Physics-Based Modelling of Lithium-ion Batteries: Degradation Implementation and Parameter Estimation

Improved Lithium-ion battery technology has allowed for a rapid increase in the application of batteries in various applications, from stationary energy storage solutions, to electric vehicles. Due to the complex electrochemical nature of Lithium-ion battery technology, modelling and control tools are needed to support the wide application of Lithium-ion batteries. For instance, battery models are used in the Battery Management System (BMS) for State-of-Charge (SoC) and capacity fade estimation, aging-aware charging or battery prognostics. While Equivalent-Circuit Models (ECM) are often used for SoC and capacity-fade estimation, an electrochemical model, such as the Doyle-Fuller-Newman (DFN) model might beneeded for more complex applications such as aging-aware charging or battery prognostics, as these applications require information on the internal states of the battery. In this thesis, we focus on improving various aspects of physics-based modelling, to allow for a better representation of the internal states of Lithium-ion cells. In particular, we focus on improved modelling by incorporating physics-based relations such as temperature dependence, and a deep analysis on ageing model implementations. Furthermore, we apply various techniques for improved parameter estimation of the DFN model, as well as the ageing models implemented in this work.In the first part of the thesis, we focus on parameter estimation of the DFN model. The DFN model has a large number of model parameters with many combinations of the model parameter values leading to the same voltage prediction. Thus, identifiability remains a key issue in estimating the model parameters of the DFN model, and can be improved in various ways. One option to improve the identifiability of the model is through improving data informativity by including a richer dataset. We propose the inclusion of physics-based temperature relations within the DFNmodel and the parameter estimation technique, in which model parameters are estimated over a wide temperature range. We evaluate the effect of including physics-based temperature relations on the identifiability of the model, as well as its voltage prediction accuracy. The implementation of physics-based relations results in parameters that are physically meaningful, and comparable model accuracies to the original parameter estimation technique, in which the model parameters are identified at individual temperatures and physics-based temperature relations are not included. To further address the poor identifiability of the DFN model, we propose an experiment design procedure for the Doyle-Fuller-Newman (DFN) model that maximises the parameter sensitivities. We consider a carefully selected class of input signals whose hyperparameters are optimised using an informativity index that uses the Fisher information matrix. As the Fisher information matrix is based on a local parameter sensitivity analysis, we compare the local sensitivity analysis with a global sensitivity analysis. This global sensitivity analysis is based on so-called Sobol indices and we will show that the global and the local sensitivity analyses lead to similar conclusions, which justifies the use of the (simpler) local sensitivity analysis. The design procedure leads to an improved model accuracy when compared to estimating the parameters using constant-current experiments or a drivecycle.In the second part of the thesis ,we focus on incorporating ageing phenomena within the DFN model and parameter estimation for ageing. Physics-based models for Lithium-ion batteries offer rich insights into the internal states of a cell, particularly as it undergoes ageing. While numerous ageing phenomena have been incorporated across various modelling approaches, a lack of in-depth analysis remains regarding the physical relevance of these implementations. In this part of the thesis, we integrate commonly-used ageing mechanisms within a DFN framework and provide a detailed analysis of each one. We demonstrate how the Tafel equation can be effectively applied to model Solid-Electrolyte-Interphase (SEI) formation, andresolve inconsistencies in the definition of the overpotential driving the SEI reaction. Additionally, we explore switching behaviours commonly employed for modelling Lithium plating, and introduce a more consistent alternative. The model is further extended to incorporate cathode oxidation, underscoring its importance in BMS strategies. Finally, we present simulation results for each individual ageing process, along with a composite model that integrates all phenomena. The final part proposes a parameter estimation procedure based on multiobjective optimization, where both objectives of having accurate voltage predictions and capacity degradation are simultaneously optimized. The implementation of the parameter estimation procedure results in parameters for both the DFN model equations as well as the ageing modelling equations. The proposed strategy shows improved prediction accuracy of both the modeled voltage as well as the modeled capacity fade. Furthermore, voltage predictions remain more accurate over the lifetime by incorporating ageing.With the contributions of this thesis, as indicated above, improved parameter estimation of the DFN model is possible through the inclusion of physics-based temperature relations, richer datasets and optimal experiment design. This results in model parameters which are more physically meaningful, with improved model accuracy. Additionally, a detailed analysis is provided on the implementation of various ageing phenomena along with the DFN model, contributing to a better standard in the field of Lithium-ion battery ageing modelling, and improved modelling of Lithium-ion batteries