Damage Prediction and Remaining Useful Lifetime Assessment of a Discrete Power Electronic Component Using a Multi-Layer Perceptron based on Mission Profile Data

In this contribution, an automated routine is presented to assess the reliability of a discrete power electronic component used in electric bikes based on a data driven approach. Real temperature profiles are acquired from the field under different loading conditions of electric bikes and its features are extracted using a custom algorithm. A Multilayer Perceptron trained with synthetic but realistic temperature loads is used to predict creep strains induced in the solder joints of a chip resistor. In addition, the remaining useful lifetime has been evaluated at various stages of the mission profile. The methodology provided a solution for real-time capable remaining useful lifetime estimation method.