Understanding the pH-responsive behavior of sodium bentonite: Experimental characterization and machine learning modeling for process optimization

Abstract

Sodium bentonite, a smectite clay widely found in boron ore deposits such as colemanite, plays a crucial role in numerous industrial processes-including ceramics, wastewater treatment, and boric acid production-due to its high surface reactivity and ion-exchange capacity. However, its pH-dependent physicochemical behavior often causes filtration inefficiencies, flocculation challenges, and product quality issues. Despite extensive research on rheological and surface properties, integrated studies that combine experimental characterization with predictive data-driven modeling remain scarce, creating a gap in understanding and optimizing pH-sensitive clay systems. This study aims to bridge this gap by providing a comprehensive analysis of the physicochemical and morphological behavior of sodium bentonite across a broad pH range (1-9). Experimental analyses included pH equilibration monitoring, viscosity measurement, turbidity assessment, filtration performance, particle size distribution (via laser diffraction), and surface morphology characterization (SEM and FTIR). The results revealed that below the zero point of charge (pHzpc approximate to 8.5), bentonite surfaces become positively charged, promoting flocculation and improving filterability, while above this point, dispersion stability and viscosity increased, resulting in yellowish filtrates. FTIR spectra confirmed the chemical stability of bentonite across pH variations. Machine learning models-Linear Regression, Random Forest, Support Vector Machines, Gaussian Process Regression, and Boosted Trees-were employed to predict key process outputs (pH equilibrium, conductivity, and ion concentrations). Among these, Boosted Trees and GPR achieved the highest prediction accuracy (R2 > 0.95), demonstrating strong potential for process optimization. This work offers a data-driven framework to understand and control pH-sensitive processes involving sodium bentonite, providing valuable insights for industrial applications such as ceramics, wastewater treatment, drilling fluids, and mineral processing. The integration of experimental and machine learning approaches not only enhances predictive accuracy but also supports the design of more sustainable and efficient filtration systems.