Learning continuous SOC-dependent thermal decomposition kinetics for Li-ion cathodes using KA-CRNNs

TL;DR

KA-CRNN introduces a physics-encoded neural framework to learn continuous SOC-dependent kinetics for cathode decomposition and electrolyte oxidation from DSC data. By embedding mechanistic reactions and representing SOC influence via Chebyshev-based activations, it yields interpretable, continuous parameters for R1, R2, and R3 across NM, NMA, and NCA cathodes. The model captures the critical SOC behavior, oxygen evolution, and electrolyte oxidation heat with good generalization to unseen SOC. This approach enables high-resolution, SOC-aware safety predictions and provides mechanistic insight into oxygen-mediated thermal runaway pathways.

Abstract

Thermal runaway in lithium-ion batteries is strongly influenced by the state of charge (SOC). Existing predictive models typically infer scalar kinetic parameters at a full SOC or a few discrete SOC levels, preventing them from capturing the continuous SOC dependence that governs exothermic behavior during abuse conditions. To address this, we apply the Kolmogorov-Arnold Chemical Reaction Neural Network (KA-CRNN) framework to learn continuous and realistic SOC-dependent exothermic cathode-electrolyte interactions. We apply a physics-encoded KA-CRNN to learn SOC-dependent kinetic parameters for cathode-electrolyte decomposition directly from differential scanning calorimetry (DSC) data. A mechanistically informed reaction pathway is embedded into the network architecture, enabling the activation energies, pre-exponential factors, enthalpies, and related parameters to be represented as continuous and fully interpretable functions of the SOC. The framework is demonstrated for NCA, NM, and NMA cathodes, yielding models that reproduce DSC heat-release features across all SOCs and provide interpretable insight into SOC-dependent oxygen-release and phase-transformation mechanisms. This approach establishes a foundation for extending kinetic parameter dependencies to additional environmental and electrochemical variables, supporting more accurate and interpretable thermal-runaway prediction and monitoring.

Figures (10)

• Figure 1: Overview of KA-CRNN framework for SOC-dependent thermal runaway kinetics. (A) KA-CRNN kinetic parameters, represented as continuous functions of SOC. (B) Standard CRNN kinetic parameters, represented as fixed scalar values without SOC dependence. (C) Example KA-CRNN prediction illustrating the continuous variation of heat-release behavior with SOC. (D) Complete KA-CRNN architecture used in this study. Reactions R1 and R2 are defined by KA-CRNN parameters as in panel (A) and are identified separately for each cathode material. Reaction R3 uses scalar CRNN parameters as in panel (B) and is learned once for all cathodes, but its effective behavior remains SOC dependent through its reliance on O$_2$ production from R2.
• Figure 2: DSC training data reproduced from cui_navigating_2025. Rows are NM, NMA, and NCA cathode materials from top to bottom. The y-axis scale changes beginning at the seventh column to improve visibility of the substantially increased heat release at elevated SOCs. Individual SOCs are indicated on each subfigure in mAh.
• Figure 3: (A) DSC reconstructions for all three studied cathodes (NM, NMA, and NCA, from top to bottom), at three example SOCs (minimum, testing, and maximum from left to right). SOCs are labeled in mAh (black for training, red for testing). Note y-axes are scaled to each column for visibility. (B) All nine SOCs for all three cathodes plotted together, showing significantly distinct trends in peak heat release, peak heat release temperature, and total heat release as the SOC increases. (C) Oxygen evolution stoichiometric coefficient KA-CRNN activations for each cathode material, highlighting critical SOCs around 210 mAh to 230 mAh across all materials
• Figure 4: Learned KA-CRNN activations for the two frequency factors (A-B) and two activation energies (C-D) of all three cathode materials. Note increase in ln$A_2$ and $E_{a, 2}$ occurs at a higher SOC for NMA cathode, corresponding to its higher critical SOC. KA-CRNN activations and discussion for all remaining kinetic parameters are available in Appendix A.
• Figure A1: Learned KA-CRNN activations for the NM cathode. Individual green Chebyshev polynomials sum to the black KA-CRNN parameter functions.