A DOE in Minitab - 3 factors and 3 levels for each factor

Z

zebras

Hi everybody,
I need to do a DOE in Minitab.
It has 3 factors and 3 levels for each factor, but I only have 15 runs instead of the 27 runs needed to do it using a 3^3 factorial design, and it also has two responses and one replica per each response.
How do I get the significant factors?
Thank you.
 

Miner

Forum Moderator
Leader
Admin
Why do you need three levels? Do you know that you have curvature, or do you want to test for curvature?

I recommend running a 2^3 factorial experiment with 3-4 center points. The center points will identify whether curvature exists. The multiple center points will estimate pure error, so you do not need to run a lot of replicates. Then, if curvature exists, you can add axial points to create a response surface design.
 
Z

zebras

Hi Miner

Thank you for your answer.

No, I do not know if I have curvature. I need to fully analyze the data, determine the significant effects including any interactions and/or quadratic terms. Maximize one of the responses.

I am knew using Minitab, but I know how to do 2 k.
the problem that I have is that my data is something like :
Factors Response
A B C P Q
1 7 10 1.1 20
2 8 20 1.3 21
3 9 30 1.4 22
it has 15 runs and the entries for the factors are the same numbers in different order. So I am assuming that there are 3 levels for each factor. But I am confussed
Thank you
 

Miner

Forum Moderator
Leader
Admin
Please post the design matrix and the accompanying results. It is difficult to understand what you mean without seeing the design structure.

You can zip the Minitab worksheet file and attach it.
 
Z

zebras

Attached is the original problem that I'm trying to fully analyze.
Thank you.
 

Attachments

  • DOE (2).zip
    3.1 KB · Views: 156

Miner

Forum Moderator
Leader
Admin
The design was not a balanced design, so you cannot analyze it using Minitab's DOE analysis, you must use regression analysis.

My analysis is attached.
 

Attachments

  • DOE.zip
    114.6 KB · Views: 164
Z

zebras

Hello Miner

Thank you for your analysis.

Can you please explain a few things?

How do you know that a design is not balanced?
What does it mean?

Is there any difference between General Regression and Regression?

Thank you for your help!
 

Miner

Forum Moderator
Leader
Admin
Balanced means that there are an equal number of experimental runs at each condition. The main issue with the experiment was that some levels were run once and others three times.

General Regression and Regression work the same way. General Regression is more user friendly for entering the model, and Regression provides greater control for an expert.
 
Z

zebras

Hello Miner

Thank you for your explanations and help!.

I am trying to learn how to do General Regression to see if I can get the same results that you did but, every time that I run it for Ressis. I get:

General Regression Analysis: Ressis. versus A, B, C

Regression Equation

Ressis. = 3.85333 - 0.002 A - 0.0375 B + 0.012 C

Which is very different to your results

General Regression Analysis: Ressis. versus A, B, C

Regression Equation

Ressis. = 0.00193632 A + 0.210163 B - 0.00385345 B*B + 0.000316726 C*C
For the Effic case it is also very different.
Evidently, I am doing something wrong.

Thank you again
 

Miner

Forum Moderator
Leader
Admin
I noticed from the effects plots that there appeared to be curvature in some of the factors, so I added quadratic terms to the model first then simplified the model by removing the non significant terms, including the coefficient term if necessary. It is an iterative process. I removed my prelimenary models, leaving the final models, so you would not be confused.
 
Top Bottom