Machine learning models accelerate deep eutectic solvent discovery for the recycling of lithium-ion battery cathodes

Abstract

Deep eutectic solvents (DESs) have been widely applied to recover spent lithium-ion batteries (LIBs); however, developing effective and efficient systems for cathode leaching via the traditional trial-and-error method requires substantial efforts. This work aims to accelerate the discovery of novel promising DESs by leveraging the conditional Generative Adversarial Network (CGAN). Three databases were constructed: (i) DESs leaching cathodes, (ii) DESs leaching metal oxides, and (iii) DES properties. The absolute Spearman's rank correlation and agglomerative hierarchical clustering analysis ensured the selection of an optimal feature set for building predictive models. An XGBoost model was developed, achieving remarkable performance (R² = 0.9702, MSE = 0.0007) in predicting cathode solubility in DESs. We employed the Shapley additive explanation (SHAP) method to quantify the importance of acidity, coordination, and reducibility of DESs and provide insights into further research. To accelerate time-consuming investigational procedures, a CGAN model was established, rapidly identifying promising DESs like ChCl : Glycolic acid, with excellent agreement between predictions and experimental results. This study offers a general data analysis framework for other metal oxides (e.g., Cu_xO, Fe_xO_y, ZnO) leaching using DESs, enabling accurate solubility prediction and deepening the understanding of cathode leaching mechanisms. The CGAN model significantly accelerates the development of a DES-based process for lithium-ion cathode recycling, saving development time and effort. Overall, this work facilitates the efficient discovery and development of effective DESs for the recovery of valuable metals from spent LIB cathodes.