Accelerated degradation tests with successive measures

International audienceAssessing the reliability of power protection systems is a major issue for companies like Schneider Electric. For these highly reliable systems, observations of lifetimes from field tests can be very long to obtain. Therefore, accelerated tests have to be performed in order to estimate reliability within a reasonable length of time. For accelerated life tests, usual accelerated failure test (AFT) models are available. These models link lifetime under nominal conditions to lifetime under accelerated conditions by means of an acceleration factor. Usual models are Arrhenius, Coffin-Manson, Peck,... However, the studied systems are dormant. They are basically not operating, but they must operate on demand. When they are activated, one can only see if they are operating or not, so their lifetime is not observable. But one can measure their level of degradation and use it to forecast their lifetime. Indeed, the failure occur when the degradation level reaches a given threshold. Then, in this study, we will perform accelerated degradation tests. The ADDT model generalizes AFT models when degradation is measured instead of lifetimes. These models are for destructive tests, that means that we have only one observation per system. In practice, for the tests performed by Schneider Electric, several successive degradation measures are observed for each system. A first approach is to use the ADDT model by keeping only the first measure of each system. We propose to adapt this model to successive degradation measures in the spirit of Hong et al (2012). We will apply this approach to the results of accelerated degradation tests for which the stress variables are temperature and humidity