Benchmarking of multi-order equivalent-circuit models with empirical ageing for Li-ion battery digital-twin workflows

Accurate and computationally efficient battery models are essential for battery management systems and battery digital-twin workflows. However, model selection requires careful evaluation of the trade-off between dynamic voltage fidelity, computational effort, energy accounting, and ageing representation. This paper presents a reproducible MATLAB-based benchmarking workflow for comparing equivalent-circuit models (ECMs) at battery-pack scale under common parameterisation and ageing assumptions. A SAFT-type RC model and first-, second-, and third-order Thevenin models are evaluated using open-access state of charge (SoC)/state of health (SoH)-dependent parameter surfaces, an empirical calendar–cycling ageing law, and automated post-processing through the DATTES toolbox. The benchmark is performed on a 64s29p LiFePO₄ battery pack under a repeated synthetic 24-h current profile over a three-year equivalent operating horizon at fixed 25 °C. The results show that higher-order ECMs improve agreement with the third-order Thevenin internal reference, but the incremental benefit diminishes beyond the second-order model. Within the investigated model family, duty cycle, and MATLAB implementation, the second-order Thevenin model provides the most favourable compromise between internal-reference voltage fidelity and per-step computational cost. The ageing outputs show identical SoH and internal-resistance-growth trajectories across ECMs because the adopted ageing law is driven by common daily stress metrics rather than ECM-specific polarization states or transient overpotential histories. Therefore, these ageing results should be interpreted as a property of the implemented modular ageing-coupling strategy, not as a general proof that ECM order has no influence on degradation in real batteries. The study provides a transparent and reproducible benchmarking workflow for controlled ECM comparison, while experimental validation, electro-thermal coupling, and estimator-in-the-loop testing remain necessary before predictive deployment in operational BMS or digital-twin systems.