This is an area with a lot of opinions, including mine.
The first problem with design verification is that many people confuse design verification with manufacturability. As a result, they look up some sample size calculations and believe they need to manufacture that many products and then test them. However, design verification seldom needs to build product. The vast majority of design verifications should use the design outputs, the drawing and specs. You can also use calculations, previous designs, etc. To take a simple case, imagine the product has a detachable power cord and the spec says it should be black, 10 feet long, plug into a North American outlet, and have a certain connector for the device. Design verification should look at one copy of the purchasing specification. You don't need to order 30 power cords and inspect them. Similarly, if you make a part, you can almost always do design verification by looking at the design output drawings, specs, etc.
For design validation you need to have an initial production unit or equivalent. Assuming you don't do destructive testing, the number you use depends on the number of design verification tests. You may be able to do all of them with one product. If you want to speed up the process, make two so you can run two sets of tests in parallel.
If you decide you need to test more than one, then you are doing a statistical test. They generally fall into two categories – attribute or variability.
In an attribute test, you test the sample size and classify each outcome as pass or fail (the attribute). The sample size depends on how many failures you are willing allow. More allowed failures increases the sample size. The best discussion I know of is in the article "Attribute Reliability and the Success Run: A Review" by Stephen N. Luko. You should be able to do a Google search to find it.
Variables sampling is more complicated because you will estimate whether the population characteristic fits in the specification by analyzing the sample. Sometimes you can make assumptions to see if the confidence intervals of the mean and variance will work.
In general, you will get smaller sample sizes with variables than with attributes. Variables used to be difficult with pen and paper calculations, but it is really easy with Excel.
------------------------------
Dan O'Leary CQA, CQE
Swanzey NH
United States
------------------------------
Original Message:
Sent: 15-Feb-2023 13:56
From: Sandra Veenstra
Subject: Sample Size for Verification and Validation Tests
Does anyone have any standards/guidance documents that you use as a reference when determining sample size for verification and validation testing as part of design control?
------------------------------
Sandra Veenstra
Director - Quality Assurance and Regulatory Affairs
Moncton NB
Canada
------------------------------