Calculating Ppk for Design Verification - Variable Sampling

[ga2qa23]

That's a great idea! (using real units in corners & bottom with equivalent weight in other locations). I never would have thought of that. I can't apply that logic to my current device shipping configuration, but that's very interesting.

May I please ask: where did you get your requirement that a sample size of one isn't satisfactory? Is it industry best practices versus a standard or government regulation?

 

[indubioush]

Please read through the previous discussion we had on this topic. Maybe it will help.
Unrealistic Packaging Validation Sample Size

 

[ga2qa23]

Thank you but that didn't really clarify. In that thread, people didn't come to an actual conclusion, and the prevailing sentiment was that that a sample size of one is OK if it's "worst case" (i.e. worst case number of devices, most sensitive devices, and worst conditions) but wasn't referencing any standard/regulation about it.

Plus, most guidances/standards/regulations about shipping only refer to sterile device packages, but not any other kinds of non-sterile packages. The standard ISO 11607-1 (for sterile packaging) has a clause about "worst case" shipping configurations, but are we even allowed to apply that logic to non-sterile shipping configurations?

 

[indubioush]

You are not going to find a proposed sample size in a standard because every device is different with different risks.

 

[ga2qa23]

True but the people who claim that a sample size of 1 being sufficient don't really talk about "risk" when justifying it. They just claim that it's "worst case" testing.

 

[Bev D]

True but the people who claim that a sample size of 1 being sufficient don't really talk about "risk" when justifying it. They just claim that it's "worst case" testing.

Of course “risk” is take into account.
1. Many test standards specify the worst case conditions and specify that a single sample can be used. We have discussed this here many times with specific examples. One example: A requirement of centrifuges is that they contain all fragments should the carrier break apart (disrupt). The test standard for centrifuges is to take one ‘carrier’ and damage it so that it will come apart when spun up to top speed. The test is to ensure that the fragments of a ‘disrupted’ carrier will not escape the enclosure. As long as the enclosure is not damaged by wear or misuse - the test is for the design. There are other requirements for wear and misuse…
2a. If risk means probability of failure: Your devices and your conditions will vary. If you take 30 devices manufactured under the same conditions and you simply send them out under basically the same conditions no calculation of failure rate is valid as your sample was not representative of the population of varying devices and conditions. Sadly this happens all to often in verification and validation testing. This is why we have OQ testing by the way to force you to test under worst case conditions For both the devices and the conditions.
2b. Thinking about the probability of failure: failure is most likely to occur at the extremes of your device specifications and under worst case conditions. (this is described by the traditional stress - strength interaction which is proven science). So if. You test the weakest devices at the strongest naturally occurring conditions and it passes physics says that failure will not occur at stronger devices and weaker conditions. Failure is not random - it must obey the laws of physics…This is exactly why we have +/- tolerances and don't specify that everything must be built exactly the same…
3. If risk means severity of failure: you dont’ test very low severity failures. If the severity is high then you should make sure you understand the conditions by characterizing them. You should have also characterized your design specifications with actual experiments and actual data. THEN in V&V testing you can be confident that either you are testing under the full range of known and allowed variation (maybe a moderate sample from a large population) OR you can test very few samples under the worst case conditions…