Fabrication of a Congo Red/Chitosan@Melamine sponge composite for efficient U(VI) removal: Adsorption behavior, mechanistic insights, and machine learning-assisted performance modeling

Organic azo dyes were employed for the first time to modify chitosan-based composite materials for uranium removal. • Pearson analysis identified initial U(VI) concentration and adsorbate volume as key features. • Good reusability (>90% after 4 cycles) and stable performance in real water matrices. • Mechanism: sulfonic acid groups coordinated with U(VI), synergistically enhanced by amino and hydroxyl groups. • Random Forest model accurately predicted adsorption capacity (R 2  = 0.9422). In the present work, a novel Congo red/Chitosan@Melamine Sponge composite (CRCS@MS) was successfully fabricated via a facile crosslinking strategy, adhering to the “waste control by waste” concept, for the efficient elimination of U(VI) from aqueous systems. Systematic physicochemical analyses confirmed the successful grafting of amino and sulfonic acid groups within the preserved three-dimensional porous skeleton of the material. Batch adsorption tests demonstrated that U(VI) adsorption onto CRCS@MS complied with the Sips model, indicating a monolayer-dominated adsorption process. A removal efficiency of 98.58% was achieved under optimized conditions. The composite maintained over 90% removal after four consecutive cycles, demonstrating good reusability, and exhibited stable adsorption performance in real seawater and river water matrices. Mechanistic studies revealed that the adsorption process was primarily dominated by electrostatic interactions between sulfonic acid groups and U(VI), with amino and hydroxyl groups synergistically enhancing the adsorption through hydrogen bonding or coordination interactions, thereby achieving efficient U(VI) uptake. Furthermore, machine learning techniques were employed to model the adsorption behavior, with seven models systematically evaluated including Support Vector Regression (SVR) with various kernels, Random Forest (RF), and Artificial Neural Network (ANN). Results demonstrated that the RF model outperformed others, achieving a test set R 2 of 0.9422 with the lowest prediction errors, validating the effectiveness of machine learning in effectively capturing the intricate relationships within adsorption processes. This work provides methodological support for the development of sustainable and efficient adsorbents for radionuclide remediation. Download: Download high-res image (144KB) Download: Download full-size image