ABSTRACT
This study evaluates activated vetiver root powder as chloride ion adsorbent for saline water treatment. Experimental results demonstrate that activated vetiver root effectively removes chloride ions from saline solutions. Higher bed heights and lower flow rates improve efficiency by providing more adsorption sites and a longer contact time. However, higher concentrations lead to rapid bed exhaustion, highlighting the need for parameter optimization. Dynamic adsorption modeling using Thomas, Yoon–Nelson, and Clark models assesses the impact of bed height, flow rates, and initial concentration on chloride removal efficiency. Breakthrough equations closely fitted experimental data in the following order: Thomas model (R2 = 0.999) > Clark (R2 = 0.996) > Yoon–Nelson (R2 = 0.994). Scanning electron microscopy (SEM) analysis identifies pores and pockets as main surface components. Insights into adsorption mechanisms were gained through infrared and X-ray spectroscopies. A novel predictive algorithm for chloride removal in a continuous mode was developed using neural networks. Artificial neural network (ANN) prediction ability was assessed using MSE and R2 values. Future commercial applications of activated vetiver powder can be investigated using real salinity-induced wastewater. Additionally, assessing the adsorbent's regeneration and reusability is vital for confirming its cost-effectiveness and sustainability in longstanding use. This study offers perceptions for developing efficient water treatment strategies by understanding adsorption dynamics and surface characteristics for optimizing chloride removal processes.
HIGHLIGHTS
Evaluation of activated vetiver root powder for chloride ion adsorption in the column study.
Insights into dynamic modeling using both linear and non-linear approaches for chloride removal.
Impact of key parameters (bed height, flow rates, initial concentration) on adsorption efficiency.
Superior fit of the Thomas model for chloride adsorption dynamics.
INTRODUCTION
Chloride contamination is a formidable challenge to environmental sustainability and public welfare worldwide (Sharma & Bhattacharya 2017). With origins spanning industrial discharge, agricultural runoff, and road de-icing practices, elevated chloride levels present a pervasive threat. Their infiltration into freshwater ecosystems induces salinization, disrupting delicate ecological balances and endangering aquatic life (Dey et al. 2024).
Saline water, often a source of chloride contamination, unveils issues in the agriculture environments wherein irrigated water needs regular monitoring. Nevertheless, not much investigation caters to alleviating water salinity for the purpose of irrigation (Kaushal et al. 2021). A high concentration of chloride in irrigation water can inhibit crop development and degrade soil quality, demonstrating the necessity for efficient chloride removal processes in water treatment (Johnson et al. 1989; Karaivazoglou et al. 2005).
Reverse osmosis (RO) and other desalination technologies have the potential to address water salinity challenges but possess significant limitations due to high capital and operating costs among many others (Hoek et al. 2022). These limitations prompt the research into exploring low-cost alternative treatments using organic compounds.
The recent upsurge of interest in natural adsorbents for specific ion removal is driven by their abundance, cost-effectiveness, and ecological sustainability (Dhumal & Sadgir 2023). Activated carbon, which consists of amorphous carbonaceous substances exhibiting elevated porosity and internal surface area (Fierro et al. 2008), is a key component in this study. Among these, activated vetiver root powder stands out as a promising candidate for chloride ion adsorption due to its substantial physical and chemical properties. Vetiver is harvested throughout numerous tropical and subtropical countries, encompassing India, Indonesia, Madagascar, etc. The production serves multiple purposes, including antimicrobial and antioxidant properties. Furthermore, vetiver roots perform a substantial part in soil conservation and assist with stabilizing the dikes (Altenor et al. 2013).
This article examines the use of activated vetiver root powder as a bioadsorbent for continuous removal of chloride ions in water treatment. By employing dynamic adsorption modeling with Thomas, Yoon–Nelson, and Clark models, it investigates the effects of bed height, flow rate, and initial concentration on chloride removal efficiency. The activated adsorbent's surface properties and adsorption mechanisms are also examined in this work. Additionally, it introduces a predictive artificial neural network (ANN) for chloride ion removal efficiency in continuous mode. Furthermore, this study provides invaluable comprehensions on strengthening the design and functioning of adsorption-based water treatment systems, advocating sustainable practices for the foreseeable future.
MATERIALS AND METHODS
Material preparation
The materials for adsorption were carefully prepared to ensure optimal performance. Initially, the adsorbent underwent thorough cleaning with tap water followed by rinsing with deionized water to eliminate dust, dirt, and impurities. After cleaning, the vetiver roots were dried at 105 °C for 5 h, ground into powder, and sieved to achieve uniform particle size.
