Use of Customer Specifications to calculate Cpk

[Dan M]

Thank you for all your responses. They have been helpful.

 

[kgott]

I am by no means an authority on SPC but from what I was taught and Demining also says this, but the control limits should be set on the basis of 3 sigma not on customer specs as the capability of the process is whatever the data is telling you it is.

As in safety, you can state any target you like but you will get nothing better that what the process is capable of delivering. Is that not the case?

 

[Jim Wynne]

I am by no means an authority on SPC but from what I was taught and Demining also says this, but the control limits should be set on the basis of 3 sigma not on customer specs as the capability of the process is whatever the data is telling you it is.

Cpk is a comparison of the specification limits to the process mean, and allegedly provides an idea of the likelihood of violating specification limits. For the same data set, the Cpk number will increase as the tolerance band widens, and will decrease as the tolerance band narrows. The OP is questioning which set of specification limits to use for the same data set--his production tolerance band is narrower than the customer's specified tolerance band, thus if he uses his internal specs the Cpk number will be lower. It has nothing to do with control limits, per se.

As in safety, you can state any target you like but you will get nothing better that what the process is capable of delivering. Is that not the case?

It's not a question of targets, but of artificial limits--the tolerance band.

 

[bobdoering]

I am by no means an authority on SPC but from what I was taught and Demining also says this, but the control limits should be set on the basis of 3 sigma not on customer specs as the capability of the process is whatever the data is telling you it is.

Traditional Shewhart charts only seek signals from the random variation found in processes that exhibit outputs of independent, random variation unless a special cause exists. So, they have nothing to do with capability - but simply process monitoring.

But, that is Shewhart charts - which are only a subset of statistical process control. Not all charting methodologies are separated from the tolerances.