Scale-Dependent hydroxyl radical generation and energy efficiency in vortex diode hydrodynamic cavitation: machine learning insights toward industrial-scale applications in water treatment

Hydrodynamic cavitation (HC) reactors are increasingly applied in the remediation of organic pollutants in water, leveraging intense shear and hydroxyl radical (OH•) generation to accelerate degradation processes. However, scaling up HC devices remains a challenge in environmental engineering due to poorly understood effects at different scales of operation. Vortex-based HC (VDs) provide superior cavitation efficiency compared to traditional orifice and venturi devices and therefore this study examines the scale-dependent generation of OH• in VDs. Coumarin dosimetry was employed as the quantification method for OH• generation. Available published experimental datasets were analysed: (i) varying inlet pressures from 100 to 400 kPa for throat diameters of 6 and 12 mm, and (ii) throat diameters ranging from 6 to 38 mm (nominal capacities of 5 to 200 L/min) at a fixed pressure drop of 280 kPa. After normalization of the available data, seven machine learning (ML) models were trained to establish relationships between operating conditions and OH• generation performance. eXtreme Gradient Boosting (XGB) and artificial neural networks (ANN) outperformed the others, with higher R² and lower RMSE. After using SHAP interpretation, these two models were used to elucidate scale effects on both radical yield and energy efficiency, resulting in actionable design guidelines for VDs at different scales. By combining experimental dosimetry with predictive ML, this work advances the fundamental understanding and practical implementation of cavitation-based advanced oxidation technologies, particularly for efficient and energy-optimized treatment of organic pollutants in wastewater.