Hi all
I was wondering if someone could maybe help. We are a medical device manufacturing company that needs to revise our sampling plans on our finished products. At the moment we have a AQL of 1% (which we were advised might be a little high), by reducing the AQL the amount of samples for testing increases.
We are currently using ISO 2859-1 for our sampling plan and according to this if we use Inspection level II and an AQL of 1% we need to send 13 samples for destructive testing, however the problem with this is that we make batches of quantity 20.
Does anybody have any advise?
Thanking you in advance.
I was wondering if someone could maybe help. We are a medical device manufacturing company that needs to revise our sampling plans on our finished products. At the moment we have a AQL of 1% (which we were advised might be a little high), by reducing the AQL the amount of samples for testing increases.
We are currently using ISO 2859-1 for our sampling plan and according to this if we use Inspection level II and an AQL of 1% we need to send 13 samples for destructive testing, however the problem with this is that we make batches of quantity 20.
Does anybody have any advise?
Thanking you in advance.
Let's start from the beginning. You validated your process originally at a certain risk level. Assuming you passed, you have confidence that your process is capable of meeting the requirements at that level of risk. With small batch sizes, you cannot commit a large sample size to destructive testing, and it is inappropriate to do sampling on small lot sizes as the AQL is a long term process average. So how do we approach this problem? Since you said you use 2859-1, it is an attribute test, can it be converted to a variable output? Are your lots of 20 truly independent? If not, can I do sampling across multiple lots to gain confidence? Can a sample be tested more than once (in other words, if it is destructive like a pull test, can I cut my sample and test it twice)?
Now, the advanced part of the discussion. Sampling plans are risk based concepts which includes both the AQL and LTPD. If we exhausted your entire lot of 20 units, we would only have 95% confidence with 86% reliability, but of course we have 100% confidence since we did 100% inspection. So sampling plans are not intended for small lots as their OC curves to not reflect the actual state. You can use the Hypergeometric, but I doubt that will help.
In general, you must ask yourself, what is the risk? If I sample one piece per lot and it passes, what is the likelihood that the other 19 will fail? What is the risk that if one piece passes the other 19 are good? I rely on my validation data to help to define my risk.