Cpk vs. Ppk - Long Term vs. Short Term - Capability vs. Stability

[DJN]

I am trying to understand the difference between Cpk and Ppk and when to use them. I think I have it sussed, but I am not sure. I have been trawling through reams and reams of literature until my head spins. OK. So far, I think I should be using Cpk when the process is in control and the data is normally distributed, and Ppk when the process is unstable and/or the data is not normally distributed. Are my assumptions right? I am also confused about the terms 'Short term' for Cpk and 'Long Term' for Ppk. I have read on this forum that, "Cpk attempts to answer the question, does my process in the long run meet specification?. So why is Cpk short term? Any help and guidance appreciated.

David

[Darius]

All that you said is right, but going to the bones Cpk uses within_sample_variation and ppk uses total_variation as estimate of standard deviation.

The "long term" and "short term" terminology, I think became in use, when at the start of production of a new product on the line, the first estimates of indicators are labeled "long term" (MIL-STD-1916), taking in account all the variation of the process (ppk). Also, you may spect little variation in a smaller time period, an so more stability (cpk).

When to use them?, as you said

CPk when the process is in control and the data is normally distributed, and PPk when the process is unstable and/or the data is not normally distributed

Many companies ask the providers for a cpk, so it's a difficult point to say "no, I don't have stability, could I give you ppk insteed?".

And for you?, being practical, both indicators are good, but keep on mind that if you can calculate Cp or pp with the Cpk/ppk, calculate them, because both indicators tell more than just one, Cpk doesn't substitute Cpk but complement it, for example if you have a Cpk value can be because of the centering, or the variation. And take also in account something else that most forget, all indicators are estimates and have variation asociated with them depending on the sample size.

[Icy Mountain]

Try this link:
More than you want to know about capability!

Also (I may get corrected since I don't have a copy yet) try the Automotive Industry Action Group Statistical Process Control Manual. U$30, at *** DEAD LINK REMOVED ***

Short course: C analyzes a system's aptitude to perform, P measures a system's actual process performance.

[tattva]

Hi! everybody,

I´m having some difficulties regarding the following:

Is there any minimum value to apply for sampling events?

I mean when you perform a study bias, stability,or so and production rate is 10,000 pieces a shift. How many pieces must I study to make it representative?

Is there a minimum or can I specify it?

Thanks in advance

[Rob Nix]

Ford Motor Company was one of the first to use the term Cpk, defined as simply process capability (determined after a process was in control). For production environments it meant by default "long term" process capability.

Ppk first appeared on initial approval requirements (a.k.a. PPAP) for first run type capability (after control was established). Since the data was pulled from the entire "first run" population it was called Ppk. But it was in reality a short term process capability, or process (P)otential.

When regular production started (i.e. unlimited population), the statistical studies could refer to Cpk, or process (C)apability. This is long term capability.

However, during the decades to follow, the quality community began to focus on the source of the sample only, and the short/long Cpk/Ppk terms got reversed. As Icy Mtn states, "C" is aptitude to perform, "P" is actual performance. You can only be convinced of actual performance in the short term. With ongoing production, all you'll ever know is "aptitude to perform".

NOTE: They both are calculated the same, and both assume a controlled process.

[Atul Khandekar]

Welcome to the Cove, Rob. Very good first post.

I think (though not quite sure), Minitab mixed up the long-short terminology. I understand they have now dropped it in favor of within & overall.

When using these terms, one must be aware of the method & period of data collection, state of the process, the way sigma is calculated...etc.

[Jeff Mathena]

I'm new to the forum, and thought I'd just pass on my limited thoughts about Cpk. In my mind, I look very analytically at the definition which I interpret as:

How many 3*(standard deviations) the mean is from the closest specification limit.

From there if you know anything about the shape of the distribution or how the "standard deviation" was calculated you can start understanding in more detail what your process might deliver.

And I always like to remember a favorite quote:

"All models are wrong, some are just more useful than others"



Jeff

[D.Scott]

Hi Jeff,
Welcome to the Cove.

I will look forward to your posts and if you ever need any help or advice in the quality field, this is the place to come.

Some of us have even been know to "give advice" outside the quality field.

Dave