Hi All,

Haven't posted in quite some time but here goes.
This is a query for a colleague so please bear with me as the details are a little thin.

I need to calculate a sample size for a population of approximetaly 5000 items to give me 95% confidence that 95% of the items will be within specification.

It is for a continuous process that generates 5000 pieces every 30 minutes. I want to carry out IPC checks at this interval.

Can anyone point me in the right direction as to how to calculate the IPC sample size.

Many thanks

Pat

I need to calculate a sample size for a population of approximately 5000 items to give me 95% confidence that 95% of the items will be within specification.

It is for a continuous process that generates 5000 pieces every 30 minutes. I want to carry out IPC checks at this interval.

IMHO you are looking at this a little backwards. First, you need a capability study to understand what the variation is. Is it normal? Is it non-normal? Understanding the rate and distribution of the variation of the process will direct you to the sampling size and frequency. You may need to do it every 10 minutes - or every 2 hours. There is not enough information here to tell.

Do not consider sampling plans - those are for aggregate populations, randomly sampled after assuming the controls in place to make the parts were adequate (as in incoming receiving or final inspection - unless that is what you are referring to. It did not sound that way from the question, you said IPC.)

Check this thread ... http://elsmar.com/Forums/showthread.php?t=33830

For a random sampling and a go/no-go test, then you need 59 item with no defects to assure a 95% confidence that at least 95% of the items are good. The sample size plays very little role here as long as it is well above 59.

If you can get an actual measurement of the variable and can assume a distribution for the data, then you can most likely get by with a smaller sample size. However, it is important to remember that the normal distribution is an idealization -- many real-life situation are no where near a normal distribution, and even if they are close to normal, the tails are the most difficult to be certain of. If the tails are off a bit from normal, you could end up with way more or way fewer defects than you would predict from a normal distribution calculation.


Tim

For a random sampling and a go/no-go test, then you need 59 item with no defects to assure a 95% confidence that at least 95% of the items are good. The sample size plays very little role here as long as it is well above 59.

That is statistically correct, but I do not recommend that approach for in process checks. Great for after the fact, or if the process offers no control other than sorting after the fact. But this approach offers no process control at all - which I would hope would be part of the goal of in process checks. It is better than nothing, but accepting 5% bad is already weak even for a final inspection. Capability study and distribution analysis is better than rubber stamping a sampling plan for IPC, in my book.

That is statistically correct, but I do not recommend that approach for in process checks. Great for after the fact, or if the process offers no control other than sorting after the fact.

Exactly. If the OP can post their data in time sequence it would be very helpful.

My recommendation is for the OP to ask their colleague to measure 3 sequential units every 30 minutes - 2 hours depending on how frequently teh process can change or is known to change (without any other knowledge there is no way for me to be more specific, we must have some observational data before we can begin to form hypotheses or draw conclusions ). This data should be collected over a period of time. (If the OP posts the first day's data, and a few more details about the process we could probably give advise as to how many days the data should be collected and if we could reduce the frequency...)

In process sampling is much more effective than post process inspection...but it does take more effort, analysis and planning to create the sample plan...