Testing sample size

[Mohamed Boualam]

[Originally Posted: 6/6/00--Transferred by ReliaSoft Moderator]

I have to meet a reliability goal of 0,98 with a confidence level of 0,5 by testing. In this case we always estimate the sample size based on non parametric distribution but the sample size is large and I can spent more time to test before retching the goal. So, is there any other manner to do that, without testing the 35 parts as obtained in this case

[RS Support]

[Originally Posted: 6/7/00--Transferred by ReliaSoft Moderator]

You can conduct a "trade-off" study by manipulating the cumulative binomial distribution in order to see how you can trade number of units for test time and still achieve your goal. This functionality is available in Weibull++ as the Design of Reliability Tests (DRT) feature. A note of caution, however: this methodology should be used as a guideline for test design, but there is no way to anticipate the actual results of the test. In many cases it would be preferrable to test more units for a shorter time than fewer units for a longer time. This is because the results of the test can usually be analyzed with greater accuracy if you have more units on test.

[James H. Vallem]

[Originally Posted: 6/21/00--Transferred by ReliaSoft Moderator]

Both the original question and the response seem to assume that a success testing plan is the only option. What about testing with a much smaller sample size, but to failure, and fitting the results to Weibull distribution? Then compare the estimate line to the R98C50 requirement.

There are several advantages to a test-to-failure plan, besides cutting the sample size way down. For one, you learn a lot more about your product's weaknesses, which will allow you to make continuous improvements (even if the parts "pass" the test). Another is that you have a better basis for ensuring that your test(s) duplicate the failure mechanisms occuring in the field.