Custom Design of Experiments Course

[#1]

We’re excited to bring you this free e-learning course on Custom Design of Experiments. This course focuses on the core principles of designing an experiment, enabling you to understand and apply those principles to achieve an optimal design using the Custom Design platform in JMP.

Custom design is an approach to designing experiments that produces optimal designs for the problem you’re trying to solve, whether that’s identifying important effects, or trying to optimize one or more responses. In addition to learning about custom design in JMP, you’ll explore key design concepts including sample size and power, balance, choice of factor ranges, blocking, and design evaluation. This course is for anyone who works in discovery, research, development, and quality assurance or control.

The demonstrations in this course were recorded in JMP 14. If you are using JMP 15 or higher, you might notice some changes to the interface, such as additional menu items or options in dialogs. These changes do not affect the course content and should not affect your ability to follow along with the course. You can go to Help > New Features to view a PDF of the new features and enhancements.

Download the course file (custom-design-of-experiments.zip) at the top of the page to use in the course.

Please send any feedback about the course to Ruth.Hummel@jmp.com with the title “Feedback on Custom DOE course”.

[AlexEdmunds]

Trying to understand the reasoning behind the supplier coefficient being 10.

[monica_beals]

Hi @AlexEdmunds, thanks for your question! In this example, we want to know the power of our experiment to detect a change in the breaking load from 30 lbs (the average for Acme) to 50 lbs (the minimum breaking load required). So we're looking for an effect of 20 pounds. The anticipated coefficient for the power analysis is entered as half of the expected effect size, so that's 10 here. The video "Scale Invariance" in the "Testing a Continuous Factor" section explains this in the context of continuous factors, but it applies to Supplier as well since there are only two levels. 

[JCabreraLuque]

Hi Monica,

When performing a custom DOE, let's say, we go through a first round of experiments and we get a set of predicted conditions that are supposed to meet our desired outcome (assuming several factors in the DOE). 

If we want to continue optimizing the conditions further, should we input the predicted conditions with new ranges for the factors, narrowing around the predicted ones in the first round?

I guess that's what you meant by resetting after each round. But wanted to make sure. My assumption is that there are conditions that can be further optimized (I'm working on ELISA methods optimization).

Thanks a lot for your response.

Best regards,

Juan (I know less than the previous person)

[gail_massari]

Hi @JCabreraLuque .  Welcome to the JMP Community.  I am mentioing @monica_beals ina copy of your post.  By doing that you will be assured she will see it:

Hi Monica,

When performing a custom DOE, let's say, we go through a first round of experiments and we get a set of predicted conditions that are supposed to meet our desired outcome (assuming several factors in the DOE). 

If we want to continue optimizing the conditions further, should we input the predicted conditions with new ranges for the factors, narrowing around the predicted ones in the first round?

I guess that's what you meant by resetting after each round. But wanted to make sure. My assumption is that there are conditions that can be further optimized (I'm working on ELISA methods optimization).

Thanks a lot for your response.

Best regards,

Juan (I know less than the previous person)

[monica_beals]

Hi @JCabreraLuque, thanks for your question!

Speaking very generally (and assuming you’ve conducted some confirmatory experiments to verify that those settings are producing the predicted outcomes), a common next step for process improvement is to conduct subsequent experiments with new ranges of the factors centered on the optimum settings you found previously. Narrowing the factor ranges might make sense in that scenario, so that you’re not conducting runs in regions of the design space that don’t produce optimum results. One caveat is that if your predicted optimum is near the edge of the original design space, narrowing the factor ranges might miss important behavior.

(The issue of resetting conditions after each run is a component of randomization in an experiment. In any given experiment, the factors should be reset from the same starting conditions before each run. When that isn't possible, we design split-plot experiments.)

Monica

[JCabreraLuque]

Thanks a lot Monica and Gail for your help.

That's what I thought I should do.

Great course, very useful.

[Fizzix96]

@monica_beals It appears that Ch. 4 of the journal is missing the details of the 7 experimental factors.  We don't know their values to use in designing the experiments, and the journal doesn't have an entry for "Design Door Seal Experiment" as suggested.  Can the journal be updated to include this information?

[monica_beals]

Hi @Fizzix96, I wrote an initial reply but then realized there's an additional issue with the journal. I will upload a new version as soon as I fix it -- watch this space

[monica_beals]

@Fizzix96 I've uploaded a new copy of the journal. The one that was there had the necessary components, but not in a spot that matched the solution. Should be fixed now, but please let me know if you run into any other issues!

[Fizzix96]

@monica_beals Consider me embarrassed...I thought I had looked everywhere in the journal before posting this, even in the other sections.   :-\\   Thanks for setting me straight.

[monica_beals]

@Fizzix96 no worries at all! I'm glad you commented, because there was a key part of that practice missing from the previous version of the journal. Even if you'd found the Design Door Seal Experiment script, you wouldn't have been able to follow the rest of the steps in the solution.