Process Capability Study - Aggregate vs. Individual Processes

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I agree, random is not good. Random only as in not rooting through to find a good one - just whatever part shows up at the sampling time, take it and report it. A sample at specific times tends to be far more effective. A large enough sample to detect the current variation is also handy.
 

v9991

Trusted Information Resource
I agree, random is not good. Random only as in not rooting through to find a good one - just whatever part shows up at the sampling time, take it and report it. A sample at specific times tends to be far more effective. A large enough sample to detect the current variation is also handy.

answer then is a "stratified sampling" which is an extension of the "specific times"...

for routine manufacturing, practice is that, teams from manufacturing & quality have different frequencies with predetermined number of samples.(this is determined by the no. of stations etc)

further, for qualification purpose, the definition of stratified sample and its assessment runs little more deeper.

in certain instances, the no. of parameters are different for different stages and samples., (depending upon variability or criticality of tests)

so this represents, entire process time, no. of samples, no. of stations, etc:2cents:
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Yes, there are a lot of things to consider when determining sample size, frequency and technique - some of which depends on the process output distribution and variation. I didn't want to go any deeper into that to wander too far away from the original issue of individual streams or aggregate. How much effort one puts into a minimal (or miniscule) value activity as determining capability index should be considered.
 

Bev D

Heretical Statistician
Leader
Super Moderator
a narrowly scoped response to the point of random vs 'stratified' sampling.

too often this rather straightforward concepts get confused even though there is ample discussion in the literature from well respected quality professionals who are/were solidly grounded in industrial statistics.

Acceptance sampling on completed material is done via random sampling. and random means random. not the convenience sampling that too many people use, even though they will claim the sample is randomly selected.

In-process monitoring and control (including SPC, PreControl, etc.) is done using stratified sampling. in other words the sampling is planned: small sample sizes taken at natural process change points. There is nothing random about these samples.

These two 'tests' have very different purposes and are therefore very different in their structure.

Capability studies should be done from 'in-line monitoring samples' but too many hacks have gotten their hands on this and tried to make it 'easy' completely gutting it's value or relation to reality - but that is a different thread
 
N

ncwalker

So I turned to Monte Carlo ... and ran a bunch of simulations to see the effects. With the understanding that my overall goal is to minimize work, I wanted to know if an aggregate study could be used to evaluate all the different subprocesses. I am pretty sure it can, if you are smart about it. You have to do a bit more than just grab parts out of the shipping boxes. And, you have to be prepared to deep dive outliers. But I think it is possible to START with an aggregate study, the answer to this study will be:

1) Everythings OK.
2) We need to look at THESE subprocesses for mean shift, or excessive variance before we sign off.
 

Attachments

  • Aggregate Process Capability - Open.pdf
    1.3 MB · Views: 274

howste

Thaumaturge
Trusted Information Resource
So I turned to Monte Carlo ... and ran a bunch of simulations to see the effects. With the understanding that my overall goal is to minimize work, I wanted to know if an aggregate study could be used to evaluate all the different subprocesses. I am pretty sure it can, if you are smart about it. You have to do a bit more than just grab parts out of the shipping boxes. And, you have to be prepared to deep dive outliers. But I think it is possible to START with an aggregate study, the answer to this study will be:

1) Everythings OK.
2) We need to look at THESE subprocesses for mean shift, or excessive variance before we sign off.

Your attachment's an interesting read. Thanks for sharing.
 
N

ncwalker

No worries.

I am toying with making a general Excel model where you can click buttons and tweak the distributions, etc. The purpose being to explain capability studies and control charts, etc. Like a tool that would make the graphs, etc, that a user could just go in with a few settings, etc.
 

howste

Thaumaturge
Trusted Information Resource
If you did that I would have to steal it so that I could explain the concept to people. :cool:
 
N

ncwalker

That's the point. I want to make something to do exactly that. :)

(My Excel Kung-Fu is pretty powerful ...)
 

Proud Liberal

Quite Involved in Discussions
Don't know where I got these from (I'd like to credit whoever did them).
 

Attachments

  • Control Chart Running Sigma with Histogram.xlsx
    1 MB · Views: 233
  • Control Chart XBar R with Histogram.xlsx
    1 MB · Views: 232
Top Bottom