Experimentally-verified thermal-electrochemical simulations of a cylindrical battery using physics-based, simplified and generalised lumped models

The parameterisation for fully physics-based, simplified, and semi-empirical electrochemical-thermal coupled models for prediction of the core temperature of a cylindrical 21,700 cell are addressed in this paper for the first time. When compared to fully physics-based models, the semi-empirical approaches require fewer input parameters overall, simplifying the necessary experimental values and reducing the burden on the battery management system. There are roughly 40 and 30 parameters needed for a Doyle-Fuller-Newman (DFN) and a single particle, respectively. The generalised lumped electrochemical-thermal model, in contrast, requires ten input and three fitting parameters. The present work, aimed at thermal management challenges, couples these electrochemical models with a 2D-axisymmetric heat transfer model to compare qualities such as computational time, accuracy, generalisation, high fidelity, and experimental parameterisations required. The core temperature and voltage predictions for a 21,700 NMC/Si-Graphite cell has been made using coupled thermal-electrochemical models that are accurate (voltage and core temperature standard deviations: 38-130 mV and 0.24-0.79 K) and have a quick response (computation time: 18-262 s).The electrochemical and thermal parameterisation for the models were conducted using the experimental measurements at 1C, 0.3, and 0.7C-rates. Measurements are the core temperature made with a fiber optic sensing system and voltage curves using a battery cycler. For the single particle and P2D models, electrode diffusivities (both negative and positive) are adjusted to calibrate the models. For the single particle model, the results revealed that the core temperature and voltage standard deviations were 0.6 K and 0.13 V, respectively. Depending on the number of mesh, computational times were between 16 and 262 s. For the P2D model, the core temperature and cell voltage standard deviations were found to be 0.038 V and 0.7 K, respectively. Each simulation takes between 27 and 52 s. This is an increase of 189 % over the time required to compute a previously developed lumped model, for which the voltage and core temperature had 0.12 V and 0.8 K standard deviations. P2D simulations in the literature take between 21 and 43 times longer than simulations in the current P2D model. For the generalised lumped model, only the 1C-rate discharge results are used in the calibration, and a generalisation test has been conducted by using a full discharge-charge curve for the first time. The results showed that the change in standard deviations of core temperature (within 21 %) and voltage (within 31 %) are acceptable when a discharge-calibrated model used for full-cycle validations.