Machine learning-guided optimization of Nb2CTx MXene preparation modified with tetrabutylammonium hydroxide: comprehensive structural characterization and enhanced performance analysis

Wei Wang, Mingrong Lu, Futian Wei, Shuju Fang, Guizhen Li, Xuewen Tan, Jianjun Wang, Yi Tang,
Machine learning-guided optimization of Nb2CTx MXene preparation modified with tetrabutylammonium hydroxide: comprehensive structural characterization and enhanced performance analysis,
Separation and Purification Technology,
Volume 378, Part 1,
2025,
134438,
ISSN 1383-5866,
https://doi.org/10.1016/j.seppur.2025.134438.
(https://www.sciencedirect.com/science/article/pii/S1383586625030357)
Abstract: Machine learning can optimize the preparation of functional materials by leveraging data-driven predictions to shorten the R&D cycle and reduce experimental repetition, thereby accelerating their application in environmental fields. In this study, the Backpropagation Neural Network Genetic Algorithm (BPNN-GA) and Random Forest Genetic Algorithm (RF-GA), the most prevalent techniques in machine learning, were utilized to optimize the preparation process of Nb2CTx MXene modified with tetrabutylammonium hydroxide (TBAOH-Nb2CTx MXene). Experimental results demonstrated that, both the BPNN-GA and RF-GA models can predict material performance effectively using a limited set of experimental data points and achieve satisfactory fitting and prediction accuracy. Under optimized conditions, TBAOH-Nb2CTx MXene exhibited a removal efficiency of 88.89 % and 88.59 % for Chlortetracycline hydrochloride (CTC). The preparation was validated through a comprehensive suite of characterization techniques, including XRD, SEM, TEM, XPS, and FTIR spectroscopy, confirming that TBAOH-Nb2CTx MXene possessed a two-dimensional crystalline structure with a multilayer arrangement and surface-terminating groups. The machine learning-optimized TBAOH-Nb2CTx MXene exhibited excellent adsorption performance for CTC (removal efficiency >88 %). The adsorption process followed pseudo-second-order kinetics and the Freundlich model, and it maintained good removal capability even after 5 cycles of reuse. The machine learning-optimized TBAOH-Nb2CTx MXene not only shortens the experimental cycle and reduces R&D costs by demonstrating excellent structure-performance relationships, but also provides a new scientific foundation for future materials research.
Keywords: Machine learning; BPNN-GA; RF-GA; TBAOH-Nb2CTx MXene; Preparation; Optimization and prediction; Characterization