Hi all!

Could someone help me with the following:

How much does CPK>1.33 represent in precentage of good pieces? Is there a matrix where I can find these values at Cpk>1; Cpk>1.67; Cpk<1; Cpk=0?

Thanks in advance! :bigwave:

I've only got a minute before I've got to go, so I'll just get you started...

First, in order for the statistics to be meaningful, you need to have a normal process that is in control. Now, the percentage can actually be different depending on if the distribution is centered or not. A perfectly centered process with Cpk = 1 will have twice as much (theoretical) nonconforming product as one with a Cpk = 1 that is not close to being centered.

What you need to get the percentages is the Z value(s) and a "standard normal" table to look up the results. When you look at the formulas you find that basically Z = 3xCpk.

I'm attaching a standard normal table - hopefully you can figure it out from here. If not I'll be back later...

Here is a very clear article with a table of the results
http://www.symphonytech.com/feature.htm

article Measuring Your Process Capability

Great link, Howard. I was trying to explain how a watch works, when all Tattva asked for was the time... :bonk:

Howard, it's a great link, tanks for the tip

For the ppm to Cpk question, both articles, Capability and Six Sigma look like a check mate (no more can be told).

From Six Sigma article (in Howard's link)

Cpk =0.8406+(29.37-LN(ppm)*2.221)^0.5

or

ppm = EXP(-((Cpk-0.8406)^2-29.37)/2.221)

But keep on mind what Don Wheeler said in te point 8.4 from Advanced Topics in Statistical Process Control, 1995.

"It's impossible to convert a capability ratio into a fraction of nonconforming product without using some probability dsitribution in the convertion... of course
, the traditional assumption is that the data are normally distributed.... in most cases the uncertainty in the fraction nonconforming will be greater than the refinement offered by such convertion"


:smokin:

Hi all!

Could someone help me with the following:

How much does CPK>1.33 represent in precentage of good pieces? Is there a matrix where I can find these values at Cpk>1; Cpk>1.67; Cpk<1; Cpk=0?

Thanks in advance! :bigwave:

I'm going to step out on a limb and say that Cpk, in and of itself, will not tell you how many good parts you have. Cpk only measures process centering.

Case in point; I have a process that has a Cpk= .8 , Cp = 1.66, Zmin = 2.386, Spec avg = .125, Process avg = .1197
There are zero nonconforming parts. All data entered was within spec.

Using the Z-value would show a "potential" for approx. .85% nonconforming.
When I want to relate percent defective to number of parts I use PPM.

I'm going to step out on a limb and say that Cpk, in and of itself, will not tell you how many good parts you have. Cpk only measures process centering.

Case in point; I have a process that has a Cpk= .8 , Cp = 1.66, Zmin = 2.386, Spec avg = .125, Process avg = .1197
There are zero nonconforming parts. All data entered was within spec.

Using the Z-value would show a "potential" for approx. .85% nonconforming.
When I want to relate percent defective to number of parts I use PPM.
The usefulness of statistics comes from measuring samples and then making inferences about the population. My question is (assuming the process is normal and in control), did you measure every part in the population? If not, then I would guess you really do have nonconforming product, you just didn't happen to find it in the samples you took.

The usefulness of statistics comes from measuring samples and then making inferences about the population. My question is (assuming the process is normal and in control), did you measure every part in the population? If not, then I would guess you really do have nonconforming product, you just didn't happen to find it in the samples you took.


I think that's what I inferred in my response when I said the reaults of the Z-value provided the "potential" nonconformances.
In response to Tattvas' question "can Cpk be related to Nonconformances"? I said no, and I stick wth that.

Sam, I very much respect your opinion, as IMO you've proven yourself to be very knowledgeable here at the Cove. I'm just trying to figure out what you're getting at. Here's what I've finally decided...

Your point: You don't know exactly how much defective product you have based on Cpk
My point: You can get a pretty good estimate if you do it right

Does that sound pretty close?

Sam, I very much respect your opinion, as IMO you've proven yourself to be very knowledgeable here at the Cove. I'm just trying to figure out what you're getting at. Here's what I've finally decided...

Your point: You don't know exactly how much defective product you have based on Cpk
My point: You can get a pretty good estimate if you do it right

Does that sound pretty close?

That sounds close enough to me.

i found this table, i can't remember where!
:o

Anilo

i found this table, i can't remember where!
:o

Anilo

This appears to come from the six sigma school of thought.

Hi.

maybe this can help you.

It comes from www.symphonytech.com

Bye.
Anilo
P.S: someone has ever heard about Statistical precontrol?
We assemble a very small batch (500 hundread pieces), 1 time in a month.
We need less than 1 day to assemble all of that.

Our Customer ask us to apply a very formal SPC, with an exctraction of 50 sample.
But this for us is an excessive cost (we assemble electronic engine control unit).

So i proposed to exctract 50 sample when we are in pre-series, and to use precontrol when we are in series production, to face the costs growth.

But from the pre-series to the series it could be passed even 3-4 months, and in the meanwhile we use the line to assemble some other, completely different, products.

Can you suggest a cost effective approch?
Thanks in advance.
Anilo

[Edited by Atul: Just corrected a typo in the link.]

This appears to come from the six sigma school of thought.

Yes, it assumes a 1.5 sigma process shift over a period of time. I have one somewhere that doesn't make the assumption and will post it for comparison if I can find it.

Dave

Hi.

someone has ever heard about Statistical precontrol?


Anilo,

The link you mentioned - symphonytech -- has an article on pre-control called "The Power of PRE-Control". Juran's Quality Handbook - 4th edition I think, also had a nice section on it. You can review it and decide if it would be helpful for you.

Anilo,
Also use the forums search function. We have some good discussion threads on pre-control.

Anilo,

The link you mentioned - symphonytech -- has an article on pre-control called "The Power of PRE-Control". Juran's Quality Handbook - 4th edition I think, also had a nice section on it. You can review it and decide if it would be helpful for you.

Yes, thanks, i read that article.

I would like to know if someone has practice about pre-control, not just theory.

According to the article, and some Professor i heard, i can use precontroll if i have a continuous production.

My production seems to small and too discontinuous for thinking that a statistical control in pre series can be useful and still valid after three months, when series starts.
Exctracting just 5 pieces at the beginning of series, saying that the process was in statistical controll 3 months before is an hard position to mantein with Audi!!

Thanks all.
Anilo

Anilo, I found it usefull if...

+You care more about the specs than to reduce variation, the SPC don't care if you are out of specs or not.
+When the variation within the specs limits does not affect the product quality, nor the production troughtput (number of units/time unit).
+When we can't reduce variation, because the source of variation is in another process that we have not direct inference and with wich specs are fixed.

Your problem seems to me like a short run (small and too discontinuous).
:thedeal: