Contour Profiler question

cweisbart

Hi all,

 

I had a question come up specific to the contour profiler in the fit model platform.

 

As you know, if you specify hi and low limits for a response, this will add shaded zones to the graph where you are falling outside those limits. The white space will be the area where your response falls within the set limits. You can designate a factor to the x axis and another to the y, and you can move the sliders for the other variables to see how the graph changes.

 

I'm trying to figure out if there is there an easy way to automate the sliding of the other factors that aren't being used as axes so that the white space is maximized and shaded regions minimized for the chosen factors on the x/y? Or is the sliding always a manual process?

 

Thank you in advance!

 

 

 

 

Phil_Kay

Hi @cweisbart ,

 

Good question but I think the answer is no. There is no solver for maximising the white space where the responses meet their spec limits (the "design space" in QbD terminology).

 

Modde (a DOE software) has some capability for doing this kind of thing and I think that Development have considered this but it is not as simple as it sounds.

 

In your example, we would change the settings of the additional factors  to maximise the white space of the 2 factors on the contour plot. There might be some situations where we would like to do this but it only really tells us what precise settings of the additional factors give us the maximum ability to flex the 2 factors on the contour plot and still meet response specifications. There are not many situations where we can precisely set all but 2 factors.

 

More generally we really want to know what is the biggest box in the dimensions of all factors, such that we can flex all factors and still achieve spec. And I think this is what Modde claims to be able to do. But, as I said before, this is more difficult that it sounds because the multi-dimension shape of the region will often be irregular. So there are different boxes that we could find and we would need to make some decisions about which factors we would want to be able to vary more than others. (In which dimensions do we want the box to be longest). I'm not sure how Modde addresses this.

 

The simulation capability in the regular prediction profiler does not directly address this but is really useful for finding the most robust settings of a process or system. This presentation is a great overview of this.

 

If you have a customer with a specific need then it would be great for them to add it to the Wish List.

 

I hope that helps,

Phil