Fractional Factorial Design in Minitab

Dear friends, this video illustrates how to create and analyze a fractional factorial design using Minitab software with an application example. You can watch our other video on basic concepts in Fractional Factorial Designs: DOE-5: Fractional Factorial Designs, Confounding and Resolution Codes Here is the link: • DOE-5: Fractional Fact... .
You can watch all our videos on DOE by clicking here to see the playlist: • DOE-2: Application of ... .
We hope you are finding our videos useful!

[@rahulbachal9004]

All vedios in DOE series are excellent and informative. Thank you sir.

[@instituteofqualityandrelia7902]

Thank you!

[@mathiasjosefweber6419]

Dear Henant, many tanks for your Videos. It helped me a lot to understand DOE better.

[@instituteofqualityandrelia7902]

I am glad you are finding the videos helpful!

[@shahana9999]

Dear sir, The video was very helpful. As a research student it helped me a lot in optimization part of my study. Thank you very much sir.

[@uhemant1]

Thanks! My best wishes!

[@sreekanthrampa3469]

very useful DOE information, I have better understanding now

[@instituteofqualityandrelia7902]

Glad it was helpful!

[@ArulRajendran]

thanks sir...

[@instituteofqualityandrelia7902]

Most welcome

[@silambarasanpalanivel618]

Nice Explanation sir

[@instituteofqualityandrelia7902]

Appreciate your feedback!

[@moisesalejandroleyvasanjua9894]

Nice Vid! Super gut gemacht

[@instituteofqualityandrelia7902]

Thank you!

[@prathameshyadav5534]

Thank you for the video. But I have one question. Is the data (response) required to follow any specific distribution like normal distribution here? If yes, then what if the data isn't normally distributed?

[@instituteofqualityandrelia7902]

As this is a designed experiment with different settings, and if the the data follows normal distribution, it shows that the effect of the factors are insignificant. In DOE, analysis is performed using ANOVA which assumes that data 'within' groups follow normal distribution. This is validated with the normal probability plot of residuals. In case the residuals do not show reasonable normality, one can consider transformations. Hope this helps!

[@prathameshyadav5534]

@@instituteofqualityandrelia7902 Thank you for the detailed answer. So if transformation is to be done (due to non-normality) should we directly paste the transformed data in response column? or there is some specific different method in Minitab to perform DOE by transforming data?

[@aohcg5120]

I am making a similar analysis and this video is helping me a lot. In my “yield” column should I use the mean of the samples I’ve taken? I have measurements of each combination (30 samples each) and don’t want to use only one data.

[@instituteofqualityandrelia7902]

I suggest that consider these as 30 replicates. Calculating and using averages will reduce the power of your analysis. One more possibility is that in addition to the regular analysis, calculate standard deviation of the replicates. Then you can optimize two responses. You may need additional help. My email id is [e-mail]. Write to me if you feel so.

[@julesbb7759]

nice

[@instituteofqualityandrelia7902]

Appreciate your feedback!

[@jyotipawar1130]

Hello sir, Your videos are really helpful. Can you please make a video on how to analyze and interpret fractional factorial design( 2^4-1) using DoE software. How to select factors that are significant from half normal plot without maintaining the hierarchy. Interpretation of interaction, contour and response surface plot.

[@instituteofqualityandrelia7902]

Principally it is similar to the procedure explained in this video. If u need any help, send mail.

[@jyotipawar1130]

Okay sir. Thank you!

[@bushrakamal2873]

Hello, I followed the exact same steps but my T and p value column has "*" in them and it says it couldn't calculate as MSE is zero and Degree of Freedom is zero . How do I fix this ? Please let me know.

[@instituteofqualityandrelia7902]

Hello Bushra Kamal, You will see "*" in the column of p-values if there are no degrees of freedom for error term. So you need to check whether you kept the original model or you reduced the model using "stepwise" comman.

