Machine Learning and Analytical Approaches to Predict the Direct Aqueous Mineral Carbonation Efficiency of Olivine-Rich Rocks

Document Type

Article

Publication Date

9-8-2025

Department

Department of Chemical Engineering

Abstract

Direct aqueous ex situ mineral carbonation of CO2-reactive silicate minerals involves the reaction of silicate minerals with carbon dioxide (CO2) to form stable carbonate minerals. Previous studies have shown that the efficiency of mineral carbonation depends on both process variables and feed mineralogy. However, modeling tools for predicting carbonation efficiency remain limited. In this study, two categories of models were developed to predict mineral carbonation efficiency and CO2uptake. These two approaches include (a) an analytical model based on a first-order reaction and (b) six data-based machine learning (ML) models. Olivine-rich rocks were used as the feed materials, and both the carbonation efficiency and the CO2uptake were determined using a direct mineral carbonation protocol. The experimental results were compared with predictions from both analytical and ML models. The analytical model showed fair agreement with the experimental data. In contrast, the ML models demonstrated superior predictive performance, provided that a sufficient data set is available for training. Accuracy further improved when multiple models were integrated into an ensemble, yielding a root mean squared error value of 7.72. Feature importance analysis from ML models identified key processes and input variables influencing carbonation efficiency. This work demonstrates the utility of both analytical and ML models for predicting mineral carbonation efficiency and highlights the relative importance of process variables in the direct ex situ mineral carbonation.

Publication Title

Energy and Fuels

Link to Full Text

Share

COinS