Taguchi DoE - strength of interactions

[GregLocock]

In real life cases where the process is already reasonably well centred, has anybody come across systems where interactions are as strong as the individual factors? That is where the sensitivity to A.B is as great as A or B alone? I'd like to exclude chemical reactions or other 'obvious' examples.



Cheers

Greg Locock


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

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

 

[btrueblood]

Yes, with fluid flow and mixing, i.e. small changes in parameters causing very nonlinear changes to outputs. Our Taguchi instructor had us use DOE for design of paper airplanes to achieve maximum distance using a rubber band for launching - you can imagine the butterfly effects inherent in that - pretty much a waste of time.

 

[GregLocock]

I've just realised I may be describing a tautology, perhaps by definition a well centred process is relatively insensitive to interactions.

Hmm, yes designing aeroplanes is an example where interactions are important, but I wonder if that is true for small perturbations in design that is already near an optimum? Once upon a time I was discussing the mad enthusiasm for DOEs that had taken over the company with a friend of mine and either me, him or the beer said that "You should only use DoEs when you know everything about the system, or nothing"

The example I'm looking at is crosswind stability for an SUV, I'm changing the yaw and roll moment coefficients and the side force coefficient, vs yaw angle, (and the other 3 but they don't matter) and I'm only getting very weak interactions, which is great because I can go and ask the aero people to change specific things, but I then started wondering how often other people see strong interactions.

Cheers

Greg Locock


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

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

 

[IRstuff]

I don't think there is a tautology, per se. "Well-centered" is likely to be due to control of the process, since it's unlikely that you'd find a process that walks itself back from the edge of failure on its own.

In the case of many systems, proportional control means that the closer you get to the control point, the less effect the feedback mechanisms have, which applies to any conventional nulling system.

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

 

[GregLocock]

I think the shape of the response surface is a nice way to visualise it.

So, in a simple 3d surface of optimisation between 2 factors, near the peak the surface tends to be well rounded, and does not have any clefts or spiky ridges near it (those are interactions), for a general mechanical system. Whereas with a chemical reactions between two reagents, calculating %age in a solution say as the response, the response surface is an almost perfect ridge along the ideal proportions for the two reagents. Sounds good.

So for general mechanical type problems, are there practical examples where the interactions are as strong as the primary effects?




Cheers

Greg Locock


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

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

 

[IRstuff]

At least, in some very niche things, possibly. It would have to result from some highly non-linear behavior, like a cam or switch.

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

 

[btrueblood]

Greg, I'd think that many people reporting strong cross-interactions in mechanics are a result of poorly posed questions, or poorly chosen variables. Think of designing/testing a bearing, where the variables include the shaft od and bearing id, obviously the gap (od minus id) would be a rather important variable on its own.

In your SUV, you'd probably see pretty strong coupling if the corners of the vehicle were sharp vs. rounded, or if steering was set up to have a lot of adverse roll coupling. Or if you put a giant rudder vane on the roof?

 

[GregLocock]

The bearing clearance example is a good one. I'm nearly tempted to go back through my old DOEs and try and assess the interaction strengths. Sounds like a matlab job- I have at least 50 !

Cheers

Greg Locock


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

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

 

[btrueblood]

50 jobs, or 50 bearings? :0

 

[GregLocock]

50 DoEs

Cheers

Greg Locock


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

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

 

[btrueblood]

Wow, cool. Maybe you could publish something that shows relative interaction strength vs. discipline (unless these are all mechanical DoE's?)

 

[GregLocock]

Without checking, I'd guess 95% or more were concerned with kinematics, compliances, spring/ sta bar set ups, or tire properties. It's a very small subset of the world!

Cheers

Greg Locock


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

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