Multiple Objectives in a Single Optimization Study? 1

[Vikingbro]

NX10.0.2 Mach 3 Industrial Design Bundle

I have a model that I have a couple of associative measurements - one linear and one of the mass of the model.

Can I go into the Optimization command (Tools tab -> Optimization icon) and set-up a study that has as its objective to get both of these measurements to targeted values using several variables (expressions)?

I can set-up a study to do either of the targeted values, but I need both situations at the same time?

Any thoughts?

Thanks in advance for your assistance!

Chris Cooper
Senior CAD Specialist
Cleveland Golf/Srixon

http://www.clevelandgolf.com

 

[BubbaK]

Optimization packages (almost) always require a single objective function that returns a number. So, if you want to minimize two functions, you have to combine them into one, somehow. So, for example, if you want to minimize both f1(x) and f2(x), you could minimize f1(x)^2 + f2(x)^2. Of course, making f1 small might be more important than making f2 small. In this case, you could minimize w1*f1(x)^2 + w2*f2(x)^2, where w1 and w2 are "weights", and w1 is much larger than w2 (to reflect the heavier emphasis on f1).

 

[Vikingbro]

I found the answer to my question and it was really simple.

I created the study to optimize the linear measurement using four variables and I set the mass measurement as an upper constraint in this study.

Chris Cooper
Senior CAD Specialist
Cleveland Golf/Srixon

http://www.clevelandgolf.com

 

[TaylorAnderson]

Nice solution, Chris. Solid thinking.

...and BubbaK is correct in assuming that the NX design optimization tool currently only supports one objective at a time. I'd like to change that down the road, but it's not going to happen in the short-term.

Thanks, Chris!

Taylor Anderson
NX Product Manager, Knowledge Reuse and NX Design
Product Engineering Software
Siemens Product Lifecycle Management Software Inc.
(Phoenix, Arizona)

 

[BubbaK]

I wasn't talking about NX, I was talking about optimisation in general. The essence of optimisation (in the form of minimization, anyway) is to make some value f small. What if you have two values f1 and f2, and you want to make them both small. What do you do? The answer is that you combine them into a single function like f1^2 + f2^2. I can't imagine what it means to have more than one objective function, so I certainly don't expect this enhancement in NX anytime soon.

 

[PrintScaffold]

What if you need to take one big, one small? Or if few big, few small?

http://www.cadroad.com

 

[BubbaK]

I'll assume that "small" means close to zero. So, suppose you want to make f1 large and f2 small. Then you minimize k1*(1/f1)^2 + k2*(f2)^2. You choose k1 >> k2 if making f1 large is more important to you than making f2 small.

 

[PrintScaffold]

Thank you! Any information source you could recommend? I'm kind of interested in this sort of stuff.

http://www.cadroad.com

 

[BubbaK]

You can look at the Wikipedia page on multi-objective optimization, here:

https://en.wikipedia.org/wiki/Multi-objective_optimization.

The most common approach is to combine the different objective functions into one, somehow. This is called "scalarizing" the problem, and it's the approach I was suggesting. I said I couldn't imagine any other approach. It turns out that other approaches do exist, but their definition of "optimal" is pretty complicated.