Welcome, Guest
Username: Password: Remember me
Forgot your password? Forgot your username? Create an account
Forum
System Modules
TELEMAC-3D
Crashing with more horizontal levels
Flat Threaded
  • Page:
  • 1
  • 2

TOPIC: Crashing with more horizontal levels

Crashing with more horizontal levels 7 years 2 months ago #27864

  • SDAC
  • SDAC's Avatar
Hello!

I'm trying to refine my mesh. My simulation requires 18 horizontal levels. However I get the error: "segmentation fault- invalid memory reference" when I prescribe 13 levels or more. Anything before that I get "exceeding maximum iterations" error message, with the number of error message increasing as I increase the number of horizontal levels. Clearly something is wrong with my mesh, but I've not had this problem before.

What might cause this to happen?

Looking at the sortie file I've identified some possible sources of the error:

1)Telemac says the maximum number of elements around a point: 11. Am I right in that the maximum number around a point is 9? I'm not sure this is the source of the error as this doesn't change as I increase the number of horizontal levels.

2) When I use "checking the mesh = yes I get told that the distance between nodes are too small. In this case, 9.99999999999909051E-003 between points 45658 and 45675. Previously when I've gotten this error the points were almost on top of each other, however in this case the distance seems large enough that is should work. What is the minimum distance between nodes?

I also have some over-constrained triangles, but in previous runs I've used a PSI scheme and it's worked fine.

I've attached the error message and mesh checking result:

Would appreciate any insight.
Many thanks!
********************************************
                   *               LECDON:                    *
                   *        BEFORE CALLING DAMOCLES           *
                   ********************************************

 THE KEY-WORD: LISTING PRINTOUT PERIOD
 APPPEARS AT LEAST TWICE , THE LAST VALUE WILL BE KEPT...

 END OF FILE FOR DAMOCLES

                   ********************************************
                   *               LECDON:                    *
                   *        AFTER CALLING DAMOCLES            *
                   *        CHECKING OF DATA  READ            *
                   *         IN THE STEERING FILE             *
                   ********************************************

 EXITING LECDON. NAME OF THE STUDY:
 TELEMAC 3D : Channel

 OPENING FILES FOR TELEMAC3D
  - FILE    T3DGEO OF TYPE BIG_ENDIAN
  - FILE    T3DGEO IS DOUBLE PRECISION

 POINT_TELEMAC3D: MEMORY ALLOCATION

 READ_MESH_INFO: TITLE= CA22
            NUMBER OF ELEMENTS:   214220
            NUMBER OF POINTS:   111569

            DOUBLE PRECISION FORMAT (R8)

 MXPTEL (BIEF) : MAXIMUM NUMBER OF ELEMENTS AROUND A POINT:  11
                 MAXIMUM NUMBER OF POINTS AROUND A POINT:  11
 CORRXY (BIEF):NO MODIFICATION OF COORDINATES

 MESH: MESH2D ALLOCATED

 READ_MESH_INFO: TITLE= CA22
            NUMBER OF ELEMENTS:   214220
            NUMBER OF POINTS:   111569

            DOUBLE PRECISION FORMAT (R8)

 MXPTEL (BIEF) : MAXIMUM NUMBER OF ELEMENTS AROUND A POINT:  11
                 MAXIMUM NUMBER OF POINTS AROUND A POINT:  11
 CORRXY (BIEF):NO MODIFICATION OF COORDINATES

 MESH: MESH3D ALLOCATED

 2D MESH
 -------

 2D ELEMENT TYPE                :       10
 NUMBER OF 2D NODES             :   111569
 NUMBER OF 2D ELEMENTS          :   214220
 NUMBER OF 2D BOUNDARY NODES    :     8916

 3D MESH
 -------

 3D ELEMENT TYPE                :       40
 NUMBER OF 3D NODES             :  2008242
 NUMBER OF 3D ELEMENTS          :  3641740
 NUMBER OF LEVELS               :       18
 NUMBER OF BOUNDARY ELEMENTS    :   151572
 TOTAL NUMBER OF BOUNDARY NODES :   383626
 INCLUDING   LATERAL BOUNDARIES :   160488
                        SURFACE :   111569
                         BOTTOM :   111569

_____________
runcode::main:
:
   |runCode: Fail to run
   |C:\TELEMAC\PH04 Telemac\PHEOX_CD0.55_MoC.cas_2017-09-22-12h02min15s\out_BlWa.exe
   |~~~~~~~~~~~~~~~~~~
   |Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
   |
   |Backtrace for this error:
   |#0  ffffffffffffffff
   |~~~~~~~~~~~~~~~~~~
EXITING LECDON. NAME OF THE STUDY:
 TELEMAC 3D : Channel

