In automotive industry a battery management system’s state prediction algorithms can change during the development cycle of a car. Moreover, even serial production cars happen to have different software versions depending on their date of production. Monitored aggregated customer data thus includes state of health (SOH) values generated by different algorithms. To ensure comparability over the collected data, databased models are needed which can be applied on selected input variables. This study aims to generate virtual customer driving data using automotive simulation models and stochastic customer driving profiles to establish a database for model training purposes. These can either be creating reference models for existing collected data but also forecasting the impact of future customer behaviour. Synthesizing aggregated driving data requires vehicle simulation models to compute the battery load profile from speed and temperature input profiles. Those load profiles can be then propagated into an ageing model which outputs the corresponding ageing factors caused by the initial driving profile. Since vehicle simulation is computationally very exhaustive, Markov chains are used to multiply extracted battery load profiles. All necessary signals are then collected and transformed into aggregated driving data format. Finally, the simulated and real customer databases are investigated towards congruence to get a better understanding of the actual customer behaviour.
Such a simulation toolchain to synthesize virtual customer driving data offers manifold opportunities as a starting point for databased ageing analysis of Li-Ion batteries in vehicle applications. In regard to state prediction it serves as a data generator to provide software status independent SoH values for training and validation purposes. Hence, a databased model can be trained on the data created by the toolchain and after successful validation applied on the actual driving data.
From a forecasting perspective the toolchain acts as a reference model to assess the estimation performance of already trained databased models. In the context of battery development, the toolchain can be applied for lifecycle prediction once an underlying ageing model has been parametrized from laboratory data.