Reliability and Degradation of Power Electronic Materials

Abstract

Testing materials and devices for a specific failure mechanism usually yields a characteristic distribution of failure times. However, failure of power electronics (PE) can be due to a manifold of mechanisms. Even in a single PE device multiple failure processes may compete. For instance, modules may contain various components that fail according to different processes. When testing a batch of devices with competing processes, the resulting failure distribution is likely to consist of a combination of distributions, also called ‘entangled’ distributions. If processes compete, the time of failure according to the one process masks the time of failure according to a competing process. The observed failure time for one process leads to a so-called censored failure time of the competing process. If there is a balance between occurring failure processes, data analytics should not only involve the observed, but also the censored failure times. Another cause for distributions to blend, is the test of inhomogeneous test batches. For instance, some test objects may contain a production flaw, while other objects don’t. Such a mix of (sub) populations will also yield a total distribution consisting of entangled distributions. Enhanced stresses in a test are likely to accelerate the various aging processes at different rates. Awareness of distribution entanglement and acceleration is useful for interpreting test results and developing effective testing methods. The present contribution reviews the statistical analysis of R&D data. It discusses competing processes, mixed populations, censored failure data, accelerated aging, and how these phenomena affect analytics by graphical analysis and parameter estimation. Finally, series and parallel system configuration and reparability aspects are discussed.