Vetiver root powder underwent activation by heating to 600 °C and applying phosphoric acid (1:4 ratio (M/V)). The outcome of this thermal and chemical treatment is the activated powder which was stored for subsequent use (Harikumar et al. 2010).
For the adsorbate preparation, a 500-ppm NaCl solution was meticulously formulated by dissolving anhydrous sodium chloride in deionized water. pH adjustments were made using 0.1 M NaOH or HCl solutions. All chemicals were of analytical grade, sourced locally from Maharashtra, India.
Characterization of adsorbent
The surface morphology and microstructure of the adsorbent were examined using scanning electron microscopy (SEM). By utilizing SEM, high-resolution images were obtained, which unveiled the surface texture, porosity, and structural characteristics of the adsorbent material. These images were instrumental in pinpointing any irregularities, cracks, or pores, providing valuable insight into the adsorption capabilities and interaction mechanisms between the activated adsorbent and Cl− ions.
The physicochemical properties of vetiver biomass were comprehensively analyzed to understand its intrinsic characteristics and potential interactions with metal ions. To elucidate detailed information about the crystallographic structure of the biomass and its structural features, X-ray diffraction (XRD) was utilized. The biomass was analyzed using Cu Kα radiation, and the scan angle ranged from 20° to 80° at a speed of 2° per minute.
Continuous column system
Adsorption dynamics
The overall chloride ion adsorption capacity for a specific initial adsorbate concentration and rate of flow corresponds to the integral of the plot depicting the concentration of adsorbed ions against the effluent time.
Synthetic chloride solutions were used in column adsorption tests. The investigations were carried out by choosing experimental settings until the activated adsorbent attained exhaustion with different bed heights (5, 7, and 10 cm), flow rates (3, 5, and 7 ml/min), and initial chloride concentrations (100, 300, and 500 mg/l). The study intends to offer an extensive comprehension of the chloride bioadsorption process based on varying operational parameters, which is crucial for optimizing practical applications for water treatment.
Uncertainties inherent in the experimental measurements were incorporated in the study with error propagation analysis. The flow rate was controlled at 3, 5, and 7 ml/min each with an uncertainty of ±0.5 ml/min. The initial chloride concentration was 100, 300, and 500 mg/l with an uncertainty of ±50 mg/l. The bed height was measured at 5, 7, and 10 cm with an uncertainty of ±0.5 cm. To obtain error estimates for derived values like the total adsorbed amount (qtotal), the total mass of chloride removed (mtotal), and the adsorption capacity (qe), these uncertainties were propagated through the calculations. The results are illustrated in Table 2.
A multitude of models, including the Thomas, Yoon–Nelson, and Clark models, are utilized to anticipate the dynamic behavior of the column. This predictive capability is vital for accurately forecasting the breakthrough curve of effluent parameters, ultimately ensuring the productive design and optimization of a column adsorption process.
Column adsorption modeling
In adsorption studies, understanding the process dynamics is crucial for optimization. Breakthrough curves, which depict concentration changes over time, are key tools for visualizing the saturation process within adsorbent materials (Hu et al. 2022). Steeper curves suggest more effective adsorption. Various adsorption dynamics models predict these curves and provide insights into adsorbent behavior, determining key parameters like kinetic coefficients and maximum adsorbent capacity (Chu 2004). The Levenberg–Marquardt method, implemented via Origin 9 software, is used to infer these parameters.
Table 1 summarizes column models used to study chloride ion adsorption on activated adsorbent. The Thomas model, widely used in column performance theory, has its equation included with KTh (ml/min/mg) as the rate constant and qo (mg/g) as the maximum solid-phase solute concentration. The Thomas model is prominently used because of its simplicity, precision in comprehending and consistent results that are dependable in predicting breakthrough curves, assuming a plug flow behavior of the adsorbate through the bed. In the Thomas model, adsorption is assumed to be governed by second-order reaction kinetics, and it is also assumed that the rate of adsorption is not restricted by external mass transfer (Benjelloun et al. 2021).