 OPENING FILES FOR TELEMAC3D
  - FILE    T3DGEO OF TYPE BIG_ENDIAN
  - FILE    T3DGEO IS DOUBLE PRECISION

 POINT_TELEMAC3D: MEMORY ALLOCATION

 READ_MESH_INFO: TITLE= CA22
            NUMBER OF ELEMENTS:   214220
            NUMBER OF POINTS:   111569

            DOUBLE PRECISION FORMAT (R8)

 MXPTEL (BIEF) : MAXIMUM NUMBER OF ELEMENTS AROUND A POINT:  11
                 MAXIMUM NUMBER OF POINTS AROUND A POINT:  11
 CORRXY (BIEF):NO MODIFICATION OF COORDINATES

 CHECKING THE MESH

 SMALLEST DISTANCE BETWEEN TWO POINTS:   9.9999999999909051E-003
 BETWEEN POINTS:        45658  AND        45675
 VALUE TOO SMALL
 POINT            1  HAS ONLY            2  NEIGHBOURS
 POINT        70298  HAS ONLY            2  NEIGHBOURS
 POINT        99851  HAS ONLY            2  NEIGHBOURS
 POINT       106840  HAS ONLY            2  NEIGHBOURS
 POINT       107598  HAS ONLY            2  NEIGHBOURS
 POINT       111567  HAS ONLY            2  NEIGHBOURS
 POINT       111569  HAS ONLY            2  NEIGHBOURS
           7  OVERCONSTRAINED TRIANGLES
 POSSIBLE BUT NOT RECOMMENDED

 PLANTE: PROGRAM STOPPED AFTER AN ERROR
 RETURNING EXIT CODE:            2
_____________
runcode::main:
:
   |runCode: Fail to run
   |C:\TELEMAC\PH04 Telemac\PHEOX_CD0.55_MoC.cas_2017-09-22-12h05min23s\out_BlWa.exe
   |~~~~~~~~~~~~~~~~~~
   |STOP 1
   |~~~~~~~~~~~~~~~~~~
The administrator has disabled public write access.

Crashing with more horizontal levels 7 years 2 months ago #27869

  • Phelype
  • Phelype's Avatar
  • OFFLINE
  • Senior Boarder
  • Posts: 140
  • Thank you received: 64
Hello,

No, the number of elements and points around a point refer to the 2D mesh, but generally do not indicate an error.

Both your program executions finished with an error, either the simulation with the segmentation fault or the mesh check with plante. There may be something with your mesh. Could you upload it, please?

Regards,

Phelype
The administrator has disabled public write access.
The following user(s) said Thank You: SDAC

Crashing with more horizontal levels 7 years 2 months ago #27871

  • SDAC
  • SDAC's Avatar
Hello,

I've attached my mesh. I'm attempting to run with a simple triangulation. I've had success with triangulated meshes before.

Thanks for looking into it!
The administrator has disabled public write access.

Crashing with more horizontal levels 7 years 2 months ago #27872

  • Phelype
  • Phelype's Avatar
  • OFFLINE
  • Senior Boarder
  • Posts: 140
  • Thank you received: 64
I think you forgot to attach the mesh : P
The administrator has disabled public write access.

Crashing with more horizontal levels 7 years 2 months ago #27873

  • SDAC
  • SDAC's Avatar
Hmm. Maybe the .slf file was too big?

Edit: finally attached it! :p

File Attachment:

File Name: Mesh2.zip
File Size: 1,133 KB
The administrator has disabled public write access.

Crashing with more horizontal levels 7 years 1 month ago #27877

  • cyamin
  • cyamin's Avatar
  • OFFLINE
  • openTELEMAC Guru
  • Posts: 997
  • Thank you received: 234
Hello,

The minimum distance between nodes is limited by the precision of the selafin format. If you are using single precision SELAFIN, then after 8 significant digits you are getting into trouble. This can be eliminated by using a mesher that can save double precision SELAFIN like the latest BlueKenue beta from here.
I had a look at your mesh, and certainly in can be improved. I am not doing channel meshes at all but I thought that the elements are supposed to be elongated at the flow direction if the mesh is to be most efficient. But that does not cause troubles. I think the problem lies at your edges, where it seems there are a lot of elongated triangles. That increases the risk of having two nodes very close to each other. I used to get edges like this with the release version of BK, but this appears to be improved in the beta releases. Have you tried the beta?

Regards,
Costas
The administrator has disabled public write access.
The following user(s) said Thank You: SDAC

Crashing with more horizontal levels 7 years 1 month ago #27882

  • Phelype
  • Phelype's Avatar
  • OFFLINE
  • Senior Boarder
  • Posts: 140
  • Thank you received: 64
Hello,

My guess on why your mesh is not working would be, as Costas said, the shape of the elements. In a general way, when numerically solving a PDE, a uniformity of the spatial discretization (the edge lengths of the elements) would make the solution easier for the model. That is, in an ideal case, if you cad a mesh composed solely of equilateral triangles, it would have an optimal geometry. Of course, unless you are simulating a triangular basin, that is impossible.

