Video Script
In this video, I'll demo JMP Custom Design and create an experiment that tests five factors that influence a response of flight time for paper helicopters.
In JMP, I'll select DOE from the top menu and then Custom Design.
As mentioned before, the response is a metric we care about and can measure well. So in Custom Design, I'll enter “Flight Time” and leave it set to maximize.
A basic paper helicopter design has two rotors that have a specified length, a body, and a single leg with a length and a width. We're also adding factors of “Paper Weight” and “Paper Clip.”
Four of the factors are continuous, or numerical, and two are categorical because those two have two states, or as JMP calls them, levels. Those levels for paper clip are Yes and No and for paper weight are Heavy or Light.
I'll enter three in "Add N Factors", then select "Add Factor," and then I'll select "Continuous" for the first three; the next two will be "Categorical." It automatically adds the number of factors my design requires.
Now, I just need to edit the names and limits. Alternatively, both the response and factors can be loaded from the linked data tables using the red triangle.
Select “Continue.” The model outline appears and is populated with an effect called “Intercept” and lists of all the main effect factors that we entered.
At this point, we need to think about our goals and the technical information or assumption that we might have. Do any of these factors affect each other? Are we seeking a quick decision about which of the factors are influential at all? Can we ignore the others in improving the response?
[If not, do we know or expect that the relationship between a factor and the response is not linear? That means there is curvature, and at different points the factor affects the response differently. Do we know or expect that factors interact with each other, either in a synergistic or antagonistic way? If so, then we need to add corresponding terms to the model. For the purposes of the demo, let's say, Rotor Length and Paper Clip could have a nonlinear impact upon response. So we mark Rotor Length and Paper Clip in both the model and in the factor list and click on the Cross button.]
[Choosing the minimal runs in a Design and examining all of the effects will not provide enough information to estimate the error. In that case, you need to do some model manipulation in the report and remove some insignificant terms. You will not receive a message that will indicate that it’s not possible to estimate the error.]
Since I already know that both gravity and mass are two major influencers in flight time, I'd like to examine how those factors are influenced by each other and the resulting change in the response. I'll choose “Leg Length” and cross with “Paper Weight” and “Paper Clip.” I'll choose “Rotor Length” and cross with “Paper Clip” and “Paper Weight.” I'll do this by selecting the factors on top and the others in the model outline and select “Cross.”
Leg Length*Paper Clip
Leg Length*Paper Weight
Rotor Length*Paper Weight
Rotor Length*Paper Clip
JMP has suggested a default of “16 Runs,” which I'll select.
Before we make the design, and so we can repeat this design, in the red triangle dropdown, select “Set the Random Seed” and enter 504054292.
Now select “Make Design.” JMP will compute the design.
Here we can view our design in the DOE dialog window; now we can select “Make Table.”
Your table should look like this.
This data table is where we get our specs to build each design run from and save the collection data from those runs. Remember, this table also has the model scripts, the DOE dialog, and other scripts in the Tables panel. So make sure to save this as you add or change things.
We'll write down each Run number on the helicopter to maintain the order and record them correctly in the data table.
Now we're off to our collection step performed outside of JMP. I dropped each helicopter from the same height and recorded the flight time until it touched the ground.
As I've mentioned, JMP has a Fit Model script with some defaults set that will compute these results and provide a model report. I'll run that by selecting the green triangle next to the model script in the Tables panel.
Now from this Fit Model, we can select “Run.”
And we now have our Model report.
At the top of the report, there's an Effect Summary which lists our factors in order of significance to the response.
We can remove factors that have little or no significance to improve our model. We can undo if necessary.
You can see here the Rotor Length is listed top for effects and is significant.
However, this does not tell us exactly the formula that we’re looking for; we'll find that in the Profiler by scrolling down.
As I scroll down the report, I'm selecting the gray disclosure arrows, hiding report information I'm not reading at the moment.
At the Prediction Profiler Red Triangle, and from Optimization and Desirability, select “Maximize Desirability.”
This now indicates the settings that result in the longest flight time.
The factors names are below the plots.
By selecting and dragging the dashed Profiler lines we can adjust the other variables to see what's needed to produce the value you’re setting.
I'd like to show you one more thing, under the red triangle, navigate into “Factor Profiling” and select “Interaction Profiles.” This set of graphs visualizes how all the factors are interacting.
So looking at the Profiler, what are the factor settings that result in the longest flight time? Or, what is the optimized response for this design?
