Machine Learning-Driven Inverse Design for Low-Carbon and Cost-Effective Organic Acid Leaching of Spent Ternary Lithium Batteries

Organic acid leaching is an effective and sustainable method for simultaneously recovering critical metals from ternary lithium batteries (T-LIBs). However, current methods overlook the structural impact of organic acids and rely on inefficient trial-and-error condition optimization. Herein, machine learning (ML) models are introduced to identify critical organic acid molecular descriptors and achieve inverse design of leaching systems through multialgorithm integration, simultaneously improving efficiency while reducing experimental costs and carbon emissions. Specifically, a comprehensive data set of 4,356 samples, incorporating acid structure fingerprints, metal properties, and operational conditions, was collected. Dimensionality reduction, fingerprint feature assessment, seed randomness, and data leakage management were conducted to enhance the robustness of the model. The optimized Extra Trees model achieved an RMSE of 13.49 and an R2 of 0.81 for leaching efficiency prediction, SHAP analysis quantitatively identified the dominant variables governing leaching efficiency and demonstrated their complex interrelationships. The inverse design conditions reduced cost and GHG emissions by 42–89% compared to reported experimental optimization based on the same set of samples, and achieved <10% relative error in experimental validation. This study conducts a systematic analysis of leaching influencing factors and offers a universally applicable protocol for chemistry-adaptive T-LIBs recycling process optimization.