Ensemble Machine Learning Reveals Key Structural and Operational Features Governing Ion Selectivity of Polyamide Nanofiltration Membranes

Dan Lu,Xuanchao Ma,Jiancong Lu,Yukun Qian,Yifang Geng,Jing Wang,Zhikan Yao,Lijun Liang,Zhilin Sun,Songmiao Liang,Lin Zhang

DOI: https://doi.org/10.1016/j.desal.2023.116748

IF: 11.211

2023-01-01

Desalination

Abstract:Diversified ion-selective separation applications have dramatically incentivized the exploitation and performance modulation of highly ion-selective nanofiltration (NF) membranes. However, the ion selectivity of NF membranes is synergistically governed by multi-scale factors of membrane structural parameters and operational parameters, with the intrinsic ion-selective mechanism still ambiguous. Herein, we proposed an ensemble machine learning (ML) method to decouple key factors affecting the ionic selectivity of polyamide NF membranes. Membrane structural parameters and operational parameters were typically extracted as input variables and linked to mono−/divalent ion selectivity by model training based on Random Forest and XGBoost algorithms. The feature importance assessment indicated the critical role of membrane structure parameters on ion selectivity, wherein pore radius dominated the mono−/divalent anionic selectivity while zeta potential for mono−/divalent cationic selectivity. Partial dependence analyses further depicted intensive insights regarding the influence of membrane structural parameters on ion selectivity. Moreover, stochastic dataset splitting measurements demonstrated the accurate predictive capability of the model simultaneously possessing excellent stability and reliability. We anticipated that the implementation of ensemble ML in explicating the intricate ion-selective mechanism created platforms for understanding the structure-membrane performance correlation and orientally manufacturing highly ion-selective NF membranes.