Individual & MR chart - Ppk or Cpk?

P

pagnonig

Dear All,

I've been through a lot of posts regarding Cpk and Ppk and the short/long term discussion on them.
Now, I would like to be sure I've catch the right practical meaning for both.

I'm collecting individual measurements to be plot in a I-MR chart.
having USL and LSL limits I would compute Ppk for this process using the Sigma (n-1) calculation and not the estimator R.avg/d2 because in this case I do not have subgroups.

Therefore I would say that Cpk is meaningless for Individual readings.

Anyway, I've noted that both the estimated and the calculated standard deviation tend to be the same value as the number of individual measurements becomes 50-60.

Any comments on my approach?

Thank you!
Giuseppe
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
It would be handy to know the process to ensure we understand the applicable distribution. :cool:
 
D

Duke Okes

Dear All,I've been through a lot of posts regarding Cpk and Ppk and the short/long term discussion on them.
Now, I would like to be sure I've catch the right practical meaning for both.
I'm collecting individual measurements to be plot in a I-MR chart.
having USL and LSL limits I would compute Ppk for this process using the Sigma (n-1) calculation and not the estimator R.avg/d2 because in this case I do not have subgroups.Therefore I would say that Cpk is meaningless for Individual readings.Anyway, I've noted that both the estimated and the calculated standard deviation tend to be the same value as the number of individual measurements becomes 50-60.Any comments on my approach?
Thank you!Giuseppe

Cpk vs. Ppk has more to do with whether all sources of variation have been included in the study, usually thought of as how many data points, or how long a time frame, have/has been used to do the calculation. It does not matter whether sigma comes from Individuals or Subgroups.

If you find that n-1 and R-bar/d2 give you the same number it probably just indicates that the process is in control.
 
P

pagnonig

It would be handy to know the process to ensure we understand the applicable distribution. :cool:

Normality test on available data passed successfully.
I'm collecting thickness measurement of a cable layer (like a tube wall).
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Normality test on available data passed successfully.
I'm collecting thickness measurement of a cable layer (like a tube wall).

If I had to choose, I would use the sigma of the individuals - hoping that my frequency was sufficient to catch the bulk of the variation. As a curious note, how did you arrive at your data acquisition frequency?

But, as Duke Okes indicated,the difference should not make an overwhelming impact on the final analysis. I would not lose sleep over either one.
 

Miner

Forum Moderator
Leader
Admin
Normality test on available data passed successfully.
I'm collecting thickness measurement of a cable layer (like a tube wall).
Is this an extrusion or drawing process? If it is and you have not already done so, I recommend performing an analysis for autocorrelation. If autocorrelation exists, the period between measurements should be longer than the period of autocorrelation.

Measurements within the period of autocorrelation are essentially pure measurement error.
 
P

pagnonig

Is this an extrusion or drawing process? If it is and you have not already done so, I recommend performing an analysis for autocorrelation. If autocorrelation exists, the period between measurements should be longer than the period of autocorrelation.

Measurements within the period of autocorrelation are essentially pure measurement error.

Thank you everyone for your support!

This is an extrusion process.
measurements are taken at the start of each produced reel (2000 meters).

I think that measurements are so distant between each other that autocorrelation should be a low risk.

anyway, how can I perform such analysis? Coudl you give me any reference on this?
thank you again!
:)
 
D

Darius

I'm collecting individual measurements to be plot in a I-MR chart.
having USL and LSL limits I would compute Ppk for this process using the Sigma (n-1) calculation and not the estimator R.avg/d2 because in this case I do not have subgroups.

Therefore I would say that Cpk is meaningless for Individual readings.
Giuseppe

Cpk is not meaningless, in the case of Individual and moving range the estimate of variation used to construct control limits is the same as the one's used to calculate cpk.

SIGMA_ESTIMATE =MR_average/d2 = MR_average/1.128

UCL = MEAN+ 3* SIGMA_ESTIMATE = MEAN + 3/1.128 * MR_AVERAGE= MEAN+ 2.66* MR_AVERAGE

LCL = MEAN- 3* SIGMA_ESTIMATE = MEAN - 3/1.128 * MR_AVERAGE= MEAN- 2.66* MR_AVERAGE

CP= (USL - LSL)/(6*SIGMA_ESTIMATE)
CPK = MIN(MEAN-LSL, USL-MEAN)/(3*SIGMA_ESTIMATE)
 
Last edited by a moderator:
Top Bottom