Surveillance Sampling Test - Determining Sample Size

S

santiko

Hei i'm san,

im new in this forum. already read a lot of topic in this forum, but this my first post.

currently i have a bit a problem in my surveillance sampling frequency.

I have a surveillance sampling frequency with 1 PCS PER SIZE every 15.000 PCS PER SIZE production or 1 PCS a Year PER SIZE. This is the old standard and did not have any statistical calculation to explain where the number come from (ISO/TS 16949). Now, i need to find out with the correct way.

Another information is the yearly production is around 9.000.000 PCS a year with random quantity per size (posible 0) and we have about 2.000 unique size. This is destructive test and cost a lot.

I have result data from previous surveillance for each size, but not much, only 5-10 data. But my idea is using as familiy size to get larger data.

The problem is,

What kind of statistical calculation approach to determine the correct surveillance sampling frequency with this case? I'm looking for the direct formulas or method to do.

I hope everyone that have an experience can help.

Thank you.

San.

Romvill2002

Involved In Discussions
Determining Sample Size-Medical Device

Hello All,

Can you interpret below determination of sample size. What would be the confidence level if 1 out of 10 samples failed?

Sample size of 10 was purposed by test engineers for the verification of the pressure switch and regulator pressure components replacement in FACSCelesta system. There were no historical or pilot data available for sample size estimation prior to the verification study. Therefore by using the vendor specification for pressure switch along with BD internal verification acceptance criteria, the power for the purposed sample size of 10 is estimated to ensure there is sufficient power to meet the test requirement with the following assumptions.
Vendor Pressure switch specification: 10 psi ± 1
BD pressure switch verification criteria: 10 psi ± 5
Assuming the vendor also tested with sample size of 10, with mean center at 10 psi, the largest standard deviation is estimate to be 1.35 .

N Mean StDev SE Mean 95% CI T P
10 10.000 1.350 0.427 (9.034, 10.966) 0.00 1.000

Using theoretical calculated standard deviation of 1.35, the power for the purposed sample size of 10 is estimated to be greater than 99% assuming the difference of the test is at most half of allowable drift (2.5) from the target mean of 10 psi with alpha of .05
Calculating power for mean = null + difference
α = 0.05 Assumed standard deviation = 1.35

Sample
Difference Size Power
2.5 10 0.999308

Mike S.

Happy to be Alive
Trusted Information Resource
Hei i'm san,

im new in this forum. already read a lot of topic in this forum, but this my first post.

currently i have a bit a problem in my surveillance sampling frequency.

I have a surveillance sampling frequency with 1 PCS PER SIZE every 15.000 PCS PER SIZE production or 1 PCS a Year PER SIZE. This is the old standard and did not have any statistical calculation to explain where the number come from (ISO/TS 16949). Now, i need to find out with the correct way.

Another information is the yearly production is around 9.000.000 PCS a year with random quantity per size (posible 0) and we have about 2.000 unique size. This is destructive test and cost a lot.

I have result data from previous surveillance for each size, but not much, only 5-10 data. But my idea is using as familiy size to get larger data.

The problem is,

What kind of statistical calculation approach to determine the correct surveillance sampling frequency with this case? I'm looking for the direct formulas or method to do.

I hope everyone that have an experience can help.

Thank you.

San.

As you say, what you are doing now has no statistical basis. You need to think about what you want your sampling to accomplish. Is there a problem you need to solve? Is your customer getting bad parts (how many bad/good?) and you want to prevent that? Do you have any estimate of the process capability now?

I don't think you have given enough information for anyone to give any suggestions.

S

santiko

As you say, what you are doing now has no statistical basis. You need to think about what you want your sampling to accomplish. Is there a problem you need to solve? Is your customer getting bad parts (how many bad/good?) and you want to prevent that? Do you have any estimate of the process capability now?

I don't think you have given enough information for anyone to give any suggestions.
Thanks for the feedback.

This sample is final product test to comply with regulation and safety assurance. There is no problem to solve and no customer getting bad part until now. But yes, I need to prevent that.

The regulation didn't state any requirement for sampling. But we need to state by ourself. The sampling approach can be quantity basis or time basis.

All special characteristic in our process manufacturing have Cpk more than 1.33.

This problem is a finding from regulation audit. I already discuss with a lot of people and still didn't get a good answer.

If you need more detail, let me know. Thanks

Mike S.

Happy to be Alive
Trusted Information Resource
I don’t know what the exact wording of the finding was, but I will take a guess and say perhaps the requirement was along the lines of “sampling plan must be statistically valid”, etc. but you had no proof of statistical validity.

One way to comply could be to pick a published sampling plan, one that has statistics such as OC curves published along with it, which meets your needs as determined by experts who understand the risks and consequences associated with failure. One option you could consider is to look at C=0 sampling plans published by Nicholas Squeglia (look at ASQ Quality Press for a copy).

I hope this helps at least get you started.