Anybody who did some hardware test in they life, eventually will face the question of sample size.
Probably it will be a tradeoff between the test duration and amount of samples to test.
So how much is enough? One, three, ten?
To answer to this question you should elaborate what is your reliability targets, or by other words how many units of the product will make it till end of life, without any failure?
Let’s take for example dishwasher, usage life is around 7 years, let’s assume that typical dishwasher working 3 times per day (3 meals), and every event is about 1.5 hours.
Target time for end of life of dishwasher in this case is: 7*365*3*1.5 hr= 11,500 hr.
From entire dishwasher population, how many units, your company need to accomplish 11,500hr ?
Reasonable number is more than 90% or maybe 95%, or 99% or 1PPM (99.9999%)?
Another time dimension should be taking in consideration, is warranty period. Your company probably want to get as less as possible dishwashers for warranties claims.
Let’s assume you know what is company targets: 11,500 hours and Reliability of 90%.
Can we answer now, how many units we need to test? Answer is still no.
We should gather one more piece of information – is this completely new product from scratch, that dos not have any previous history, or it is upgraded version of product that there on market for years?
If you have previous information about failures of similar in design products typically the tested units will be in the range of 6 to 12 units.
If there is not any hystorical data and you want get high reliability the numbers going to direction of 20 – 50 units.
As reliability engineer I always suggest to run more units as possible.
Neeraj Dubey says
Hi Semion,
THe numbers you suggested, are they a rule of thumb, something empirical only? Success run theorems, weibayes approach etc assume zero faiures so how do we know 12 will work even with historical data?
Semion Gengrinovich says
Hi Neeraj,
There is no one rule of thumb. It can be variety different factors, that can affect sample size.
For instance test type, product type, in that regard you can read next my LinkedIn articles here:
https://www.linkedin.com/in/semion-gengrinovich-19235243/recent-activity/articles/
Specifically those ones:
https://www.linkedin.com/pulse/sample-size-during-feasibilityprototyping-semion-gengrinovich-6zx3c/?trackingId=nlEs72rMRRmowKph8fT8UA%3D%3D
https://www.linkedin.com/pulse/sample-size-during-design-validation-dv-semion-gengrinovich-cagac/?trackingId=nlEs72rMRRmowKph8fT8UA%3D%3D
https://www.linkedin.com/pulse/sample-size-during-product-validation-semion-gengrinovich-axzic/?trackingId=nlEs72rMRRmowKph8fT8UA%3D%3D