Capability for one sided specification

Regarding item 2 above: The first step is to identify rational subgroups. This will be highly dependent on your particular process.

Include similar things into the subgroup, put dissimilar things between the subgroups. In essence, this means do not mix process streams (i.e., different machines, mold cavities, etc.) unless you can statistically show that the different process streams are not different.

Next, determine the size of the subgroup. If there is no logical basis for a rational subgroup that subgroup size may = 1. Otherwise, select a subgroup size that economically balances your need to detect a process shift with the risk of nuisance adjustments. If your process is stable and highly capable, use a smaller subgroup size. If less capable, you may want to use a larger subgroup size. Do not automatically use a subgroup of n=5. The basis for this was that it was easy to calculate the mean by taking the sum of the 5 values, multiplying it by 2 and shifting the decimal 1 space to the left. Prior to having calculators this was a great solution.

Finally, determine the frequency of sampling. This also should be rationally based. What is the major source of process variation? Material? How often do you change batches/lots of material? Every 2 hours? Set the frequency of sampling at 2 hour intervals. Is the major source of variation due to ambient temperature? Select a frequency based on how rapidly the ambient temperature changes.

 

And regarding item 1 above. This is very controversial. The easy way, which is not necessarily the best way, is to just use Cpk and forget about Cp. However, this has the potential to drive some bad behavior. For example, you have a maximum specification with no minimum, but it is more economical to hold to a target value. It is usually easier to move the process average than to reduce the variation, so chasing the Cpk may drive a more costly behavior of moving the process average below the target rather than reducing variation.

 

Why are you calculating a capability index? does your customer require it? If so, what is the requirement specifically. See Bob Doerring's resource on this topic. (in the resource section - tab at the upper left)
If you don't have a customer requirement, consider NOT doing it. simply plot your data against the specification limit and see where your process is relative to the specification. if you have defects, understanding the actual defect rate also has value...

 

Boundaries are natural limits as opposed to a specification. GD&T specifications that take absolute values rather than negative numbers (e.g., position, roundness, etc.) have a natural boundary of zero even though there is a 1-sided max spec.

 

I think Bob meant that you have a boundary plus a specification. His Roundness example would have a lower boundary of zero plus a maximum roundness spec.

 

both sides are bounded for fill volumes