Improvements on physics-informed models for lithium batteries

The fast adoption of battery electric vehicles (BEV) has resulted in a demand for rapid technological advancements. Strategic areas undergoing this development include lithium-ion energy storage. This is inclusive of electrochemical design improvements and advanced battery management control architectures. Field objectives for these developments include but are not limited to, reductions in cell degradation, improvements in fast charging capabilities, increases in system-level energy densities, and a reduction in energy storage costs. Improvements in online predictive models provide a path for realising these objectives through informed control interactions, reduced degradation effects, and decreased vehicle costs. This thesis contributes to these developments through improvements in fast physics-informed battery models for both lithium-ion and lithium-metal batteries. The key novelty presented is the improvement of real-time, physics-based electrochemical model generation for lithium-ion batteries. A computationally informed realisation algorithm is developed and expands on the previously published realisation algorithm methods. An open-source Julia-based architecture is presented and provides a high-performance implementation while maintaining dynamic language capabilities for fast code development, and readability. A performance improvement of 21.7\% was shown over the previous discrete realisation algorithm, with an additional framework improvement of 3.51 times when compared to the previously published framework. A methodology for the creation and modification of the reduced order models via in-vehicle hardware is presented and validated through an ARM-based model generation investigation. This addition provides a versatile method for cell degradation prediction over the battery life and can provide an interface for improved prediction of cell-to-cell variations. This methodology is applied to intercalation-based NMC/graphite batteries and is both numerically and experimentally validated. A further element of novelty produced in this thesis includes advancements in lithium-metal phase-field representations through the creation of a Julia-based numerical framework optimised for high-performance predictions. This framework is then utilised as a ground truth model for the development of an autoregressive physics-informed neural solver aimed to predict lithium-metal evolution. Through the implementation of the physics-informed neural solver, a reduction in the numerical prediction time of 40.3\% compared to the underlying phase-field representation was achieved. This methodology enables fast lithium-morphology predictions for improved design space explorations, online deployment, and advancements in electrodeposition material discovery for lithium-metal batteries