Machine learning surrogate modeling of materials extrusion for lithium-ion battery electrode dry manufacturing

Dry electrode manufacturing, such as extrusion and 3D-printing, offers a promising path toward more sustainable and cost-efficient lithium-ion battery production. However, optimizing extrusion-induced microstructure formation using only high-fidelity particle simulations remains computationally expensive. In this study, we develop a machine learning surrogate modeling framework to accelerate discrete element method (DEM)-based extrusion process analysis and inverse design. Six material and process parameters were selected using Latin Hypercube Sampling, and the corresponding evolutions of microstructure were generated from DEM simulations. The simulated time-dependent microstructure descriptors were further processed into quasi-steady state metrics, including void fraction, tortuosity factor, and effective (Binder and Carbon nanofiber)-network connectivity descriptors. Multiple regression algorithms were systematically benchmarked to construct fast surrogate models for predicting extrusion-induced microstructural states. The optimized models showed good predictive capability across the target descriptors, enabling rapid exploration of material-process-microstructure relationships at a substantially lower computational cost than direct DEM simulation. Shapley additive explanations were further employed to interpret parameter contributions and identify the dominant factors. The trained surrogate models were then coupled with single- and multi-objective optimization strategies to demonstrate inverse process design for targeted pore volume fraction, reduced tortuosity, and enhanced electronic connectivity. In addition, the time-series microstructural data were leveraged to further improve the predictive capability of the surrogate framework. This work establishes a data-efficient and interpretable workflow that integrates particle-scale simulations, machine learning, and process optimization, providing a scalable pathway for AI-accelerated manufacturing design of next-generation energy storage electrodes.