Sample Size for Short Term & Long Term Process Capability

Dear All

It is said that sample size for short term process capability study shall be min 30 consecutive readings with subgroup size 1 & long Term process capability study it should be min 100-125 readings, with subgroup size 2-4. But I could not find any reference.

Looking for valid reference from AIAG manuals, ASQ OR SPC/SQC book.

Pl help

Kedar

The 30 is fairly common for initial capability studies (e.g. PPAP), but it does depend on the process. On the long-term studies, it is not a question of number but of time. You want enough time to pass so that all sources of variation have an opportunity to be present (e.g., different suppliers, different operators, different temperatures, etc.). That might be a month, it might be much longer. A "long term" study with 125 widgets made right in a row is not long term.

kedarg6500 said:

Dear All

It is said that sample size for short term process capability study shall be min 30 consecutive readings with subgroup size 1 & long Term process capability study it should be min 100-125 readings, with subgroup size 2-4. But I could not find any reference.

Looking for valid reference from AIAG manuals, ASQ OR SPC/SQC book.

Pl help

Kedar

Click to expand...

As I have readed, 100, sample size 5 means 25 cosecutive samples (of 5 as sample size), but what is on this number?..., well, I think, It's just normality asumption (plus something else to be sure), rememeber that the normality test has a table until 60 samples, so with 120 samples (the double), is just to be sure that normality is accomplished.

You can just get rid of outliers and obtain the sample (not contaminated by outliers), and so the parameters must represent the process. There are several outilers detection methods, like Chauvenet, Grubbs, interquartile distance, or something else. calculate non parametric capacity indexes or take for grantes normality (if the process has demostrated such behaviour).

note: I have done studies, were I used outlier elimination, and obtain the mean with a few data points (less that 200) that I need 10,000 or more samples to obtain (without outlier elimination).