Abstract
<jats:p>The escalating global demand for renewable energy integration and electric vehicles necessitates the development of cost-effective, safe, and high-performance energy storage systems capable of operating under extreme conditions. While Lithium-ion technologies dominate the market, resource scarcity and safety concerns have spurred intense interest in Sodium-ion capacitors (SICs) as a strategic alternative. Despite their potential for high power and energy density, the practical deployment of SICs is severely impeded by the uncontrolled growth of sodium dendrites, particularly across wide temperature ranges (−40 °C to 60 °C). This complex interfacial instability leads to capacity fading and catastrophic safety hazards, posing a significant challenge for precise lifetime prediction and risk assessment.In this work, we propose a physics-informed deep learning framework that integrates high-fidelity multiphysics modeling with deep operator networks (DeepONet) to predict sodium dendrite growth and capacity retention. A fully coupled electrochemical–thermal–mechanical model with arbitrary Lagrangian-Eulerian (ALE) moving mesh is established to accurately simulate dendrite nucleation, propagation, and morphological evolution. To address the scarcity of experimental data and mitigate model overfitting, we adopt a variational autoencoder (VAE)-assisted strategy to construct high-fidelity synthetic datasets. significantly strengthening generalization performance. Benchmark evaluations demonstrate that the DeepONet model effectively learns the infinite-dimensional operator mapping among temperature, current density, cycle number, dendrite length, and capacity retention. Furthermore, we successfully realize the inverse inference of dendrite size from capacity fading, providing a quantitative tool for non-invasive safety assessment. This work offers a universal paradigm for modeling interfacial instability in complex electrochemical systems, paving the way for the reliable design of wide-temperature energy storage devices.</jats:p>