Learning continuous SOC-dependent thermal decomposition kinetics for Li-ion cathodes using KA-CRNNs
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.
Paper Prompts
Sign up for free to create and run prompts on this paper using GPT-5.
Top Community Prompts
Explain it Like I'm 14
Overview
This paper is about making lithium-ion batteries safer. The authors study how batteries heat up and can sometimes catch fire (called “thermal runaway”), and they build a smart, physics-aware computer model that predicts how this danger changes as the battery’s state of charge (SOC) goes from low to high. Think of SOC like how full a fuel tank is: higher SOC often means more energy and potentially more risk.
What questions does the paper try to answer?
To keep things simple, the paper focuses on these questions:
- How does the battery’s SOC change the way the cathode (one of the battery’s main parts) breaks down and releases heat?
- Why is there a “critical” SOC where the heat release suddenly becomes much worse?
- Can we build a model that smoothly follows these changes across all SOCs (not just at 100% SOC) and still make sense scientifically?
How did they study it? Methods explained simply
The authors use lab measurements and a physics-guided machine learning model.
- The lab tool: Differential Scanning Calorimetry (DSC). You can think of DSC like a super-precise oven thermometer that tells you how much heat a sample releases as you warm it up. They tested mixtures of cathode material and liquid electrolyte at different SOCs.
- The materials: They tested three popular nickel-rich cathodes:
- NM (Nickel-Manganese)
- NMA (Nickel-Manganese-Aluminum)
- NCA (Nickel-Cobalt-Aluminum)
- The reactions they model:
- The cathode changes from a “layered” structure to a “spinel” structure.
- Then it changes from “spinel” to “rock salt,” and this step can release oxygen (O2).
- The released oxygen reacts with the electrolyte (the liquid in the battery), which produces a lot of heat—like oxygen fueling a fire.
Here’s a simple way to see those steps:
1 2 3 4 5 |
R1: Layered cathode -> Spinel cathode + heat R2: Spinel cathode -> Rock salt cathode + oxygen + heat R3: Oxygen + Electrolyte -> Combustion products + a lot of heat |
- The model: KA-CRNN (Kolmogorov–Arnold Chemical Reaction Neural Network)
- “CRNN” means the model uses real chemical rules (like Arrhenius and mass-action laws) and machine learning together. It respects physics while learning from data.
- “KA” means the model lets key parameters (like activation energy, reaction speed, and how much oxygen is released) change smoothly with SOC. Imagine knobs that turn a little more as the battery’s charge increases. The model learns exactly how much to turn each knob.
- Because the oxygen released in R2 feeds R3, changes in oxygen release strongly affect the heat from the electrolyte reaction. That’s how the model captures sudden jumps in heat at high SOCs.
- They also add gentle rules during training to keep the model realistic (for example, not letting reaction “orders” go to silly values, and encouraging oxygen release to grow with SOC when evidence says it should).
What did they find, and why does it matter?
- Continuous, SOC-dependent behavior: The model didn’t just work at 100% SOC. It predicted how heat release and peak temperature change across all SOCs, which is how batteries are used in real life.
- A critical SOC “tipping point”: The model matches experimental results showing that above a certain SOC (roughly around 210–230 mAh for these materials), the battery releases much more heat and the heat peaks become sharp and dangerous. This is linked to a sudden increase in oxygen release from the cathode at high SOCs. More oxygen means more reaction with the electrolyte, which means more heat—like throwing fuel on a fire.
- Differences among cathodes:
- NMA and NCA are generally more stable than NM at lower SOCs (they are less “hot” and start heating later).
- The critical SOC (where things get much worse) is higher for NMA than for NCA, showing NMA tends to stay safer longer.
- Shared electrolyte behavior: The model learned one common set of parameters for the electrolyte oxidation (R3) across all cathodes, while letting the cathode steps (R1 and R2) vary with SOC. This makes the model simpler and more general, and it still works accurately.
- Works on unseen data: The model predicted heat release at an SOC that was not used during training, showing it wasn’t just memorizing the data.
Why is this important? Impact and future possibilities
- Better battery safety: Instead of assuming the worst case (100% SOC) all the time, this model can predict risk at any SOC—useful for day-to-day operation, storage, and charging.
- Early warning and monitoring: Devices could use models like this to watch for dangerous behavior in real time and take action (like cooling, limiting charging, or communicating risks).
- Design and policy guidance: Engineers and safety regulators can better set SOC limits for safer use and storage, especially in electric vehicles and energy storage systems.
- Expandable to more conditions: The same approach could be extended to include other factors (like pressure, temperature, or voltage), or to full battery systems including the anode and separator, helping build comprehensive and interpretable safety models.
In short, this paper shows how a physics-aware AI model can learn and explain how battery danger ramps up with charge level—capturing a critical tipping point—and it lays the groundwork for safer, smarter batteries.
Collections
Sign up for free to add this paper to one or more collections.