777703_en-US

liliconcom

Optimizing multiple functional responses?

Hi, 

a customer develops products which undergo different tests resulting in functional data. While they are really happy with FDE, they get stuck when finding the best conditions for all characteristics simultaneously, i.e. optimizing multiple functional responses.

While we can analyze each individually and create some profilers we can link, it would be nice to have something like "Fit Group" for FDE.

 

Does anybody have a solution/example/ideas/tips? 

 

Thanks

Florian

Re: Optimizing multiple functional responses?

To view multiple responses at one time, I've had success using the Functional DOE Analysis red triangle ... Save Columns ... Save Prediction Formula for each of the curves, then using the Graph...Profiler to view all these curves in one window. Example output below from an upcoming Discovery talk, with the functional data/curves in yellow on the left:

 

 

Re: Optimizing multiple functional responses?

Thanks @Jed_Campbell ! Looks like an interesting talk

Did you also try optimizing multiple functional curves?

 

Re: Optimizing multiple functional responses?

Ah, I see what you mean now. I hadn't tried using the optimize functions for multiple simultaneous curves. I took some time just now to try, using the Mill DOE file from sample data, since it has multiple curves in it. I hoped to save the desirability formulas for each curve (FDOE Profiler red triangle...Optimization and Desirability...Save Desirability Formula), then use the Graph...Profiler tool to find the best desirability, but when I try to save the desirability formula, JMP gives me a column formula error, then doesn't create the column.

That was a long way to say "I agree there needs to be a better way to optimize multiple curves."

Re: Optimizing multiple functional responses?

Of course, after hitting the 'reply' button, I had another idea, which ended up working:

1. In the profiler for each curve, use the simulator to output a random table, with all factors as random/uniform.
2. The random tables contain difference from target and integrated error from target. Rename these as unique to each curve.
3. Concatenate the data tables
4. Fit a model to the integrated errors for each curve
5. Set Desirabilities to minimize, and potentially prioritize each curve
6. Make sure profilers are linked, and Optimize

I'd call it a bit of a hack, but it seems to work. I've attached the final file w/ the model attached. Happy to explain further if needed.

Re: Optimizing multiple functional responses?

Hi Jed,

 

thanks, this works well! I made another example using completely different types of functional responses with the NMR + Mill dataset:

The file is very large and I am trying to create a better example in the next days. But I just heard from another customer as well, that they often take multiple functional measurements of a sample, which they then try to optimize for.

 

Thanks

Florian

Re: Optimizing multiple functional responses?

I'm also working on a blog post that will explain this a bit better. I just need to carve out time to finish/post it.

Re: Optimizing multiple functional responses?

blog post is here, for anyone needing to do this in the future, and internal wish list is here, for making this work natively.