Analysis of multiply censored run-in data

Test data from burn-in of complex electromechanical systems were analyzed using hazard plots in order to determine whether the systems were being tested for the optimum length of time. The problems encountered were how to determine the failure distribution and its parameters for multiply censored data and how to put it in a form easily understood by those outside the reliability field. A BASIC computer code generated plots of multiply censored data, consisting of test results for nine systems, for five common distributions to determine the best fit. It was found that the effects of burn-in at the module level could best be seen as the system level using the three-parameter Weibull distribution. From a so-called bathtub curve plot showing the change in instantaneous failure rate as a function of run-in time (wafers), intuitive conclusions can be drawn regarding the adequacy of run-in. Alternatively, the asymptotic instantaneous failure rate can be used as the useful life failure rate in cost equations, which should provide an objective measure of the adequacy of run-in.<>