Dear all, i have a question for you. does anyone hear this:
CPK Averages
Individual CPK part values must be combined and averaged to report a “Plant CPK Average” at each plant.
The Plant CPK average will be reported on a monthly basis.

i always know apply Cpk or Ppk to one process, for statistically controlling the processs of manufacturing some part, so, what the plant Cpk means and for what to do this?

I presume this is a request from a current or prospective customer. It appears from the request the folks making the request want a simple measuring stick to determine if the entire plant is running efficiently.

I don't necessarily agree with the concept, but it seems simple enough to do the arithmetic if your plant is already calculating Cpk for each individual part you manufacture.

For those who may like (or need) a quick refresher in Process Capability and Process Performance, you might start here for a clear definition of each:
http://www.isixsigma.com/library/content/c010806a.asp

I don't necessarily agree with the conceptNor do I... It sounds a bit strange imo. Maybe I'm dense, but I can't really see the use of such a figure? Anyone else? Where is Steve when we need him? ;)

/Claes

Nor do I... It sounds a bit strange imo. Maybe I'm dense, but I can't really see the use of such a figure? Anyone else? Where is Steve when we need him? ;)

/Claes
Actually, one could infer a lot of things from such a number:

Do the folks at this plant KNOW how to calculate a Cpk?
If the number seems too "pat," we can ask for the raw data - folks have been known to "fudge" reports, but it takes a real expert to fudge raw data by working in reverse from a phony Cpk.
This may be a test of Supplier Responsiveness - will they comply with a relatively simple request?
If the number reported is between 1.33 and 2.0, we might conclude the supplier is using adequate equipment and processes. If the number is much higher, we might conclude the capability is much higher than current customer requirements - we would wonder why - do they pursue scut work with wide open tolerances? do they really report a factual number or is it "created" to look good? In any case, we would investigate in more depth.
If the number is near or below 1.0, we would seriously wonder why these folks were in business with machines or processes that can't maintain control for the tolerances required by customers.

Actually, one could infer a lot of things from such a number:As you say, but one would have to look deeper to understand what this average figure actually is worth.

It's just that I could think of easier ways of finding out whether the supplier knows how to calculate a Cpk or not, and I would much rather like to see individual Cpk figures. This average could be pushed up, for instance by throwing in results like you suggest in no. 4, and a low average could mean that some tolerances are nigh on impossible to live up to with the best available equipment.

A test of Supplier Responsiveness? Perhaps, but why not test that with something more useful?

Anyway, has anyone else come across this?

/Claes

As you say, but one would have to look deeper to understand what this average figure actually is worth.

It's just that I could think of easier ways of finding out whether the supplier knows how to calculate a Cpk or not, and I would much rather like to see individual Cpk figures. This average could be pushed up, for instance by throwing in results like you suggest in no. 4, and a low average could mean that some tolerances are nigh on impossible to live up to with the best available equipment.

A test of Supplier Responsiveness? Perhaps, but why not test that with something more useful?

Anyway, has anyone else come across this?

/Claes
We have both commented it is NOT the way we would do things. That said, a lot of folks have their own quirky "rules of thumb" in which they have a lot of faith. One lady known to me has a way of checking out a new restaurant which may seem bizarre, but she swears by it:
Before allowing herself to be seated, she goes to the restroom (toilet) and looks in. If it isn't spotless, she leaves.
Obviously, she is not from New York City, where restaurant bathroom facilities can be pretty harrowing even in some otherwise nice-seeming places. She would NEVER eat in a restaurant there.

Hi lydielu
I am not sure it was proposed by your customer? if yes, let it alone just like nonsense, the so-called new conception must be concocted by your customer, as far as know, we can only calculate the monthly CPK for each preocess and then take the relevant measures to improve the processes to achieve the goal of >1.33.
FORGET IT!!
:mad:

Whether or not you like the idea, there is also a statistical question as to just how to average.

As an example, suppose you you produce 1,000,000 pieces of Part A on 10 production lines each with a Cpk of 2, and 1,000 pieces of Part B on one production line with a Cpk of 1.

The average of the two types of part is clearly 1.5.

The average of the 11 production lines is 1.9

The average of the 1,001,000 parts is very close to 2.


Clearly, this could get complicated. You and the customer would both have to know what method you used. Or just use the method that gives the best looking result and make the customer ask if they want details ;)

Tim F


P.S. This reminds me of an old statistics puzzle. Two competing suppliers - Vendor X and Vendor Y - produce the same set of products: Part A, Part B, Part C, .... Vendor X claims that his factory produces fewer ppm defective overall. Vendor Y claims that his factory makes each and every part with fewer ppm defective. Both are correct!

I have been asked the same question by a new customer. I was completely stumped by the question and didn't have a clue as to the answer. However a look on the IsixSigma web site gave me an answer. It is this.

Convert each Cpk to a standard normal deviate; i.e., Z=Cpk*3. Next, you must convert Z to equivalent yield (in Excel use normsdist(Z)). After this, you need to multiply all of the yield values to get the rolled throughput yield ... Y.rt = Y1 * Y2 * ... * Yn, where n is the number of yield values (i.e., the number of Cpks you converted). Then normalize the rolled yield by computting Y.norm= Y.rt^(1/n). Finally, convert Y.norm to a Z value (using Excel) and then divide by 3 to get the final answer ... Cpk.norm = Z.norm / 3. The resulting normalized Cpk is "kind of like an average" but is statistically valid.

