Global sensitivity analysis on parameter identifications of electrochemical Li-ion cell model using transient test data scaled from battery electric vehicle experiments

Accurate performance prediction of lithium-ion batteries at a cell level is crucial before the cell can be scaled to a pack for a system-level simulation of battery electric vehicles (BEV). The Doyle-Fuller-Newman (DFN) model is commonly used to predict the thermal-electrochemical performance of a Li-ion cell. The model has numerous parameter identifications, which is challenging when selecting important parameters for model optimizations and calibrations. The cell model parameters are not transferable due to different materials and properties. Related literature studies on parameter identifications only used measured cell response data from cell testing chambers, which did not consider the impact of real vehicle thermal management systems (VTMS). This work presents a thorough global sensitivity analysis to identify the most suitable NCA/Gr.-SiO x cell parameters before optimization. Firstly, the Elementary Effect (EE) method was utilized to evaluate (un)important 42 global parameters, of which 16 parameters can be reasonably neglected due to their low mean EE and standard deviations. Experiments of a battery electric SUV were performed under repeated Worldwide harmonized Light vehicles Test Cycle (WLTC) and combined Highway Fuel Economy Test Cycle (HWFET) with Federal Test Procedure (FTP75) driving. Measured transient performances (voltage, state-of-charge, and cell temperature) of a 75-kWh Li-ion battery pack (4416 cells) are scaled to a cell level for model validations. Then, the remaining 26 parameters are optimized for the cylindrical 21700 cell model to reasonably validate the dynamic cell performances. The sensitivity of the important DFN parameters is reported, providing a guideline for future parameter identifications in Li-ion pack model development with actual VTMS. • Electrochemical battery cell parameterization using global sensitivity analysis. • Cell performance data scaling from Tesla Model Y tests under various driving cycles • Parameter identifications are reduced from 42 global parameters to 26. • Parameter rankings from the most to least influential on 21700 NCA/Gr-SiOx cell. • Accurate predictions of transient Li-ion battery cell performances.