Problems with step size

- May 31, 2023 at 6:23 am
  
  Julen
  Subscriber
  
  Hello everybody,
  I am simulating a polymer melt inside of an extruder. I started out with the simulation being static, so no moving parts. The next step was to apply the rotation. For this I am using the kinematic reversal, so the cylinder around the screw is rotating instead of the screw itself.
  The static simulaitons work fine, but since I added the dynamic part, I run into the following problem. After a total of 39 hours Ansys Polyflow stops the computation because the step size gets decreased below the minimum I am setting. So naturally I lowered the limit (that simulaiton is still running).
  I am posting the error code as well. I am worried, that my setup will just get closer and closer to the solution but never actually reach it. Do you have any suggestions on what I could aditionally change to get a solution?
  |================> |
  Preprocessors
  Convergence assumed : Rel. var. LT 1.00000E-03
  Preprocessors
  Convergence assumed : Rel. var. LT 1.00000E-03
  F.E.M. Task
  *** WARNING *** from [SOLVR]
  Convergence or divergence has not been detected.
  Divergence is assumed.
  
  Divergence of influent solver occurred !
  Worst solver is :F.E.M. Task
  
  ==================================
  At S = 9.9929286E-01,
  dSnew is smaller than dSmin. Stop.
  dSnew = 7.5765381E-05
  dSmin = 1.0000000E-04
  ==================================
  
  *************************
  * *
  * STATISTICS *
  * *
  *************************
  
  Memory information : Global usage before end
  PID : 37180
  Max Virtual Memory : 156269 Mbytes
  including
  - Initial Memory : 86993 Mbytes
  - Additional Memory : 69276 Mbytes
  Time information :
  CPU time : 139852.3 sec.
  Elapsed time : 139852.0 sec.
  Parallelism information :
  Algo. parallelism : 0.0
  Max procs : 0.0
  Cpu Monitor :
  POLYFLOW cpu/elapsed time in seconds: 139852.2920/139852.2913 (1 calls)
  TOPO cpu/elapsed time in seconds: 28.9960/28.9953 (1 calls)
  FIELDS cpu/elapsed time in seconds: 19.2390/19.2392 (1 calls)
  Check license cpu/elapsed time in seconds: 0.9870/0.9877 (13 calls)
  SOLVR cpu/elapsed time in seconds: 0.0150/0.0148 (1 calls)
  EVOLUT cpu/elapsed time in seconds: 139802.1310/139802.1309 (1 calls)
  CCONTR cpu/elapsed time in seconds: 575.5230/575.5221 (93 calls)
  SOLVR cpu/elapsed time in seconds: 0.5670/0.5705 (42 calls)
  SOLVR cpu/elapsed time in seconds: 137851.5760/137851.5780 (113 calls)
  Prediag cpu/elapsed time in seconds: 372.8410/372.8393 (51 calls)
  DescrElement cpu/elapsed time in seconds: 368.0450/368.0453 (51 calls)
  diagonalSolver cpu/elapsed time in seconds: 125.7790/125.7759 (51 calls)
  MatrixConstruction cpu/elapsed time in seconds: 110.7410/110.7420 (51 calls)
  Preitr cpu/elapsed time in seconds: 943.3730/943.3746 (62 calls)
  DescrElement cpu/elapsed time in seconds: 635.0920/635.0882 (62 calls)
  describeAnElementLS cpu/elapsed time in seconds: 75.3560/75.4204 (68923890 calls)
  createConnectivityBlocksLS cpu/elapsed time in seconds: 297.4790/297.4764 (62 calls)
  Itrslv cpu/elapsed time in seconds: 136336.6460/136336.6397 (317 calls)
  MatrixConstruction cpu/elapsed time in seconds: 35648.7700/35648.7694 (317 calls)
  ItrslvSolve cpu/elapsed time in seconds: 100065.8360/100065.8337 (308 calls)
  SolveGlobalAssembly cpu/elapsed time in seconds: 1912.6130/1912.6265 (308 calls)
  DLRU cpu/elapsed time in seconds: 10275.6700/10275.6647 (308 calls)
  Optimization cpu/elapsed time in seconds: 161.5800/161.5718 (308 calls)
  Permutation cpu/elapsed time in seconds: 143.3230/143.3232 (22 calls)
  Build elimination order cpu/elapsed time in seconds: 143.2240/143.2251 (22 calls)
  Symmetrisation cpu/elapsed time in seconds: 3.1940/3.1934 (22 calls)
  Metis cpu/elapsed time in seconds: 134.0980/134.0969 (22 calls)
  Delaying cpu/elapsed time in seconds: 12.4480/12.4459 (22 calls)
  Apply delaying cpu/elapsed time in seconds: 11.7220/11.7217 (2 calls)
  delayLastDualAfterLastPrimal cpu/elapsed time in seconds: 0.2070/0.2066 (2 calls)
  Factorize cpu/elapsed time in seconds: 83687.7930/83687.7964 (308 calls)
  Symbolic pass cpu/elapsed time in seconds: 1399.1610/1399.6750 (250665188 calls)
  Optimize active matrix ordering cpu/elapsed time in seconds: 421.3170/421.3119 (616 calls)
  Merge matrix to reach AM size cpu/elapsed time in seconds: 1125.1390/1125.1455 (308 calls)
  BackSub cpu/elapsed time in seconds: 2701.6370/2701.6273 (308 calls)
  ItrslvEnd cpu/elapsed time in seconds: 82.8870/82.8915 (308 calls)
  generate outputs cpu/elapsed time in seconds: 880.5990/880.6024 (20 calls)
  Stat cpu/elapsed time in seconds: 0.3150/0.3144 (1 calls)
  Stop. Normal end of Polyflow
  
