0

TOUGH2; grids; running time too long

Hi!

I used TOUGH2 for a transient fluid flow analysis (total time for 1 year) using grids showed in the figure.

Element volumes range from 1.0e-04 to 2.0e-1.  The total number of elements is around 160000. After 24 hours, TOUGH2 is still running. The time step seems keep constant with a value of 0.843750E+02. 

I think it took quite a long time.  I checked some information showed in the output file. It shows that TOUGH2 reduced time step at element APF78. The element is the smallest element in the mesh. I changed types of solvers provided in the code but almost no change.

I cannot change the mesh because I still need the mesh for modeling fracture propagation. 

Is there any way to shorten the simulation time without changing mesh or changing PC? 

 

Thanks!

Best regards,

Jin

 

The information showed in the output file is:

 ...ITERATING... AT [    1, 1] --- DELTEX = 0.864000E+05  MAX. RES. = 0.260381E+08 AT ELEMENT APF78  EQUATION   2

 !!!!!!!!!!!!!! EXCESSIVE RESIDUAL =  2.60381E+07
 +++++++++ REDUCE TIME STEP AT (    1, 1) ++++++++ ++++++++   NEW DELT =0.216000E+05
 ...ITERATING... AT [    1, 1] --- DELTEX = 0.216000E+05  MAX. RES. = 0.650952E+07 AT ELEMENT APF78  EQUATION   2

 !!!!!!!!!!!!!! EXCESSIVE RESIDUAL =  6.50952E+06
 +++++++++ REDUCE TIME STEP AT (    1, 1) ++++++++ ++++++++   NEW DELT =0.540000E+04
 ...ITERATING... AT [    1, 1] --- DELTEX = 0.540000E+04  MAX. RES. = 0.162738E+07 AT ELEMENT APF78  EQUATION   2

 !!!!!!!!!!!!!! EXCESSIVE RESIDUAL =  1.62738E+06
 +++++++++ REDUCE TIME STEP AT (    1, 1) ++++++++ ++++++++   NEW DELT =0.135000E+04
 ...ITERATING... AT [    1, 1] --- DELTEX = 0.135000E+04  MAX. RES. = 0.406845E+06 AT ELEMENT APF78  EQUATION   2

 !!!!!!!!!!!!!! EXCESSIVE RESIDUAL =  4.06845E+05
 +++++++++ REDUCE TIME STEP AT (    1, 1) ++++++++ ++++++++   NEW DELT =0.337500E+03
 ...ITERATING... AT [    1, 1] --- DELTEX = 0.337500E+03  MAX. RES. = 0.101711E+06 AT ELEMENT APF78  EQUATION   2

 !!!!!!!!!!!!!! EXCESSIVE RESIDUAL =  1.01711E+05
 +++++++++ REDUCE TIME STEP AT (    1, 1) ++++++++ ++++++++   NEW DELT =0.843750E+02
 ...ITERATING... AT [    1, 1] --- DELTEX = 0.843750E+02  MAX. RES. = 0.254278E+05 AT ELEMENT APF78  EQUATION   2

8 replies

null
    • Mikey_Hannon
    • 3 yrs ago
    • Reported - view

    Jin Wu,

    It’s entirely possible that a model this big would take more than 24 hours to run.  It’s hard to say, specifically, what measures you can take to shorten this time.  In general, numerical stability limits the size of the time steps when the model introduces inherently unstable flow phenomena like large pressure differences that take place over small distances.  The best advice I can give, without looking too closely at the model you’re generating, is to test different strategies (e.g., higher residual tolerances in Jacobian time step calculations) on a simpler model that incorporates most of the physics/conditions you would like to include and see which approaches give faster results before applying them to the large model.

    If you find there are no shortcuts, I’d encourage you to run this model with a parallel simulator like TOUGH3.

    Sorry, I wish I could be more helpful.

      • Jin_Wu
      • 3 yrs ago
      • Reported - view

      Mikey Hannon 

      Thanks for your reply.

      I really appreciate your comments. 

      I am modifying the model following your advice. 

      Thanks!

      Best regards,

      Jin.

      • Jin_Wu
      • 3 yrs ago
      • Reported - view

      Mikey Hannon 

      Hi,

      After  tried several times, I set scale factor  to 1000 to change the size of the mesh (default = 1.0). It ran quite well. The model quicks converged and finished after several time steps.

