Predictive Analysis and Data-Driven Modeling for Electrochemical Degradation of Li-ion Batteries

Lithium-ion (Li-ion) batteries are widely used across various industries, ranging from portable
electronics and electromobility to battery energy storage systems. Stationary energy storage systems, in
particular, play a crucial role in stabilizing power grids, integrating renewable energy sources and
providing reliable energy supply. Li-ion batteries are characterized primarily for their exceptional
energy density, low self-discharge rate, and long lifespan. Despite their longevity, Li-ion batteries are
prone to capacity degradation, reflected in their State of Health (SOH), which is a crucial metric for
assessing their remaining capacity and overall reliability, during both calendar and cyclic aging, leading
to a decline in performance and reliability. These batteries represent complex electrochemical systems,
characterized by numerous degradation phenomena. Consequently, various approaches are employed,
including physics-based models [1], empirical models, equivalent circuit models, or data-driven models
leveraging machine learning techniques [2, 3].
In this experiment, 14 Li-ion pouch cells of the lithium-nickel-manganese-cobalt oxide (NMC) type
were tested. The testing was conducted at temperatures of 5, 25, and 45 °C and Depth of Discharge
(DoD) levels of 70%, 90%, and 100%, as Li-ion batteries exhibit varying degradation mechanisms and
rates under different conditions, as described by Ch. Birkl et al. [4]. The batteries were charged using a
Constant Current Constant Voltage (CCCV) protocol and discharged using a Constant Current (CC)
method, with each battery undergoing a total of 500 cycles.
The experiment focuses especially on developing a data-driven model based on machine learning,
utilizing 12 Li-ion batteries as training dataset while dedicating the remaining 2 for validation by
estimating their SOH. While battery state prediction using such a dataset may require processing time,
the primary aim of this research is to enable real-time analysis through a Reduced-Order Model (ROM).
This model selectively incorporates key data points rather than the full dataset, ensuring accurate
estimation of the battery's condition based on specific input parameters.
The work contributes to evaluating battery performance during cycles not included in the training
data, reducing costs associated with testing batteries of similar composition. Moreover, this work will
provide a solid foundation for a digital twin designed to monitor stationary battery system conditions in
real-time, enabling strategies to minimize degradation and enhance reliability.