DOE Analysis - Experiments with 5 Factors

[seasonal]

I did experiments with 5 factors:
1- Materials --(4 materials)
2- Coolants ---(3 coolants)
3- Cutting depth --(2 depths <25,50>)
4- Cutting numbers--(2 times <1 cut--2 cuts>)
5- Grit size -- 3 sizes(Fine --Medium --Coarse)

Response ;
1- strength
2- roughness

I run 48 samples using DOE and then run the analysis .I got the result attached but doesn't know which factor is more significant? I don't know how to save imp fie and send it here.


Fit Model(
Y( :Flexural Strength ),
Effects(
:Material,
:Bur Grit Size,
:Coolant Fluid,
:Cutting Depth,
:Number of Cuts,
:Material * :Bur Grit Size,
:Material * :Coolant Fluid,
:Material * :Cutting Depth,
:Material * :Number of Cuts,
:Bur Grit Size * :Coolant Fluid,
:Bur Grit Size * :Cutting Depth,
:Bur Grit Size * :Number of Cuts,
:Coolant Fluid * :Cutting Depth,
:Coolant Fluid * :Number of Cuts,
:Cutting Depth * :Number of Cuts
),
Random Effects( :Whole Plots ),
Personality( Standard Least Squares ),
Method( REML ),
Emphasis( Minimal Report ),
Run(
:Flexural Strength << {Analysis of Variance( 0 ), Lack of Fit( 0 ),
Plot Actual by Predicted( 0 ), Plot Regression( 0 ),
Plot Residual by Predicted( 0 ), Plot Effect Leverage( 0 )}
)
)

 

[v9991]

its little bit difficult(almost dangerous) to interpret&draw inference without knowing the product/process!!!! (cause, although data is used for inferences, but it only leads hypothesis!!!...the point is...dont blindly follow the data !!!)

if you could post the design+results(atleast the excel file), we could probably help you out step by step !!! (again, we can only infer from the available data, but this has to be done in more rigorous way, with functional & technical inputs!!!)

having said that, here are few steps...
1. build the model by selecting the appropriate steps, based on half normal plots OR the p-value of the possible models!
2. ANOVA table :- look out for p-value of model & lack of fit (p<0.05), (+ p-value of the above selected terms. there must be specific rationale for inclusion or exclusion of not meeting p<0.05 criteria)
3. Regression table :- parallelly look out for R-Sq values ( esp...Adj & Pred R-Sq not to differ by 0.2 (rule of thumb!!!)
4. look out for regression equation, and the co-effiients reflect the amount of +ve or -ve effect on the response!
5. evaluate the need for augmentation!!!

also, i find following ref to be extremely usefull & relevant..
Analysis of DOE data


hope this helps.

 

[seasonal]

Thank you for your replay .
I attached an excel file to all the data .

Your help is highly appreciated .

 

[Miner]

What does the column entitled Whole Plots signify? Whole plots typically indicate a Split Plot design structure, but your factors do not align with the whole plot numbers. Does this indicate how the design was blocked?

 

[seasonal]

Thank you for your post.
The whole plot means how the design was blocked.
I did each plot in a single day.
Each plot contains 12 different sample with different interaction.

Thank you

 

[Miner]

What software are you using? If you have Minitab, I can share my analysis directly.

First, you will have a fully saturated model, which means that you cannot test for every possible interaction. There is no way to test for 3-way and higher interactions, so you must remove them from the model in order to test the block effect, 2-way interactions and main effects.

Next, test for the block effects and remove it from the model, if not significant. Then non-significant 2-way interactions and lastly non-significant main effects.

I attached the interaction plots blow.

 

[seasonal]

Thank you for your answers.
I use JMP to do the analysis . `Do you suggest to download Minitab?

Thanks

 

[Miner]

No. If it was one that I was familiar with, I could tailor my response to the software.

 

[seasonal]

I downloaded Minitab and start doing some analysis but for now I couldn't have the same results I got from JMP !!
I don't know why .. May I did something wrong .

 

[Miner]

I am attaching my project file for reference.