      So I think maybe the problem is due to small elements in the model. Large pressure differences may take place over small distances. I tried to change RE1 (relative error from 1.0e-5 to 1.0e-10) or RE2 (absolute error to from 1.0 to 100.0). It seemed that there was no improvement. The time step got much smaller (say 1.0e-3s). It stopped much earlier before the total time arrived.

      Do you have any advice to solve the issue without changing mesh size?

      Thanks!

      Best regards,

      Jin

    • yqzhang
    • 3 yrs ago
    • Reported - view

    Hi Jin Wu,

    Thank you for providing some description of the problem. In addition to Mikey's advice, one needs more information to help: e.g., What is the eos module you used? You showed the first time step information but after some time, was the code still iterating on the same element APF78? What is happening for the element (e.g., any phase change) that the code is iterating on?

    Also, can you start your initial time step as one second and let TOUGH decides how to grow time steps?

    Yingqi

      • Jin_Wu
      • 3 yrs ago
      • Reported - view

      Yingqi Zhang 

      Hi.

      Thanks for your reply. I use EOS3. 

      After some time ( I think quite a long time), the code iterated on other elements. I haven't checked the final result. Because the output file is too large.

      Part of the input file is:

      THM Analysis
      ROCKS
      MATRI    01.8400e+034.0000e-011.3300e-151.3300e-151.3300e-152.1000e+001.0000e+03
      ATMOS    01.8400e+039.9000e-011.3300e-151.3300e-151.3300e-152.1000e+001.0000e+03
      MAEDZ    01.8400e+034.0000e-011.3300e-151.3300e-151.3300e-152.1000e+001.0000e+03
      
      PARAM
           999        100000000040000500100000
       0.000e+00 3.154e+07 8.640e+04                    0.0000e+00
      1.0000e-5
      2.40000000000000e+061.00000000000000e-031.50000000000000e+01
      START
      ELEME
          1          MATRI3.6560e-02          1.0000e+00-4.373e+00 7.507e+00 0.000e+00
      ...
      BDT93          ATMOS1.0000e+50          1.0000e+00 1.750e+00-6.334e-01 0.000e+00
      OX17          MATRI1.0000e+50          1.0000e+00 1.145e+01 1.223e+01 0.000e+000
      ...
      OX17
      2.4000000000000e+061.00000000000000e-031.50000000000000e+01
      BDT93
      1.01300000000000e+051.50000000000000e+019.90000000000000e-01
      
      ...

      The attached file is my input file. Please have a look if you have any time.

      Thanks.

      Best regards,

      Jin

    • Keurfon_Luu
    • 3 yrs ago
    • Reported - view

    I concur with   Mikey Hannon , you should start with a lower resolution mesh to make sure that your input files are running correctly. I don't know what meshing software you are using (although it looks like Gmsh), but you should be able to apply a general mesh size factor.

    Besides, how many Newtonian iterations did you set? I find it weird that it always failed after the first one. Did you set MOP(16) (usually half the number of Newtonian iterations)?

      • Jin_Wu
      • 3 yrs ago
      • Reported - view

      Keurfon Luu 

      Hi!

      Thanks for your reply.

      You're right. I used a mesh generated by Gmsh. I started with a much larger uniform mesh generated by Flac3d (the total number of grids is around 9300). It ran quite well and finished in 10 minutes. Then, I used a finer mesh generated by Gmsh. The same input parameters are used  for the finer mesh. 

      I think I used default Newtonian iterations set in Tough2 (EOS3). I set MOP(16) > 1 in both models with different meshes. I changed to MOP(16)=1. It seemed that there is  no improvement. The model is still running. I will check results later.

      Best regards,

      Jin  Wu

      • Keurfon_Luu
      • 3 yrs ago
      • Reported - view

      Jin Wu If you left the number of Newtonian iterations to its default value, it should be 8 then. I would set MOP(16) to 4 to double the time step size if one cycle converges in less than 4 Newtonian iterations.

      I would still lower the resolution of the mesh at the time being to find out what could be the cause of the issue. In Gmsh, you can simply add the following line in your script (e.g. to double the size of the mesh):

      Mesh.CharacteristicLengthFactor = 2;

Content aside

  • 3 yrs agoLast active
  • 8Replies
  • 227Views
  • 4 Following