Deep learning modeling of kaolin-activated carbon (derived from cigarette waste) hybrid ceramic membrane performance for wastewater treatment applications

Water pollution remains a critical global concern, underscoring the urgent need for efficient wastewater treatment technologies. In response, there is growing interest in using ceramic membranes, especially those modified with various materials, in separation processes due to their effectiveness in treating large volumes of wastewater. The models were evaluated using a comprehensive set of performance metrics. The primary objective is to predict membrane flux and rejection rates, addressing the challenges of nonlinearities in membrane filtration processes. The study developed experimental and computational scenarios that investigate the predictive capabilities of advanced models, including deep learning Long Short-Term Memory (LSTM), hybrid Adaptive Neuro-Fuzzy Inference System (ANFIS), Neural Networks (NN), and Boosted Trees (BT) to simulate the performance of kaolin ceramic membranes modified with active biochar. The experimental modelling results revealed that the ANFIS-M1 model achieved the highest predictive performance signifying near-perfect alignment between predicted and observed values. Similarly, the NN models demonstrated excellent predictive accuracy (RMSE = 0.0002), while the LSTM models showed moderate performance (DC = 0.9534) and the BT models exhibited comparatively lower predictive efficiency (DC = 0.7467). Experimentally, the hybrid kaolin–activated carbon membrane demonstrated an oil rejection rate of 99.99 ± 0.01 %% and maintained high flux stability, confirming its superior separation capability and antifouling behaviour. The findings confirm the effectiveness of AI models, particularly ANFIS and NN, in improving the accuracy of membrane performance prediction and advancing sustainable wastewater treatment technologies.