LSL of Zero...or not 1

[powerhound]

I'm doing a capability study on the flatness of a group of parts. The IDI indicated the Weibull as the best plot use. The LSL on flatness is always 0. What I noticed is if I enter a 0 for the LSL I get a Ppk of 1.08. If I leave the LSL blank I get a Ppk of 2.72. Why this difference and which one should I use.

Thanks folks,

John Acosta, GDTP Senior Level
Manufacturing Engineering Tech

 

[GregLocock]

Using which analysis package? What is an IDI?

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[powerhound]

Oops, sorry. At some point I was on a Minitab page. I must have made my way out of it.

I'm using Minitab 16 and an IDI is Indvidual Distribution Identification. It's a tool into which you can bring your data and it will apply the different distributions and transformations to it to see which it fits best. The you can do your analysis using the correct plot.

John Acosta, GDTP Senior Level
Manufacturing Engineering Tech

 

[GregLocock]

I don't have minitab installed on this computer. So, create a small dataset with the same mean and s as your dataset, work out Ppk by hand, with an LSL of zero, and then fiddle with LSL til you get the same Ppk as in your first post.

Then you'll know what Minitab thinks a blank LSL means (my guess would be -USL, but that is wrong, that would still give the same Ppk. That would be the likely case for process capability (Cpk)). Then shoot the minitab developer for making unwarranted assumptions.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[GregLocock]

Hmm, I've overcomplicated my solution. Try typing different values into that box til you get the 2.72 number as an answer.

I must admit I've rarely had to think about single sided distributions, the hack answer of reflecting it about the mean seems awful.

In reality you are concerned with what proportion exceed your USL. Quite obviously you don't have a normal distribution, so any assumptions about 3 sigma =99.6% or whatever are misleading. I can't be bothered to research it for you but one sided distributions should be in any reputable text. In the bad old days the statisticians used to transform the x axis, using any old function that worked, until they had a normal distribution, and then they could just use the normal tools that other people with brains had worked out for them.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[powerhound]

Thanks for the insight, Greg. Turns out that if you use a 0 for the LSL, minitab automatically calculates a certain amount of times that the value should statistically drop below it. The Ppk winds up suffering for it. Dropping below 0 is not possible for this distribution so leaving the LSL field blank, thus there technically is no LSL, you don't get penalized for the value statistically dropping below it.

Thank you again for your help.

John Acosta, GDTP Senior Level
Manufacturing Engineering Tech

 

[GregLocock]

Thanks for the reply, nice to know

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?