CCD design optimization

Question

Is there a way to output a table of solutions from a CCD using the profile after fitting the reduced model? Or you can only select maximize desirability and get one result? i'm aware if you move around the starting point maybe the algo will find an alternate solution.

I'm trying to get a table of outputs showing possible optimums within the response surface (an example from Minitab is attached). The dataset is a saddle shape. I'd like to show students how to find the peaks of the saddle. Attached is an example solutions table I know how to get from Minitab. The highlighted rows are similar results to what the paper found with ridge and canonical.

context if it helps:

I'm an engineering consultant helping companies apply DOE approaches to learn about their processes. I'm currently delivering classes using Minitab and I'm creating alternate modules to teach the same concepts using JMP. I'm working on an introduction to steepest ascent and CCD module. we run through the Box and Liu paper "Statistics as a Catalyst to Learning by Scientific Method Part 1. The JMP dataset is attached.

Box Helicopter Catalyst (Optional).pdf

Intro to Response Surface Methodologies.pdf

RSM DOE3b.jmp

awelsh · Answer

Not quite what I was looking for, but I do like this feature and will start integrating it into my analysis. I was not previously aware of this. Thanks

awelsh · Answer

I have used the simulator before in Monte Carlo discussions so I'm familiar with that feature. I did try some stuff with that but it seemed not right. I do like this visualization of the datapoints though. I haven't seen that before.

What I decided to do was just forget about getting a table of predictions and just show how to save the prediction formula to a column and then open it up with Graphs/Surface Plot. Seeing the response surface graphically and being able to spin it around and slice it with sliders I think is enough for this introduction. I'm not showing them any ridge analysis or getting into the canonical eigenvalues or anything with them at this point. So having an interactive response surface for them to spin around would be better anyway than looking at a table.

Victor_G · Answer

Hi @awelsh,

I might have a different suggestion.

You can try to fit a model, use the Profiler and open the Simulator.

From here, specify Uniform (or other distributions types) over your experimental space, any noise added to the response and the number of simulated points, and click on "Make Table" :

You will then have a new JMP datatable, and on this table there is one column "Obj" linked to the desirability of your response. You can then sort the simulated points by this column and highlight/export the best points :

You can also use the scripts added to the datatable to visualize the points of interest :

Hope this answer will help you,