DOE- Custom Design with Blocking

[#1]

I am a very new to JMP software and need help to develop a table for DOE. I have three factors at two levels with two blockings, Operators at four level (Operator, 1 to 4) and Cavities at four level (Cavity, 1 to 4). Each operator will check three samples from each cavity from each run for breakage or no breakage. So, with three factors at two leveles, four operatores, four cavities, there will be total 384 runs. How to make the table from Custome Design

[#2]

*I've changed my recommendation a couple times here.  I think the current recommendation is the cleanest approach.  Also, previously stated specifying 96 runs, should be 128

Hi @JG,

Welcome to the community!  It sounds like you have a specific full factorial design in mind (4x4x3x2x2x2), correct?  In each run, each operator should be taking 12 total samples correct?  And across all operators, each run should have 48 samples.

If that's all true, I think you actually have 3 levels of blocking: 8 blocks for "run" (48 samples in each run), 32 blocks for Operator (operator is held constant for 12 samples at a time, 4 operator blocks within each "run" block), and 128 blocks for samples from the same cavity (4 blocks of size 3 within each operator block).  This would be a split-split plot design with repeated measures.  That's pretty complex!

There's a couple ways to approach this.  If you want to use Custom Design, you can do that by specifying a split-split plot structure with your 3 2-level factors as "Very Hard" to change, Operator as "Hard" to change, and Cavity as "Easy."  Model should probably contain all 2-way interactions.  You would then specify 8 whole plots, and 32 subplots, and 128 total runs (you'll need to cross this design with your 3 repeat measures from each cavity to get your 384 total data points).  Your design dialog should look something like this:

I was able to get the full factorial for the 128 runs with the default settings, but you can increase the number of random starts or design search time if that doesn't work out for you the first time.  You can tell if you got it if the Correlation Map looks like this:

 Now, you'll just need to add your 3 repeat measures to your design.  You can do this a couple ways.  Easiest is probably just add 3 new columns (e.g. Sample 1, Sample 2, Sample 3) and then stack those 3 columns.  Your table will then have 384 rows.

I attached my result using this method.

[#3]

Hi Cameron,

Thank you very much for your prompt response!  But it seems to be little more complicated than I thought. But the experiment is not so complex. I try to explain what I am doing.

There is a molded plastic part being manufacured by varying three factors at two levels .There are four cavities in the die. There are four inspectors, who will inspect the finished parts for any breakage. Each inspector will inspect 3 samples from each cavities from each run (12 samples/run for  four (4) cavities). I am using inspectors and cavities as blocking factors.  I was thinking that I would be able to generate that easily. But it is not so as you mentioned.

JG

[#4]

You're right, what you're actually doing is not that complex.  However, to properly account for the structure you are creating in the data, that part is actually a little complex.  At the highest level, you have 8 runs for your 3 2-level factors.  Below that, you have multiple levels of repeat measures.  That is where the modeling and design get interesting.

Do you actually want to model the operator and cavity fixed effects or do you just want to account for their variation?  You can just make operator and cavity random effects.  That would allow you to get rid of the WP and SP blocks.  To make that table, you could generate your 8-run design for your 2-level factors and cross it with your repeat measure structure (Operator, Cavity, 3 repeat measures).

[#5]

Hi Cameron,

Thanks once again! I want to use the blockings as a fixed effects. Let me try to make the table what you suggested. If I have any further issues, I will post again in the site.

Regards,

JG