Optimization, Abaqus CAE, Python scripting 2

[AK_93]

Hello,

I am doing a project, in which I am running the Abaqus simulation through python scripting.
I have some experimental values and I am using these values to get some initial assumption. Then these initial assumptions are used as input in Abaqus simulation, after running the simulation the output i.e. the simulated values are compared with the experimental one that I have used to generate the input.
So after comparing experimental and simulated values I will get some error. I want to reduce this error by optimizing the input values of the simulation.
I am not getting any idea, how to optimize these input values in order to reduce the error in a certain limit.
Please suggest the possible solution to this query.
I will definitely appreciate any suggestions.

Thank you.

 

[IRstuff]

You need to do optimization:

https://en.wikipedia.org/wiki/Mathematical_optimization.

That presumes you have some knowledge of the shape of the response surface so that you can determine which direction to go to find the set of values that minimizes your "error." Alternately, you can brute force it by using Monte Carlo or similar scattershot approaches, particularly if you do not know how the system function really behaves mathematically

This brief thread might point you to something: thread724-477642 BFGS

https://en.wikipedia.org/wiki/Broyden–Fletcher–Goldfarb–Shanno_algorithm

does not require derivatives

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[AK_93]

Thanks for your response!

I am a little bit confused, I want to clear something, My objective function is error = simulated - experimental and to reduce this error I want to optimize the input values (two input variables) for that I want to predict these values.
I had made an initial guess and had obtained the error for initial prediction, but now how should I predict the next input values.
So is there any way to predict these input values?

 

[GregLocock]

if your new inputs reduced the error by 50% then a next guess would be twice as big a change.

But that is usually a bad approach, you are much better off finding out the sensitivity to each of your scaling factors alone, and whether there is an additional effect of both together.

So, ideally you run a matric of

lo lo
hi hi
hi lo
lo hi


Of course in practice we don't know how close we are to the peak, so the level of factor chosen may skip past the local optimum. So, given runs are cheap,and our time is not, we'd run the full 9 element matrix of L M H for each factor.


Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?