[@bushrakamal2873]

@@instituteofqualityandrelia7902 Is it necessary to reduce the model using stepwise command ? I only want to see the normal probability plot, the main and interaction effects plots. I don't want to do regression at this step.

[@uhemant1]

@@bushrakamal2873 not necessary. You can reduce the model manually as well. But without reducing the model, the degrees of freedom will be zero and you will not get p-vallues.

[@bushrakamal2873]

@@uhemant1 Okay, is there a reason why you chose backward elimination" in stepwise ? Also, my confidence level is 95%, so alpha to remove would be "0.05" then, right ? Thankyou .

[@amiraaboelnour2532]

this video is excellent. I insert 19 factors for 20 run and it keep show me this "Backward Elimination of Terms α to remove = 0.1 The initial model was saturated. The stepwise procedure removed the following terms in order to obtain sufficient degrees of freedom to begin: X4, X8, X10, X14, X16" could you please explain me why?

[@instituteofqualityandrelia7902]

Hello Amira, as it was a saturated model, perhaps Minitab omitted the terms mentioned by you as the effects were probably negligible. Another possbility is Minitab calculates Lengths PSE (pseudo standard error) to estimate p-values. So it may have calculated the p-values with Length PSE and omitted these terms. You can check in the Pareto charts with all factors (without stepwise procedure). Hope this helps. Good luck!

[@Evelina_Fani]

Hello! I have a question for a 2^4 full factorial design.The factors are A,B,C and D and I have values for 2 replicates.My problem is that I have seen some examples where they sum the values in one column and in other examples they put the values in different columns and proceed to the design using both as responses.Is there any difference in these two methods?Which one is correct? Thank you in advance!

[@instituteofqualityandrelia7902]

I have not understood your question. Do you mean two rsponses or two replicates? It will be better if you send the data to me on email [e-mail].

[@rahulbachal9004]

Dear sir , pl add what is the use of surface plot and contour plot .

[@instituteofqualityandrelia7902]

Thanks. Will do it sometime. Need to check whether I covered it n previous videos.

[@zelihaylmaz3037]

Hi, sir I am an industrial engineering student. In my 2(5-1) experimental designs, it does not calculate t value and v value. What should I do? What is wrong?

[@instituteofqualityandrelia7902]

I need to know your data before I make my comments.

[@zelihaylmaz3037]

@@instituteofqualityandrelia7902 thank you sir

[@ovdickson]

Hello, Who has an idea how the "Yield" values used here were obtained? The values copied from the Excel file.

[@instituteofqualityandrelia7902]

The example is taken from our training material and this is not a real experiment. But it must be obtained by conducting trials as per the treatment combination of each row in the experiment. You may like to watch our video where data is captured using a catapult. kzbin.info/www/bejne/eYjLq5uCo82ee5o

[@bhaskaransundaram3287]

Sir, any particular reason using fractional factorial doe... why not another method. which one is good method and easily get the final response optimizing result?

[@instituteofqualityandrelia7902]

The purpose of Fractional Factorial DOE is to reduce efforts, time and budget. Which other method are you suggesting?

[@bhaskaransundaram3287]

@@instituteofqualityandrelia7902 i use always Taguchi method, thats the reason i asked, plz dont mind, your videos really good, just for knowledge purpose i asked.....

[@instituteofqualityandrelia7902]

I have no problem if you want to use Taguchi Designs! But please be aware that many of the Taguchi Designs are Fractional Factorial Designs. The selection of Taguchi Designs is based on the Linear Graphs. Also, suppose you are using the L8 array for three factors it will be equivalent of the Full Factorial Design. But if you use 4, 5, 6, or 7 columns for Factors in the Taguchi L8 Design, it would mean confounding of some of the effects and reduce the resolution of the design and will make it Fractional Factorial. In Classical Designs, one needs to use the resolution codes to understand the confounding structure in Fractional Factorials. In Taguchi Designs, the Linear Graphs and Triangular Table provides input to understand the extent of confounding. Incidentally, the video is on Classical DOE and therefore I have not discussed any details of Taguchi Designs.