Abnormal result with disallowed combination script in split plot design

Question

When I discussed with my customer about split plot design we found some abnormal result.
You can find my sample as below, and we try to build the design that with some constrains about factor.
In our case, we have 7 hard factors and 6 easy factors.

Basically, we want to design out the experiment have below criterion
1. when A=100, B=C
2. when A=10 or 50, B>C

In order to satisfy those 2 criterion, we put below script in disallowed combination script area(you also can find the script in attachment file ).

A == 100 & B != C | A != 100 & B - C <= 0

however, we found there are some experiments that obey the constrain we set, such as the row 29 with (A,B,C)=(10,50,50).

Then we design another experiment that only 3 factors A,B,C with above 2 criterion, and the result is perfect, so it look like the abnormal occur because we consider some harder factors in split plot design.
Can you help me to realize is this a bug or I miss some important split plot design information lead to this abnormal design result?

We try to design in 16.1.0 and 15.2.1, both with the same abnormal design result.

Split plot constrain question.jmp

Phil_Kay · Accepted Answer

Hi Eric,

I ran the same DOE Dialog in JMP Pro 17 EA7 and got a much nicer design (attached).

I suggest that you log this with technical support to see if this is a known defect that has been fixed in version 17.

Cheers!

Phil

Eric_Wang · Accepted Answer

Thanks Ryan, the information  is helpful.

And I want to check do you plan to fix the issue in JMP 16 series (such as 16.2.1 )?

Our customer use split plot design a lot and hope it can fix(or the work around method ) for current version.  
Thanks

Ryan_Lekivetz · Accepted Answer

Hi @Eric_Wang , unfortunately the date has passed unless we're made aware.

The issue should only arise with discrete numeric, so at least the customer can be aware when the issue may arise. One possibility is to treat those as categorical in the design phase to see how the design looks. Another alternative is to do the "fixing" approach, since it typically looks to be one or two rows that have the issue. If all Easy to change, yet another is to load a candidate set with allowable combinations only as covariates and allow the rows to be repeated.

Cheers,

Ryan

Ryan_Lekivetz · Accepted Answer

Hi Eric,

If you use the attached jsl in 16.2, where you have the attached data table open, it doesn't "know" about the disallowed combinations, but will run faster since it does not have to keep checking if runs are allowable. I replicated the allowable combinations for the discrete numeric 10 times and loaded them as covariates.

Cheers,

Ryan