Model . | Non-linear equation . | Plot . | Model parameters . |
---|---|---|---|
Thomas | KTh and qe | ||
Yoon-Nelson | KY–N and τ | ||
Clark | A and r |
Model . | Non-linear equation . | Plot . | Model parameters . |
---|---|---|---|
Thomas | KTh and qe | ||
Yoon-Nelson | KY–N and τ | ||
Clark | A and r |
Experimental condition . | qtotal (mg) . | mtotal (mg) . | % Cl− removed . | V (ml) . | qe (mg/g) . | |||
---|---|---|---|---|---|---|---|---|
Bed height (cm) . | Flow rate (ml/min) . | Initial concentration (mg/l) . | Vb . | VT . | ||||
5 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 341.76 ± 54 | 625 ± 88 | 54.682 ± 11 | 200 ± 5 | 1,250 ± 5 | 85.441 ± 13 |
7 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 429.76 ± 65 | 750 ± 106 | 57.302 ± 12 | 210 ± 5 | 1,500 ± 5 | 71.627 ± 11 |
10 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 552.12 ± 81 | 950 ± 134 | 58.118 ± 12 | 250 ± 5 | 2,100 ± 5 | 69.015 ± 10 |
7 ± 0.5 | 3 ± 0.5 | 500 ± 50 | 421.335 ± 85 | 510 ± 99 | 82.615 ± 23 | 60 ± 5 | 1,020 ± 5 | 70.223 ± 14 |
7 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 420 ± 64 | 750 ± 106 | 56 ± 11 | 50 ± 5 | 1,500 ± 5 | 70 ± 10 |
7 ± 0.5 | 7 ± 0.5 | 500 ± 50 | 419.468 ± 60 | 840 ± 103 | 49.93 ± 9 | 35 ± 5 | 1,680 ± 5 | 69.911 ± 10 |
7 ± 0.5 | 5 ± 0.5 | 100 ± 50 | 591.173 ± 302 | 1,000 ± 509 | 59.117 ± 42 | 200 ± 5 | 2,100 ± 5 | 98.529 ± 50 |
7 ± 0.5 | 5 ± 0.5 | 300 ± 50 | 484.175 ± 98 | 825 ± 160 | 58.688 ± 16 | 180 ± 5 | 1,650 ± 5 | 80.696 ± 16 |
7 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 429.75 ± 65 | 750 ± 106 | 57.3 ± 11 | 150 ± 5 | 1,500 ± 5 | 71.625 ± 11 |
Experimental condition . | qtotal (mg) . | mtotal (mg) . | % Cl− removed . | V (ml) . | qe (mg/g) . | |||
---|---|---|---|---|---|---|---|---|
Bed height (cm) . | Flow rate (ml/min) . | Initial concentration (mg/l) . | Vb . | VT . | ||||
5 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 341.76 ± 54 | 625 ± 88 | 54.682 ± 11 | 200 ± 5 | 1,250 ± 5 | 85.441 ± 13 |
7 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 429.76 ± 65 | 750 ± 106 | 57.302 ± 12 | 210 ± 5 | 1,500 ± 5 | 71.627 ± 11 |
10 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 552.12 ± 81 | 950 ± 134 | 58.118 ± 12 | 250 ± 5 | 2,100 ± 5 | 69.015 ± 10 |
7 ± 0.5 | 3 ± 0.5 | 500 ± 50 | 421.335 ± 85 | 510 ± 99 | 82.615 ± 23 | 60 ± 5 | 1,020 ± 5 | 70.223 ± 14 |
7 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 420 ± 64 | 750 ± 106 | 56 ± 11 | 50 ± 5 | 1,500 ± 5 | 70 ± 10 |
7 ± 0.5 | 7 ± 0.5 | 500 ± 50 | 419.468 ± 60 | 840 ± 103 | 49.93 ± 9 | 35 ± 5 | 1,680 ± 5 | 69.911 ± 10 |
7 ± 0.5 | 5 ± 0.5 | 100 ± 50 | 591.173 ± 302 | 1,000 ± 509 | 59.117 ± 42 | 200 ± 5 | 2,100 ± 5 | 98.529 ± 50 |
7 ± 0.5 | 5 ± 0.5 | 300 ± 50 | 484.175 ± 98 | 825 ± 160 | 58.688 ± 16 | 180 ± 5 | 1,650 ± 5 | 80.696 ± 16 |
7 ± 0.5 | 5 ± 0.5 | 500 ± 50 | 429.75 ± 65 | 750 ± 106 | 57.3 ± 11 | 150 ± 5 | 1,500 ± 5 | 71.625 ± 11 |
qtotal = overall chloride ion adsorption capacity (mg), mtotal = overall mass of adsorbed chloride ions (mg), Vb = volumetric output at breakthrough point (ml), VT = total volumetric output at saturation (ml), qe = adsorption capacity (mg/g).
