topic Re: Prediction Profiler with factors and Responces in Discussions

Prediction Profiler with factors and Responces

lzeng2 — Fri, 29 Jul 2016 15:16:26 GMT

Hi Dear Feiends,

I am looking for some assistance. I have set of data with strength vs heat treatment temperature an material compositions. With various composition and different heat treatment temperature, you will get diferent strength. Through combination of composition and heat treatment, we can measure strength UTS. However, we are trying select desired composition so we can desired strength at dedicated heat treatment temperature. That is the goal. Right now, due to composition variation, we will get different strength at same heat treatment. That caused issue for us. For example, at 1100F heat treatment, we need strength UTS at 140, always. While at 1175F, we need strength UTS at 135, always. Then the question is, what is the ideal composition?

I tried Prediction Profiler. But I am not sure how to lock responce that is strength. I want to lock factor, heat treatment and responce, strength, so I can see the desired composition.

Really appreciate your help!

Liang

Re: Prediction Profiler with factors and Responces

Dan_Obermiller — Fri, 29 Jul 2016 18:25:07 GMT

I am not sure I completely understand your question. The data table that you provided has two models, both of which have problems since the model hierarchy is not followed and neither one of them has the heat treatment as a factor in the model. Both of your models are fitting across all heat treatments since it does not appear in the model. However, there are a few things that MIGHT be what you are looking for.

For a given response (strength), what are the factors that can give that response? There may be MANY compositions that could achieve a desired Strength. (think of a contour plot where one of the contour lines is your desired strength -- lots of possible combinations). You could actually use the contour profiler and set a contour line at the desired strength to see all of the possible combinations of the factors that could give you that response. You could also Fit Model and using the red popup menu choose Estimates > Inverse Prediction. That will force you pick settings for all of the factors but one so that JMP can determine that proper setting (remember, there may be many possible combinations, not just one).

If you were trying to use the Prediction Profiler and just need to lock a factor setting, hold down the ALT key and click on a profile picture. That will bring up a dialog where you can set the level of a factor and there is a checkbox to lock that factor at that setting so that it will not change when looking for an optimum.

I hope something here is helpful to you.

Re: Prediction Profiler with factors and Responces

Bill_Worley — Wed, 19 Oct 2016 03:31:30 GMT

Hi Liang,

Go to the Prediction Profiler red hot spot and select Desirability Functions. You will see an added set of blocks at the bottom of your profiler. To lock a variable value hold the control key and right click in the desirability box for the variable of interest and tick the Lock button and click OK. You will now see a solid red line for that variable. Lock any other variables that you want to be static the same way. Now you need to set a goal for the Strength UTS. You do this by control clicking in the far right box where the desirability goal is. In your case you can set a Match Target or None as the goal. I would suggest Match Target to start with. See the images below with examples.

Once you have this done go back to the red hot spot and select Maximize and Remember.

Reset your target value in your desirability function and then select Maximize and Remember. Remember will put a new row of data under your profiler and each time you do the Remember step a new row will be added. You can toggle back and forth between settings to see how things change.

If you want to save the Remember data right click anywhere in the Remember values and select Make Into Data Table.

Hope this helps,

Bill

Re: Prediction Profiler with factors and Responces

Dan_Obermiller — Fri, 29 Jul 2016 18:29:18 GMT

Ah, and another great (and better!) reply from BillW!

Re: Prediction Profiler with factors and Responces

lzeng2 — Sun, 31 Jul 2016 02:00:24 GMT

Dano and BillW,

Highly appreciated!

It works!

Have a nice weekend!

Liang

Re: Prediction Profiler with factors and Responces

lzeng2 — Sun, 31 Jul 2016 06:51:49 GMT

Hi DanO,

Is there way I can have multiple contour profiler at the same time? In reality, the required composition at dedicated temperature for desired strength is different. I am trying to put together to find common space. I also re-attach data.

Appreciated.