  ** WARNING IN THE COMMAND LINE ARGUMENTS OF POLYFLOW **
  Polyflow has been launched with the option -acc nvidia
  however, IKT-SIM-WIN-5 machine is equipped with a non-efficient device.
  
  GPU usage
  Number of available GPUs on IKT-SIM-WIN-5 : 1
  Device IDs of available GPUs : 0 Quadro RTX 4000
  Polyflow has been launched with the option -na 1.
  Polyflow has been launched in parallel with the option -th 8, a maximum of 8 GPUs will be
  used.
  With the current set up, the GPU with the following device ID will be used with the
  preference order :
  0: Quadro RTX 4000
  
  **************************
  * Expert tool messages *
  **************************
  At step 1, at least one bad element has been detected in the mesh.
  This could occur during divergence and disappear when the failed step
  is restarted. For this reason, the simulation has not been aborted.
  However, the corresponding geometry (boundaries) and results have
  been saved into the files bad_mesh.stl and bad_mesh.csv respectively.
  The faces of the bad element have been saved in the stl file, so they
  can be displayed in Ansys SpaceClaim. If needed, and in order to
  resume the simulation with a better mesh, import the bad_mesh.stl
  file in Ansys SpaceClaim for creating a new geometry, mesh that
  geometry and then, restart Ansys Polyflow using the bad_mesh.csv
  file.
  The worst element has the following feature(s):
  Too small angle in element 1247064, (x,y,z) = (0.026723,
  -0.0011204, 0.6139).
  This can lead to local inaccuracies and possibly convergence
  difficulties. Please check quality indicators as flow balance, heat
  fluxes, ... It is strongly recommended to fix the mesh.
  
  *******************************
  * Summary of the simulation *
  *******************************
  The computation failed.
  
  *****************************
  * Expert tool diagnostics *
  *****************************
  Out of convergence disk
  
  *****************************
  * Expert tool diagnostics *
  *****************************
  The problem F.E.M. Task has not converged.
  The evolution scheme can not reach the final value of the evolution
  parameter 1. Evolution stops for a value of the evolution parameter
  equals to 0.9994 because the step size has been reduced below the
  minimum (as assigned in Polydata).
  
