Is there a minimum industry standard for PpK ?
- Thread starter SGquality
- Start date
C
Curtis317
1.67 PpK per my memory. However I cannot find any reference in the AIAG SPC manual for either CpK 1.33 either. Just the equations.
The minimum requiremet for Ppk to be 1.67 was removed some time ago I believe and it may have come from a single automotive supplier rather than AIAG. My memory may be very faulty on this, so if others remember more accurately, pease chime in.
What I do remember precisely is that the Ppk min applies when the capability study is done on early builds and the samples are from a very short time. In this context Ppk is an estimate of short term capabilty, but uses the total standard deviation since there are no subgroups: a single sample is taken from a short build that usually uses only one raw material lot(s), one operator, one piece of equipment, etc. This limited build condition provides the 'short term' condition. the minimum value is set at 1.67 to allow for build to build variation that would increase the overall variation reducing the long term capabiilty estimate.
in the beginning there was only one formula and one index: Cpk and the formula used the total standard deviation.
Then someone decided to get cute adn they added a short term capability estimate. Cpk became the short term estimate using the within subgroup standard deviation in the formula. Ppk became the long term estimate usign the total standard deviation in the formula. Most statistical software follows this convention. but most of industry has dropped the short term aspect as it has no real value. so often Cpk requirements expect you to use the total standard deviation...
very confusing adn overly complicated for an estimate that provides no real value or insight. There are many threads here that discuss this topic...should you want to learn more a search is a great place to start...
What I do remember precisely is that the Ppk min applies when the capability study is done on early builds and the samples are from a very short time. In this context Ppk is an estimate of short term capabilty, but uses the total standard deviation since there are no subgroups: a single sample is taken from a short build that usually uses only one raw material lot(s), one operator, one piece of equipment, etc. This limited build condition provides the 'short term' condition. the minimum value is set at 1.67 to allow for build to build variation that would increase the overall variation reducing the long term capabiilty estimate.
in the beginning there was only one formula and one index: Cpk and the formula used the total standard deviation.
Then someone decided to get cute adn they added a short term capability estimate. Cpk became the short term estimate using the within subgroup standard deviation in the formula. Ppk became the long term estimate usign the total standard deviation in the formula. Most statistical software follows this convention. but most of industry has dropped the short term aspect as it has no real value. so often Cpk requirements expect you to use the total standard deviation...
very confusing adn overly complicated for an estimate that provides no real value or insight. There are many threads here that discuss this topic...should you want to learn more a search is a great place to start...
These can be found in AIAG PPAP 4th edition.
As are the exceptions to its usage!!!
D
Darius
Agree, without the environment variables, no value is a "real standard", if the sample size is not defined nor the presence of patterns and sampling methodology, is there a meaning in such number? is as good as 42.
Agree, without the environment variables, no value is a "real standard", if the sample size is not defined nor the presence of patterns and sampling methodology, is there a meaning in such number? is as good as 42.
Agree or disagree with the use of Cpk/Ppk....
I just answered the OP's question.
D
Darius
Sort off,
BUT Not as an standard, not to measure is worse course of action to measure, as this article saysYou Can't Manage What You Don't Measure by John Reh
The issue here is to apply the same metric to all, and this is NOT right.
If you compare a metric against new data from the same system, if there are no new weir trends on the data (behave rougthly the same way), is possible to determine the trend of this system, using Cpk or ppk is fine but if you compare against the same system performance.
There is the saying that you shall not use performance or capability indicators if you are not stable, but if this is the case how could you expect to improve your system to such state of art behaviour, you are entering to the cycle of which is first the egg or the chicken. If your system has a non gaussian distribution, the same thing; but if your system behave as a log-normal distribution, Gamma, etc. managing that can led you to improvements, but only if you compare to the same condition.
In some other posts I tried to show that cpk/ppk are estimates obtained from a set of data, so you shouldn't take them as a true value which if you increase by 0.01 something was going on for good and decreased by 0.1 something was don wrong (I put 0.1 because is there the believe that a change no matter how big or small is a change
). And how can you compare the precision of 10 data sample against a 1000 data sample if you take just the cpk/ppk value?, and the sample size is almost always taken as a negligible piece of information, "the indicator came from a very famous statistical package and must be true", really I saw this many times.note: I put 42, because is the answer to everything, the answer is fine but what is the frame that make this "magical number" give it's meaning is missing.
Last edited by a moderator:
Oh, trust me, I am in your camp.
The Cpk/Ppk addiction is rampant in automotive land. If I had my way, I would eliminate them all together. But in the mean time, OEM's continue to refer to their own and PPAP requirements, and I am not in a position to change that. 
Similar threads
