|
Post by abdussamad77 on Sept 23, 2007 22:34:12 GMT -5
I have one set data with 3 variables (v1,v2 and v3). I was trying to use surrogate (RSA,KRG,RBNN and PBA) approach for optimization. [The surrogate are described here: pdf.aiaa.org/jaPreview/JTHT/2007/PVJA30211.pdf] I can not accept optimum design which produces 2*v1>v2, because of geometric constraint. I am using fmincon function (MATLAB), I tried to applied the constraint in fmincon: fmincon(@fun, x0, [2 -1 0], [0], [], [], lb,ub) but it did not produce desired result. If I do not apply constraint criteria, all the surrogates are producing optimum design with v2 touching the lower boundary of design space and this is violating the 2v1<v2 criteria. Any suggestion?
|
|
|
Post by Rafi Haftka on Oct 9, 2007 16:04:55 GMT -5
As I noted in a previous post fmincon is not a good optimizer. If you have to use it, you may consider transforming the variables so that the constraint will be simplified to a bound constraint. So, for example, if your constraint is 2v1-v2>0, you can replace v2 by v2'=2v1-v2, so that the constraint is v2'>0.
Rafi
|
|
|
Post by abdussamad77 on Oct 10, 2007 9:41:58 GMT -5
Dear Prof Haftka, Thanks for the suggestion. I will try as per your instruction.
Abdus Samad Inha University Republic of Korea
|
|