The Yoon and Nelson model for a single-component system includes KY–N (1/min) as the rate constant, τ (min) as the time for 50% breakthrough, and t (min) as the breakthrough time. These parameters are derived from experimental data to compute theoretical breakthrough curves. The rate constants involved in the adsorption process are predicted using time-based data. The Yoon–Nelson model does not put emphasis upon the nature of the adsorbent or properties of adsorbate (Djelloul & Hamdaoui 2015). Experimental characteristics such as flow rate and initial concentration have a major impact on the accuracy of the model (Darweesh & Ahmed 2017).
Clark's model, which integrates mass transfer concepts with the Freundlich isotherm, includes parameters A and r for the kinetic equation and n as the Freundlich isotherm exponent. The Clark model's capability to represent adsorption kinetics and equilibrium scenarios is significant, such as on heterogeneous adsorbent surfaces or in distinct pore structures. Studies have shown that within these complex systems, the Clark model effectively forecasts breakthrough curves (Hu et al. 2020).
Computation modeling
ANNs have been employed for modeling of various ion removals using bioadsorbents (Khandanlou et al. 2016; Fawzy et al. 2018). Similarly, ANN studies in continuous column mode were reported to show accuracy for predictive modeling (Yetilmezsoy & Demirel 2008). The ANN model was created to mimic the adsorption process and predict the adsorption capacities (qe) using input factors like initial concentration, flow rate, contact time, % of chloride removed, qtotal, mtotal, and volume of treated saline water at the breakthrough and saturation points as shown in Table 4 (Das & Mishra 2021). The ANN model was constructed using a multi-layer perceptron (MLP) architecture. It included an input layer corresponding to the number of input variables, hidden layers, and an output layer representing the predicted adsorption performance. Rectified Linear Unit (ReLU) activation functions were used for the hidden layers and a linear activation function for the output layer (Yusuf et al. 2020). The Levenberg–Marquardt algorithm was used for network training. The metrics used for analyzing the robustness and reliability ANN model included R2 values and RMSE values (Parsaei et al. 2022).
RESULTS AND DISCUSSIONS
Characterization of adsorbent
Column-operating parameters
Effect of the column bed height
The increase in bed height for chloride ion removal results in a greater number of accessible sites for adsorption and an extended contact time involving the chloride ions and the adsorbent material. This leads to higher removal efficiency of chloride ions from the saline solution (Han et al. 2009; Ahmad & Hameed 2010). Higher bed columns exhibit improved performance, as reported in previous literature (Omitola et al. 2022).
Moreover, the slower exhaustion of the adsorbent bed is desirable for the effective removal of chloride ions from the saline solution. Therefore, higher bed heights are preferable for the operation, as they provide a longer duration of effective adsorption (Mondal 2009). This implies that water treatment plants can enhance their column designs by using taller adsorption beds, leading to substantial improvements in the removal of contaminants from water (Al-dhawi et al. 2024).
Effect of the flow rate
A corresponding enhancement in removal efficiency was observed with the decrease in flow rate, indicating that lower flow rates are more suitable for the adsorption process. This is attributed to the fact that at lower flow rates, chloride ions are afforded additional time to permeate into the pores of the adsorbent substance, allowing for better utilization of the available binding sites and achieving adsorption equilibrium.
Conversely, at higher flow rates, chloride ions have less residence time in the column, resulting in incomplete utilization of the adsorbent material's capacity and potentially leading to desorption of previously adsorbed ions. This rapid passage of chloride ions through the column results in earlier breakthrough times and higher concentrations of chloride ions in the effluent.
Therefore, low flow rates are beneficial for chloride ion removal in fixed-bed columns due to their higher adsorption capacity, as observed in similar studies (Han et al. 2009; Albadarin et al. 2012). However, in real-world water treatment environments, it is essential to regulate the flow rate to ensure effective removal of contaminants. Slower flow rates can improve adsorption but may decrease the volume of treated water per time unit (Abdiyev et al. 2023).
Effect of the initial concentration of adsorbate
Therefore, striking a balance in choosing the operational concentration is crucial for achieving optimal column performance, considering both adsorption efficiency and breakthrough time. The initial concentration of chloride ions in the inlet flow is a critical factor affecting column performance. An increase in the inlet chloride ion concentration directs to a steeper slope of the breakthrough curve, resulting in a reduced volume treated before saturation of adsorbent bed.
