Parameter Identification and Sensitivity Analysis for Zero-dimensional Physics-based Lithium-Sulfur Battery Models

TL;DR

This paper develops and compares zero-dimensional Lithium-Sulfur battery models by estimating parameters from experimental data and analyzing their sensitivity, providing insights into model fidelity and practical applicability.

Contribution

It introduces a systematic approach for parameter estimation and sensitivity analysis of zero-dimensional Li-S models using experimental cycling data.

Findings

Parameter estimation for four Li-S models from cycling data.

Sensitivity analysis highlights key parameters affecting model outputs.

Comparison of model fidelities based on experimental data.

Abstract

This paper examines the problem of estimating the parameters of a Lithium-Sulfur (LiS) battery from experimental cycling data. LiS batteries are attractive compared to traditional Lithium-Ion batteries, thanks largely to their potential to provide higher energy densities. The literature presents a number of different LiS battery models, with different fidelities and complexities. This includes both higher-fidelity diffusion-reaction models as well as "zero-dimensional" models that neglect diffusion dynamics while capturing the physics of the underlying reduction-oxidation reactions. The paper focuses on zero-dimensional LiS battery models, and develops four such models from the literature, reflecting different choices of which redox reactions to model. There is a growing need for using experimental cycling datasets to both parameterize these models and compare their fidelities. To address this need, we fabricated LiS coin cells and performed charge/discharge cycling tests on these cells. In parallel, we analyzed the sensitivity of simulated LiS battery charge/discharge characteristics to underlying model parameters. Using this sensitivity analysis, we selected a subset of model parameters for identification, and estimated these parameters for all four LiS battery models from cycling data, thereby arriving at a consistent experimental comparison and assessment of these models' respective fidelities.