01   2.043E+00  < 13002- 3|  12> [   0 -   0]
     3 1.400E+01 9.370E-02  (   0)   6.671E-02   4.192E-03  <   898- 3|  15> [   0 -   0]
.
.
.
  1996 1.002E+04 1.157E-02  (  -8)   5.724E-03   3.589E-05  <  3309- 3|  12> [   0 -   0]
  1996 1.002E+04 1.723E-03  ( -17)   5.566E-04   3.488E-05  <  4177- 1|  16> [   0 -   0]
  1997 1.002E+04 1.156E-02  (  -8)   5.788E-03   2.186E-05  <  3309- 3|  12> [   0 -   0]
  1997 1.002E+04 1.668E-03  ( -17)   6.330E-04   5.019E-05  <   343- 1|  16> [   0 -   0]
  1998 1.003E+04 1.162E-02  (  -8)   5.750E-03   3.685E-05  <  4037- 3|  12> [   0 -   0]
  1998 1.003E+04 1.784E-03  ( -16)   5.619E-04   3.556E-05  <  4177- 1|  15> [   0 -   0]
  1999 1.003E+04 1.160E-02  (  -8)   5.808E-03   2.207E-05  <  4037- 3|  12> [   0 -   0]
  1999 1.003E+04 1.722E-03  ( -17)   7.187E-04   6.869E-05  <   343- 1|  15> [   0 -   0]
  2000 1.003E+04 1.166E-02  (  -8)   5.761E-03   3.725E-05  <  4037- 3|  12> [   0 -   0]
  2000 1.004E+04 1.825E-03  ( -16)   5.665E-04   3.607E-05  <  4177- 1|  15> [   0 -   0]

Be sure to use MPI when running the simulation, I don't trust using the svsolver no-mpi version. What platform are you running on?

Cheers,
Dave

[gang fang]

Hi, Dave
Thank you so much for your time.
I ran mpi mode with 6 logic cpus, and the platform is:
OS : win 10 x64
CPU: i7 8750H @2.2GHz physics 6core logic 12cores
RAM: 32G
Also , I use your recommandation step size (0.001), the steady mode is successful for 2000 steps(last residual is 1.453E-02, not to best E-03 state).
Here I have some confusions:
This model inlet diamter is about 0.33 cm, and input Q is 6cc/s , so the estimated vmax is 140cm/s.
Let CFL to be 1 and the global max edge mesh size is 0.05, so estimatd step size is 0.000357(deltaH/Vmax).
In my original setting is 0.0001, I think is enough for the coverage(for it is much smaller than estimated value), but Failed, I even decrease to 0.00005, also failed.
Your trial size is 0.001, it is much bigger than estimated value by CFL, but succeeded. From my point of view, I think if the step size is smaller enough , simulation should be always coveraged. But it may be wrong.
1. Is there some other things needed to consider for step size setting beside CFL?
2. Is there a general and robust rule for simulation setting(include mesh size generation, simulation parameters setting) to make simualtion run successfully? While it is not to try different parameters for different cases?(we usrally simulate the incrannial aneursyms )
3. what about steady and unsteady mode setting different for stepsize?
4. Mesh generation step , mesh statics show: shortest edge is 0.00028 and the longest edge is 0.079 while I set the global max edge size is 0.05. So which one we should use for estimating the step size ? https://drive.google.com/file/d/1mLrpFB ... sp=sharing
5.Also how to confirm if the mesh quality is good or not for CFD? And How to improve?

Sorry for so many questions, May be useful for the CFD beginners.
Thank you so much again!

Best,
Gang

[David Parker]

Hi Gang,

The CFL condition is just a guide and does not guarantee convergence.

Additional parameters that are used to improve convergence are the Step Construction and Maximum Number of Iterations for svLS NS Solver parameters.

The typical CFD workflow is to start with a medium mesh size for the problem you are solving and refine the mesh until the results (i.e. pressures) converge. You also need to examine the results and see if they make sense. The simulation residual may look ok but the results could still be wrong.

Cheers,
Dave

[gang fang]

Hi, Dave
Thank you so much!
So the coverage is try process, no standard for parameters setting.
So if I test steay mode is ok , run same parameters except inlet input for pulse mode, should be guaranteed to be coveraged?

[David Parker]

Hi Gang,

You are correct. The CFD analysis is a trial and error process to discover the correct solver parameters for a given mesh size and boundary conditions. Remember that a converged solution does not mean a good solution! You must always check your results to see if they make sense.

Cheers,
Dave