Scholarlink

<p>This study investigates the use of activated carbon derived from Phumdi biomass for the removal of Fe (II) from water in both batch and fixed-bed adsorption systems. Characterization of the adsorbent using SEM (scanning electron microscope) and EDX (energy-dispersive X-ray) showed that Phumdi-activated carbon (PAC) had a porous structure and functional groups conducive to Fe (II) removal. The PAC exhibited a BET surface area of 4.474&#xa0;m<sup>2</sup>/g and a mesoporous structure, confirmed by nitrogen adsorption–desorption isotherms. FTIR and EDX revealed the presence of functional groups (O–H, C = O) and successful Fe (II) adsorption onto the PAC surface. In batch system, optimal efficiency was achieved at a pH of 6, a contact time of 120&#xa0;min, and dosage of 3.5&#xa0;g/L. The adsorption process was best described by the Langmuir isotherm, indicating monolayer adsorption, and the kinetics followed a pseudo-second-order model. In fixed-bed studies, the impact of flow rates, bed heights, and influent concentrations on Fe (II) removal was examined. To complement the experimental findings, a comprehensive dataset was developed, and two machine learning models, Random Forest (RF) and Extreme Gradient Boosting (XGBoost), were applied to predict removal efficiency. These ML models provided accurate predictions, offering a reliable alternative for optimizing adsorption processes and reducing experimental workload. The XGBoost models achieved higher prediction performance <i>R</i><sup>2</sup> value (0.920), RMSE value (7.90), and MAE (5.54) in case of batch studies and <i>R</i><sup>2</sup> value (0.97), RMSE value (5.44), and MAE (3.97) in fixed bed. The evaluation and comparison of the two ML models revealed that the XGBoost model exhibited the highest test <i>R</i><sup>2</sup> value and the lowest RMSE in both the studies, significantly outperforming RF model. The feature importance analysis revealed the order of input feature types as Time (45%) &gt; Dose (32%) &gt; pH (21%) &gt; Fe (II) initial concentration (6%) for batch analysis, and Time (41%) &gt; Dose (37%) &gt; pH (11%) &gt; initial concentration (11%) for fixed-bed analysis. These results demonstrate the potential of PAC for efficient Fe (II) removal in water treatment systems, with machine learning models providing valuable insights for optimizing adsorption processes, reducing experimental effort, and enhancing environmental sustainability.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Sustainable water purification: evaluating Phumdi biomass adsorbent performance through machine learning-based feature analysis

  • 期刊论文

  • 发表日期：

    2025年2月13日

查看全文
  • Lairenlakpam Helena,
  • Sudhakar Ningthoujam,
  • Potsangbam Albino Kumar

摘要

This study investigates the use of activated carbon derived from Phumdi biomass for the removal of Fe (II) from water in both batch and fixed-bed adsorption systems. Characterization of the adsorbent using SEM (scanning electron microscope) and EDX (energy-dispersive X-ray) showed that Phumdi-activated carbon (PAC) had a porous structure and functional groups conducive to Fe (II) removal. The PAC exhibited a BET surface area of 4.474 m2/g and a mesoporous structure, confirmed by nitrogen adsorption–desorption isotherms. FTIR and EDX revealed the presence of functional groups (O–H, C = O) and successful Fe (II) adsorption onto the PAC surface. In batch system, optimal efficiency was achieved at a pH of 6, a contact time of 120 min, and dosage of 3.5 g/L. The adsorption process was best described by the Langmuir isotherm, indicating monolayer adsorption, and the kinetics followed a pseudo-second-order model. In fixed-bed studies, the impact of flow rates, bed heights, and influent concentrations on Fe (II) removal was examined. To complement the experimental findings, a comprehensive dataset was developed, and two machine learning models, Random Forest (RF) and Extreme Gradient Boosting (XGBoost), were applied to predict removal efficiency. These ML models provided accurate predictions, offering a reliable alternative for optimizing adsorption processes and reducing experimental workload. The XGBoost models achieved higher prediction performance R2 value (0.920), RMSE value (7.90), and MAE (5.54) in case of batch studies and R2 value (0.97), RMSE value (5.44), and MAE (3.97) in fixed bed. The evaluation and comparison of the two ML models revealed that the XGBoost model exhibited the highest test R2 value and the lowest RMSE in both the studies, significantly outperforming RF model. The feature importance analysis revealed the order of input feature types as Time (45%) > Dose (32%) > pH (21%) > Fe (II) initial concentration (6%) for batch analysis, and Time (41%) > Dose (37%) > pH (11%) > initial concentration (11%) for fixed-bed analysis. These results demonstrate the potential of PAC for efficient Fe (II) removal in water treatment systems, with machine learning models providing valuable insights for optimizing adsorption processes, reducing experimental effort, and enhancing environmental sustainability.

Graphical Abstract