(3cu) Implementation of Physics-Based Battery Models: Review, Analysis, and Applications- Towards Physics-Informed Data-Driven Models Beyond Physics-Based Models | AIChE

(3cu) Implementation of Physics-Based Battery Models: Review, Analysis, and Applications- Towards Physics-Informed Data-Driven Models Beyond Physics-Based Models

Authors 

Lee, S. - Presenter, University of Washington
Research Interests:

Today, lithium-ion batteries are considered the best means to store energy in renewable energy applications. When lithium-ion battery systems are deployed in energy applications, mathematical battery models are used to understand, operate, and optimize battery performance. Therefore, it is essential to use adequate and relevant battery models to improve performance and usability of energy applications. Typically, battery models can be classified into various groups, from empirical to molecular-level models. Among many other models, physics-based battery models hold the potential to simultaneously provide systems engineers, battery designers, and material scientists with an optimal computational framework that maximizes safety and usability for renewable energy applications and sustainable mobility. To maximize the safety and usability of the battery system, physics-based models have been used in combination with the degradation mechanism, such as the SEI layer growth, lithium plating, active material loss, and impedance increase. While the mechanism-specific models are validated over well-defined operating conditions, implementing various degradation modes in battery models remains a challenging task, which requires coupling with thermal and mechanical effects. For this reason, data-driven models with machine learning (ML) techniques recently gain a great attention as an alternative tool of predicting battery performance due to their potential in achieving high accuracy with low computational cost. However, one of the critical limitations of current data-driven models is that they provide limited information related to capacity/power fade parameters as a black box. The experimental data-driven battery models conventionally collect current-voltage data, which do not contain kinetic/transport/thermal parameter information on state of health estimation.

In this poster session, pseudo-two-dimensional (P2D) and its simplified models will be reviewed. The P2D model is the most popular and acceptable physics-based model, and several simplified models have been proposed. Next, the current participation strategy of physics-based models for renewable grid applications will be presented. Precise estimations of state of health with system parameter information are essential to maximize the safety and durability of lithium batteries, for example, in order to prevent them from thermal runaway caused by internal temperature rise, short circuit, and material degradation. Lastly, the concept of physics-informed ML will be introduced. Although physics-based models provide physical insights, characterizing and simulating all degradation mechanisms is a challenging and probably not realistic task. The goal of the physics-informed ML is to infer the correct parameters for the voltage data from experimental measurements and model outputs. Physics-informed ML training of the voltage data sets can have the potential to offer accurate predictions with kinetic/transport/thermal parameter information.

Topics