Design Verification Sample Size - What philosophies are out there

[MNPIKE]

All -

Recently we have done device design verification testing on sample size of 3. Recently to achieve a 90/90 (passing rate/confidence level), that has been increased to a much larger number. To my knowledge the 90/90 is more of a manufacturing tool, not design. UL for example only requires a "representative sample size of one". I am curious what philosophies are out there on verification sample size and how those were determined as well as any other knowledge out there on this topic.

Thanks!

 

[Jim Wynne]

All -

Recently we have done device design verification testing on sample size of 3. Recently to achieve a 90/90 (passing rate/confidence level), that has been increased to a much larger number. To my knowledge the 90/90 is more of a manufacturing tool, not design. UL for example only requires a "representative sample size of one". I am curious what philosophies are out there on verification sample size and how those were determined as well as any other knowledge out there on this topic.

Thanks!

Welcome to the Cove, MNPIKE

It might be good to provide a little more information, starting with defining terms. What do you mean by "design verification"? Under what circumstances are the individuals being built? Are they design prototypes, or assemblies made using the same materials, processes and machines intended for use in regular production?

The quote from UL looks like one of those Wheel of Fortune "Before and After" puzzles; it might be "representative sample size," or "sample size of 1," but strung together it doesn't make sense.

 

[MNPIKE]

During design verification we are testing to requirements. Thus the output is either pass or fail. The devices used in these tests are representative of production devices. "Tests are made on one representative sample of the item being tested.", is the entire string from UL 60601-1.

Testing 90/90 to me makes sense if one were verifying production but not design. I was curious what others are doing.

Thanks for the help!

 

[MEANMC]

I work with medical devices, but I do use LTPD for design verification also.
We use LTPD to allow us make confidence and reliability statements such as 95/95 or 95/99.

 

[sagai]

Hello there,
I have just faced with this sampling kind of questions and the reference I have found useful was the standard called ISO2859.
There is a legacy copy available over the internet here, may be helpful.

I am also still a bit struggling how to accommodate such statistical method to areas are not directly relate to incoming and outgoing real product inspections and I am not certain if this level of scrutiny really count or not.

Meanmc, can you give a little bit insight about the method you apply to encapsulate this statistical sampling into design verification acivities? ... of course on the level of detail we can know more about.

Cheers!

 

[MEANMC]

For areas of our quality management system that require lot to lot control like incoming inspection I use AQL.
As Design Verification is concerned with verifying the design ( I.e. lot to lot variation is not a focus). Then tools like LTPD or MTBF are appropriate. MTBF is useful for machines etc. I use LTPD especially where the device is single use and the test is destructive. Using LTPD allows you make certain confidence and reliability statements in relation to the design.
I have successfully used these plans on numerous FDA and European medical device submissions.
Obviously using variable data helps offer opportunities in relation to smaller sample sizes.