EUVIMEDEUVIMED entdecken
Crossref

Learning Discharge Behavior from Electrochemical Impedance Spectroscopy Data: A Deep Learning Framework for Rapid Li-Ion Battery Aging Diagnostics

Bryce De Muth, Tom-Eliot Jullien, Sai Thatipamula, Simona Onori

ECS Meeting Abstracts · 07.07.2026 · MA2026-01 (7) · 2883-2883

DOI: 10.1149/ma2026-0172883mtgabs ↗

Kurz gefasst

Fachartikel in der Zeitschrift ECS Meeting Abstracts, erschienen am 07.07.2026. Verfasst von Bryce De Muth und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Successful deployment of lithium-ion battery (LIB) systems requires accurate, real-time insight into internal states and degradation mechanisms. However, quantifying battery aging remains challenging due to limited observability. Existing methods involve tradeoffs between accuracy, diagnostic time, and practicality for field deployment. This work presents a novel deep learning approach for aging estimation using electrochemical impedance spectroscopy (EIS) data. Traditional capacity measurements (e.g., a C/20 discharge) can take a full day, whereas EIS provides feature-rich electrochemical data in minutes. EIS captures key aging indicators such as charge-transfer resistance, double-layer capacitance, and solid-state diffusion resistance. 1 Consequently, frequency-based equivalent circuit model (ECM) parameterization has proven effective for state of health (SOH) estimation. 2 However, generalizing these methods across cell chemistries, aging conditions, and datasets introduces complexity that data-driven machine learning (ML) approaches are well positioned to address. While ML models have already proven effective for battery state estimation, predicting discharge behavior directly from EIS data remains largely unexplored. This work addresses this gap by presenting a novel ML pipeline that predicts full voltage-vs.-capacity (V(Q)) and differential capacity (dQ/dV) curves from measured impedance. Full discharge curve prediction enables deeper understanding of the interplay between aging and performance losses. Notably, SOH can be directly extracted from the predicted discharge curves. This enables scalable, EIS-based SOH estimation without requiring a full discharge. Mapping low-dimensional EIS inputs to high-dimensional discharge curves is an ill-conditioned inverse problem, which we address using principal component analysis (PCA). Target discharge curves are compressed into a small number of principal components, selected to balance reconstruction accuracy and computational efficiency. A multilayer perceptron (MLP) neural network, utilizing a 32-dimensional bottleneck layer, is then trained to map a 38-dimensional EIS input vector (comprising real and imaginary impedances across 19 frequencies) to these PCA coefficients. The pipeline was validated using a cell-based train/validation/test split on two datasets with distinct form factors and cycling conditions. The first dataset comprises data from LiCoO 2 (LCO) coin cells cycled at a 1C rate. 3 The second dataset contains data collected from LiNiMnCoO 2 (NMC) cylindrical cells with diagnostic cycles conducted at a C/20 rate. 4 The unique characteristics of each dataset help support the robustness and applicability of the method. Results across distinct state of charge (SOC) conditions confirm the method is SOC-agnostic. The ML model significantly outperforms a mean-curve baseline, reducing V(Q) discharge curve prediction root mean squared error (RMSE) by 80.5% and 60.7% on the LCO and NMC datasets, respectively. On the more feature-rich dQ/dV curves, error was reduced by 68.9% (LCO) and 36.6% (NMC). SOH estimation achieved a mean absolute percentage error (MAPE) of less than 1% across both datasets (0.72% for LCO and 0.58% for NMC). This work suggests that EIS-based ML pipelines can serve as a rapid and accurate alternative to discharge-based aging characterization across diverse cell chemistries and cycling conditions. References: 1. P. Iurilli, C. Brivio, and V. Wood, J. Power Sources , 505 , 229860 (2021). 2. C. J. Chu, S. Thatipamula, and S. Onori, J. Electrochem. Soc. , 172 , 100540 (2025). 3. Y. Zhang et al., Nat. Commun. , 11 , 1706 (2020). 4. G. Pozzato, A. Allam, and S. Onori, Data Brief , 41 , 107995 (2022). Figure 1

Zitieren

APA

Muth, B. D., Jullien, T., Thatipamula, S., & Onori, S. (2026). Learning Discharge Behavior from Electrochemical Impedance Spectroscopy Data: A Deep Learning Framework for Rapid Li-Ion Battery Aging Diagnostics. ECS Meeting Abstracts, MA2026-01(7), 2883-2883. https://doi.org/10.1149/ma2026-0172883mtgabs

Vancouver

Muth BD, Jullien T, Thatipamula S, Onori S. Learning Discharge Behavior from Electrochemical Impedance Spectroscopy Data: A Deep Learning Framework for Rapid Li-Ion Battery Aging Diagnostics. ECS Meeting Abstracts. 2026;MA2026-01(7):2883-2883. doi:10.1149/ma2026-0172883mtgabs

BibTeX
@article{Muth2026,
  author = {Muth, Bryce De and Jullien, Tom-Eliot and Thatipamula, Sai and Onori, Simona},
  title = {Learning Discharge Behavior from Electrochemical Impedance Spectroscopy Data: A Deep Learning Framework for Rapid Li-Ion Battery Aging Diagnostics},
  journal = {ECS Meeting Abstracts},
  year = {2026},
  volume = {MA2026-01},
  number = {7},
  pages = {2883--2883},
  doi = {10.1149/ma2026-0172883mtgabs}
}
Kostenlos starten →