Moreover, at a steady flow rate, an elevation in the inlet concentration of chloride ions reduces the throughput until reaching the breakthrough point. This phenomenon is ascribed to the rapid saturation of adsorbent sites by high chloride ion concentrations, thereby shortening the breakthrough time (Hussein & Mayer 2022).
The propagated uncertainty in the experimental data demonstrates the accuracy of adsorption capacity predictions (Badalyan & Pendleton 2003). For instance, it was discovered that the adsorption capacity (qe) under standard experimental conditions was 85.441 ± 13 mg/g. When assessing the predictive models' goodness of fit, this degree of uncertainty was taken into consideration. This emphasizes the necessity of precise measurement and thorough error analysis in investigations involving adsorption.
In summary, the data signifies that the initial concentration of adsorbate ions significantly influences the breakthrough curve and column performance, emphasizing the importance of carefully selecting the operating concentration to achieve optimal removal efficiency and breakthrough time. In real water treatment scenarios, to increase the effectiveness of the mechanism of adsorption, pretreatment procedures like dilution or the preliminary removal of other interfering ions may be necessary for waters with high initial contamination concentrations (Osman et al. 2023).
Table 2 represents different experimental conditions studied along with breakthrough curve parameters.
Mathematical modeling of breakthrough curves
The data were fitted to the Thomas model to determine the values of both qe (adsorption capacity, (mg/g)) and KTh (rate constant, (ml/min/mg)) which were calculated from the intercept and slope of a plot of ln [Ct/Co] against t as shown in Table 1. The results for the dynamic modeling for column adsorption are shown in the Table 3. Accordingly, the study found that as the initial concentration of the adsorbate rose from 100 to 500 mg/l, the adsorption capacity (qe) decreased, indicating a decline in accessible adsorption sites. In contrast, the Thomas constant (k) increased with initial concentration. Furthermore, increasing bed height improved adsorption capacity by adhering to increased adsorbent mass, whereas a decrease in the Thomas constant indicated slower kinetics owing to increased mass transfer resistance. This aligns with the assumptions drafted in the Thomas model stating and the rate of adsorption is directly proportional to the available adsorption sites (Mustafa & Ebrahim 2010). Notably, as the flow rate increased from 3 to 7 ml/min, adsorption capacity reduced, indicating decreased efficiency with higher flow rates, implying that higher flow rates decrease the duration of contact between the adsorbate and adsorbent, resulting in diminished adsorption capacity. Hence, it is possible to infer that as the flow rate rises, the driving force of mass transfer in the liquid film also increases (Tor et al. 2009). These findings emphasize the complex relationship between concentration gradients, mass transfer processes, and adsorption kinetics, providing important insights for optimizing chloride removal (Li et al. 2018).
Effect of bed height variation . | |||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Thomas model . | Yoon-Nelson model . | Clark's model . | |||||||||||||
Height (cm) . | qe . | KTh . | R2 . | SSE . | RMSE . | KY–N . | τ . | R2 . | SSE . | RMSE . | A . | r . | R2 . | SSE . | RMSE . |
5 cm | 2.80 | 1.72 | 0.995 | 0.05 | 0.04 | 0.027 | 138.09 | 0.986 | 0.05 | 0.043 | 37.82 | 0.024 | 0.989 | 0.03 | 0.03 |
7 cm | 94.75 | 0.03 | 0.998 | 0.03 | 0.03 | 0.024 | 175.01 | 0.99 | 0.03 | 0.034 | 51.91 | 0.020 | 0.996 | 0.01 | 0.02 |
