Discussion Groups

Robot Structural Analysis

Reply
Distinguished Contributor
Tuctas
Posts: 172
Registered: ‎10-10-2012

Solvers

433 Views, 4 Replies
01-22-2013 10:39 PM

  I would like to have an explanation that has to do with the obtaining results of modal analysis when using different types of solvers. 

 

  In a not really complex structure that I work (a 3D frame), I firstly used the “automatic” option when choosing the solver but there was no convergence and the analysis was not completed, i.e was interrupted (after a few warnings of instability type2 e.tc). I didn’t notice in the report which solver was automatically chosen by the program and the next step was just to choose manually the Skyline method after which the analysis was completed (a “simple” warning about instability of type3 of a node occurred). In the results of modal analysis, the period of the first mode was abnormally high (1865 sec! although the participation percentage was very small). After that, I choose “Multi-threaded solver” and the analysis was ended very fast and without any warning at all. The results of modal analysis were normal and generally everything was o.k.   

 

  When I made one more analysis while choosing again the Skyline method, the results were normal (the abnormal fundamental period that occurred at the first time was disappeared) and they were the same with the results of Multi-threaded method!

 

  Is generally the Multi-threaded method the most preferable option in modal analysis as concerns small to middle range structures?

 

  In the attached picture, at the bottom, there is a value of "precision". Which value should be acceptable? 

Please use plain text.
Product Support
Artur.Kosakowski
Posts: 4,744
Registered: ‎12-17-2010

Re: Solvers

01-23-2013 12:09 AM in reply to: Tuctas
For Auto settings Robot uses Sparse_M solver for larger and Skyline for smaller size models. One should not run modal analysis for models with reported instabilities before determining their exact causes and making sure they can be ignored (not influence the results).
Precision displayed in main part of full calculation note corresponds to the number of significant digits during
stiffness matrix decomposition.
Its definition is:
integer part of (17.0-log(MaxVal/MinVal)-log(sqrt(Neq))
where:
MaxVal, MinVal - max and min values on the diagonal of stiffness matrix after
decomposition (given in calculation report)
Neq - number of equations i.e. number of static degrees of freedom
If you find your post answered press the Accept as Solution button please. This will help other users to find solutions much faster. Thank you.


Artur Kosakowski
Please use plain text.
Distinguished Contributor
Tuctas
Posts: 172
Registered: ‎10-10-2012

Re: Solvers

01-23-2013 02:41 AM in reply to: Tuctas

  Thank you Artur,

 

  Gernerally in Robot, sometimes some instability messages (of type 3) occur that are not always justified so we just ignore them, and some other times if we change the solver they don’t appear again.

  In the model that i was reffering there wasn’t any kind of complexity, it was a simple 3D frame with some pinned releases at their ends (the instability of type 2 couldn’t be justified at all). I could send you the model but i am not sure that you will reproduce the initial messages because after i choose Multithreaded method everything was o.k even if i returned and used the first solver (Skyline)..

 

  Anyway, i would like to know about the Multithreaded method which times it is suggested to be used and if we can use it in general, because i noticed that it is more stable and quick (are there any “constrains” in using it?).

 

  Furthermore, just from your experience, do you think that a value of precision equal to -4 could be a warning that the modal results aren’t precise?

 

    

Please use plain text.
Product Support
Artur.Kosakowski
Posts: 4,744
Registered: ‎12-17-2010

Re: Solvers

01-23-2013 05:22 AM in reply to: Tuctas

  Gernerally in Robot, sometimes some instability messages (of type 3) occur that are not always justified so we just ignore them, and some other times if we change the solver they don’t appear again.

 

If you check what are their reasons and then determining them ignore then OK :smileyhappy:

 

 

  In the model that i was reffering there wasn’t any kind of complexity, it was a simple 3D frame with some pinned releases at their ends (the instability of type 2 couldn’t be justified at all). I could send you the model but i am not sure that you will reproduce the initial messages because after i choose Multithreaded method everything was o.k even if i returned and used the first solver (Skyline)..

 

Send me this model please.

 

  Anyway, i would like to know about the Multithreaded method which times it is suggested to be used and if we can use it in general, because i noticed that it is more stable and quick (are there any “constrains” in using it?).

 

Due to its nature it has less capability of reporting instabilities so I would run the test (e.g. one load calculations) with other solver first unless you are positive that the model is correct.

 

  Furthermore, just from your experience, do you think that a value of precision equal to -4 could be a warning that the modal results aren’t precise?

The negative value is suspicious.

 

    


 



Artur Kosakowski
Please use plain text.
Distinguished Contributor
Tuctas
Posts: 172
Registered: ‎10-10-2012

Re: Solvers

01-25-2013 05:43 AM in reply to: Tuctas

  I post our last discussion with Artur, maybe will help someone with similar questions.

 

“The result that it is still strange to me is that the (torsional) 1st mode has completely zero mass participation (percentage of massUx=massUy=0.00%)?!”

Artur: That was the surprise for me too when I noticed that for the first time but looking at the formula that describes the participation mass calculation this is the correct result.

 

“Do you generally suggest using firstly a solver like Skyline just to investigate if there are instabilities and after that to use the Multithreaded method?”

Artur: Correct.

 

“Does the Multithreaded method have any other “disadvantage”?”

Artur: I’m not aware of any others.

 

 And a simpler question: how can we just run the calculations for a selected only load case and not for all of the previously defined cases/combinations?”

Artur: Set the other as auxiliary. Mind that for the linear static you cannot set the cases that are included in the definition of the combinations as the auxiliary for the results of this combination are based on the superposition of the results of its components.

 

“Do you mean that if a load case is included in a combination, it cannot be set (the load case component) as an auxiliary case, or that in such case the results of the combination will not be correct?”

Artur: If this is linear static analysis (combination) then the result of combination will be N/A. You can do so if you have the non-linear analysis.

 

Please use plain text.