Integration of COMSOL Multiphysics Simulations and Machine Learning Algorithms for Predictive Optimization of Lithium-Ion Battery Efficiency

High-energy lithium-ion batteries demand cathode architectures capable of sustaining high areal capacities without compromising electrochemical or thermal stability. Among layered oxide cathodes, LiNi 0.8 Mn 0.1 Co 0.1 O 2 (NMC811) offers high specific capacity (∼200 mAhg −1 ) and reduced cobalt content, but thick, high-mass-loading electrodes typically exhibit severe Li + transport limitations, polarization, and thermal heterogeneity. To address these issues, an integrated framework is developed, combining COMSOL-based pseudo-two-dimensional (P2D) simulations with machine learning algorithms to enable rapid prediction and optimization of structure-property-performance relationships. High-fidelity simulations are performed across electrode thicknesses of 60–240 μm and porosities of 20%–45%, generating datasets that capture coupled ion diffusion, charge transport, and heat generation. Trained ensemble learning and deep neural network models achieve R 2 > 0.95 for key electrochemical and thermal outputs. Bayesian multi-objective optimization identifies Pareto-optimal electrode architectures, achieving 178 mAhg −1 with minimized overpotential (<45 mV) and temperature rise (<8 °C). This hybrid physics-ML strategy establishes a scalable platform for accelerating thick-electrode design and provides quantitative insights into the critical role of transport and thermal dynamics in next-generation high-energy lithium-ion batteries.