Integrated adsorption using ultrahigh-porosity magnesium oxide with multi-output predictive deep belief networks: A robust approach for fluoride treatment

Duc Anh Nguyen,Viet Bac Nguyen,Am Jang
DOI: https://doi.org/10.1016/j.cej.2024.149586
IF: 15.1
2024-02-01
Chemical Engineering Journal
Abstract:The simultaneous recovery and modelling of aqueous fluoride treatment have become urgent due to the complicated characteristics of fluoride-containing wastewater and its nonrenewable nature. This study demonstrates the exceptional potential of Ranunculus-like MgO calcined at 400–600 °C (i.e., M4–M6) with ultrahigh porosity and crystallinity as superior adsorbents for fluoride recovery. M6 outperforms the top 100 most recent fluoride adsorbents, with an adsorption capacity of 405.76 mg/g and high effectiveness across a wide pH range and co-existing ion concentrations. M6 also meets WHO drinking water standards for residual contaminants after five cycles achieving 80 % reusability and 97 % fluoride recovery. FTIR and XPS analyses show ion exchange, H-bonding, and electrostatic attraction as the underlying mechanisms for fluoride adsorption. All the conventional predictive modelling is unsatisfactory, with significant limitations in removal efficiency and magnesium leakage predictability. In such challenging situations, the true revelation lies in the developed deep belief network (DBN), which delivers robust prediction performance (MAE = 0.919, RMSE = 2.140, R2 = 0.998) across all output features and circumstances, surpassing all existing studies using thousands of data points for an output prediction. This M6 adsorbent and DBN model also has immense potential for effectively predicting fluoride treatment in the natural water test with less than 5 % error, paving the way for transformative applications in the near future.
engineering, chemical, environmental
What problem does this paper attempt to address?