topic Re: Custom DOE - Issues with Continuous Factors, Runs only at Extreme Value in Discussions

Custom DOE - Issues with Continuous Factors, Runs only at Extreme Value

csf — Tue, 19 May 2020 20:56:49 GMT

I have a custom experiment I'm setting up.

I have 5 distinct surfaces I'm testing on. A, B, C, D, E.

I added a second factor, "load" as continuous from 5 to 15

I added a third factor "hold time" as 5 to 30

When I run "Make Design" Im surprised to see multiple runs of the same combination with all values at the extreme of the 2nd and 3rd factor. Is there a reason it doesn't have me test any intermediate combination?

Thanks in advance.

Re: Custom DOE - Issues with Continuous Factors, Runs only at Extreme Value

statman — Tue, 19 May 2020 21:26:33 GMT

If you set the 2 additional factors to 2-levels (5, 15 and 5, 30), then the experiment will only provide treatment combinations at those 2-levels. If you want to test the factors at more than 2-levels, you must specify this when you add N Factors.

Level setting depends a lot on where you are in your study. 2-level designs are quite typical in early investigations (often called screening designs). The levels are chosen far enough apart to have a wide inference space. Essentially you are increasing the chance of seeing a factor effect, of course, this is done for all of the factors to mitigate bias. 2-level designs will allow for first order model development as well as the ability to estimate interactions provided the design has enough resolution. As your study approaches "optimal", then typically some sort of RSM is used often with more than 3-levels for factors thereby being able to estimate complex non-linear relationships.

Re: Custom DOE - Issues with Continuous Factors, Runs only at Extreme Value

Dan_Obermiller — Wed, 20 May 2020 13:10:08 GMT

A slightly different way to look at this: JMP will choose the design according to the model that you want to estimate. If you have only main effects, then you only need 2 levels to estimate those straight line relationships. If you require curvature (like squared terms), then you will get 3 levels for the factors.This is where you get efficiency: only perform the levels required to estimate the desired models.

Re: Custom DOE - Issues with Continuous Factors, Runs only at Extreme Value

csf — Wed, 20 May 2020 14:53:27 GMT

If you want to test the factors at more than 2-levels, you must specify this when you add N Factors.

Do you mean manually adding additional level at the mid point? So instead of 2-level, I should choose 3-level?

Re: Custom DOE - Issues with Continuous Factors, Runs only at Extreme Value

statman — Wed, 20 May 2020 17:05:00 GMT

As Dan suggests, you can add quadratic terms to the model you want to estimate, or you can add Discrete Numeric 3-level as in attached pic.

Re: Custom DOE - Issues with Continuous Factors, Runs only at Extreme Value

statman — Wed, 20 May 2020 17:04:15 GMT

Screen shot of adding "powers"..2nd.