10 cm | 333.63 | −0.007 | 0.996 | 0.09 | 0.04 | 0.017 | 254.33 | 0.982 | 0.09 | 0.046 | 58.05 | 0.014 | 0.994 | 0.04 | 0.03 |
Effect of flow rate variation . | |||||||||||||||
Thomas model . | Yoon-Nelson model . | Clark's model . | |||||||||||||
Flow Rate (ml/min) . | qe . | KTh . | R2 . | SSE . | RMSE . | KY–N . | τ . | R2 . | SSE . | RMSE . | A . | r . | R2 . | SSE . | RMSE . |
3 ml/min | 319.17 | 0.010 | 0.996 | 0.05 | 0.036 | 0.01 | 174.28 | 0.986 | 0.05 | 0.04 | 21.56 | 0.016 | 0.989 | 0.0722 | 0.04 |
5 ml/min | 228.56 | 0.016 | 0.999 | 0.02 | 0.028 | 0.02 | 167.76 | 0.993 | 0.02 | 0.02 | 42.25 | 0.020 | 0.996 | 0.01 | 0.02 |
7 ml/min | 203.71 | 0.010 | 0.997 | 0.03 | 0.037 | 0.03 | 129.80 | 0.989 | 0.03 | 0.03 | 48.31 | 0.026 | 0.990 | 0.03 | 0.03 |
Effect of initial concentration variation . | |||||||||||||||
Thomas model . | Yoon-Nelson model . | Clark's model . | |||||||||||||
Initial Concentration (mg/l) . | qe . | KTh . | R2 . | SSE . | RMSE . | KY–N . | τ . | R2 . | SSE . | RMSE . | A . | r . | R2 . | SSE . | RMSE . |
100 mg/l | 526.83 | 0.01 | 0.996 | 0.05 | 0.03 | 0.01 | 223.54 | 0.991 | 0.05 | 0.03 | 38.16 | 0.01 | 0.994 | 0.03 | 0.02 |
300 mg/l | 200.54 | 0.01 | 0.996 | 0.02 | 0.02 | 0.02 | 174.06 | 0.994 | 0.02 | 0.02 | 51.59 | 0.02 | 0.996 | 0.02 | 0.02 |
500 mg/l | 71.06 | 0.03 | 0.996 | 0.03 | 0.03 | 0.02 | 175.01 | 0.991 | 0.03 | 0.03 | 52.91 | 0.02 | 0.996 | 0.01 | 0.02 |
Effect of bed height variation . | |||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Thomas model . | Yoon-Nelson model . | Clark's model . | |||||||||||||
Height (cm) . | qe . | KTh . | R2 . | SSE . | RMSE . | KY–N . | τ . | R2 . | SSE . | RMSE . | A . | r . | R2 . | SSE . | RMSE . |
5 cm | 2.80 | 1.72 | 0.995 | 0.05 | 0.04 | 0.027 | 138.09 | 0.986 | 0.05 | 0.043 | 37.82 | 0.024 | 0.989 | 0.03 | 0.03 |
7 cm | 94.75 | 0.03 | 0.998 | 0.03 | 0.03 | 0.024 | 175.01 | 0.99 | 0.03 | 0.034 | 51.91 | 0.020 | 0.996 | 0.01 | 0.02 |
10 cm | 333.63 | −0.007 | 0.996 | 0.09 | 0.04 | 0.017 | 254.33 | 0.982 | 0.09 | 0.046 | 58.05 | 0.014 | 0.994 | 0.04 | 0.03 |
Effect of flow rate variation . | |||||||||||||||
Thomas model . | Yoon-Nelson model . | Clark's model . | |||||||||||||
Flow Rate (ml/min) . | qe . | KTh . | R2 . | SSE . | RMSE . | KY–N . | τ . | R2 . | SSE . | RMSE . | A . | r . | R2 . | SSE . | RMSE . |
3 ml/min | 319.17 | 0.010 | 0.996 | 0.05 | 0.036 | 0.01 | 174.28 | 0.986 | 0.05 | 0.04 | 21.56 | 0.016 | 0.989 | 0.0722 | 0.04 |
5 ml/min | 228.56 | 0.016 | 0.999 | 0.02 | 0.028 | 0.02 | 167.76 | 0.993 | 0.02 | 0.02 | 42.25 | 0.020 | 0.996 | 0.01 | 0.02 |
7 ml/min | 203.71 | 0.010 | 0.997 | 0.03 | 0.037 | 0.03 | 129.80 | 0.989 | 0.03 | 0.03 | 48.31 | 0.026 | 0.990 | 0.03 | 0.03 |
Effect of initial concentration variation . | |||||||||||||||
Thomas model . | Yoon-Nelson model . | Clark's model . | |||||||||||||
Initial Concentration (mg/l) . | qe . | KTh . | R2 . | SSE . | RMSE . | KY–N . | τ . | R2 . | SSE . | RMSE . | A . | r . | R2 . | SSE . | RMSE . |
100 mg/l | 526.83 | 0.01 | 0.996 | 0.05 | 0.03 | 0.01 | 223.54 | 0.991 | 0.05 | 0.03 | 38.16 | 0.01 | 0.994 | 0.03 | 0.02 |
300 mg/l | 200.54 | 0.01 | 0.996 | 0.02 | 0.02 | 0.02 | 174.06 | 0.994 | 0.02 | 0.02 | 51.59 | 0.02 | 0.996 | 0.02 | 0.02 |
500 mg/l | 71.06 | 0.03 | 0.996 | 0.03 | 0.03 | 0.02 | 175.01 | 0.991 | 0.03 | 0.03 | 52.91 | 0.02 | 0.996 | 0.01 | 0.02 |
q0 = Thomas Model adsorption capacity (mg/g), KTh = Thomas constant (ml/mg/min), R2 = correlation coefficient, KY–N = Yoon–Nelson Constant (min−1), τ = time needed to achieve 50% adsorbate breakthrough (min), A & r = Clark's constants, SSE = sum of squares error, RMSE = root mean square error.
