A Physics-driven and Data-driven Digital Twin for Vehicle Immunity Testing

We present a digital twin approach for the systematic evaluation and prediction of critical configurations in vehicle immunity testing. Based on parameterized simulation models, training and test data for regression and classification models (supervised learning) are generated. White box machine learning models are used so that the user can interpret and explain the results and relationships between the attributes and responses. This makes it possible to predict at an early vehicle development phase which components could prove critical during final vehicle testing. It is shown how to use this approach to evaluate in which vehicle variant a component will be exposed to worst-case conditions.