Cpk roughly double Ppk How to fix this?

So the sheer number of knives/lanes doesn’t really lend itself to a classical Cpk/Ppk study with subgroups.

You are better off calculating Cpk from the short term variation (between adjustments) and Ppk from the long term variation (over several adjustment periods). You would use the total standard deviation of all points in each case. The differentiator is time and not within & between subgroups. Alternatively you might calculate the total variation within each lane and ‘pool’ it. This will be your short term SD. then use the grand average of all of the lanes as your average. This will be the ‘best’ ‘capability’ (Cpk) of your process. This is what you would get if you centered all of the knives together.

Hello,

I ran the study with three subgroups (three lanes) and achieved passing results for CpK and Ppk. I appreciate all of your help with this study. Thank you!

What are you measuring? A diameter? Length? Were there any process adjustments anywhere within the data set? What kind of process is it? (Maybe this IS stable for that process! It depends.) There could be measurement error - not gage error (big difference).

Christopher Lear said:

Hello,

I ran the study with three subgroups (three lanes) and achieved passing results for CpK and Ppk. I appreciate all of your help with this study. Thank you!

Click to expand...

Why not use a deviation from 0 to look at process capability? You set up for 4.0" (for example). Some may be 4.0005", some maybe 3.9995". If you are doing a capability study to tolerance, this would show short and long term capability to hold a tolerance you guarantee.
Maybe I'm wrong, but thats my thinking.

This is the classical web process problem. People look at samples across the web, then compare them over time. This is not correct. All data needs to be a function of time. So, looking at each lane individually, e.g. with the samples being perhaps 3 every 10 minutes along that lane, is the correct way to look at it. Then, whichever one is the worst Cpk is your process capability. And, Bev D is correct, if the data shows the samples to not be random and independent, then standard Cpk and Ppk calculations may not apply. So, it is important to determine if a process is random or a function.