Questions regarding sensitivity output

Torben0159

Hello,

I have been doing an optimization with Optistruct and now I want to do a sensitivity analysis. I optimized the gap between desired and current frequencies of some modes, using this formula:

delta_f = sum[(f_now_1 - f_desired_1)^2+...+(f_now_10 - f_desired_10)^2]

 

I used the OUTPUT cards to create .slk files for all iterations. I used normal orthotropic material behaviour, so I have 5 design variables and delta_f as the response. What I am looking for is something like "I change variable 1 by 10% and delta_f changes for xx%". If I do this for all 5 variables I will find out which has the most impact on the solution, right? How I understand the .slk file, I can say something like "the new value of variable 1 = 1,1* old_value of variable 1" and then it will calculate the new estimated response and I can see how much it would change.

Is my train of thought correct? If yes, which Iteration would i use for my calculations, since the sensitivities are changing? My guess would be the first iteration, because it usually has the highest change?

I have a mixture of positive and negative sensitivities, so I should use the conservative response?

Do the sensitivities have a specific unit, e.g. in %? Below is the slk file for Iteration 0, how do I interpret a sensitivity of 244 for E_P?

 

 

Community-Pasted-Attachment-1607609067494.png

Community-Pasted-Attachment-1607609511114.png

Adriano Koga_20259

Hi,

 

sensitivities are the variation of the responses with respect to your design variables change.

Their unit is equivalent to: delta Reponse/delta desvar >. if your response is stress and your variable is thickness, it will be MPa/mm, for example. How much stress will a change in thickness bring to my design.

 

You need to understand that sensitivities change in each part of your design space. Imagine them as the slope of a montain when you're climbing it. It will change as you approach the peak/valley.

So sensitivities are calculated close to a certain reference value/point and they're valid around this point.

 

In the end, close to the reference point, they will give you what you want: if I increase E_N by a unit, I will get a descrease (-16.9) in my Response. For E_P, if I increase by a unit, it would increase my Response by 244. Again, notice that this is valid for the surroudings of your reference point, which will change in each iteration.