Data-Driven Health Prognostics of NMC Lithium-Ion Batteries via Impedance Spectroscopy Using a Hybrid CNN-BiLSTM...

Area of Science:

• Electrochemistry
• Materials Science
• Artificial Intelligence

Background:

• Accurate battery health prognostics are vital for electric vehicles and electronics.
• Electrochemical Impedance Spectroscopy (EIS) combined with AI shows promise for LiCoO2 cells.
• Broader validation across chemistries and formats is needed for practical energy storage systems.

Purpose of the Study:

• To develop and validate an EIS-AI framework for battery prognostics.
• To assess performance on NMC111 cylindrical and NMC811 pouch cells.
• To demonstrate the framework's scalability and interpretability for real-world applications.

Main Methods:

• Developed a hybrid Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) model.
• Trained the model on over 13,000 EIS spectra from NMC111 and NMC811 cells.
• Analyzed salient features to identify degradation-related frequency regimes.

Main Results:

• Achieved R2 > 0.92 for state of health estimation and R2 > 0.90 for remaining useful life prediction.
• Demonstrated robust performance across different battery chemistries and cycling protocols.
• Identified chemistry- and protocol-dependent impedance features linked to battery degradation.

Conclusions:

• EIS spectra are physically informative for data-driven battery prognostics.
• The developed EIS-AI framework is scalable and interpretable.
• This approach offers a viable pathway for advanced battery management systems (BMSs).