It's good question, and an interesting subject ...
Yes, it's not only FDA, but in Europe the directives refer to compliance with standard for both design
and production, a point which is often overlooked.
Also, IEC 60601-1:2005 uses the following definition of a type test (bold is my emphasis):
3.135 TYPE TEST
test on a representative sample of the equipment with the objective of determining if the equipment, as designed and manufactured, can meet the requirements of this standard
From fundamentals, number of samples depends on variability of results and the margin to an applicable limit. That depends a lot on the specific design aspect being tested. So you cannot decide for the whole standard.
For example, for earth leakage current, if the manufacturer uses 1nF EMC capacitors, including capacitor tolerance (say 20% including temperature effects), and stray leakage (varies usually 20-40uA) we can expect the actual production results to be variable between 100-140uA. Since the limit is 500uA, a single test result of say 127uA proves that the design is implemented as expected, and "representative" of production given the huge margin to the limit. Additional samples are not going to increase confidence in a pass / fail result.
Many safety tests are like this, in practice there will be mild to moderate variability in the test results on each production sample, but the margin to the limits will be far higher than the variability.
But, sometimes there will be a close result, or large variability. In that case there may be a few options:
- redesign with a greater margin or reduced variability
- run several tests until you get enough data to be confident of pass/fail result
- determine that the test method/criteria is already conservative, so there is no significant risk due to production variations.
Standards are usually based on a series of worst case assumptions which are necessary for consistency but rarely occur in the real world. Also the allowable limits often include some margin to allow for production variations. In that case, worrying about whether a test result is representative of production might be overkill. For example, in the temperature rise test, we assume worst case supply voltage, cooling, operating mode, loading, operating conditions, length of loading, room ambient etc etc. If under all of those conditions a test result is marginally below the limit, production variations are unlikely to introduce any significant risk in the real world.
On the other hand, we may have a test result where a PCB track fused during a switching transformer short circuit test. Since it was an unintended part of the circuit that provided protection, it's reasonable to ask if the result is representative of production. In that case, running the test on 10 samples might be a reasonable approach.
So to summarize, it's a case by case, taking into account the test, the design, expected variability of the result, margin to the limit, and how critical the limit/test is in the real world.
Because it is built in to the standard (via the definition of a type test), this is something that the test laboratory should consider when each test is performed. They should inform the applicant if they feel a test result will not be representative of production.