Machine Learning-Based Prediction and Optimization of Membrane Fouling in a Hybrid Biochar-Membrane System for Treating Maize Flour Mill Wastewater

Abstract

Agro-processing mill wastewater has a high organic load, which negatively impacts the environment. In this study, a hybrid coconut-shell-derived biochar (CSDB) membrane system was developed, and machine learning models were used to predict and optimize the removal of pollutants and membrane fouling. CSDB with a porosity of 68.7 ± 1.2%, a pH of 9.2 ± 0.1, a BET surface area of 512 ± 18 m2/g, and a total carbon content of 83 ± 1.5% was prepared by pyrolysis at 500 °C and steam activation at 700 °C. Batch experiments demonstrated maximum removal efficiencies of 91.30 ± 1.8% for Cd (pH 7.5), 90.91 ± 2.1% for Pb (pH 6.5), and 67.71 ± 2.4% for Cr (pH 3.5). The adsorption kinetics were best described by the pseudo-second-order model (R2 > 0.98), while the pseudo-first-order model yielded lower R2 values (0.72–0.85), supporting a chemisorption-dominated mechanism. Compared with membrane-only operation, fouling resistance was reduced by 35–45% when coupled with CSDB pretreatment. Experimental data from a two-stage experimental design were used to train artificial neural network (ANN), extreme gradient boosting (XGBoost), random forest (RF), and support-vector regression (SVR) models. The best predictive accuracy (test-set R2 = 0.91–0.96, RMSE = 1.8–3.4%, MAE = 1.4–2.7%) was obtained by the XGBoost and ANN models. The optimal conditions for removing 85–92% of the metals and simultaneously reducing fouling were determined by multi-objective optimization (CSDB dosage of 12–15 g/L, contact time of 90–105 min, and transmembrane pressure of 1.0–1.1 bar). The predictions were validated experimentally, with relative errors of ≤ ±6%. The preliminary technoeconomic assessment, which is based on a 10 m3/day modular unit, a membrane lifetime of 18–24 months, and an energy consumption of 0.85–1.1 kWh m-3, estimates a treatment cost of approximately ₦450/US$0.27 m-3. This hybrid CSDB–membrane process coupled with machine learning offers a convenient and cost-effective pathway for the sustainable treatment of maize flour mill effluents and agricultural reuse.