Nastran optimization with preload (centrifugal)

[Arnaud G.]

Hello,

I have a Nastran FE model of a blade, which I first managed to optimize in order to fit the best as possible experimental ground measurements (no rotation of the blade = no preload).
The .bdf file was roughly as follows:

SOL200
...
SUBCASE 1 $ subcase for which responses (DRESPi) are used in the objective function (DESOBJ) : minimization of combined errors of modal frequencies & shapes
ANALYSIS=MODES
DESSUB=120 $ I don't want more than +/-6% error on frequencies and I want a minimum MAC of 90% for every mode
...
SUBCASE 2 $ constraints on displacement for a specific static loadcase (at ground = no rotation)
ANALYSIS=STATIC
DESSUB=121 $ I don't want more than +/-10% of error on displacement @ some grid points
...

Now, here is the problem:

I know the modal frequencies of the part while rotating, and I would like to include this additional data in the problem.
I have tried several things:

#1 : SOL200 with a preloading static subcase + STATSUB

SOL200
...
SUBCASE 1 $ static preloading of the blade = rotation
ANALYSIS=STATIC
LOAD=99
...
SUBCASE 2 $ Modal analysis trying to restart from previous subcase to get the modified stiffness
ANALYSIS=MODES
STATSUB=1
DESSUB=110 $ I don't want more than +/-10% error of modal frequencies while rotating
...
SUBCASE 3 $ subcase for which responses (DRESPi) are used in the objective function (DESOBJ) : minimization of combined errors of modal frequencies & shapes
ANALYSIS=MODES
DESSUB=120 $ I don't want more than +/-6% error on frequencies and I want a minimum MAC of 90% for every mode
...
SUBCASE 4 $ constraints on displacement for a specific static loadcase (at ground = no rotation)
ANALYSIS=STATIC
DESSUB=121 $ I don't want more than +/-10% of error on displacement @ some grid points
...
...
RFORCE,99,etc...

This doesn't work, since in optimization solution, STATSUB is only permitted for a BUCKLING optimization. [I don't understand why this couldn't be possible for a preloaded modal optimization also...]


#2 : SOL400 with SUBCASE including two STEPs

SOL400
...
SUBCASE 1 $ static preloading of the blade = rotation
STEP 1
ANALYSIS=STATIC
LOAD=99
...
STEP 2 $ Modal analysis trying to restart from previous step to get the modified stiffness
ANALYSIS=MODES
DESSUB=110 $ I don't want more than +/-10% error of modal frequencies while rotating
...
SUBCASE 3 $ subcase for which responses (DRESPi) are used in the objective function (DESOBJ) : minimization of combined errors of modal frequencies & shapes
ANALYSIS=MODES
DESSUB=120 $ I don't want more than +/-6% error on frequencies and I want a minimum MAC of 90% for every mode
...
SUBCASE 4 $ constraints on displacement for a specific static loadcase (at ground = no rotation)
ANALYSIS=STATIC
DESSUB=121 $ I don't want more than +/-10% of error on displacement @ some grid points
...
...
RFORCE,99,etc...

This runs, It doesn't crash, but from a certain time, nothing happens (no modification of any file), and I have to kill the process.


#3 : DMAP

I am currently digging into DMAPs, and this is the trickiest way for the moment... I did not find the solution yet.
I am trying to modify the stiffness matrix for a particular subcase (the subcase for which Analysis=Modes & with RFORCE). However I am facing a problem with the design sensitivity phase, leading the variations of the design variables...If I do not modify the right SubDMAP at the right position (ALTER), this won't obviously work...


#4 : Running the optimization outside of Nastran, and just calling several jobs of Nastran, and reading results in f06 or f04 files

This is the last solution I would attempt, only in case nothing else works...




So...Does anyone can help me for this ???

Thanks in advance !


Arnaud

Nota : I have found several examples of centrifugal forces integration during optimization, but It was for quite older versions of Nastran, and with an external optimizer...

 

[dmapguru]

Hi Arnaud,

You wrote "I know the modal frequencies of the part while rotating, and I would like to include this additional data in the problem.". So, my question is, is the blade the only component in the model and if not, do the design variables change the modes of the blade, or just the other parts of the structure?

 

[Arnaud G.]

Yes, the blade is the only component. I consider the blade foot clamped, and FE results are very close to ground tests (no rotation) results.

The global reason of my problem is that I could find several FE Models as results of optimization (by selecting different design variables, with different bounds, and playing with the optimization parameters) to best fit ground experimental results, but not all of them would sufficiently match the experimental results with rotation...