Enhancing lithium-ion battery slurry extrusion through neuro-adaptive controller and predictive modeling

Abstract

Industrial-scale extrusion of lithium-ion battery slurries is a complex process where maintaining stability is critical for final product quality. This study introduces a neuro-adaptive controller designed to optimize and stabilize cathode slurry extrusion, addressing the common industrial challenge of limited data availability. The core innovation of our work is the integration of a data-efficient predictive model, which functions as a high-fidelity AI simulator, directly into a feedforward control loop. This hybrid approach uniquely addresses the challenge of process optimization in data-scarce environments by enabling virtual exploration of the entire operational parameter space, eliminating the need for costly and time-consuming physical experiments. We first generated a dataset (68 data points) through systematic experiments varying solids content, screw speed, and mass flow rate. We then demonstrate that a gradient boosting regressor model outperforms a more complex deep neural network for this sparse industrial dataset, establishing it as the ideal foundation for our AI simulator. The neuro-adaptive controller leverages this trained AI simulator to proactively predict critical process outputs (specific energy and torque) and identify optimal input parameters that minimize process deviations. This integrated approach, validated in a laboratory setting, confirms that our neuro-adaptive controller framework enhances the stability and efficiency of slurry extrusion, presenting a practical and data-efficient pathway toward smart battery manufacturing.