Early Detection of Lithium Plating in Lithium Ion Batteries: Using Multiple Physics-Based Electrochemical Signatures to Construct a Machine Learning Framework

Abstract

A key challenge for Lithium-ion batteries (LiBs) is the capability to manage battery performance and predict lifetime, which is becoming critical as electric-vehicle (EV) adoption continues to increase. Early detection of battery-aging phenomena and the implications for the performance are crucial for maintaining warranty and avoiding safety-related liabilities. Among all of the aging phenomena in LiBs, loss of Li inventory (LLI) is a major reason behind capacity loss and shortened battery lifetime. LLI can be further separated into solid electrolyte interface (SEI) formation and Li plating. The growth of SEI layer is caused by continued usage of a LiB, in which the SEI layer thickens and leads to gradual capacity fade. Unlike SEI formation, Li plating causes drastic reduction of performance and safety issues in the LiB1, such as accelerated capacity fading and short circuiting. Therefore, Li plating should be identified and avoided early on.We established a machine learning (ML)-based algorithm framework for early identification of SEI and Li plating. This framework merges the strength of two mainstream approaches for estimating the state-of-health of LiB, namely the data-driven approach and physics-based modeling2. Meanwhile, we aim to identify the aging modes or mechanisms during battery cycling. In our approach, we analyze electrochemical signatures that are recorded along with battery aging cycles, then identify physically meaningful trends that are directly related to SEI growth or Li plating. Through this approach, multiple pieces of data can be analyzed in a data-driven approach, while the physical meanings are preserved as in the physics-based modeling.Here, we use graphite/NMC532 pouch cells with a variety of cell designs (anode loading, porosity) and charging protocols (CC-CV, 2-step, multi-step from 4C – 9C) to construct this classification framework. We discover that the evolving trend of capacity fade, end of charge voltage after rest (EOCV), and Coulombic efficiency (CE) can be combined into a logistic elastic net to build a ML classification model separating SEI growth and Li plating, as shown in Figure 1. The results also show that this framework differentiates the two cases as early as within the first 25 aging cycles. Additionally, we show that multiple electrochemical signatures have coherent response to aging phenomena and need to be integrated together to improve the accuracy in the classification. Because of the preservation of physical meanings, this approach has the potential to be extended to general cell designs and other types aging phenomena, such as degradation of electrodes. Ultimately, this early-decision framework will potentially shorten the battery design and test cycles, as well as provide insights into cell-design parameters and use guidelines to extend cell lifetime and delay the aging process. Waldmann, T., Hogg, B.I., and Wohlfahrt-Mehrens, M. (2018). Li plating as unwanted side reaction in commercial Li-ion cells – A review. J. Power Sources 384, 107–124.Hu, X., Xu, L., Lin, X., and Pecht, M. (2020). Battery Lifetime Prognostics. Joule 4, 310–346. Figure 1