Functional Inputs for Non-functional outputs in FDOE Examples?

Question

Hi All,

I have a customer building a business case for JMP Pro, they've got lots of value statements built up for FDE in terms of FDOE of non-functional inputs and functional outputs (specifically for bioreactor process development). Has anyone got any good examples for functional inputs being used (or even functional in and functional out?) that I can share where it has driven value/new understanding?

Jed_Campbell · Accepted Answer

Here's an example of how to do it.

Areas where it is useful include particle size distributions on incoming powders for sintering processes and for mining ore recovery processes.

Peter_Hersh · Accepted Answer

Hey Ben, Great question.   The case I have heard most for functional Machine learning or functional factors is determining if a curve is normal or a certain curve can be predictive of a result.  The first time I sold FDE the company was looking at a vacuum pressure gauge during a process.   They knew what the curve looked like when they made a good product and how that was different when they made a bad product.  This allowed them to flag a "bad" batch early in the process.  Here is a quick video that @Jason_Wiggins and I did for mastering JMP.  Identifying the Impact Curved Factor Shapes have on Responses - JMP User Community

For the functional factors and responses that is a challenge @HadleyMyers  and I put together a video on this, but it is a pretty complex process.  We used an example of someone monitoring their glucose level while fasting then after eating a selected snack.  This was a real example from General Mills who was trying to develop diabetic friendly snacks.

New FDE demo - Continuous inputs and outputs - JMP User Community

martindemel · Accepted Answer

Hi Ben,

where FDE also is used with functional inputs and outputs is in  bioprocessing. Think about a Titer Process where you have Online and Offline data. Usually you just catch the starting values or several time point specific snapshots, e.g. after the initial phase where the growing finally starts.

Lonza used both data streams to find a model predicting online data from the offline data good enough. Then they could predict just from the offline data the online data much earlier than the full process (after 40h vs. 120h). Not sure they finally implemented it but they did test runs and verified that the model worked surprisingly accurate. They should present at Discovery Munich 2020, but did not get the final approval to share material or record due to a customer not providing permission and the event went virtual.

If you do not name Lonza and keep it quite abstract this should be no problem to tell. They also had other presentations where the scenario could have fit to FDE with functional inputs and scalar output, though they had a different focus (andreas trautmann, claire baril, anne-catherine portman where some of the presenters and I think in DSE24 anne-catherine has another presentation).