Telemac has a lot of very sophisticated methods to alleviate this condition and allow you to use very irregular meshes without major problems at the cost of a bigger computational time.

What is really strange (and makes me doubt the hypothesis of the elongated elements) is that your simulation does not stop in the solver due to convergence problems, instead something is causing a segmentation violation. This shouldn't be caused by poorly shaped meshed. But, since it's the only hypothesis I have for now...

As Costas said, your mesh has very long, thin triangles at the edges and could be improved. There is a tool in BlueKenue that can help. You can go in "View -> T3 Mesh Toolbar". In this toolbar there is a button "Auto-Adjust Nodes" that will try to reshape the elements of your mesh to try to follow a Delaunay Triangulation. Along with this you can split and delete elements and edges to modify your mesh.

I modified you mesh a little to try to see a better result. Only then I realized I don't have the .cli and the other necessary files to run your simulation. So I'm attaching you mesh, with a few modifications, with the elements less elongated. Try to run your simulation with this version to see what happens. If the problem persists, please attach also the files necessary to run the simulation.

P.S.: Could not attach, so use this Drive link:
drive.google.com/open?id=0B8r99pzQjmbQSEFoV0QydlBLWGM

Regards,

Phelype
The administrator has disabled public write access.
The following user(s) said Thank You: SDAC

Crashing with more horizontal levels 7 years 1 month ago #27884

  • SDAC
  • SDAC's Avatar
Thanks to you both!

I'm using the BlueKenue beta because of the single precision SELAFIN problem mentioned above.

The segmentation error was resolved this morning when I updated my computer. Previously, the only variable I changed to cause the error was the mesh however following the update to my computer the error was stopped. Odd, but since Telemac now accepts my mesh I've not investigated it further.

It still isn't optimum, as I have to drop to low time-steps to get it running (about 0.01.) Currently I'm testing what degree of mesh size Telemac can handle before it fails.

What were the modifications you made, Phelype? Thankyou for your edited version. Comparing the two I see the meshes are less stretched and that certain boundary nodes are reduced.

Whilst testing my meshes I found another error that I've not found on the forum:

ELEBD: POINT 9601
APPEARS ON
DIFFERENT BOUNDARIES
WRONG MESH

Investigating, I have altered the boundaries on my *.bc2 and deleted the same point on my mesh to see how it changes but the error remains. Phelype's mesh works fine, so I imagine it's nothing to do with how I'm prescribing the boundary conditions. Visually I couldn't see anything that appears wrong to me.
What could cause this?
The administrator has disabled public write access.

Crashing with more horizontal levels 7 years 1 month ago #27885

  • Phelype
  • Phelype's Avatar
  • OFFLINE
  • Senior Boarder
  • Posts: 140
  • Thank you received: 64
Hello,

You have to reduce your time step most probably because of the poorly shaped elements that are making the solution more difficult. When you improve the shape of your mesh the solver will allow a larger time step. I would suggest you try using the "T3 Mesh Generator" instead of the channel generator.

I used the "Auto-Adjust nodes" feature until clicking on it would stop having an effect (a few hundred times). Then I split a few triangles of the boundary to reduce their size and make them similar to the adjacent ones. Finally I removed the overconstrained triangles. Although they are not a problem, I think they are pointless, since they are pointless to the solution.

The error you are getting probably means that your .cli file is incorrect. It appears that some node appears twice on the .cli. Ty generating it again to see if the problem persists.

P.S.: If this error appears when you use my mesh, I suggest you discard it. I made a lot of modifications, so there may be something wrong with it.

Regards,

Phelype
The administrator has disabled public write access.
The following user(s) said Thank You: SDAC

Crashing with more horizontal levels 7 years 1 month ago #27916

  • SDAC
  • SDAC's Avatar
Hello,

I still get the segmentation fault with some of my meshes but Telemac works well on my other meshes it's not a concern for me for now. I still find it an odd error to have in relation to the mesh.

Phelype, your mesh worked fine. I'm using the more regular mesh above because I've specified the number of nodes at each cross-section of the channel. I've noticed that it takes longer to run and the mass doesn't quite converge. Reading around the forum, I take it the triangular meshes you get from "T3 Mesh Generator" are considered better overall compared to regular, triangle meshes (which is essentially what I've done by specifying the nodes)? e.g. greater flexibility in element generation.
The administrator has disabled public write access.
Flat Threaded
  • Page:
  • 1
  • 2
Moderators: pham
Forum
System Modules
TELEMAC-3D
Crashing with more horizontal levels

The open TELEMAC-MASCARET template for Joomla!2.5, the HTML 4 version.