A Physics-Informed Fourier Neural Operator Framework for Zero-Plating Fast Charging of Lithium-Ion Batteries via Real-Time Electrochemical State Estimation and Barrier-Riding Optimal Control
- 1Institute of Chemistry
- 2Islamia University Bahawalpur
Abstract
Extreme fast charging of lithium-ion batteries is critically limited by internal lithium plating at the negative electrode-separator interface, driving rapid degradation and thermal hazards. Real-time mitigation is fundamentally impeded because the interfacial plating overpotential (ηplating) is inaccessible to non-invasive sensors, leaving conventional protocols blind to metallic lithium deposition. While high-fidelity Doyle–Fuller–Newman (DFN) models resolve these dynamics, their computational cost (∼ 8 s) precludes embedded closed-loop control, whereas empirical equivalent circuit models lack the spatial resolution to detect plating. To resolve this trade-off, we develop a physics-informed 1D Fourier Neural Operator (FNO-1D) surrogate incorporating an explicit cumulative charge integral Q(t) = ∫ I dt to preserve mass conservation. Trained on 7,771 continuum DFN trajectories, the surrogate achieves a 4, 349× speedup (1.84 ms inference) with high fidelity (V RMSE = 5.40 mV, η plating , RMSE = 2.24 mV). Coupled with an Ensemble Smoother with Multiple Data Assimilation (ESMDA), the framework filters sensor noise (±10 mV to 5.94 mV) and reconstructs unobservable plating dynamics (RMSE = 4.12 mV, 95% CI = ±3.22 mV), establishing continuous closed-loop observability. A Barrier-Riding Nonlinear Model Predictive Control (NMPC) policy then maximizes charging current along the degradation boundary. Across a 10% - 80% SOC window, it cuts charging time by 21.7% (33.7 min vs. 43.0 min CC-CV)while strictly preventing lithium plating (η plating ≥ +14.81 mV > +5.0 mV), limiting peak temperature to 38.5◦C, and decreasing SEI growth by 21.5%. Finally, transferability is experimentally validated across two phys ical cycler benchmarks: cross-chemistry transfer on Oxford pouch cells, and chemistry-matched validation on commercial RWTH Aachen Sanyo 18650 NMCcells, achieving a terminal voltage tracking RMSE of 33.02 mV (86.0% error reduction over zero-shot transfer) providing an experimentally grounded paradigm for degradation-free fast charging.
Related
- Physics-Informed Explainable Machine Learning for Hybrid State of Health and Remaining Useful Life Estimation of Lithium-Ion Batteries
- A Physics-Informed TCN-LSTM Approach for Interpretable Internal Temperature Estimation of NMC811 Lithium-Ion Batteries
- Al-doped C3N4 composite cathodes suppress polysulfide shuttle and enable long-lifetime Li-S batteries with physics-informed NeuralODE lifetime forecasting
- Why Electrochemical–Thermal–Mechanical Understanding is Needed to Explain Knees in Lithium-Ion Batteries Degradation
- Electrochemical Properties and Changes of Crystal and Electronic Structures of Spinel-Type Mg1.33– y V1.67– x + y Mn x O4 ( x = 0.15, 0.06 ≤ y ≤ 0.13) for Magnesium Secondary Batteries
BibTeX
@misc{tariq2026physicsinformed,
title = {{A Physics-Informed Fourier Neural Operator Framework for Zero-Plating Fast Charging of Lithium-Ion Batteries via Real-Time Electrochemical State Estimation and Barrier-Riding Optimal Control}},
author = {Hussnain Tariq},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010874/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010874/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - A Physics-Informed Fourier Neural Operator Framework for Zero-Plating Fast Charging of Lithium-Ion Batteries via Real-Time Electrochemical State Estimation and Barrier-Riding Optimal Control
AU - Hussnain Tariq
PY - 2026
DA - 2026/10/22
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010874/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010874/v1
AB - Extreme fast charging of lithium-ion batteries is critically limited by internal lithium plating at the negative electrode-separator interface, driving rapid degradation and thermal hazards. Real-time mitigation is fundamentally impeded because the interfacial plating overpotential (ηplating) is inaccessible to non-invasive sensors, leaving conventional protocols blind to metallic lithium deposition. While high-fidelity Doyle–Fuller–Newman (DFN) models resolve these dynamics, their computational cost (∼ 8 s) precludes embedded closed-loop control, whereas empirical equivalent circuit models lack the spatial resolution to detect plating. To resolve this trade-off, we develop a physics-informed 1D Fourier Neural Operator (FNO-1D) surrogate incorporating an explicit cumulative charge integral Q(t) = ∫ I dt to preserve mass conservation. Trained on 7,771 continuum DFN trajectories, the surrogate achieves a 4, 349× speedup (1.84 ms inference) with high fidelity (V RMSE = 5.40 mV, η plating , RMSE = 2.24 mV). Coupled with an Ensemble Smoother with Multiple Data Assimilation (ESMDA), the framework filters sensor noise (±10 mV to 5.94 mV) and reconstructs unobservable plating dynamics (RMSE = 4.12 mV, 95% CI = ±3.22 mV), establishing continuous closed-loop observability. A Barrier-Riding Nonlinear Model Predictive Control (NMPC) policy then maximizes charging current along the degradation boundary. Across a 10% - 80% SOC window, it cuts charging time by 21.7% (33.7 min vs. 43.0 min CC-CV)while strictly preventing lithium plating (η plating ≥ +14.81 mV > +5.0 mV), limiting peak temperature to 38.5◦C, and decreasing SEI growth by 21.5%. Finally, transferability is experimentally validated across two phys ical cycler benchmarks: cross-chemistry transfer on Oxford pouch cells, and chemistry-matched validation on commercial RWTH Aachen Sanyo 18650 NMCcells, achieving a terminal voltage tracking RMSE of 33.02 mV (86.0% error reduction over zero-shot transfer) providing an experimentally grounded paradigm for degradation-free fast charging.
ER -