DOE CCD

[#1]

Dear community, 

I am working with 7 factors, and one of them is a quality factor. I was planning to start with a screening analysis using a fractional factorial design (2^7-2) resolution (IV). I would like to add some center points to test curvature, but one of my factors is not a number but is quality factor (like material), I am not pretty sure how to deal with this factor and how to process it. Any advice will be valuable to me 

[#2]

Hi @William29,

Welcome in the Community !

Working with fractional factorial design seems a good idea regarding the number of factors you have.
Concerning your question with your quality factor, here are some ideas, options and questions to help you :

• A simple and straightforward approach would be to consider the quality factor as a categorical factor, with various levels : Low, Acceptable, Good, High for example. Depending on the number of levels, using this factor as a categorical one may greatly increase the required number of runs in your design.
• As the quality may be linked to some type of order (see previously the list) or ranking (1st quality, 2nd quality, 3rd quality), you could also use a discrete numeric factor for this variable. You would have to "code" and order the quality levels as 1, 2, 3 or other numbers (as the "gaps" between various quality levels may not be equidistant/homogeneous).
• Finally, I would perhaps consider examining what is driving this quality variable. Are there measurements or numerical characteristics to estimate the quality level ? If yes, you could perhaps use these direct measurements information in your design as continuous factors (possibly covariates as you may not have full control over the possible measured values and related quality levels).

No matter which option(s) you choose, using the platform Custom Design will help you include various factors types (numeric continuous with categorical, discrete numeric and/or covariates). You will also have full flexibility on the assumed model complexity, and you may add centre points (and replicate runs if necessary/possible), and choose an acceptable number of runs.

Hope this first discussion starter will help you,

[#3]

It is, of course, non-sensical to have center points for the categorical factor.  Here are some options:

1.  Start with just the 2-level fractional factorial.  This easily accommodates both categorical and continuous factors.  Analyze the results, then decide if you need to add center points, or move the space before augmenting the space.

2. You can use the continuous variables to create a center point and then use one level of the categorical.  This is then replicated for the other level of the categorical.  It's not ideal, but it will give you an idea about curvature.  Analyze the data graphically.

[#4]

before you go down the road of using a resolution IV design, maybe check out definitive screening designs. 

https://community.jmp.com/t5/R-D-Blog/Introducing-Definitive-Screening-Designs/ba-p/636147

A Custom Design would work too.