Design sensitivity analysis and design optimization 1

[frztrb]

Hi All

Since I am very new to the field of optimization , despite I have a conceptual and general idea , that design sensitivity analysis studies the rate of change in the response of a regards to one or more desing parameters , then optimization might be used , to find the best point where we have constrints on

I would like to know the methodology used in both methods , are the same ? I mean optimization has just added constraints to the derivatives and DSA is just one step before optimization ?

Also for a system I usually see only a DSA is done , but an optimization is not done , what is the approach for decision making ?

 

[GregLocock]

Depends. In non linear cases a gradient based optimisation may find a local optimum, but not the global optimum. So there are many strategies for automated optimisation, and performing a formal DSA may not be part of them. I don't know why you'd do a DSA unless you were intending to use it to optimise the design.

The key weakness with any optimisation is that ultimately you have to identify a single number to describe your design. Figuring out how to generate that number is probably the biggest part of an optimisation. However don't despair if you can't do that the first time round, you often learn a lot from failed attempts at optimisation (typically you get the answer, work out what it looks like, and say that can't be right because..., so you come up with a new scoring system that takes the new factor into account).

Cheers

Greg Locock


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

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

 

[frztrb]

Very Clear explanation , thanks a lot Greg , I am reading introductory papers about optimization , one of them mentioned that DSA can shorten the computation efforts of optimization , is it always correct , in my case , I would need to optimize the constitutive matrix according to the deflection(or stress?) they have , by changing the geometry parameters , like thickness and some few (2 or 3 ) others

 

[GregLocock]

"DSA can shorten the computation efforts of optimization , is it always correct"

From a practical perspective yes I'd probably include some form of sensitivity analysis both at the start and the end of an optimisation. At the start to eliminate the no-hoper factors, so I can reduce the experiment size, and at the end to ensure that whatever solution I have found by whatever method is actually optimised, that is, assume that I have found a near optimum solution and use sensitivity to pick the local optimum.

In well mannered experiments an approach like Taguchi DoE does not explicitly perform a single factor DSA, but the specific sensitivity to each factor actually forms the basis of the solution, and is available for inspection.



Cheers

Greg Locock


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

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

 

[frztrb]

thank you so much Greg