A tailored and rapid approach for ozonation catalyst design.

Catalytic ozonation is widely employed in advanced wastewater treatment owing to its high mineralization of refractory organics. The key to high mineralization is the compatibility between catalyst formulation and wastewater quality. Machine learning can greatly improve experimental efficiency, while fluorescence data can provide additional wastewater quality information on the composition and concentration of organics, which is conducive to optimizing catalyst formulation. In this study, machine learning combined with fluorescence spectroscopy was applied to develop ozonation catalysts (Mn/γ-AlO catalyst was used as an example). Based on the data collected from 52 different catalysts, a machine-learning model was established to predict catalyst performance. The correlation coefficient between the experimental and model-predicted values was 0.9659, demonstrating the robustness and good generalization ability of the model. The range of the catalyst formulations was preliminarily screened by fluorescence spectroscopy. When the wastewater was dominated by tryptophan-like and soluble microbial products, the impregnation concentration and time of Mn(NO) were less than 0.3 mol L and 10 h, respectively. Furthermore, the optimized Mn/γ-AlO formulation obtained by the model was impregnation with 0.155 mol L Mn(NO) solution for 8.5 h and calcination at 600 °C for 3.5 h. The model-predicted and experimental values for total organic carbon removal were 54.48% and 53.96%, respectively. Finally, the improved catalytic performance was attributed to the synergistic effect of oxidation (•OH and O) and the Mn/γ-AlO catalyst. This study provides a rapid approach to catalyst design based on the characteristics of wastewater quality using machine learning combined with fluorescence spectroscopy.