we are currently using p charts to monitor and control attributes for glass manufacturing. We push about 2,000 discrete glass panels in one day (2 shifts).
we are currently spot sampling 5
glasses a day.
is this sufficient sampling?
how many samples do we need to measure per day or shift to make sure with 99% confidence that the numbers are giving us the good "state of quality".
Thanks.
In order to answer the question of "is this sufficient sampling" we have to have some discussion of risk versus costs. First, p-charts (Pass/Fail) data are not as sensitive as actual measurement data. I assume that you are testing to some form of visual inspection criteria, not a measurement. If you are testing a measurement against certain limits, you would reduce the risk of missing a shift in the process by trending the actual measurement data on either an Xbar R set of charts, or X individual moving range. If this involved machining, you may want to look at Bob Doering's work here on
SPC with tool wear.
Assuming p-charts are appropriate, we then have to answer what do you mean by "good 'state of quality'"? For Pass/Fail sampling, we usually say things like "I want to be 95% confident that no more than 5% of the items are defective". By the way - it takes 59 samples with no failures to be able to make that statement.
You have said you want to be 99% confident of something. Let us say that is that no more than 5% are defective. I can only make that statement after completing 90 samples with no failures.
There are fancy statistics (operating characteristic curves) to help answer this question, but just as a rough estimate, it could take you 90/5 or 18 days to detect a shift to a defect level greater than 5%. Can you wait 18 days?
I'd suggest that 5 per day is probably too small if you are putting out 2,000 per day. I'd up that to about 25 which would give a better daily percentage to plot (with only 5 samples, you have only 6 possible results each day - 0%, 20%, 40%, 60%, 80%, or 100% defective).
