Skip to content
Open access

Multiscale Digital-Twin Modeling of Energy-Storage Materials by Integrating First-Principles Calculations and Machine-Learning Interatomic Potentials

Oct 2026 · EAI Endorsed Transactions on Energy Web · 0 citations · 45 references

Abstract

Conventional modeling of energy-storage materials is constrained by the computational cost of first-principles calculations, limited electrochemical-state awareness in machine-learning interatomic potentials, and weak coupling between materials-scale models and online battery decision-making. This study proposes a first-principles-constrained electrochemically conditioned machine-learning interatomic potential (FP-EMLP) integrated with a multiscale digital-twin framework (FP-EMLP-DT). A multistate configuration space spanning ion concentration, temperature, defects, strain, and phase boundaries is combined with active learning based on committee uncertainty, structural novelty, and physics-based risk. The equivariant graph model incorporates electrochemical-state gating, charge-conserving projection, and differentiable long-range electrostatics, and is trained using energies, forces, stresses, equilibrium voltages, and migration barriers. Predicted diffusion coefficients, free energies, elastic tensors, and chemical expansion coefficients are transferred to an electrochemical-mechanical-thermal-damage model. The resulting model is further integrated with a physics-based reduced-order network, uncertainty-gated residual learning, online data assimilation, and degradation-aware control. FP-EMLP achieves mean absolute errors of 6.5 meV atom−1, 33 meV Å−1, 0.051 V, and 0.038 eV for energy, force, voltage, and migration barrier, respectively, with a 4.8% aggregate relative error across multiscale fields. Active learning reduces required first-principles labels by 33.3% while covering 91.2% of high-risk configurations. FP-EMLP-DT achieves mean absolute errors of 0.93% for state of charge and 1.21% for state of health. Degradation-aware control reduces maximum material stress and damage increment by 16.5% and 36.8%, respectively, demonstrating an effective multiscale pathway from electronic structure to online battery management.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.