Re: Can CPK be a good manufacturing processor’s quality objective?
My boss chooses CPK to be the manufacturing processor’s quality objective. However I think it’s not a good idea because CPK doesn’t cover variance between groups, and doesn’t reflect the processes’ performance. I intend to tell my boss to use PPK or DPMO instead, but he seems not to know the two indicators.
I want to make sure that CPK is not proper for quality objective, so I’m here for confirmation. Is there someone who uses CPK as quality objective?
I think a big part of the problem is that Cpk has much more name recognition that Ppk. From what I have heard around the Cove here, it isn't unusual for people to say Cpk, but actually calculate Ppk. If you really want to understand your process, you really ought to look at Cpk, Cp, Ppk AND Pp. At one extreme is Cp, which tells you how well you could do if the process was constent and centered. At the other extreme is PPK, which tells you how well your are actually doing, given that the process isn't completely consistent ande isn't perfectly centered. Cpk & Pp fall inbetween.
I agree with Sammy76 that the single best indicator of the process is Ppk.
I have two competing opinions about DPMO. At one level it simply tells the fraction of defects found - count the defects and divide by the number of opportunities. You could just as well express this as DPHO (per Hundred, ie percentage) - no difference other than the location of the decimal place. Either form is an excellent indicator of performance.
My problem comes when people try to extrapolate the data to predict DPMO. They start with imprecise estimates of mean and standard deviation, make questionable assumptions about normal distribution, then proceed to predict the defect rate to 2 or 3 digits. In many cases I wouldn't trust these calculations to within an order of magnitude, let alone distinguishing between 3.4 DPMO and 3.5 DMPO.
As for your boss, I'm not sure how to convince him to use Ppk rather than Cpk. Perhaps you could subtly lead him toward the conclusion you want. Ask him what it is that he really wants to measure - the "hypothetically"

possible capability or the "actual"

capability. Or ask how he wants to calculate the standard deviation - using an "estimate"

where you "have to"

average together a bunch of individual calculations, or "directly"

calculating the standard deviation from all the available data. Just like a good magician, you can probably force him into a specific response while making him think he was acting freely.

Then tell him that what he chose is really known as Ppk rather than Cpk.
Or just keep reporting the Cpk that he asked for and the Ppk (and perhaps Cp as well) that you think are important, and interpret what these two number tell you. He can always prune out the Ppk value if he really wants, but at least you have given him the numbers that he really should have.
Tim F