DOE-7: Analyse Factorial Design with Minitab: Case Study in Maximizing Fatigue Strength

Dear friends, this is part-2 of our video on Design of Experiments using Minitab. In part-1, Hemant Urdhwareshe had explained how to create design using Minitab. In this part-2 of the video, Hemant has illustrated complete details of analysis of the design using Minitab. Analysis includes interpretation of Minitab output in voice and text!
We are sure, you will find this video extremely useful for Quality Practitioners, Reliability Engineers, Design and Process Engineers, Six Sigma Professionals, Design for Six Sigma practitioners.
Some corrections at 5:21 -- The R-sq value is 95.45, R-sq Adjusted is 92.53 and R-sq Predicted is 86.63%. There is some error in these figures in the video. Apologise for the oversight.
We recommend that viewers should see our previous videos on Design of Experiments before viewing this video. Here are links to our previous videos on DOE:
Introduction to DOE: • DOE-1: Introduction to...
Application of DOE to Spot Welding using Excel: • DOE-2: Application of ...
Coded and Uncoded Values and Developing Regression Equation: • DOE-3: Design of Exper...
Fractional Factorial Designs: • DOE-5: Fractional Fact...
DOE-6: Case Study in Creating Full Factorial Design in Minitab: Optimization of Fatigue Strength
• DOE-6: Case Study in C...

[@yipinghe8092]

Thank you for your tutorial! I watch the whole series from the very first one and I understand DoE a lot better!

[@instituteofqualityandrelia7902]

Happy to hear that!

[@kevinwaycaster]

Thank you for sharing this explanation! I also deeply appreciate the captions to help us understand what is being said. I’m in a MBA program and stats aren’t my strongest skill so thanks again.

[@instituteofqualityandrelia7902]

Welcome! I am glad to know!

[@jcournoyer2]

Just watched the whole DOE series. One of the few videos that actually looks under the hood of what minitab offers. great job

[@instituteofqualityandrelia7902]

Thank you! Glad to know you watched the full DOE series! Do you work for Minitab? Regards..Hemant

[@jcournoyer2]

@@instituteofqualityandrelia7902 No, I am a Mech. Engineer that needed a refresher.

[@pravinshewale5235]

Thank You for sharing. These are very simple videos and good explanations to understand.

[@instituteofqualityandrelia7902]

Thank you Pravin!

[@subasa83]

Thank you for your video. However, there is a statement you made which mentioned the highest average response is 8.3 but I saw in one corner there is also a value of 9.1. Why is that?

[@instituteofqualityandrelia7902]

Thank you for your comment. You are correct. It should be 9.1 instead. Apologise for my oversight!

[@omarmtz74]

Thank you so much for your very clear explanation.

[@instituteofqualityandrelia7902]

Welcome! I am glad you liked it!

[@zainabkazim2974]

Thank you sir, the explanation is easy and understandable

[@instituteofqualityandrelia7902]

Appreciate your feedback! Thank you.

[@darksword98]

I passed my course thanks to you. Thank you very much!

[@instituteofqualityandrelia7902]

I am glad it helped! Congratulations!

[@Chiyosan777]

hello, thanks for the explanation, but in my factorial design i don't have to option to do contour or surface plots, can i use graph option instead?

[@instituteofqualityandrelia7902]

Hello friend, you may have defined the factors as text. In case these are numerical, you can create the design with numerical factors and analyse. If your factors are discrete, then contour surface plots are not possible.

[@jibinjosemathewjibinjosema7617]

Excellent tutorial Sir, Tank you!

[@instituteofqualityandrelia7902]

Thanks. Appreciate your feedback.

[@mathuuma]

Dear sir, What is the red reference line in pareto effect diagram? How its calculated by minitab? Also, is it possible to optimize design by using the results of finite element analysis to minimise the stress or displacement in the structure we analysed?

