Sample Subgroup Size for NP/P chart - nPbar>=5

[Stevenli]

Fellows,

For Subgroup size of NP/P chart, in SPC manual, it's just simply stated that "Since the control limits are based on a normal approximation, the sample size used should be such that nPbar>=5." I am wonder whether anyone can share more experience on how to determine the sample size for each subgroup.

My real life is that one of feature to be monitored is a dimension but with go/no go gauge to check, then the inspector will take 5 pcs per half day. And # of rejected is always as 0, so he used those data to map the chart and always looks good.

But I have a concern about this sample size and the process itself. So would appreciate your any comments.

Thanks
Steven

 

[Steve Prevette]

Since you are dealing with a dimension, you would be much better off measuring the dimension and plotting the data on a control chart. You lose a lot of information by going "go no go" and yes, since you have a lot of zeroes, sounds like you are potentially losing a lot of information.

The reference you found on nPbar>=5 sounds like a good rule of thumb, but you obviously aren't there if you are only sampling 5 per day and usually getting zeroes. What is your average number of samples between failures?

 

[Matt33]

You need the sample size sufficiently large so that there is a good chance of catching defects. As you have seen, collecting five parts and checking them always shows 0 defects, thus this is not a good solution. Some possible options include:
1.) Increasing your sample size from 5 to a much larger number such as 25, 50, or 100.
2.) Instead of measuring the count or proportion of defective parts, you might try counting the parts produced since the last time you saw a defective (no-go) part. This would lead to a rare-event chart such as a g-chart.
3.) Stop doing what you are doing since there is cost in obtaining and measuring parts and you are not getting value from the results.

 

[Stevenli]

Very useful reccomendation, thank you. In this case, rare-event chart could be a good option.

 

[Steve Prevette]

Very useful reccomendation, thank you. In this case, rare-event chart could be a good option.

True, very true.