DoE Workflow/Tool-Chain

[#1]

Hi everybody,

I am evaluating JMP. Using the trial version.

Before I used JMP, I used R+MiniTab for a DoE with about a Million data points.

So I summarized the data in R to get 20 medians as response.

MiniTab shows run orders, which I used to join my results. 

But how does that works in JMP?

My DoE table has no run order. Neither I am able to export this table. Is that a limitation to the trail version?

I figured out, that I could make join with a different table, but because of the repetitions, I still need a unique key to join with.

So what is the workflow in JMP for exporting a DoE-table, run the DoE and reimport the results?

[#2 · Accepted solution]

It is not really clear to me what you are trying to do. But maybe some of this information will help.

You can create a design in JMP under the DOE menu. There are many choices for the type of design. In the design creation process, just before you create the data table, there is a checkbox to add a run order column.

Checking that and clicking Make Table will give you a DOE with a run column. You can then enter your response data directly into the JMP table. There is no need to export or "re-import" the table. It can all be done in JMP, including the summarizing of the data.

Being new to JMP, I would strongly suggest going through the tutorial that is provided under the Help menu. There are also MANY free online videos and tutorials under jmp.com. Please explore some of those videos to learn more about what JMP can do. Best of luck! 

[#3]

Hi Dan,

done my very best to help my self. But after 2 hours of reading and watching tutorials just for that last point on the list, I gave up.

I think that´s really a bit the tricky thing with JMP.

Of course there is a solution for your problem. Just in very, very few times you are the one with a new problem. But where to find the checkboxes or how it must be done is the issue.

Well - if I have seen the dialog from our snapshot somewhere... But I really didn´t.

But now I know it exists and is exactly what I was look for. For the moment that is good enough. So if we buy the software, it will work.

Thank you very much for your help!

[#4]

As Dan says, there is no need to setup the design in R.  Create the design in JMP (this can be exported in Excel if you prefer), analyze the design in JMP and then if you want you can export the analysis into Word or PowerPoint or other file types.

[#5]

Hello Statman,

didn´t plan to use R to setup the design. 

Last time I used Minitab to setup the design. But I done the data wrangling in R, because Minitab is not very handy for such big datasets.

At the end I exported the summary and pasted them into the result column of Minitab.

So because of my experience with R, Minitab, Statgrafics and DesignExpert, it was quite easy for me to create a design in JMP too. Even without any tutorial. But the export drove me crazy...    

[#6]

@EDS : A designed experiment with a million observations? I'm curious, can you articulate additional details around how you went from apparently 20 unique treatment combinations to a million observations? I've never heard of a designed experiment like this so I'm always interested to learn about unique approaches to DOE in general.

[#7]

Hi P.,

well. It´s quite simple.

It is a printing process. The stencil has ~20.000 openings. 

The optical inspection unit measures area, height and volume of every opening. So three values per opening.

Of course I focus on special areas with a weak performance. And there are just 20 real repetitions.

Goal of the DoE is to reduce outliers and raise the median of the volume. But the height and area might be interesting for further investigations.

My first DoE (R+Minitab) brought me impressive results. 

Now I´m planning a DoE with the same concept, but with 80 to 140 real repetitions.(It´s a bit tricky. A lot of hard to change factors...)

There for a I looking for a software to handle the whole data at once and filter the named areas interactively.

So the only puzzle piece I´m missing to reach my goal is the way to link the measured data to the DoE table...

[#8]

@EDS : Interesting. Thanks you for sharing those additional details.

Back, many, many years ago, during my tenure at Eastman Kodak Company, I worked on a project team where our goal was to demonstrate proof of concept for using DOE as a central method to optimize and make breakthrough improvements in image quality for 4 color process printing on white opaque polyethylene where flexography was the printing method. Pretty much anything you buy in the store on a flexible substrate (doesn't have to be just white opaque poly) uses flexo to print the images you see on the package. So this is a common process for packaging applications.

Key to the process is a cylinder called an anilox roll. Essentially an anilox roll is a laser engraved roll with cells anywhere from 100 to 900 cells per inch and each cell can be variably engraved for number of cells per inch, depth, opening, and overall volume. The roll is a critical to process factor for the volume of ink that is transferred to the printed substrate. The project was very cool and I learned so much along the way. Both about statistics and team dynamics.

If you would like to read some reports of our work, send me a message here on the Discussion Forum with an appropriate email address to send them too...if I can find them on my computer.

Good luck in your future endeavors.

[#9]

You are welcome.

Actually it is a solder paste printing process.

Lets see if I can share something on the end. Specially considering the non-disclosure.
The whole experiment/evaluation will take until the end of the year. Still a lot to prepare and discuss...