Hi,
attached is a table of characteristics, the special thing is their profiles have a different profile on the right of the x-Axes. This is a behaviour of the product. The goal is to predict such profile based on DOE factors.
The issue is that FDE does not fit a model for the profiles without aproxiating also the behaviour to the right of a profile: E.g. the red most right profile forces a knot to be quite to the right. The most left profile (greenish) does not have values and also is not expected to have values larger than about 1.6 . But the model will of course try to fit a piecewise spline to between the knows also to the last know which is probably way mor to the right on the axes. But this is not the true behaviour.
Using just 1 knot I get some decent good results, but some of the curves will have overshoot profiles, depending how far the knot is from first profile increase (for cubic). quadratic model does not reflect the steepness of the profile, linear does not fullfill the profile itself.
I wonder if there is a better way than using just one knot or a general better approach. I tested some of the data prep options but these don't help here as we do not want to have the profiles aligned or overlapped (using DTW).
Thanks, Martin
