BookmarkSubscribe
Choose Language Hide Translation Bar
Occasional Contributor

## JMP DOE Response Surface Model, why less combinations near upper limit region

see attachment, this is a results from custom design with response surface model. it seems the software generate less experimental combination at the region near the upper limiter of the factors. why this happens?

6 REPLIES 6
Staff

## Re: JMP DOE Response Surface Model, why less combinations near upper limit region

I don't believe that this custom design is for the RMS model for two factors.

• There would be only three levels for each factor.
• There are way more runs than are necessary for reasonable level of power and estimability unless the anticipated effects are much smaller than the anticipated standard deviation.

• How did you define the factors and the model?
• How many runs do you ask for?
• Did you request center points or replicates?
• Did you ask for a space filling design?
• Are Fe and Cr mixture components in the midst of other components and factors?

This design just doesn't make sense with the limited background information provided.

Learn it once, use it forever!
Super User

## Re: JMP DOE Response Surface Model, why less combinations near upper limit region

I would say that looks like a space filling design, but that doesn't even look right. Did you use covariate factors by chance? Can you post screenshots of what you did to generate the design?
-- Cameron Willden
Occasional Contributor

## Re: JMP DOE Response Surface Model, why less combinations near upper limit region

The attached picture show how I did it. But it is a example. in real case, there are 15 factors.

DOE setup I am searching for new alloys. There are 15 continuous factors, i.e the compositions of the elements. We have a very good model to predict the properties of the alloys. What I want to do with JMP DOE is to generate a lot of alloy compositions and then test it in our model to see which composition can give the best properties.

I am starter of JMP DOE. I found it seems Mixuture design with space filling may be a better choice. But I am wondering technically why less experiment points in the region near the upper limit were generated.

Occasional Contributor

## Re: JMP DOE Response Surface Model, why less combinations near upper limit region

The picture is just a example. I am designing a new alloys, there are 15 continous factors, i.e the compositions of the elments.

I used custon design. After define the ranges of each factor and the constraints, I simply click the RSM button in Model. and ask for 5000 runs (My idea is to fill the space as much as possible).

Occasional Contributor

## Re: JMP DOE Response Surface Model, why less combinations near upper limit region

Maybe I should be give more background.

I am searching for new alloys. We have a very good model to predict the properties of the alloys. What I want to do with JMP DOE is to generate a lot of alloy compositions and then test it in our model to see which composition can give the best properties.

Super User

## Re: JMP DOE Response Surface Model, why less combinations near upper limit region

I'm guessing it's because Cr and Fe are in your linear constraints.  There probably aren't many conditions, considering all the factors subject to your constraints, where you are able to have both of those mixture components in such high proportions to the total mixture at the same time.  That would mean you can't have much of your other metals in the alloy in that top-right region of Cr and Fe.

-- Cameron Willden