Session: 17-08-01: Renewable Energy Systems

Paper Number: 174027

An Advanced Machine Learning Framework to Predict Faulty Batteries From Ageing and Electrochemical Impedance Spectroscopy Data 

Sudden acceleration in battery ageing observed during long-duration battery cycling can provide useful insights into cell degradation and may also indicate the existence of defects in batteries. The ageing and degradation process of cells involves different mechanisms, such as the formation of a solid-electrolyte interphase (SEI), loss of active material, and lithium plating, etc. Rigorous battery cycling under various operating conditions to generate meaningful ageing data can be expensive and time-consuming. Moreover, traditional physics-based simulations to predict cell ageing suffer from computational complexity, long run-time requirements. Accurate early prediction of cell lifetime and ageing is therefore crucial to designing safe and reliable battery packs. Various techniques of battery ageing prediction models proposed in recent years involves equivalent circuit models (ECM), electrochemical models, and machine learning frameworks. While ECMs provide fast estimations at low accuracy due to a lack of electrochemical information, high-fidelity electrochemical models are complex and may have limited success due to the non-linear ageing pattern of cells. In contrast, machine learning models do not require an in-depth understanding of electrochemistry or ageing dynamics, and can provide a fast and meaningful prediction of capacity fading with feature extraction and pattern recognition techniques. 

Recent studies in published literature [1] reported changes in electrochemical impedance spectroscopy (EIS) and capacity fade as batteries age. Studies at Sandia National Labs [2] further demonstrated EIS signals as an early indicator of battery degradation near thermal runaway. Such results motivate us strongly to explore possible correlations between EIS data with cell capacity fade to predict battery ageing and failure. To start with, we use the dataset published in [1] where the authors reported EIS information and battery cell capacity (Ah) of twenty 18650 Graphite/LFP 1.1 Ah commercial cells. These measurements were recorded at every 100 cycles over thousands of charge/discharge aging cycles for each cell at three different state-of-charge (SOC = 0%, 50%, and 100%). Using this dataset, we will use machine learning (ML) to learn the mapping between the current cycle’s EIS curve and cell capacity to the cell capacity at the next 100 cycles. This work would demonstrate the connection between EIS curves and cell capacity fading, indicating there is a learnable relationship between the two and that EIS can be used as an early indicator of battery degradation. To do this, we plan to utilize a 3-channel 1-D convolutional neural network (CNN) where the real and imaginary parts of the EIS curve are input along with the current cell capacity, and the cell capacity at the next 100 cycles is predicted. We also plan to incorporate uncertainty quantification (UQ) for the neural network using the implementation in [3] which will provide us confidence bounds on our cell capacity predictions. The success of this modeling effort will guide us for further extension of such predictive approach to different battery chemistries.

References:

[1] Wheeler, William & Venet, Pascal & Sari, Ali. (2025). An ageing study of twenty 18650 lithium-ion Graphite/LFP cells in first and second life use. Scientific Data. 12. 10.1038/s41597-025-04712-7.

[2] Loraine Torres-Castro et al. (2024). Early Detection of Li-Ion Battery Thermal Runaway Using Commercial Diagnostic Technologies. J. Electrochem. Soc. 171 020520

[2] Winovich, Nick, Karthik Ramani, and Guang Lin (2019). ConvPDE-UQ: Convolutional neural networks with quantified uncertainty for heterogeneous elliptic partial differential equations on varied domains. Journal of Computational Physics 394: 263-279.

Presenting Author: Rishi Roy Sandia National Laboratories

Presenting Author Biography: Rishi Roy is currently a researcher at Sandia National Labs, Livermore, California. His primarily interest is in battery safety, laser diagnostics in combustion, hydrogen, and machine learning. He has co-authored several peer reviewed journals, conference articles, and technical reports in these areas. He regularly serves as a reviewer to several ASME, AIAA, Elsevier journals and different funding oppotunity proposals for the Department of Energy. He is a receipent of different funding awards at Sandia, and various prestigious awards and fellowships by the AIAA and NSF.

Authors: