Data-Driven Modeling Approach for the Virtual Conversion of a Hybridized Passenger Car

Physics-based modeling is an important and cost-efficient tool within the design process in vehicular technology. Creating and validating predictive 0D/1D models is a time-consuming process that requires extensive domain knowledge and specific experimental data for each sub-system to be modeled. To handle increasing complexity and variant diversity in the design process of hybrid vehicles, a data-driven modeling approach based on real driving data is introduced. A digital twin is derived using a power-split Ford Galaxy FHEV as an exemplary use case to validate the methodology. The digital twin is divided into four individually trained Long Short-Term Memory (LSTM) networks. Training data is acquired using a ROSI Dongle OBD data logger.