ANN architecture | |
Input parameters | Bed height (cm), Flow rate (ml/min), Initial metal ions concentration (mg/l), qtotal (mg), mtotal (mg), % chloride removal (%), Volume of treated saline water at breakthrough point, Vb (ml) and volume of treated saline water at saturation point, VT (ml), adsorption capacity (mg/g) |
No. of layers | 3 |
Training algorithm | Levenberg–Marquardt (trainlm) |
Transfer function | Logsig |
Max epochs | 1,000 |
Training data, testing data | 85%, 15% |
RMSE | 0.0405 |
R2 | 0.99 |
ANN architecture | |
Input parameters | Bed height (cm), Flow rate (ml/min), Initial metal ions concentration (mg/l), qtotal (mg), mtotal (mg), % chloride removal (%), Volume of treated saline water at breakthrough point, Vb (ml) and volume of treated saline water at saturation point, VT (ml), adsorption capacity (mg/g) |
No. of layers | 3 |
Training algorithm | Levenberg–Marquardt (trainlm) |
Transfer function | Logsig |
Max epochs | 1,000 |
Training data, testing data | 85%, 15% |
RMSE | 0.0405 |
R2 | 0.99 |
In the Yoon–Nelson Model, it was found that the time needed to achieve 50% adsorbate breakthrough (τ) decreased with higher initial adsorbate concentrations, from 100 to 500 mg/l. Conversely, the Yoon–Nelson constant (KY–N) increased in proportion to the initial concentration. This can be attributed to faster saturation of adsorption sites leading to earlier breakthrough (Bakhta et al. 2024). Interestingly, raising the height of the adsorbent layer increased the breakthrough time, though not in direct proportion to the height (Ajmani et al. 2020). This phenomenon could be explained by a greater probability of preferential drainage channels developing when there is more adsorbent material present. It aligns with other findings that suggesting that the probability of an adsorbate breakthrough and the amount of adsorption capacity available determine the likelihood that adsorbate molecules will either be adsorbed or desorbed (Myers et al. 2023). Furthermore, increasing the flow rate from 3 to 7 ml/min resulted in increase in KY–N, whereas τ tends to decrease, indicating a slower breakthrough time. These findings align with similar observations reported in previous studies (Chowdhury et al. 2013).
The study also utilized the Clark model, which integrates mass transfer principles with the Freundlich isotherm to describe adsorption kinetics. Within this model, parameters A and r characterize the kinetic equation, with their values determined by the slope and intercept of the Clark model equation. The increase in A value as bed height rises shows that greater bed heights improve the adsorption capacity of the system (Yahya et al. 2020). Similarly, the increase in parameters A and r with higher flow rates indicates that higher flow rates enhance adsorption capacity, possibly due to increased mass transfer rates (Sun et al. 2014). However, the slight decrease in R2 at the highest flow rate suggests that the model's accuracy may be impacted by rapid flow, possibly due to insufficient contact time between the adsorbate and adsorbent. Furthermore, the increase in A value with higher initial concentrations suggests that higher concentrations enhance adsorption capacity, likely due to a stronger driving force for mass transfer (Patel 2019). The consistent R2 values across different concentrations demonstrate that the Clark model effectively captures adsorption behavior, even at high concentrations. This approach offers insights into the dynamics of chloride adsorption onto the activated vetiver root powder and provides a framework for understanding the underlying mechanisms driving the process (Długosz & Banach 2018).
Table 3 presents the column parameters obtained for the adsorption process under dynamic conditions using non-linear regressions. It is evident that all three models adequately captured the variation in the flow rate of adsorbate, initial concentration variation of adsorbate, and bed height.
