1
$\begingroup$

I have a finite element model of a geometric structure which is computationally expensive to solve. The model is parameterized with 3 parameters. Each configuration is solved for an increasing load between 0 and load_max at n discrete loads. So the output (a specific property of the geometric structure) per configuration is a series of n values. I evaluated 3^3=27 combinations of these parameters.

Given the input combinations and the 27 output value series, how can I perform a sensitivity analysis to quantify the influence of each parameter on the output value and to determine which parameter combination can minimize the output value?

$\endgroup$
2
  • 1
    $\begingroup$ I'm not sure if this is properly described as "sensitivity-analysis". It seems more like standard experimental design issues in engineering. Are you assuming / worried about the possibility of interactions? If not, fractional-factorial designs are worth exploring. $\endgroup$ Apr 4 '15 at 15:56
  • $\begingroup$ Yes, I assume that there are interactions between the parameters. $\endgroup$
    – m.s.
    Apr 4 '15 at 16:01
1
$\begingroup$

With only 27 data points (and you haven't said how precise these values are), you should be concerned about overfitting the data. I'd start by fitting a simple quadratic response surface model and see whether the predictions look reasonable.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.