Ppk Issue - Some Products have Unilateral Tolerances

abhipatel

Involved In Discussions
Here is a brief intent on 1 of the projects am working on and need feedback from the experts here

Intent: Reduce variation within product and from product to product
Measurable : Height @ Facing Operation
Sampling : 5 pcs every 30 mins
Gauge : Dial Indicator
Measures : Max & Min Height of product

Issues --- some products have unilateral tolerances and make it hard for me to conduct a capability analysis as data is log-normal rather than normal (atleast that is what I interpret :notme:)

Right now am using a box plot to characterize the variations existing within 1 product (min max - range) and using a X-R chart to deduce variations across lots(of 5)

Process capability analysis of the data has left me stumped ....

Any recommendations on how I should proceed in this case?
 

Bev D

Heretical Statistician
Leader
Super Moderator
why do you need process capability data? I assume you mean a Cpk (or Ppk) value).

typically these are not worth the effort and provide little to no useful information.

A simple plot (a multi-vari) of the 5 values for each lot plotted against the spec limits will provide all the data you really need to get started. you wil see the within sampel variation and the between sample variation as well as the spread of the process vs the specifications...
 

abhipatel

Involved In Discussions
customer needs a value of 1.67 and being a sheet metal operation..it is driving me insane..jeez......:mad::mad:

a conventional lathe operation of facing + sheet metal work + Ppk is a nightmare for me right now.....have no clue that it would even drive my value higher from what it is now...since repeatability in sheet metal work is very hard to achieve considering all the variances involved..

appreciate the help..keep the feedback coming..will post a sample data from office tommorrow
 
K

kaikai

One of a solution is this;
Assuming that the data is log-normal, applying logarithmic transformation on the data and specification limit. After that Cpk of the transfoemed data will be able to calculate.
 

abhipatel

Involved In Discussions
1 of my major issues is currently lets say we have a drawing specification of 20 +0.10 +0.30....the production would just convert it into a "bilateral" and run it with a mean of 20.2 +0.10 -0.10...

Not that it is an issue but according to what I read on the forum with "regards to precision machining" (in my case CNC facing) mean is of not much significance. Driving to it would promote overcontrol....

it would follow a rectangular/uniform distribution with tool wear slope as a factor....

I am having hard time convincing management though, they saysby doing the 'normal thing with bilateral' we would be able to drive to the mean and reduce variance.....

my processes really show low ppk value and from what i interpret(from the capability of precision machining) compressing control limits too much is going to promote over-control.....

counter statement from my boss is we are not aware of the process location, spread and he tends to think that taking "hi-lo heights" (which i agree) and plotting its mean (hi-lo mean) will let me know that the process is behaving as converted to normal/???:mg::mg::mg::mg::confused::confused:
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Not that it is an issue but according to what I read on the forum with "regards to precision machining" (in my case CNC facing) mean is of not much significance. Driving to it would promote overcontrol....

it would follow a rectangular/uniform distribution with tool wear slope as a factor....

I am having hard time convincing management though, they saysby doing the 'normal thing with bilateral' we would be able to drive to the mean and reduce variance.....

You are right, "management" is wrong. There is no economic or functional benefit to increase the adjustment for tool wear (overcontrol to 'run to the mean'), causing more operator intervention in the process.


my processes really show low ppk value and from what i interpret(from the capability of precision machining) compressing control limits too much is going to promote over-control.....

Ppk is statistically not applicable. Correct capability for precision machining is [USL-LSL]/[UCL-LCL]


counter statement from my boss is we are not aware of the process location, spread and he tends to think that taking "hi-lo heights" (which i agree) and plotting its mean (hi-lo mean) will let me know that the process is behaving as converted to normal/???:mg::mg::mg::mg::confused::confused:

He is mixing methodologies for different distributions, and applying statistics incorrectly. The net result is confusion and incorrect interpretation. Not sure why he is trying to re-invent the wheel.
 

abhipatel

Involved In Discussions
Yes...I am going to go on basis of "spc for precision machining" and plotting X hi-lo charts and running capability analysis as stated (or something like Cpp or Cpt which i also read from the forums).

My question though would be how do I derive the control limits for my operation?

I do not want to put a start at 0.6 or 0.75 of spec. limits as a start as I am unaware of the tool wear slope/rate.

assuming the unilateral tolerance example i have above how should i start the study? my plan:

1. subgroup = 125 (samples 5 each) data collection
2. calculate control limits
3. identify & tackle special causes
4. re-define control limits (shrinking them till i get to 0.75 of Spec limits) while constantly monitoring process

How would I be able to detect tool wear rate & replacement frequency statistically? I might be able to do it for the product that I am sampling but how would I parallely deploy it to jobs say with different dimensions/diameters???

Go elsmar...have learnt a lot from here...thank u all...another new thing added to my arsenal with regards to process capability of non-normal data with unilateral tolerances...... THANK YOU!! :applause::applause::applause:
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I do not want to put a start at 0.6 or 0.75 of spec. limits as a start as I am unaware of the tool wear slope/rate.

75% of the tolerance is acceptable.
The only effect tool wear rate has is the frequency of checks. For your study, do a very significant frequency of checks (100% is best), and determine the frequency from the slope.

How would I be able to detect tool wear rate & replacement frequency statistically? I might be able to do it for the product that I am sampling but how would I parallel deploy it to jobs say with different dimensions/diameters???

Each product characteristic should have its own chart (recommended a chart for each finish tool), and a linear regression of either the Hi or Lo data should give you the slope (if your gaging is adequate).
 

abhipatel

Involved In Discussions
thank u bobdoering...

maybe a naive question...but how do i do a 100% frequency of checks (100% sampling? for 1 whole cycle? - that might be a lot of parts :rolleyes:)? can u please detail it out...how would i be able to tell it to the operator in common man language when to check it?

i have only 1 product characteristic and that is height but the tool wear would vary as when product OD becomes larger tool would have to remove more material throughout the periphery.

That is the only 1 characteristic am measuring across all products but I dont know how to correlate say a tool wear rate determined from 2 cm dia and equalize it when i use the same tool for a 5 cm dia job.
 
Top Bottom