Optimizing Water Purification: Adsorbent Performance of Phumdi Biomass Activated Carbon for Fe(II) Removal Using Artificial Neural Network

This study evaluates the performance of artificial neural networks (ANNs) in predicting Fe (II) adsorption using activated carbon derived from Phumdi biomass (PAC) in batch and continuous fixed-bed setups. Phumdi, a unique biomass from the Loktak Lake ecosystem, serves as a sustainable and cost-effective precursor for activated carbon production due to its rich organic composition and functional groups. Large accumulations of Phumdi are commonly found along the periphery of Loktak Lake, where they are often regarded as waste. To account for variations in biomass properties, samples were collected from three different sites: a national park, an agricultural area, and a residential area. BET surface area analysis confirmed the porous nature of the activated carbons, with values ranging from 2.722 to 5.940 m(2)/g across different biomass sources. FTIR characterization identified key functional groups, including hydroxyl, alkyl, and carbon-carbon bonds, which play a crucial role in Fe (II) adsorption. Amongst the batch analysis parameters, the agitation speed was found to be optimum at 250 rpm, and the temperature at 298 K, with an equilibrium time of 120 min. Kinetic studies followed a pseudo-second-order model, indicating chemisorption, while isotherm analysis confirmed Langmuir model conformity, with a maximum adsorption capacity ranging from 1.12 to 6.50 mg/g. Thermodynamic studies confirmed that the adsorption process is exothermic and spontaneous, driven by energy release and a decrease in free energy. Fixed-bed experiments using activated carbon from phumdi biomass from an agricultural area were conducted at varying flow rates (2 mL/min and 4 mL/min), bed depths (20 cm, 40 cm, and 60 cm), and influent concentrations. The maximum throughput of 12 L was achieved before significant breakthrough at 5 mg/L, 4 mL/min, and 60 cm, indicating optimal adsorption performance under these conditions. ANNs demonstrated high predictive accuracy, with R-2 values of 1.00 for training, 0.99 for testing, and 0.95 for validation in the batch system, and 0.99 for training, 0.98 for testing, and 0.95 for validation in the fixed-bed system. The optimal ANN architectures were 6-6-1 for batch adsorption and 4-12-1 for fixed-bed adsorption, with mean squared errors (MSE) of 0.004645 and 0.000856, respectively. This study highlights the potential of Phumdi-derived PAC as a sustainable adsorbent and showcases the effectiveness of ANN modeling in optimizing adsorption efficiency and predictive accuracy, offering an environmentally friendly solution for water treatment.