No. of samples and Duration in a Reliability Test

[bhattalok]

Experts,
I am very new to the theory of reliability engineering and seeking answer for the following question from you SME's.

I was asked this question in an interview:
"How do you determine the number of samples to test and how long (or how many cycles) you need to perform the reliability for a electronic device.

Say for example: you are testing the glass on the LCD display of device to survive dust and water. Assume that there is no established std in the industry by any organization(MIL, IEEE, IEC, ASTM etc.) How do you determine in this case how many samples do you need to test and how long the test should be conducted?

I understand the it might depend on the reliability you are targeting with what confidence level? Lets say 90% reliability with 95% confidence.

Also, I am looking for some basic Failure Analysis Techniques reference material. Any direction is welcome.

Thanks,
BA

[David]

One option is to use the Cumulative Binomial distribution. Weibull++ (http://www.reliasoft.com/Weibull/index.htm) has a utility called the Design of Reliability Tests which uses this distribution to estimate sample size or test duration based on given inputs. An article which presents the Design of Reliability Tests can be found at http://www.weibull.com/hotwire/issue24/relbasics24.htm. Additional information on test design is presented at http://www.weibull.com/LifeDataWeb/test_design.htm.

I hope this helps.

[bhattalok]

Hi David,
Thanks for your response. It appears that in order to determine time or no.of samples required, you need any one of them as an input.

Do we randomly provide input values as:
- 10 test units or 100 hrs of test time? or is there is a way you can also determine these values that are input.

Alok

[David]

Correct. You can either estimate the number of samples based on the testing time or you can estimate the testing time based on the number of samples. One of these values must be known and the value is not random. When designing a test you might know, for example, how many samples are available for testing. The number of samples might be limited based on cost, time to reproduce, resources, etc. So given this limited number of testing samples, the Cumulative Binomial distribution can return the required test time to demonstrate a certain reliability with a given confidence level. The same concept can be applied if you have limited time to conduct the testing.

I hope this helps.

[bhattalok]

David,
This is really helpful.
Appreciate your response.

Regards,
Alok