DOE (Design of Experiments) for 3 Variables, Many Levels

Mike S.

Happy to be Alive
Trusted Information Resource
This one goes beyond what I remember from my DOE training so long ago!

Let's say I wanted to optimize a process that was driven by 3 variables. Variable A could be at 3 levels, Variable B could be at 8 levels and Variable C could be at 9 levels. Other than doing 216 experiments to cover every combination, is there a good designed experiment methodology that I could use to cover this territory in the most efficient manner possible?

Thanks in advance!
 

Miner

Forum Moderator
Leader
Admin
Re: DOE for 3 variables, many levels

Are the three variables continuous, and is there a reason that you want to evaluate ALL of these levels (other than being interested in curvature)?

If they are continuous variables and you are only interested in detecting curvature, I recommend starting with a 2-level factorial design with a few centerpoints. If the centerpoint is not significant, there is no curvature and no need to test all of the levels. If there is curvature, you can add axial points and convert it into a response surface design.

If the factors are discrete variables such as supplier A, B and C, and you actually need to test every single level you will need to use a general full factorial design.
 

Mike S.

Happy to be Alive
Trusted Information Resource
The variables are continuous, with some concern about curvature.

Thanks!
 
Top Bottom