Practical aspects of optimization

[hky20]

I think optimization is great and I love doing them but in the back of my mind there are always a few questions about how useful the results actually are in the context of a greater design programme.

Say someone is designing the casing for a combustor in an aircraft engine and he uses optimization to minimize the casing's weight subject to some stiffness constraints. If a substructure/breakout FEA model is used to compute the casing stiffness, and you get your boundary conditions from a global loads analysis, wouldn't those boundary conditions be invalid as your design moves away from the state of the system at which the GLA was done? Do you have to run a new GLA for every new casing design you want to assess? It seems incredibly inefficient to make design decisions that way/feed an optimization process with inaccurate data.

The other question I had was how do you account for the possibility of breaking another component when carrying out your optimization. Say the combustor casing is connected to a turbine casing, and the turbine casing also has its own stiffness constraints. How do you know if your optimum combustor casing design won't violate those constraints if they are not in your FEA model? Is the only way to just have additional constraints from a systems engineer to prevent this from happening?

 

[rb1957]

i gag, literally gag, whenever anyone says "optimization". The most you can do is to qualify that with "at this point in time, with this set of data, for these constraints".
how do you optimize for production ? when prices and trade-offs change daily ?
how do you optimize for maintenance ? when maintenance costs change daily ?

2nd para ... yes, this is a real world issue with "optimization" ... any change changes the optimal solution. As you say "incredibly inefficient". but "inaccurate data" is misleading ... there is no truth out there, only the data we use as inputs.

3rd para .. you should run intact structure and failed structure. intact has ultimate load applied, failed has limit load.



another day in paradise, or is paradise one day closer ?

 

[IRstuff]

The "assumption" for this type of process is that any optimization is locally flat-ish, i.e., the optimum "point" is not a spike, so within some range of parameters, you haven't fallen off the optimized point. You then iterate on the new set of parameters to confirm that you're still relatively optimized. At the end of the day, that's the basis of most approaches to solution finding, even Newton-Raphson, In fact, many routines "break" if the "optimum" solution is a spike, since that results in discontinuities and solvers tend to fair poorly in those cases.

3rd para -- yes, you should have the constraint of whatever you attach to; otherwise, that's a bit equivalent to design an "optimized" building without regard to foundations or bedrock.

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

 

[SWComposites]

Yes, you need to rerun the GLA.
Yes, you need appropriate constraints or boundary conditions, or the resultswill be rubbish.

 

[hky20]

Thanks everyone! I think the bigger question I had in mind for the 3rd paragraph is that where does the buck stop when it comes to assessing a design change in a component. What I can gather from the answers is that even if you make a design change only to the combustor casing you have to simulate the entire engine, and possibly even the entire aircraft, in order to see if it breaks any other component, which seems quite extreme to me.

Is this more of a design issue, where perhaps the scope of the design change is restricted and a safety factor is introduced to account for that possibility of breaking the rest of the system?

 

[rb1957]

Where I work, we consider the structure as far as we feel that the change is having a significant affect. Yes, I know ... a lot of subjective words.

Sometimes, like mounting an antenna, you can reasonably say the antenna would affect the local fuselage only, and distribute the effect of the antenna into the adjacent frames and fuselage skin, and show these effects are small. And, yes, there's a bunch of work to address buffet.

Sometimes a more extensive mod would require us to consider the entire fuselage.

If you're changing a combustor casing then the immediate engine is important. Probably most important is the impact the design change has on the combustor's performance. I'd expect that a combustor has negligible effect outside of the engine casing (unless you've effed up the engine performance which is clearly a different story). How optimal is this design ? Then depends on your design constraints and the design data you have. You could say its an optimal design 'cause you minimised the design time to do it.

another day in paradise, or is paradise one day closer ?

 

[hky20]

Ahh I see, so something like a staggered approach to doing the assessment, where if the nth "adjacent" component isn't significantly affected then you assume that anything beyond that is also not affected? That does sound quite practical!

 

[IRstuff]

As a point of reference, we recently tested an external antenna to a system for vibration, and it had been tested before, albeit rattling pretty hard, but modifications were made, so we shook it and it rattled as before, but this time it broke a secondary metal mount.

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

 

[SWComposites]

Optimization is a tool. It is not a substitute for thinking. Sadly this all too often not well understood or acknowledged.