[@instituteofqualityandrelia7902]

The red line represents limit of randomness at the specified confidence level. It is calculated using t-distribution and error degrees of freedom. For example, at 8:43, the value above the line is shown as 2.14. The error degrees of freedom are 14. So this 2.14 is critical value of two-tailed t-distribution with 0.05 alpha risk and 14 degrees of freedom. Use function =T.INV.2T(0.05,14) to get this value on XL. Try in some other cases. Watch our video on t-test. Link to this video: kzbin.info/www/bejne/m6bdq4SLaNNji7c.

[@Dr.NagendraJayaram]

Very well explained sir. Great job....

[@instituteofqualityandrelia7902]

Thank you!😊

[@luigi88953]

Hello, is it possible to use response optimizer without considering any interaction effects? I ask you this since when I studied response surface, the only way to optimize was considering curvature (interactions, quadratic effects). I was wondering if even in the factorial response optimizer, interactions are mandatory for optimization.

[@instituteofqualityandrelia7902]

I would say consider interactions if these are statistically significant. Else, you can omit it from the model.

[@luigi88953]

@@instituteofqualityandrelia7902 thank you. Actually my question was about optimization. Can I still optimize without considering interactions (if they are not significant)? I ask you this because in response surface the prerogative for optimization was to have interaction and/or quadratic effects. I wonder why this would be different in pure factorial doe.

[@instituteofqualityandrelia7902]

@@luigi88953 yes. If you omit from the model, these will not be used for optimization.

[@luigi88953]

@@instituteofqualityandrelia7902 thank you!

[@laravalstar4814]

This helped me pass my course. Many thanks sir!

[@instituteofqualityandrelia7902]

I am glad it helped! Congratulations!

[@RosydePalma]

Thank you so much, Sir. Really helpful, congratulations!

[@instituteofqualityandrelia7902]

Glad it was helpful!

[@juanmiguelguzmanmunoz4053]

Great explanation; the best i have seen in the web. Thank you Sir!!

[@instituteofqualityandrelia7902]

Wow, thanks a lot for your feedback and appreciation! I get motivated to do more with such feedback!

[@GyanVisharad]

It's worth watching sir

[@instituteofqualityandrelia7902]

Thank you!

[@udayanibinduhewa5644]

Thank you. it was very clear

[@instituteofqualityandrelia7902]

Good to know!

[@omBrumSoha]

Why is the threshold p value for significant factor is p

[@instituteofqualityandrelia7902]

Hello Eric, The threshold p-value is decided by the confidence level as you have mentioned. You can change confidence level in Minitab in Stat>DOE>Analyse Factorial Design and Options (Ref Minitab 17 version). It may vary for other versions. And yes, the R-sq (Adj) 86% is an error. Apologise for that. Will trty to correct it if possible in the video. I have added the correction in the description. KZbin does not allow editing the video. Thanks for your keen interest.

[@chaitanyapawar3970]

if p-value of block is significant, does that mean there is unacceptable lot to lot variation? if yes, what action can be usually taken to resolve the issue?

[@instituteofqualityandrelia7902]

It depends on what the blocking factor is. If p-value shows that it is significant, then we need to consider the block in the model and prediction. There is no question of accepting or not accepting!

[@betulerdogan2236]

this is so good thanks for it. but i dont understant block? what is it? is it same value with our factor?

[@uhemant1]

A block is a discrete factor that is not under control of the experimenter. Examples are different batches, diffferent machines. It is not of primary interest to the experimenter but can result in additional variation. Please see our previous videos on DOE to understand the videos better. This is the seventh video on DOE.

[@GyanVisharad]

Please explain block and replicate in detail sir...why no. of run increased from 8 to 24 after adding 3 block and 3 replicate

[@instituteofqualityandrelia7902]

Thanks Ravish. For blocking please watch our other videos. On this case, three blocks are for three batches. Three replicates also provide better power and lower beta risk.

[@sakthi-ganesh]

Why three blocks are chosen? What effect will we observe if we choose only one block?

