Modal Analysis with CE - how to choose Slave Node?

[meshparts]

Hello!

I'm dooing a modal analysis of a machine-tool. Due to gearing mechanism I use a total number of 5 constraint equations. I found very interesting that the block lanczos solver does not converge unless I reverse the initial contraint equations. That means the left side becomes the wright side and the left side becomes the wright side.

Theoretically and practically it should make no deference, since each constraint equation couple two equivalent machine parts.

Can anybody explain this to me?

Regards,
Alex

 

[Hup]

Hi Alex,
do you create the CE manually? I don't follow the explanation "the left side becomes..." I just know, that the form of a general CE is

Const = sum(ki*ui)

But probably your problem is not so simple...

 

[meshparts]

Hi Hup,

Thanks for your reply. I create the CE manualy. For example:

ce,1,0,n1,rotz,7,n2,rotz,-1

By reverting the CE I mean:

ce,1,0,n2,rotz,-1,n1,rotz,7

I don'n understand, why this makes such a big difference between convergence and not convergence. Do you have an idea?

Regards
Alex

 

[Hup]

Hi Alex,

then I do not understand either. I use similar CEs in harmonic response analyses of a ball screw drive, but I have never experienced such problems.

I'm sorry.
H-up

 

[meshparts]

Perhaps I should reformulate my problem. The output of the block laczos solution looks like this:

  LANCZOS CYCLE NUMBER =   1

     new shift:  3.9478D+03     modes still needed:        5

  FREQUENCIES AT CURRENT LANCZOS CYCLE
     1  0.54395877E+02     2  0.53812397E+02     3  0.40685055E+02
     4  0.39247235E+02     5  0.36196551E+02     6  0.28922018E+02

     number of steps      :     10
     eigenvalues found    :      6
     total no. eigenvalues:      6

  LANCZOS CYCLE NUMBER =   2

     new shift:  1.1557D+05     modes still needed:        0
 curEqn=     30  totEqn= 141778 Job CP sec=   1925.011
      Factor Done=   0% Factor Wall sec=      0.000 rate=    0.0 Mflops

It is obvious that the eigenfrequencies are computed but then LANCZOS CYCLE NUMBER 2 follows and the simulation don't goes further.

Sometimes I allso get a lot of LANCZOS CYCLES and the message "eigenvalues found : 0" appears, although at the firs cicle all eigenfrequencies were computed!

Had anybody experienced such a thing??

Regards,
Alex