Zero-valent iron based materials selection for permeable reactive barrier using machine learning

Area of Science:

• Environmental Engineering
• Materials Science
• Data Science

Background:

• Zero-valent iron (ZVI) based reactive materials are crucial for permeable reactive barriers (PRBs) in groundwater remediation.
• Selecting effective ZVI materials is challenging due to the vast number of emerging iron-based materials and the need for long-term stability.
• Existing data limitations hinder the practical application of advanced screening methods for PRB reactive materials.

Purpose of the Study:

• To develop and validate a machine learning (ML) approach for efficiently screening ZVI-based reactive materials for PRBs.
• To integrate ML models with evaluation indices and experimental data to enhance material selection practicality.
• To identify key material properties influencing the performance of ZVI in remediation applications.

Main Methods:

• Utilized the XGBoost model to predict kinetic data for ZVI-based materials.
• Employed SHAP (SHapley Additive exPlanations) to enhance model accuracy and interpret feature importance.
• Conducted batch and column tests to evaluate geochemical characteristics and validate ML predictions.
• Investigated mechanistic pathways and endpoint products of iron compound transformations.

Main Results:

• Machine learning, combined with experimental data, significantly improved the prediction accuracy of ZVI material performance (RMSE reduced from 1.84 to 0.6).
• SHAP analysis identified specific surface area as a fundamental factor correlating with kinetic constants of ZVI-based materials.
• Experimental results demonstrated that ZVI exhibited superior performance compared to AC-ZVI, with 3.2 times higher anaerobic corrosion reaction kinetic constants and 3.8 times lower selectivity.

Conclusions:

• This study presents a successful initial application of machine learning for the selection of reactive materials in PRBs.
• The developed ML approach enhances the efficiency and practicality of identifying optimal ZVI-based materials for environmental remediation.
• Specific surface area is a critical parameter for optimizing ZVI material selection, guiding future material design and application.