  *****************************
  * Expert tool suggestions *
  *****************************
  There is a non-linearity in problem Navier-Stokes 3D introduced by
  the power law index (n) of the viscosity law. The value of the power
  law index is 0.755.
  The expert system suggests to use a Picard iterative scheme with a
  number of iterations about 30 or 40.
  There is a non-linearity in problem Navier-Stokes 3D introduced by
  the power law index (n) of the viscosity law. The value of the power
  law index is 0.755.
  The expert system suggests to use an evolution for this parameter
  coupling with Newton-Raphson iterative scheme. The evolution scheme
  must be the following :
  - f(s) = 1/s
  - S-ini = n = 0.755
  - S-final = 1.0
  - Initial increment = 0.9437 = n/0.8.
  If you can not define the evolution with those typical values of the
  evolution parameters, maybe your problem is not well defined. Please
  check the setup of your simulations.
  The problem F.E.M. Task does not converge because there are not
  enough iterations. The expert system suggests to change the maximum
  number of iterations from 30 to 56.
  
  Thanks to everyone in advance! If I didn't specify enough please let me know, I will provide you with as much information as needed.
  Julen
- June 1, 2023 at 9:16 am
  
  Prashanth
  Ansys Employee
  
  Hello,
  What parameter do you have under evolution? Also, the listing file made some suggestions it seems. Did you try any of those?
- June 2, 2023 at 6:44 am
  
  Julen
  Subscriber
  
  Hello,
  the Simulation that gave the error message in the original post had exactly the evolution parameters that the expert system suggests in the listing file.
  I’ve run two simulations with different parameters since, in both of those I aditionally turned on viscous heating (because I forgot in the previous one), the the evolution parameters are not the only thing I changed:
  
  1. Evol parameters at f(s) = 1/s // S-ini = 0.3 // Delta_S-ini = 0.2 // Delta_S-max = 0.25
  
  I still get an error message, with the following suggestions:
  *****************************
  * Expert tool diagnostics *
  *****************************
  The problem F.E.M. Task has not converged.
  The evolution scheme can not reach the final value of the evolution
  parameter 1. Evolution stops for a value of the evolution parameter
  equals to 0.9984 because the step size has been reduced below the
  minimum (as assigned in Polydata).
  
  *****************************
  * Expert tool suggestions *
  *****************************
  The grid Brinkman number (Br), which is defined as follows:
  Viscous heating/ Thermal diffusion (ONLY IF TEMPERATURE IS NOT
  CONSTANT) ,
  is equal to 7.97e+08 (greather than 20.0).
  It could be too large and induce convergence difficulties. Please
  check the setup of the simulation and more specifically :
  – the flow rate or the entry velocity,
  – the viscosity,
  – the thermal conductivity,
  – the coordinates,
  – the coherence of system units.
  There is a non-linearity in the problem Navier-Stokes 3D introduced
  by the viscous heating.
  The expert system suggests to use an evolution for the scaling factor
  related to the viscous heating. The evolution scheme must be the
  following :
  – f(s) = s
  – S-ini = order of 0.001 such as the viscous heating could be
  neglected
  – S-final = 1.0
  – Initial increment = S-in
  There is a non-linearity in the problem Navier-Stokes 3D introduced
  by the viscous heating.
  The expert system suggests to use an evolution for the thermal
  conductivity. The evolution scheme must be the following :
  – f(s) = 1/s
  – S-ini = order of 0.001 such as the diffusivity is dominant.
  – S-final = 1.0
  – Initial increment = S-ini
  If you can not define the evolution with those typical values of the
  evolution parameters, maybe your problem is not well defined. Please
  check the setup of your simulations.
  