The correlation coefficient values for all models using non-linear regression analysis were satisfactory, indicating their reliability in this study. Notably, the Thomas model exhibited adsorption capacity values (about 71.06 mg/g) across different variations close to the experimentally reported value (71.62 mg/g). Similar results have previously been reported by (Nwabanne et al. 2022). This also suggests that the non-linear regression analysis using the Thomas model may offer the most accurate description of the adsorption mechanism under dynamic conditions. Chosen error metrics, such as RMSE and SSE, were employed to evaluate the discrepancy between the experimental data and the model's non-linear predictions. Across all instances, the error metric values were below zero, suggesting minimal error dispersion in the model parameter predictions.
Computational modeling
As shown in Table 4, this investigation constructed a three-layer ANN model for assessing chloride ion biosorption employing activated vetiver powder in a continuous operating mode. A subset of the data was segregated comprising 70% for training, 15% for validation and 15% for evaluating the predicted accuracy of the model (Nighojkar et al. 2022). Validation was performed to fine-tune the model and prevent overfitting. The testing subset provided an impartial evaluation of the performance of the model, while optimization of the model parameters was done using training data subset (Maleki et al. 2020). The ideal design was ascertained by training multiple feedforward networks to minimize mean squared error (MSE) (Chen et al. 2020). The chosen architecture for the ANN comprised a feedforward neural network with eight inputs, known as MLP (8:3:1:1), an output layer with one neuron, and a hidden layer with three neurons.
CONCLUSION
The experimental results showed that activated vetiver root have promising ability to remove chloride ions from saline solutions. The study underscores the importance of bed height, flow rates, and initial concentration of adsorbate in chloride ion removal using activated vetiver root powder. Increased bed height enhances adsorption site availability and contact time, improving removal efficiency. Lower flow rates favor higher adsorption capacity, while rapid bed exhaustion at higher concentrations compromises performance. Balancing these factors is crucial for optimizing chloride removal in water treatment applications. Adsorption dynamic tests revealed that the breakthrough equations aligned closely with the experimental data, consistent with the principles of Thomas, Yoon–Nelson, and Clark. The non-linear equations of the models helped accurately determine the model parameters.
An excellent fit with the model's R2 value of 0.999 and RMSE of 0.028 was obtained with the Thomas model, which postulates that the rate of adsorption is directly proportional to the number of empty sites. The Thomas model demonstrated that kinetic factors principally govern adsorption. The breakthrough time of the Yoon–Nelson model, which predicts precisely that 50% of the adsorbate chloride ions are eliminated, decreased as the initial concentration was raised. Yoon–Nelson showed a high R2 value of 0.994 and an RMSE of 0.02. The Freundlich equilibrium isotherm and mass transfer principles are combined in the Clark model, which produced an R2 value of 0.996 and an RMSE of 0.02. The adsorption capacity of the model is significantly influenced by parameter A, particularly at greater flow rates and higher bed heights. Sensitivity analysis of parameters indicated that variations in parameter A had a greater impact on the Clark model compared to parameter r. In the Yoon–Nelson model, variation in τ had more influence on the model than KY–N.
SEM analysis revealed that pores, fibers, hills, and pockets were the main components of the surfaces. Additionally, infrared and X-ray spectroscopies can deliver further insights into the adsorption mechanism by exploring the functional groups accountable for chloride ion uptake from saline solutions.
Precise forecasts for the elimination of chloride ions from saline solutions have been made possible using ANNs with R2 value of 0.99 using the Levenberg–Marquardt algorithm. The approach can effectively anticipate the outcomes of bioadsorption, resulting in substantial savings of effort, time and expenses, as demonstrated by the accurate match between the predictions and the findings of the experiment.
Subsequent investigations should concentrate on examining the enduring stability and recyclable nature of powdered activated vetiver root in real-world water or wastewater treatment scenarios. Further research could examine the possibility of blending activated vetiver root powder with other nanomaterials or bioadsorbents to improve its selectivity and adsorption ability. Forecasting adsorption behavior under different circumstances by employing machine learning techniques could potentially be utilized for understanding the optimization of adsorption processes for upscaling or industrial applications.
ACKNOWLEDGEMENT
The authors extend their sincere gratitude to COEP Technological University Pune for their invaluable support and resources provided throughout this research endeavor.
AUTHOR CONTRIBUTIONS
R.C.D. and P.S. conceptualized the whole article and developed the methodology. R.C.D. performed the experimentation and presented the data for the first draft of the manuscript. P.S. provided guidance and reviewed the manuscript. All authors read and approved the final manuscript.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.