Particle-scale design of lithium-ion battery cathodes for manufacture by coupling physics-based modelling with machine learning interpretation

Lithium-ion battery performance is significantly affected by electrode manufacturing processes and corresponding electrode microstructures. This work proposes a digital workflow that integrates physics-based high-fidelity models and light machine learning (ML) surrogates to provide insights into particle-scale microstructural evolutions impacting battery performance. For this purpose, Discrete Element Method (DEM) simulations combined with a Carbon Binder Domain (CBD) phase algorithm are developed to model the 3D battery electrode microstructure evolution during the calendering process of electrode manufacture with a sub-micron resolution. Through the detailed model, a comprehensive dataset of electrode properties, including tortuosity, electrical conductivity, and electrochemical performance, is synthesized. Following on this, an accurate ML surrogate is trained to quantify the impact of manufacturing and microstructural key factors and parameters on the battery performance. Interdependency analysis demonstrates that a specific feature space exists to determine battery energy capacity in manufacturing. Specifically, the synergistic combination of reduced calendering (from 0.65 to 0.54) and increased CBD micro-porosity (from 0.3 to 0.6) enhances the specific capacity by 20–60 %. The proposed computational strategy bridges high-fidelity modelling and lightweight surrogates to accelerate electrode design. This workflow supports predictive analysis and reduces the experimental workload, in the process of the electrode and battery cell performance optimisation of the next generation lithium-ion batteries.