  2. Evol parameters of the second simulation f(s) = 1/s // S-ini = 0.755 // Delta_S-ini = 0.09437 // Delta_S-max = 0.25
  *****************************
  * Expert tool diagnostics *
  *****************************
  The problem F.E.M. Task has not converged.
  The evolution scheme can not reach the final value of the evolution
  parameter 1. Evolution stops for a value of the evolution parameter
  equals to 0.9993 because the step size has been reduced below the
  minimum (as assigned in Polydata).
  
  *****************************
  * Expert tool suggestions *
  *****************************
  The grid Brinkman number (Br), which is defined as follows:
  Viscous heating/ Thermal diffusion (ONLY IF TEMPERATURE IS NOT
  CONSTANT) ,
  is equal to 9.47e+08 (greather than 20.0).
  It could be too large and induce convergence difficulties. Please
  check the setup of the simulation and more specifically :
  – the flow rate or the entry velocity,
  – the viscosity,
  – the thermal conductivity,
  – the coordinates,
  – the coherence of system units.
  There is a non-linearity in the problem Navier-Stokes 3D introduced
  by the viscous heating.
  The expert system suggests to use an evolution for the scaling factor
  related to the viscous heating. The evolution scheme must be the
  following :
  – f(s) = s
  – S-ini = order of 0.001 such as the viscous heating could be
  neglected
  – S-final = 1.0
  – Initial increment = S-ini
  There is a non-linearity in the problem Navier-Stokes 3D introduced
  by the viscous heating.
  The expert system suggests to use an evolution for the thermal
  conductivity. The evolution scheme must be the following :
  – f(s) = 1/s
  – S-ini = order of 0.001 such as the diffusivity is dominant.
  – S-final = 1.0
  – Initial increment = S-ini
  There is a non-linearity in problem Navier-Stokes 3D introduced by
  the power law index (n) of the viscosity law. The value of the power
  law index is 0.755.
  The expert system suggests to use a Picard iterative scheme with a
  number of iterations about 30 or 40.
  There is a non-linearity in problem Navier-Stokes 3D introduced by
  the power law index (n) of the viscosity law. The value of the power
  law index is 0.755.
  The expert system suggests to use an evolution for this parameter
  coupling with Newton-Raphson iterative scheme. The evolution scheme
  must be the following :
  – f(s) = 1/s
  – S-ini = n = 0.755
  – S-final = 1.0
  – Initial increment = 0.9437 = n/0.8.
  If you can not define the evolution with those typical values of the
  evolution parameters, maybe your problem is not well defined. Please
  check the setup of your simulations.
  
  What I don’t understand is the expert system suggesting multiple different evolution parameters, ranging from the f(s) being different and even the values for the different s being different, as it is one FEM Task. Doesn’t this mean that I can only ever have the same parameters?
  What should I do next? I don’t know where I can change the Brinkman number, I don’t even know if I can change the visous heating bit…
  
  I hope the answer isn’t too much all over the place. Thank you in adavance for your help!
- June 2, 2023 at 2:24 pm
  
  Prashanth
  Ansys Employee
  
  These are just suggestions from the solver and you cannot have these many parameters under evolution. You need to find out which parameter out of these is causing the issue and focus on just that.
  For example, if you think viscous wall heating is the issue, put a simple f(s)=s evolution scheme on flow rate.
- June 5, 2023 at 7:29 am
  
  Julen
  Subscriber
  
  How many evolution parameters can I have in one system?
- June 5, 2023 at 7:55 am
  
  Julen
  Subscriber
  
  Also another question. If the simulation stops at 0.9988 but gives me a result file to work with, how accurate could those results be? Does that mean, that it only simulates 998.8 milliseconds instead of the full 1 second? If that were the case, I would be abele to draw some conclusions from the simulatoins or not?
- June 6, 2023 at 9:22 am
  
  Prashanth
  Ansys Employee
  
  You can use it more than one parameter, but it won't be effective. It is best to use it on only the parameters that needs it.
  For more information, go through the Polyflow user's docs: Chapter 29: Evolution
  You can read through the fields and convergence behavior in listing-max file and get an idea why it failed or stopped at 0.9988.