Global Sensitivity Analysis of a Digital-Twin for Battery Electric Vehicles -Data Pipeline Development for Large-scale Simulations

Abstract

Modeling is a central tool in the development of battery electric vehicles, particularly in applications of validation and system optimization. This thesis aims to assist the development of simulation platforms through a large-scale global sensitivity analysis of a digital twin representing a heavy duty battery electric vehicle, using energy consumption as the primary output. The Elementary Effects Test was implemented to screen the model input space, after which a Sobol’ sensitivity analysis was conducted on the most influential input parameters. The sensitivities were further explored through derivative-based measures, linear regression and Monte Carlo filtering. The results consistently identified the gross combination weight as the overwhelmingly most influential parameter. The Sobol’ analysis further indicated that, depending on the drive cycle, this parameter accounted for between 25% and 77% of the total output variance. In addition the initial temperature, the aerodynamic drag area and the rolling resistance coefficient were concluded as influential input parameters but not in the same magnitude. The relative ranking of parameter importance was found to vary with the drive cycle used for the simulations. The linearity of the relation between input parameter and output was also investigated. Gross combination weight, rolling resistance coefficient and the aerodynamic drag area were found to have mostly linear effects on the output, thus suggesting that these relationships can be represented by simplified models in future work.