System identification of lithium-ion battery dynamics : from characterisation to application

To alleviate range anxiety among electric vehicle (EV) owners, the accuracy of lithium-ion battery (LIB) mathematical models in the low state of charge (SOC) range must be enhanced. A battery model that is easy to parameterise while maintaining accuracy over the entire SOC range is required in sophisticated battery management algorithms. This thesis addresses this knowledge gap via system identification methods of characterisation, identification, and application. The level of non-linearity over different SOCs is first studied by using random phase odd-multisine signals, and applied on the Doyle-Fuller-Newman (DFN) model and a three-electrode experimental set-up of a commercial 5Ah cylindrical 21700 LIB cell. The charge transfer coefficient is determined as the most sensitive parameter towards battery nonlinearity and with an asymmetrical Butler-Volmer kinetic the model nonlinear response provided good agreement against experimental data. The cathode even order nonlinearity is the main contributor towards the battery voltage nonlinearity while the anode starts to dominate at very low SOC. Utilising the newly proposed characterisation method, a non-linear equivalent circuit model with diffusion dynamics (NLECM-di↵), which phenomenologically describes the main electrochemical behaviours, such as ohmic, charge-transfer kinetics, and diffusion processes, is identified. Compared to the parameterisation challenge of electrochemical models, the NLECM-di↵ does not rely on geometrical parameter and all parameters are determined from the measured current and voltage signals. The NLECM-di↵ is around 50% more accurate than a conventional ECM and is comparable to the single particle model with electrolyte model (SPMe). When simulating driving cycles and long duration discharges, the dominant voltage loss changes from ohmic to the diffusion losses, and the characteristic of the negative electrode is determined as the primary reason for the low-SOC-error. The last part of this thesis presents three case studies of model application as part of the project ‘Virtually Connected Hybrid Vehicle (VCHV)’. The SPMe and the NLECM-di↵ models were demonstrated in Hardware-in-the-Loop (HIL) and therefore merit consideration for EV applications