[@instituteofqualityandrelia7902]

Hello! Thank you for your question. I should have explained that in the first part of the video. The three blocks are for the three batches of material. Apologise for delay in my respones. Link to part 1 of the video: kzbin.info/www/bejne/mmLPhoGohLCWaMksi=k67m9ztysCFZkCUe

[@abderhmanmanchawi5277]

Thank you so much

[@instituteofqualityandrelia7902]

Welcome! Appreciate your feedback!

[@theluis469]

At 1:29 the highest average is 9.13 is in the corner with high case depth and radius and low roughness

[@instituteofqualityandrelia7902]

Yes. True. But with the regression equation that includes interaction, you may find the optimum settings somewhere else. This in only illustration, and what you say may not be true in other cases. Thanks for your keen interest.

[@ayanamsam6931]

Please Sir. Please...I'm in dire need of assistance. I have data that needs to be analysed using full factorial method but I don't know how to go about it. Please can you help me? I'm frustrated Please

[@instituteofqualityandrelia7902]

Hello Ayanam! Watch my foundation videos on the subject of DOE. Here are links! 1.kzbin.info/www/bejne/poWkhpRspr2JbKc 2. kzbin.info/www/bejne/mXq7in6vo9ZkY9E 3.kzbin.info/www/bejne/nmeokoNppJlgqbc

[@maheshroyal3607]

SIr, when we need to use factorial design and RSM response surface mode design in minitab , both methods are answering for data with results.

[@instituteofqualityandrelia7902]

I have not understood your comment. Kindly clarify.

[@jayasimhasln4615]

sir what is the difference between full factorial design, half factorial design and Response surface methodology

[@instituteofqualityandrelia7902]

Thanks for your ineterst! Please watch our video on Fractional Factorial Designs. Here is the link: kzbin.info/www/bejne/g4LLl2SBe82Medk. The Resposne Surface Designs are used when response is nonlinear and requires second or higher order models.

[@Lunat29]

Hi.. Could you please help me? The contour plot option is unactive and I can't plot it

[@instituteofqualityandrelia7902]

Hi Natalia! It is possible that you defined the factors as text. Please check. You must define the factors as numeric to get contour plots.

[@shehzadkhanlucky7477]

Hello sir , I follow your video for the analysis of DOE on minitab 18 and it doesn't give p-values and t-values. What can be the possible reason?

[@instituteofqualityandrelia7902]

Hello Shehzadkhan, Thanks for your question! This is most likely because of insufficient degrees of freedom (DF) for the error term. Calculation of p-values in ANOVA requires at least one DF for error. You will be seeing * instead of p-values. Am I correct?

[@shehzadkhanlucky7477]

@@instituteofqualityandrelia7902 Absolutely right sir * sign appears

[@shehzadkhanlucky7477]

@@instituteofqualityandrelia7902 Sir what am i supposed to do now?

[@instituteofqualityandrelia7902]

@@shehzadkhanlucky7477 You can either omit terms that have very little contribution. You can identify these by looking at Sum of Squares (SeqSS) values. If this is not possible, then you can replicate experiment but this is additional time and budget.

[@shehzadkhanlucky7477]

@@instituteofqualityandrelia7902 Ok, thank you sir.

[@alkabanchhor352]

Can anyone help?.in analysing factorial design my residuals are always coming 0 and the normality plot have all points at 0.please tell the mistake i made

[@instituteofqualityandrelia7902]

Hi Alka! This happens in saturated designs. Ths measn that the number of treatments in the experiment equals number of effects that you are analysing, and there is no degree of freedom left for Mean-square error in ANOVA and thus F-ratio cannot be calculated. You can see my other videos on Degrees of freedom, fractional factoril design and other videos on DOE. Good luck!

[@ankitverma8820]

hello sir ! Hemant sir as great as always :)

[@instituteofqualityandrelia7902]

So nice of you

[@instituteofqualityandrelia7902]

Thanks Ankit!