Amorphous-dominated MgO hollow spheres enhanced fluoride adsorption: Mechanism analysis and machine learning prediction

Lin Fan; Dexi Wang; Honglei Yu; Jinyuan Guo; Yan He

doi:10.1063/5.0246431

Amorphous-dominated MgO hollow spheres enhanced fluoride adsorption: Mechanism analysis and machine learning prediction

J Chem Phys. 2025 Jan 7;162(1):014702. doi: 10.1063/5.0246431.

Authors

Lin Fan^{1

2}, Dexi Wang¹, Honglei Yu^{1

3}, Jinyuan Guo¹, Yan He⁴

Affiliations

¹ School of Mechanical Engineering, Shenyang University of Technology, Shenyang 110870, Liaoning, China.
² Engineering Training Centre, Liaoning Petrochemical University, Fushun 113001, Liaoning, China.
³ School of Chemical Equipment, Shenyang University of Technology, Liaoyang 111000, Liaoning, China.
⁴ School of Mechanical Engineering, Liaoning Petrochemical University, Fushun 113001, Liaoning, China.

PMID: 39745152
DOI: 10.1063/5.0246431

Abstract

Amorphous-dominated magnesium oxide hollow spheres (A-MgO) were prepared using a spray-drying method in this study. These hollow spheres exhibited excellent sphericity, large specific surface areas, and abundant porosity. A-MgO exhibited outstanding fluoride adsorption properties, with a maximum adsorption capacity of 260.4 mg/g. When the pH value was less than 8, the fluoride removal percentage remained more than 87.4%. Moreover, the removal percentage remained above 75% even after five application cycles. In addition, the research revealed that SO42-, CO32-, and PO43- exerted a more pronounced effect on fluoride removal, whereas coexisting ions such as Br-, Cl-, NO3-, and HCO3- had minimal impact on this process. An in-depth analysis of the adsorption mechanism demonstrated that the process of fluoride adsorption by A-MgO involves various synergistic mechanisms, such as electrostatic adsorption, ion exchange, oxygen vacancy adsorption, physical adsorption, and pore filling. To predict the fluoride ion adsorption performance of A-MgO under complex conditions, a high-performance machine learning model, GBDT-S, was developed using hyperparameter optimization. The R2 of 0.99 and 0.80 for the training and testing datasets, respectively, with the RMSE of 3.26 and 3.89. Interpretative analysis using SHapley Additive exPlanations technology indicated that reaction time, PO43- concentration, and pH were key factors influencing the fluoride ion removal percentage.