Using Prediction Profiling to Maximize Model Proficiency – Part 2
See how to use a prediction model to find optimum conditions (set points and ranges) to use the model to make decisions.
- Chapters
- descriptions off, selected
- captions settings, opens captions settings dialog
- captions off, selected
- en (Main), selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
This is a modal window. This modal can be closed by pressing the Escape key or activating the close button.
See how to:
-
Find optimum set points
-
Optimize desirabilities
-
Identify the combination of factor settings that will optimize the response
-
Define desirabilities
-
Remember settings
-
Lock factor settings (reset factor settings)
-
Simultaneously optimize multiple responses, including when one response is more important than the others
-
Simulate
-
- Find optimum ranges
- Use Design Space Profiler to set ranges on process parameters and then narrow factor space based on projected failure rate
- Make a Connected Table and Scatterplot Matrix X
Questions answered by @andreacoombs1 and @cweisbart at the live webinar demo.
Q: Can you edit the values in the Columns Manager?
A: You can change some of the values such as column names, data type, format, etc.
Q: When you optimized your response, how do you make sure percentages don’t go over 100%?
A: The reason it's going over 100% is assuming that, because it is a linear model, the distribution is normal, which may not be the case. In JMP Pro, you can change the distribution of your response to confine it from zero to one. You can use the beta distribution and generalized regression, which is used for responses that are, for example, between zero and one or between 0% and 100%. That will prevent you from having responses above one or 100%, however you're treating the data.
Q: When you were looking at the design space profiler, there was a volume portion percentage. What is that?
A: When you begin, you are starting with a hundred percent of your design space. As you do the full simulation of 10,000 runs, you are looking at a hundred percent of those 10,000 runs and as I'm narrowing my design space with what percentage of those 10,000 run are being unselected from my design space. This is shown as the proportion or volume of your design space.
Q: Does the design space profiler take into account the flatness of the response?
A: Yes, indirectly. Again, you are trying to find the smallest change in X. That results in the biggest increase in your specification rate. The slope of your response is driving if you are going to be in specification. See video below.
- Chapters
- descriptions off, selected
- captions settings, opens captions settings dialog
- captions off, selected
- en (Main), selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
This is a modal window. This modal can be closed by pressing the Escape key or activating the close button.
Q: Can you extrapolate outside your design space to find somewhere where you might have a higher inspect rate?
A: If you are running model from DOE, your design space and prediction profiler range are constrained by the range of factors tested in the DOE. The Prediction Profiler limits you to factor space. Otherwise, we can use our model to test a wider range and then extrapolate. You would have to do it manually. There's no there's no automated way to do that.
Resources
- Part 1 - Using Prediction Profiling video.
- Mastering video and Q&A on Design Space Profiling
- Prediction Profiler documentation
- Prediction Profiler options, including JMP Pro Extrapolation Control
- Statistical Details on Extrapolation Control in Prediction Profiler
Upcoming Events
-
Characterizing and Comparing Dissolution Profiles for New Drug Formulations
Feb 21Do you need to formulate drug products to match a specific dissolution profile? Do you need to re-formulate or validate that changes in ra... -
Making the Best Choice for Defect Count Models
Feb 7Least Squares Regression is often the go-to method for building statistical models. But should it be? Quality measures, like defect counts that are us... -
EMEA Mastering JMP: Understanding and Modeling Response Curves
Jan 31Learn how to analyze sequential measurement data where measurements you want to analyze as responses are not single points, but a range of points pres... -
Designing Process Optimization Experiments Using Models Based on Production Data
Jan 24Do you have historical or real time production data? Have you wondered how to leverage these data to design high value, efficient, cost-effective expe...