Synergistic interfacial engineering of mesoporous magnetic metal oxide TiOnanocomposites for sustainable visible-light photocatalysis: Experimental insights and ML-based performance prediction

Abstract

This study investigates the structural, optical, morphological, magnetic, and photocatalytic properties of FeO/TiOnanocomposites (FeT NCs), synthesized through a modified sol-gel method for the photodegradation of Reactive Yellow 145 (RY145). Characterization of FeT NCs (PL, XRD, FTIR, VSM, DRUV-Vis, DLS, Zeta potential, XPS, BET, SEM, TEM, TGA) revealed that FeOincorporation into TiOenhances charge separation, suppresses electron–hole recombination through Ti–O–Fe linkages, and improves photocatalytic efficiency. The calcined 0.025FeT3 exhibited high crystallinity with dominant anatase TiOand no rutile transition. SEM and TEM revealed a core–shell morphology with FeOcores encapsulated by TiO, while aggregation was minimized by synthesis conditions. Optimal photocatalytic performance (84.51% % RY145 removal at neutral pH) was achieved using 1 mg mL0.025FeT3 following pseudo-first-order kinetics. The Langmuir–Hinshelwood model yielded rate and equilibrium constants of 2.80 mg.Lminand 2.42 L mg, respectively. Mechanistic and scavenging experiments indicated that photogenerated holes and •OH radicals dominated the degradation process. The FeT catalyst maintained high stability over six cycles. Magnetic measurements showed soft magnetic behavior with low coercivity and remanence, favoring easy recovery. The reduced bandgap (2.62 eV) facilitated visible-light activation, while BET analysis confirmed a mesoporous structure with high surface area. XPS verified the oxidation states of Fe and Ti, and HPLC confirmed RY145 decomposition via azo bond cleavage and oxidation to carboxylic acids, demonstrating efficient and sustainable photocatalytic activity. 0.025FeT3 demonstrated efficient, stable, and magnetically retrievable photocatalytic activity under visible light, highlighting its potential for sustainable treatment of textile wastewater. To optimize the batch experimental data, a novel ML-driven predictive framework was tested to model and map the relationships between the selected optimization parameters (FeT contents, FeT dose, reaction time), to predict RY145 photodegradation efficiency, and to identify the optimal operating window for improved photocatalytic performance (using three regression measures R, MAE, and RMSE). The CNN models outperformed with a predicted accuracy and Rvalue of 0.91. Based on the results, ML-based evaluation outperformed manual optimization and traditional statistical methods, delivering a more efficient and reliable way for process optimization.