|
[Sponsors] |
June 9, 2017, 11:42 |
Residuals and convergeance
|
#1 |
Senior Member
dilaw meda
Join Date: Jun 2017
Location: algeria
Posts: 145
Rep Power: 8 |
hello every one, the residuals after 1144 iterations still like this , without any message of "the solution is converged" , what should i do ?
|
|
June 12, 2017, 03:24 |
|
#2 |
Senior Member
Join Date: Aug 2014
Location: Germany
Posts: 292
Rep Power: 13 |
Residuals do look OK. So whats the problem exactly?
|
|
June 12, 2017, 08:31 |
|
#3 |
Senior Member
dilaw meda
Join Date: Jun 2017
Location: algeria
Posts: 145
Rep Power: 8 |
||
June 12, 2017, 08:52 |
|
#4 |
Senior Member
Join Date: Aug 2014
Location: Germany
Posts: 292
Rep Power: 13 |
Are you doing steady stat or transient simulation here?
|
|
June 12, 2017, 08:54 |
|
#5 |
Senior Member
dilaw meda
Join Date: Jun 2017
Location: algeria
Posts: 145
Rep Power: 8 |
||
June 12, 2017, 09:11 |
|
#6 |
Senior Member
Join Date: Aug 2014
Location: Germany
Posts: 292
Rep Power: 13 |
Then it's converged. You see it from the residuals not going down anymore. It's not saying "Convergence reached" because you may have not defined Automatic Convergence Detection or the values that are defined are to low to reach with your current setup. You need to check also other important parameters in your simulation if they're converged or not. If they have, that's the result of your simulation. You can now compare with experiment, find agreement or redo simulation with better mesh, better BC, better Solution Shemes etc. It's an iterative process also, you see..
|
|
June 12, 2017, 19:55 |
|
#7 |
Senior Member
Svetlana Tkachenko
Join Date: Oct 2013
Location: Australia, Sydney
Posts: 407
Rep Power: 14 |
Just like BlnPhoenix said, there are different things to try (and ways to check them)
- improve your mesh, then the flow would be resolved more accurately (do mesh convergence plot called 'H-convergence plot' or grid convergence index to quantify this effect), - try a different solver in case it suits your problem better (compare results from different models and solvers of different order), - try a different boundary condition or change the computational domain size in case (compare results from different boundary conditions), - compare your results with experiments, so that you know that you're solving the same problem on the computer and in the experimental data and the problem set on the computer is not irrelevant. These processes are explained at, for example, https://www.grc.nasa.gov/WWW/wind/va.../overview.html validation and verification https://www.grc.nasa.gov/WWW/wind/va.../tutorial.html the full tutorial |
|
June 12, 2017, 20:25 |
|
#8 |
Senior Member
dilaw meda
Join Date: Jun 2017
Location: algeria
Posts: 145
Rep Power: 8 |
What parameter to improve in the mesh, i have 1milion tetrahedral cells, skewnes is 0.1 . Also , if i want to use the LES model the problem will be the time step, i read that the best time step should be less than "the minimal size of mesh/velocity" so in my case the time step should be less than 0.0001 , so if visualising a result of 10 minutes flow i need 6 milion time step :O !!!!
|
|
June 13, 2017, 03:33 |
|
#9 |
Senior Member
Join Date: Aug 2014
Location: Germany
Posts: 292
Rep Power: 13 |
LES is very demanding! I would not go down that road if your limited with your resources (CPU speed, HDD space etc.). Instead use RANS it works with fewer cells and is accurate enough for engineers
A tip to reduce cell count: Try to use a hexaeder mesh type. This saves a lot of cells and is also more accurate than tet mesh at the same time. Ansys Workbench has the "Cutcell" Mesh method which generates hexader-dominant mesh for your geometry. |
|
June 13, 2017, 06:00 |
|
#10 | |
Senior Member
dilaw meda
Join Date: Jun 2017
Location: algeria
Posts: 145
Rep Power: 8 |
Quote:
|
||
June 13, 2017, 06:07 |
|
#11 |
Senior Member
Join Date: Aug 2014
Location: Germany
Posts: 292
Rep Power: 13 |
1 min total simulation time with a time step of 1e-5 secs means:
60s / 1e-5 secs = 6e6 iterations. Yes. |
|
June 13, 2017, 06:13 |
|
#12 |
Senior Member
dilaw meda
Join Date: Jun 2017
Location: algeria
Posts: 145
Rep Power: 8 |
Hahaha , 6 milion iteration if the solution converges after 1 iterations, but if for example it converges after 9 iterations , the times step will increments every 9 iterations so i need 9*6 = 54 milions iterations to see just 1 minute of the simulation !!!!!! I need a nasa stations to do that . if i want to see just 1 seconde, it will not be sufficient ?
|
|
|
|