Onesided specification or Nonnormal
Six Sigma – iSixSigma › Forums › Old Forums › General › Onesided specification or Nonnormal
 This topic has 11 replies, 8 voices, and was last updated 18 years, 11 months ago by Mikel.

AuthorPosts

September 19, 2003 at 1:30 am #33352
Hello.
This is a question in real life. We want our one supplier provide CPK to us, but they refused to do that, and their reason is as belowing(also their email):
“…….I refer you toQS9000, Production Part Approval Process (PPAP, 3rd edition)I.2.2.9.5 Processes with Onesided specification or NonnormalDistribution.The supplier shall determine with the customer an alternative acceptancecriteria [alternative to Cpk] for processes with onesided specificationsor nonnormal distribution……It is our conclusion that QS9000 documentation supports our policy thatcalculation of Cpk values for OUR PRODUCT is not appropriate……”What should we do now? Can you help me?
0September 19, 2003 at 2:25 am #90070what about CpU and CPl;
If there’s no lower spec limit – Cpk = CpU;0September 19, 2003 at 2:26 am #90071Cpk=CpU – In case there is not lower spec limit
Cpk=CpL – In case there is not upper spec limt0September 19, 2003 at 3:08 am #90073Cesar.
I do agree with you, and I know waht should we do if we have only onside specification or nonnormal data, but, our supplier do not agree that, their reason is QS9000 which detail as my above email.0September 19, 2003 at 11:57 am #90085
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.Tom,
Both you and your supplier need to sit down and read what you wrote in your first posting. In particular note the sentence in bold:
“The supplier shall determine with the customer an alternative acceptancecriteria [alternative to Cpk] for processes with onesided specificationsor nonnormal distribution……It is our conclusion that QS9000 documentation supports our policy thatcalculation of Cpk values for OUR PRODUCT is not appropriate……”What should we do now? Can you help me?”
I agree with the assessment that Cpk is incorrect. Just because it is incorrect does not mean that an appropriate process capability computation is out of the question. If they are particularly intransigent then the two of you should sit down with Bothe’s book Measuring Process Capability and read section 4.2 – Process Capability Defined for Unilateral Specifications and perhaps Chapter 8 Measuring Capability for Non Normal Data.0September 19, 2003 at 12:02 pm #90086Robert:
Thanks for your help. But, I do not know the book you reffered. Now, the urgent thing facing to us is to persuade our supplier to provide CPK to us, but , we do not familiar with the item of QS9000 they said in email . Can you help me?0September 19, 2003 at 12:34 pm #90087Is your supplier willing to provide you with Ppk? What about a modified Cpk using Johnson transformations or Pearson curvefitting? Both are intended to help evaluate data that is not bellshaped.
Matt0September 19, 2003 at 1:01 pm #90088Tom,
Your supplier’s knowledge of QS9000 and Statistics seems to exceed your company’s knowledge. Why do you want Cpk or any other statistical index? You should answer that question first.
Based on my experience, a frequent answer among Automotive Suppliers is “because we / our customer want(s) a PPAP” – which indicates a severe lack of ownership for quality planning.
If I am correct, provide the answer to the question and we posters will help you improve your supplier management system.0September 19, 2003 at 2:57 pm #90094
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.If we assume that you really have a need for identifying the supplier’s process capability then the simplest approach would be to follow the advice on pp.55 of the Bothe book:
” A process producing a characteristic with a unilateral specification meets the minimum requirement of capability when it is stable, and has no more than .135 percent of its output for this characteristic outside the single specification limit.”
Note that there is no reference to the shape of the distribution in this quote.
If a plot of the data reveals that it is nonnormal and you are in need of more than the above then you should take the data, plot it on normal probability paper and determine from the graph the .135 and 99.865 percentiles. This will permit a computation of the equivalent six sigma spread and thus a computation of P’pu – a measure of capability which is not based on an assumption of normality.
The equivalent 6*sigma spread is 99.865 percentile – .135 percentile. Thus the equivalent 3*sigma is this value divided by 2. For a unilateral tolerance and a nonnormal distribution the equivalent Cpk computation is
P’pu = Maximum [ {(USL T)/(99.865percentile – 50 percentile)}, {(USL – 50 percentile)/(99.865 percentile – 50 percentile)}]
where the percentiles are read from the normal probability plot, the USL is your single spec and T is the target value.
If the data is nonnormal you could try transforming it using the Johnson Transform and then attempt to use the transformed data to compute a capability based on a unilateral tolerance but if you do this you will have to do a lot of investigation concerning the stability of the nonnormality of the existing distribution. If the shape of the distribution isn’t stable you will have to keep checking the Beta1 and Beta2 values (the skewness and kurtosis measurements scaled by the biased estimate of the variance) to make sure that the Johnson transform you are using is the correct one.
In light of what you have written I would have to agree with the other people who have posted to this thread. It does appear that there is a disconnect between what you and your supplier know about the statistics of the situation and it may be the case that your management is asking for some of these things just because everyone else is asking as well. Neither of these is a good thing. I would strongly recommend that you read the book I cited or at least read other papers on the issues surrounding unilateral specifications and their impact on capability calculations before making any decisions.
0September 21, 2003 at 12:53 pm #90113Tom,
You have asked your supplier for Cpk. You don’t say why. I am assuming you want to understand their defect rate or want to know the process distribution. The supplier referenced the automotive supplier book(AIAG) PPAP indicating that for a onesided specification, these indices needed to be treated differently. That is true. A real concern with nonnormal distributions which frequently occur with onesided specifications.The statements in the PPAP manual do not excuse a supplier from determining the capability. It is merely suggesting that the commonly used indices Cp and Cpk may not be the best method for summarizing the capability.
You need to decide for your company, how will you handle onesided specifications. Several suggestions have been made to you. You can simply have your supplier calculate the percentage of the distribution outside of the specification, based on the appropriate statistical model. You may want a smaller percentage than what has been stated in these messages.
Eileen0September 21, 2003 at 9:31 pm #90116There are many ways to normalize a nonnormal distribution and compute process capability for onesided specification.
Your supplier seems has a real attitude problem by using QS9000 as excuse. Obviously your supplier is lacking profound knowledge which is a key element in QS9000. :(0September 21, 2003 at 10:29 pm #90117Tell your supplier that you agree anc could care less about Cpk. Ask them for a copy of all their data as an alternative. Do your own analysis and respond appropriately.
0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.