Now there are nonpositive jacobian

[sirpete83]

Hallo,
i have a new problem.
The story behind.... some months ago I simulated a part with its eigenmodes. It worked perfect, but i do not remember the exact version of freecad.
Because of other projects I use always the newest daily build.

Now I have to simulate the old part again, no problem i thought. But it is ... there are now non positive jacobians !?! in the same file... whats going on .. spend 8 hours on it now I'm here.

gmsh wtihout errors or warnings.
cgx without bad elements
ccx nonpositive jacobians

Any ideas?

Thanks

sono1.FCStd

eigenmode problems

(669.23 KiB) Downloaded 42 times

[bernd]

your file runs fine for me on ccx 2.11

screen.jpg (196.73 KiB) Viewed 2379 times

But I had some problems to on simple shapes too with frequency. See https://forum.freecadweb.org/viewtopic.php?f=18&t=22673 Try a static analyais by fixing some face and applying a force or a pressure. Does it helps.

bernd

[bernd]

There is another one you could try. Use the netgen mesher. It produces better tetra meshes than gmsh, ccx has less problems with.

[sirpete83]

Hallo bernd,

do you have changed anything on the mash, that was in the file, or did you only start a new simulation?

Yes, that is the result I had before, now it doesn't work any more... I tried different versions of ccx. 32 bit - 64 bit MT an non MT.
The fact is, that it works a few weeks ago. But not at the moment.

I testet it on 2 different machines, later on Linux.

Now I think the problem is in front of the Screen

I will check it again.. and will check out, how to use the netgen mesher

Thanks a lot for this moment.

[bernd]

download your file, start FC, load your file, activate the analysis, select the solver, press the run button.

BTW: you are not the first one spending hours on non-positive jacobians. I implemented some highliting. If you use the run button the non-positve should be green. See screen here: https://forum.freecadweb.org/download/file.php?id=38080 Does this word for you. Do you get lots of them?

Mostly it helps to make a finer mesh or add some loads an a fixed face and have a look if a static analysis works.


Bernd

[fandaL]

I can reproduce the problem. It helped when I changed position of the mesh in property editor to move out of axes (to have mesh without nodes with some coordinate being 0). It also helped when I changed positions coordinates like

Code: Select all

*Node, NSET=Nall
...
130, 85.06, -14.75, -3.09452848545541e-014

to

Code: Select all

*Node, NSET=Nall
...
130, 85.06, -14.75, 0

From this perspective, it looks like problem is in the formatting of very low numbers in the inp file.

OS: Windows 7
Word size of OS: 64-bit
Word size of FreeCAD: 64-bit
Version: 0.17.11203 (Git)
Build type: Release
Branch: master
Hash: e2e1f44b45737f8b900a4ac7766f8f064c434638
Python version: 2.7.8
Qt version: 4.8.7
Coin version: 4.0.0a
OCC version: 7.1.0

[UR_]

Code: Select all

*Node, NSET=Nall
...
130, 85.06, -14.75, -3.09452848545541e-014

Hello fandaL,
"-3.09452848545541e-014" is an illegal floating point format
ccx reads only F20.0 (i. e. Fortran floating point field 20 chars wide)
pending chars are not accepted, that means -3.09452848545541e-014 becomes -3.09452848545541e-0
That should be avoided.

BTW, i got these jacobians all day long and spend a lot of time figuring out whats going on.
But i never counted chars

[bernd]

Does it mean we changed the prececision to a to much hight value here: https://github.com/FreeCAD/FreeCAD/blob ... .cpp#L1191 done git commit e58b7479

[sirpete83]

Hey all,
thanks a lot for your fast support!
This workaround works fine for me!

Peter

[UR_]

Does it mean we changed the precision to a too much height

Hello bernd,
there is a overhead in iostream caused by signs, decimal separator and exponents.
please have a look at this example

Code: Select all

// floatfield
#include <iostream>     // std::cout, std::defaultfloat

int main () {
  double a = .1234567890123456789;
  double b = -.1234567890123456789;
  double c = 100000000000000000000.;
  double d = -100000000000000000000.;
  double e = -0.1234567890123456789e-101;
  double f = +0.1234567890123456789e+101;
  
  std::cout.precision(13);

  std::cout << "defaultfloat:\n" << std::defaultfloat;
  std::cout << "12345678901234567890\n";
  std::cout << a << '\n' << b << '\n' << c << '\n' << d << '\n' << e << '\n' << f << '\n';

  return 0;
}

output:

Code: Select all

defaultfloat:
12345678901234567890
0.1234567890123
-0.1234567890123
1e+020
-1e+020
-1.234567890123e-102
1.234567890123e+100

for reference please have a look at: http://www.cplusplus.com/reference/ios/defaultfloat/