LouV & Ryan,
In my follow-up series of experiments, I'm considering two options:
hiding/excluding the worst X% results, and then starting an Augmented Design plan with the most skewed results not "clouding the picture"... and I need to explain the "hiding/excluding" wording. The worst results are remarkably visually unacceptable. Therefore the combination of settings that made them would, in hindsight, be Disallowed Combinations.
So the question I have is: Should I clear the design-space picture, exclude/hide really bad resulting runs, then Augment?
As a point of historical reference, a visual rating of "2" (1-5 scale, w/1=best, 5=worst ever), is better than about 95% of everything we've ever made. The 1-5 scale was set up with the understanding that a "3" was currently acceptable product, and anything rated < 3 would be considered an improvement.
Have a great weekend,
Thanks for the further detail. My only concern would be that when you narrowed down the likely factors was it done in DOE fashion? Since some factors that may be important due to interactions may have been dismissed. I only harp on this point because after a long career of running DOE's my only regret in retrospect was not to include more factors in my initial design and I wanted to share that experience with you. Also, if appropriate, always remember to include water as a factor since its understanding is probably responsible for 50% of my raises over the course of my career. If you are using JMP 7 I am surprised that you are able to run the DSD but if so that's great.
We narrowed down the factors available on the Ishikawa diagram (originally had 24 possibilities) to the most likely candidates through engineer's process knowledge + multi-voting. The multi-voting gave each team member 100 points to "assign" to factors based on their knowledge of the process under consideration. Then the highest voted factors were selected for DOE-Screening. Just after that I stumbled upon DSD. Brilliant stuff indeed!
I am not understanding your comment, "Also, if appropriate, always remember to include water as a factor since its understanding is probably responsible for 50% of my raises over the course of my career." We aren't using water in our process.
To build the 4 factor DSD, I made a (JMP7) Screening Design for the 4 factors. Then added one center point. After [Make Table], I manually changed two of the rows to match the DSD 'pattern'. However, since Ryan Lekivetz (see above) suggested using the 6 factor DSD, I am going to revisit the DSD design I was using and try to wrangle it (Augment Design?) into a 6 factor version DSD. Can JMP handle this?
Thanks for all the helpful information!
Sorry for the confusion about my "water" comment. Although water was not in many of my chemical processes it certainly reared it's head as a factor coming in with some of the raw materials or environment. With a such a low molecular weight versus some of the complex chemicals we were synthesizing knowing our sensitivity to water was important. So for instance if we were running our chemical process in Ethyl Acetate we would spike water in at low levels to understand it's impact at low levels. All of his may not apply to you but wanted to share some of life's experiences with you.
I'm with Ryan on adding two additional factors based on your multi-voting process since you will be running 13 runs and would gain information about them for free.