Subscribe Bookmark RSS Feed

Interactions in L27 Taguchi Array

tyagi10

Community Trekker

Joined:

Apr 19, 2016

Hey everyone,

I am building a model in JMP where I am testing 4 factors and their interactions. I want to be able to use the L27 array for the same by placing the main factors in columns 1,2,5 and 8 and then putting the interactions in the others according to the interaction table for an L27. The issue is that whenever I select the L27 in JMP and input 4 factors it shows me a table with only 4 columns and does not show the 13 column table which the L27 actually shows. Moreover I can't see any interactions.

If I try using custom designer and then assign interactions and factors and run 27 tries, the levels that generate don't match an L27. I even tried manually editing the levels in the table but it still doesn't seem to work. Does anyone know a method where one can implement the L27 with factors and interactions and run the whole table and assign certain factors to columns of our choice?

Thank you

6 REPLIES
M_Anderson

Staff

Joined:

Nov 21, 2014

Hi!  Can you provide a screen shot of what your set up is for the design?  When I provide the Taguchi platform with 4 (3-level) factors I get the standard L27 set up that would be expected. 

11362_DOE__Taguchi_Arrays.png

tyagi10

Community Trekker

Joined:

Apr 19, 2016

Hey Mike,

Thank you for your response. You're absolutely correct. I get that option when I enter four factors. But the issue is that there is nothing to enter the three interactions I want to study. Moreover, I am not so sure if there is a way to assign certain factors to specific columns. As of right now, my factors are automatically assigned to the first 4 columns of the L27 and I'd like to shift them to 1,2,5 and 8 and study the interactions in the others. Please see the pictures below.

[Missing image]

[Missing image]

Also, it doesn't display the whole table. not sure if thats an issue though.

[Missing image]

Here, I'd like to study interactions under construct model effects too but it only allows me to select the main  factors.

Also, here's a picture of an L27 I manually edited to match an actual L27. I did the same step to create the table but then manually entered all those values to generate a complete table and then used all of them as inputs to the model effects. I have used only the first 11 column of the table and left out the error terms. But this is probably incorrect and gives me a result I am not so sure is right.

[Missing image]

[Missing image]

Thank you for your help. Really appreciate it.

M_Anderson

Staff

Joined:

Nov 21, 2014

My understanding is that Taguchi Designs are only suitable for main effects, due to the design being saturated.  I suspect this is why the interactions don't appear in the model when you run the model script.  The rule I've found is that if a term or interaction is not appropriate for the given design then it won't appear in the modeling platform.  Maybe someone with some more experience with them can chime in on that point.

That said, based on what you appear to be doing, the custom designer would be the better choice.  You have a lot more visibility to what your design can do there.

Best,

M

tyagi10

Community Trekker

Joined:

Apr 19, 2016

Hey Mike,

Thanks for your prompt reply. You do have a point. I did try using custom designer too but in that case I wasn't able to test the same levels as the L27 array. It generates a random order of levels and I'm not so sure if I can edit those. Is it possible to change the table to suit what I need it to do?

Thank you

M_Anderson

Staff

Joined:

Nov 21, 2014

Sure, you can edit the table after the fact.  If you do, I would check the as generated design and the modified design using the Evaluate Design feature under the DOE menu.  You just want to make sure that you're not degrading the design quality with the modifications.

You might also look to setting up things like linear constraints if you need more control over where the algorithm is putting the experimental runs in the design space.

best,

M

tyagi10

Community Trekker

Joined:

Apr 19, 2016

Hey Mike,

Thanks for your response. Based on what you mentioned, I ran a custom design array, used 4 factors and 3 interactions. Edited the table generated and the evaluated the design. I chose three outputs as chosen and it generated a result which looked more consistent with what I would expect. Anything you see that looks amiss?

[Missing image]

[Missing image]

[Missing image]

Thank you so much for all your help. Appreciate it. You saved my day.