Hope this helps.

Since the customer requires you to provide him the CPK average, then he MUST KNOW the concept of it, you can ask him to give you an example or reference, otherwise he maybe only learn the title from others (ha-ha), please let it alone, I often seek the simialar supports from my customers if they require us to do any analysis which we have never heard before. please have a try!;)
Buffalo

Actually, one could infer a lot of things from such a number:

If the number is near or below 1.0, we would seriously wonder why these folks were in business with machines or processes that can't maintain control for the tolerances required by customers.
Wes

Ah my favorite! "tolerances required by customers" assumes that the customer knows what they need.

We are automotive suppliers. We have many features with poor index numbers. We have state of the art equipment. The parts manage to fit on cars and drive down the road. There are ZERO warranty claims. Our competitors have the same situation.

Tight specs loosely enforced = death of Quality

It has always been a pet project of mine to investigate where print specs came from, and to try to understand the engineering analysis behind them.

Especially in the last decade, the big 3 have "right sized" their engineering staff and we tend to find recent grads making prints. "I just copied the last print" is a typical answer to why the spec was selected.

An aggregate index like plant Cpk is of little excitement to me, since it just begs analysis of the underlying data.

A plant Cpk of 1.67 could be hiding a whole lot of poor capability values, offset by some very high values, leading to a false sense of security.

As Claes and I both stated, "this is not a request we would make."

I agree an organization has a duty to question tolerances as part of Contract Review. I burned more than one butt in my time when I found out a customer's draftsman left in four decimal tolerances (0.0000) because that was the default on his CAD program, when the true tolerance required for a product was in the two decimal (0.00) category. I simply told the guy in charge - "You can have it any way you want, 4 decimals or 2 decimals, because our equipment will handle either. The only difference is you will pay nearly twice as much for the 4 decimal tolerances because we have to change our tooling more frequently to account for wear and we will have to use more expensive gaging equipment to inspect the product."
The situation here is different - the supplier in question is obviously not a sophisticated high tech manufacturer used to dealing with the sharks on the Purchasing side of the table who will take every advantage they can of a naive supplier.

You are also correct when you write
An aggregate index like plant Cpk is of little excitement to me, since it just begs analysis of the underlying data.

A plant Cpk of 1.67 could be hiding a whole lot of poor capability values, offset by some very high values, leading to a false sense of security.

My impression of a customer who made such an inane [that's very close to "insane"] request is that he has no idea of what to expect or what to do with the information when he gets it than to plug it into some lame a$$ "supplier scorecard" program foisted on him by an equally lame a$$ consultant.

Convert each Cpk to a standard normal deviate; i.e., Z=Cpk*3. Next, you must convert Z to equivalent yield (in Excel use normsdist(Z)). After this, you need to multiply all of the yield values to get the rolled throughput yield ... Y.rt = Y1 * Y2 * ... * Yn, where n is the number of yield values (i.e., the number of Cpks you converted). Then normalize the rolled yield by computting Y.norm= Y.rt^(1/n). Finally, convert Y.norm to a Z value (using Excel) and then divide by 3 to get the final answer ... Cpk.norm = Z.norm / 3. The resulting normalized Cpk is "kind of like an average" but is statistically valid.

Hope this helps.
I disagree with this method of calculation. Let me explain:
The concept of Rolled throughput yield is usually computed to give visibility to process steps that have high defect rates. It does that measuring the probability that a unit will go through an entire process or procedure without receiving any defects.

It uses the multiplication of probabilities, answering the question "what is the probability that the unit will pass step one, and step two, and step three, etc."
The normalization of the RTY is just a way of detecting the overall yield of the process.

First problem:
The quoted procedure "transforms" cpk in RTY, assuming that all the characteristics are normally distributed, AND that they have the same weight to customer satisfaction.

Second problem:
What is the statistical/probability question that this procedure is answering? I think that this is the only way to give statistical validity to the procedure, but I can't see clearly what this "untransformed Cpk from normalized Yield from transformed Cpk" is measuring.


Just my comments. Any ideas?

thank you all! i only read the conception from a english training material, which is really confusing me. so, i ask you and hope to get real answer.

i never know cpk could be used like that. for all of us, cpk reflects capability for one process,it focus on a "processs",not for a part, let alone a plant. if the processes are different from each other, what real sense the average plant cpk is, only a number reflects plant's capability,but a lot of things could be used to reflect this.

is it used within a group? the group hopes to get average capability of each plant in it?

thank you all! i only read the conception from a english training material, which is really confusing me. so, i ask you and hope to get real answer.Well, I really shouldn't knock something I know so little about, but this plant CPK thing still seems exceedingly strange to me... I can offer but one piece of advice here: Contact the customer and ask them to clarify matters.

Hopefully, they know what they are talking about, and if not, perhaps you'll be able to work something useful out together.

/Claes

Another problem with "average Cpk" is that Cpk depends on specs from a particular customer. If you made the same part for two different customers but the customers had specified different tolerances, then you would have two values of Cpk for the same part!

Tim F