Ppk for Combined Data Sets

sinned

Registered Visitor
I've got a set of data from the process engineer with its Ppk calculated to be 1.80. However, it turns out that the set of data are a combination of four sources, each of which from ID measurement of an individual hole. (As usual, they can't be distinguished anymore)

In this case, what can be said about the stdev and Ppk for those individual holes ? Should the Ppk for each hole be ~1.8 at worst case ?
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I've got a set of data from the process engineer with its Ppk calculated to be 1.80. However, it turns out that the set of data are a combination of four sources, each of which from ID measurement of an individual hole. (As usual, they can't be distinguished anymore)

The point of Ppk is to see if the process is centered and falls within the specifications. That said, how does combining four sources represent a "process"? What question would that answer? A "Grand Ppk" of a part? I suppose if it passes, the lesser levels of capability should be academic. But a process needs to first be analyzed with time-order data plotting (both the highest and lowest of each part's diameters, in the case of a circular feature), then generalized with capability indices – if applicable.

The other point that arises from data being presented in this manner is the training of the process engineer on capability. It sounds like they barely know the basic concept of rubber stamping Ppk, and very little about the underlying concepts. Ultimately, they are not using the tool to make good decisions.

In this case, what can be said about the stdev and Ppk for those individual holes ? Should the Ppk for each hole be ~1.8 at worst case ?

I believe that would be the case, but the other question is whether Ppk is even applicable. What is the process making the hole? How is the process controlled (SPC)…or is it? Is the measurement system accurate to truly detect the variation?
 

Bev D

Heretical Statistician
Leader
Super Moderator
mathematically the individual Ppk values can't be greater than the whole for that data set.

IF you plot the histogram or trend the values and see distinct peaks that might be an indication that there is a difference in the hole IDs but this difference - if it were to exist -would over estimate teh standard deviation giving you a worse overall Ppk...
 

sinned

Registered Visitor
This is also what I've observed: when individual data set are combined into a mix of different data stream, the Cpk would always get worse than calcuated seperately.
Can i argue that if i am getting a mix data set with Cpk>1.33, the individual data stream will not worse than that figure ?
 

sinned

Registered Visitor
Yes, bobdoering. Thanks.
I can't say the process engineers understand SPC or Ppk. But when they heard that 100 ID data Ppk has to be done in few days on a part with 4 holes, that was the report turned in to the quality guy right before the deadline.
 

Bev D

Heretical Statistician
Leader
Super Moderator
This is also what I've observed: when individual data set are combined into a mix of different data stream, the Cpk would always get worse than calcuated seperately.
Can i argue that if i am getting a mix data set with Cpk>1.33, the individual data stream will not worse than that figure ?

You don't have to argue it - you can state it with confidence
 

Bev D

Heretical Statistician
Leader
Super Moderator
Yes, bobdoering. Thanks.
I can't say the process engineers understand SPC or Ppk. But when they heard that 100 ID data Ppk has to be done in few days on a part with 4 holes, that was the report turned in to the quality guy right before the deadline.

On the other hand, this statement makes the confidence in your true capability an issue of practicality rather than statistical math. 100 parts in a short time gives almost no practical confidence in your stability or capability over time. And if your sample is composed of 4 holes across 25 sequential parts you have even less.

Of course it sounds like your organization is just 'checking the box' of a short term PPk study for your customer and you really aren't trying to understand your process?
 
Top Bottom