The results v9 vs. v8 1

[AndyLake]

We have phased in a new version of NeNastran (V9). Earlier we used V8. The new version gives totally different maximum stress compared with version 8. We have used the same NAS-file for both Nastran versions.

How that is possible?!

 

[SWComposites]

Probably because the output settings in the Model Initialization file are different. These can be accesed in the NEiEditor. On you could be using different contour plot options. Be sure you are comparing the same thing - not element vs nodal stresses, element internal vs corner stresses, etc. An d don't just look at the contour plot output, rather look at individual element stresses.

What type of elemetns are you using? How are you making the comparisons? How much different are the stresses?

 

[AndyLake]

Ok, I have to check the output settings. The contour plot options are the same so it is not the problem. I am sure I comparing the same thing. I have made tests by using a few different models. The problem is the same with CHEXA or CQUAD4 –elements. Error depends on model. The model A maximum stress was 300MPa vs. 295MPa (quite little, but still…). The model B maximum stress was 200MPa vs. 350MPa !!! Also the stress distribution is different.

 

[fkmeyers]

What solution are you running? If it is nonlinear statics they changed the default for BISECT from OFF to ON and that may be causing the difference.

 

[dmacx]

What did Noran say? They may want your model or at least the details of it.

We have not observed any significant differences in the V8.4 verses V9 results. If you are comparing V8.0 to V9.0 then you are comparing a version that is over 6 years old and I would expect to see of differences in the QUAD results as I know they enhanced them a while back. If you are talking about V8.4 compared to V9.0 then try forcing QUADINODE and HEXINODE to ON in both the V8.4 and V9 runs and then compare. Like SWComposites said the differences are usually in the INI settings and they have a bunch of AUTO settings in there which may have changed behavior. BISECT was mentioned earlier.

 

[AndyLake]

What solution are you running?

-> Linear static

 

[AndyLake]

Earlier we used V8.4 not V8.0. Sorry.

The stress results are exactly same when the number of nodes is not more than about 50 000. Then I use the same model and increase the number of nodes by copying the elements. Now the results are not the same!

Small model -> exactly same results
Big model -> differences between V8.4 and V9.0

 

[AndyLake]

The newest test: A model has 110 000 Nodes and the type of elements is CHEXA.

The maximum stress when V8.4 is used: 732MPa
The maximum stress when V9.0 is used: 777MPa

Both of runs HEXENODE was OFF.

Then I turned HEXENODE -> ON and the maximum stress by using V9.0 was 754MPa.

We are going to take contact with noran next week.

 

[dmacx]

It is HEXINODE not HEXENODE. Turn ON HEXINODE in both cases. I would not use HEXENODE.

 

[dmacx]

Do you have any other element types in the model?

 

[AndyLake]

HEXINODE was AUTO in all runs. The AUTO setting will use the ON setting for linear static.

In the lastest test the element type was CHEXA. I have made the tests also with CQUAD4 elements and the problem was the same. The difference of maximum stress between V8.4 and V9.0 depend on the size of the model.

 

[dmacx]

Have you tried comparing runs with HEXINODE and QUADINODE set to ON for both V8.4 and V9? It would be interesting to see if that makes any difference.

It is strange that the error increases with a finer mesh density. In your test models, is there a large variation in stress over a few elements?

 

[AndyLake]

Have you tried comparing runs with HEXINODE and QUADINODE set to ON for both V8.4 and V9?
-> I have not tried that yet.

I noticed the difference between V8.4 and V9.0 when the version changed during the project. The model was large. Then I made a small test model (the plate and hole in the center of the plate) with CHEXA elements and I was surprised because the results were exactly same. Then I increased the model size by copying the elements (-> 40 plates connected each other and 40 holes). Now the maximum stress was different (V8.4 vs. V9.0).

In your test models, is there a large variation in stress over a few elements?
-> Yes. The stress gradient is high.

 

[NoranTech]

Dear Beginner77,

In NEiNastran V9 the HEXINODE/QUADINODE=AUTO setting will set HEXINODE/QUADINODE=ON for models under 100,000 degrees of freedom (this threshold can be changed by the analyst using the DECOMPAUTOSIZE setting in the .ini file). For models above this size, NEiNastran V9 will still use the ‘INODE=ON’ until after the stiffness matrix calculation where it is then turned off. In NEiNastran V8.4 HEXINODE/QUADINODE would default to ON regardless of the model size.

The HEXINODE and QUADINODE options add an internal node to the elements which will increase accuracy. However, for finely meshed models with small stress gradients through the elements, the INODE setting will not make a significant difference in the results, yet it will reduce the solution time with it set to OFF. If you want to keep consistent with V8.4 results you can permanently set QUADINODE and HEXINODE to ON in the Nastra.ini file (or Setup-Default Analysis Options in the Editor).

Thank you,

Jonas Dalidd
Noran Engineering, Inc.

http://www.nenastran.com

 

[AndyLake]

Dear Jonas,

Thanks, it helped!

We turned HEXINODE and QUADINODE to ON. Now the results are exactly same (V8.4 vs. V9.0). I made tests for several models. I thought earlier that HEXINODE and QUADINODE are always ON when AUTO is used. I didn’t know about the limits of 100 000 dof’s.

One model is made of CHEXA and CQUAD4 –elements (over 100 000 dof’s). We ran the model by V9.0 and HEXINODE and QUADINODE were ON. The results were consistent with V8.4. If HEXINODE and QUADINODE were OFF or AUTO, the maximum stress of the CQUAD-elements was lower and the maximum stress of the CHEXA-elements was higher. Illogical.

Usually the element with “extra nodes” is more flexible (more dof´s) so it’s illogical that the maximum stress of the CHEXA-elements rose when HEXINODE and QUADINODE were turned OFF or AUTO. The maximum stress of the CQUAD-element became lower so it behaves as expected.