A practical application of the response surface methodology to power electronics failure prediction

This study deals with a simulation and data processing method for health monitoring of a power electronics device. A typical approach to power electronics lifetime assessment is to run accelerated ageing tests and statistical methods to determine a mean time to failure for a given mission profile. Using safety coefficients, an expected service life is then known for a given unit. Depending on the consistency of the test protocol with the in-use ageing rate, lifetime predictions can be substantially different from the component actual lifespan. To manage these differences, safety coefficients and maintenance plans are implemented to assure a reliable operation and avoid failures. At due date, a device is then replaced regardless of its working ability and is considered a pure loss. To overcome such a drawback, it would be interesting to monitor on a real time basis the module ageing, thus replacing it only when approaching its end-of-life and before it fails. There is hence a need to find an indicator whose evolution is representative of the device ageing. This indicator also has to be attainable, that is to say it can be measured and interpreted so failure can be anticipated. This study proposes a method based on multiphysics simulation and statistics to achieve this objective. Once the chosen ageing indicator is presented along with hypothesis related to the multiphysics model used, this paper suggests a data processing method aiming toward real-time health monitoring of a power electronics module through this indicator.