Accelerated testing for qualification

[vietv le]

[Originally Posted: 7/21/00--Transferred by ReliaSoft Moderator]

This is how we perform qualification: our product spec calls out for the operating temperature between -10 degree C and 60 degree C and the voltage at 3V and 5V. We test the units at the four "corners" for 2000 hours such as (-10 degree C and 3V, -10 degree C and 5V, 60 degree C and 3V, 60 degree C and 5V). what do you think about this qualification plan ? Any suggestion on some different methods that i can try ?

appreciate all the inputs

[RS Support]

[Originally Posted: 7/25/00--Transferred by ReliaSoft Moderator]

This test seems to be more of an acceptance test at different use conditions rather than an actual accelerated life test. An accelerated life test involves testing to failure at conditions above and beyond the stated operating conditions and then making estimations of the life characteristics at the use conditions. My experience with this sort of "four corners" testing is that it is mostly zero-failure acceptance testing rather than life testing designed to gather useful data about the life of the product being tested. What sort of information is being gathered in your test? What is the overall goal of the test?

[RS Tech Support]

[Originally Posted: 7/25/00--Transferred by ReliaSoft Moderator]

I am familiar with this type of testing. Strictly speaking, this is not really an accelerated life test, but rather - as you mentioned - a qualification test to ensure that the product is capable of operating at the specified operating conditions. What sort of information is being collected from this test? Is there any time-to-failure information being gathered, or rather is the no-failure type of assurance test?

[Viet le]

[Originally Posted: 7/26/00--Transferred by ReliaSoft Moderator]

Thanks for reply, we collect time between failure for all test samples. We multiply individual time between failure by an acceleration factor to yield the projected MTBF. We sum up all the projected MTBF's to give the total time between failure (T) and use Chi-square to calculate the confidence interval for MTBF for the purpose of waranty. Confidence interval of MTBF = 2*T/Chi-square(2*n+2). I personally did not agree this is the way to perform accelerated test but I agree with you this is more like an acceptance test for qualification. Any insights to how I can plan an accelerated testing along with this qualification test so I can kill two birds with one stone.

Thanks, Viet

[RS Tech Support]

[Originally Posted: 7/27/00--Transferred by ReliaSoft Moderator]

It sounds like you are making two fairly sizable assumptions in your testing: 1)the products under consideration have a constant failure rate, and 2)you know what the acceleration factor is. Let me address these two assumptions seperately.

The use of the chi-squared method for calculating the confidence interval for the MTBF is only applicable for the exponential distribution. This means the assumption of a constant failure rate, which might not be a valid assumption, particularly if you're dealing with mechanical parts or components. If you do not feel that is a valid assumption, you should consider using a different method of calculating your confidence intervals. (Note that many of these advanced calculations can be carried out using ReliaSoft's ALTA software.)

Since you are assuming an acceleration factor, the need for an accelerated life test is somewhat superfluos, since one can argue that the entire point of an accelerated test is to calculate such an acceleration factor and use it to determine the life characteristics of your product at a use stress that is lower than the stress at which you tested.

You may want to take a step back and consider exactly what you are attempting to accomplish with your testing. If you are just trying to make sure that there are no egregious faults with your product at the range of stated use conditions, it sounds like you have accomplished that already. However, you should be cautious about trying to slap any sort of statistical results on your test, as this type of testing doesn't really lend itself to a vigorous statistical analysis. If, on the other hand, you are actually trying to quantify how changing use conditions will affect the reliability, you should consider exactly what you will be attempting to demonstrate under which conditions. Wayne Nelson's accelerated life testing text is an excellent starting place for someone who wants to learn more about accelerated testing. See also http://www.reliasoft.com/Books/index.html for more references.

[RS Tech Support]

[Originally Posted: 8/1/00--Transferred by ReliaSoft Moderator]

You may also want to start out by looking at: http://www.weibull.com/acceltestwebcontents.htm