Practical identifiability of electrochemical P2D models for lithium-ion batteries

Abstract

Electrochemical models play a significant role in today’s rapid development and enhancement of lithium-ion batteries. For instance, they are applied for design and process optimization. More recently, model and parameter identifiability are gaining interest as thorough model parameterization is key to reliable simulation results. Especially electrochemical models are often prone to unidentifiability and overfitting due to their high number of adjustable parameters. In this article, the most common electrochemical peudo-2D model of a lithium-ion battery is parameterized. A three-step procedure is applied which considers quasi-static 3-electrode measurements of the open-circuit potential, C-rate tests, and electrochemical impedance spectra. Identifiability of each step is discussed in-depth and a general guidance for future parameterizations is derived. The conducted study reveals the insufficiency of open-circuit potential and C-rate tests to fully parameterize the electrochemical model. Highly dynamic tests, e.g., impedance spectroscopy, are required to resolve the ambiguity of diffusive and electric processes under quasi-static conditions. Any parameterization of electrochemical models requires experimental data of electrode-resolved tests, as well as a combination of quasi-static and highly dynamic tests. The results of this study provide guidance for the use of electrochemical models in applied sciences and industry. Graphic abstract

Analysis of Lithium‐Ion Battery State and Degradation via Physicochemical Cell and SEI Modeling

“…The fundamental parameterization procedure is similar to our previous work. [20] It highlighted that the commonly chosen approach for P2D cell model parameterization based on discharge curves cannot ensure good parameter identifiability. Although the experimental data may be represented accurately, a non-unique parameter set could lead to wrong conclusions with respect to limiting processes in the cell.…”

“…The results further illustrate the importance of a multi-step parameter estimation similar to previous works. [20,71] For instance, the SEI thickness in our model is directly related to the loss of lithium inventory. The analysis of a discharge curve at a low C-rate enables a parameter estimation without significant parameter interactions due to negligible kinetic and transport limitations.…”

“…For example, a double-layer capacitance is likely better identifiable from EIS measurements than C-rate tests, whereas a solid diffusion coefficient may be better accessible from discharge curves. [20,21] Importantly, there are also parameters like reaction rate constants that influence both C-rate and EIS data. Without a model for the joint analysis of both measurement types, a robust and unambiguous cell diagnosis cannot be ensured.…”

“…Without a model for the joint analysis of both measurement types, a robust and unambiguous cell diagnosis cannot be ensured. [20,22] Furthermore, SEI-related insights from aging models that use only cycling data [17,23] are commonly limited to the loss of cyclable lithium, which is attributed to SEI growth. With an EIS-based cell state estimation, changes in interfacial SEI properties like process kinetics and double-layer capacitance become accessible.…”

“…Such a restricted diagnosis is disadvantageous for multiple reasons. For example, a double‐layer capacitance is likely better identifiable from EIS measurements than C‐rate tests, whereas a solid diffusion coefficient may be better accessible from discharge curves [20,21] . Importantly, there are also parameters like reaction rate constants that influence both C‐rate and EIS data.…”