Boron removal from water by adsorption onto activated carbon prepared from palm bark: kinetic, isotherms, optimisation and breakthrough curves modeling

Naturally present in the environment, boron is a chemical element, which can combine with oxygen or other elements in compounds that one names borates. The latter are present in waters, sedimentary rocks or in certain soils. In water, boron exists in the form of boric acid B(OH)₃. In sea water, its concentration reaches approximately 5 mg L⁻¹. Boron is known as one of the essential elements for living plants, animals and humans. However, in greater amounts, boron can be harmful to human, animal and plant life. In humans, the sign of acute toxicity includes nausea, vomiting, diarrhea, dermatitis and lethargy (Shaaban 2010). The World Health Organization guidelines for drinking water quality proposed 2.4 mg L⁻¹ as standard for boron in drinking water (Liu et al. 2009).

Boron is one of the elements most problematic in water. There does not exist any effective method for its elimination. Main processes that have been studied for boron removal are chemical precipitation, adsorption (Lyu et al. 2017), reverse osmosis (Dydo et al. 2012) and electrodialysis. Among these methods, an adsorption method is promising for the boron removal because it requires simple operating conditions and can be applied for the water treatment at low concentrations. For this reason, numerous studies on the boron adsorption have been conducted so far using various adsorbents such as natural and synthesized clay minerals, ion-exchange resins, activated carbons, metal oxides, cellulose and ashes.

The aim of this study was to prepare activated carbon from palm tree and to carry out the structural, functional and surface chemistry of the prepared adsorbent. We investigated the adsorption behavior of boron onto this adsorbent. Kinetic and isotherm studies were made. Thermodynamic parameters, and batch and column capacities were calculated. Thomas, Bohart–Adams, Yan and Yoon–Nelson models were applied to experimental data obtained from the column study. A fixed-bed system was determined in laboratory scale through a two-level factorial experimental design. The influences of initial boron concentration, adsorbent dosage and volumetric flow rate were studied regarding breakthrough and saturation times, volume treated and fractional bed utilization.

The model aqueous solution for the experiments was prepared by dissolving an appropriate amount of boric acid (Acros Organics) in distilled water to the concentration of 100 mg L^–1. All other solutions used were freshly prepared for each experimental run.

Palm tree bark was used for preparation of granular activated carbon, and was obtained from the oasis of Gabes, Tunisia. It was washed with hot distilled water to remove dust-like impurities. After that, it was dried in an oven at a temperature of 105 °C for 24 hours and then crushed and sieved to the desired particle size (1–2 mm). The carbonization process was performed in a muffle furnace at 700 °C under inert atmosphere, with a ramping rate of 10 °C min⁻¹ (Foo & Hameed 2012). The sample was held at the carbonization temperature for 2 hours. The carbon produced was soaked in a potassium hydroxide (KOH) solution with an impregnation ratio of 1:1. The mixture was then dehydrated in an oven overnight at 105 °C to remove moisture and then activated under the same conditions as carbonization, to a final temperature of 600 °C. After activation, the samples were cooled to room temperature. Then, they were washed sequentially several times with hot distilled water (70 °C) until the pH of the washing solution reached 6–7. Finally, samples were dried in an oven at 110 °C for 24 hours and then stored in plastic containers (Ahmad & Hameed 2010).

The pore structural characteristic of this activated carbon was investigated by nitrogen adsorption using an automatic Micromeritics ASAP-2020 volumetric adsorption analyzer. Prior to analysis, the sample was degassed for 4 h under vacuum at 423 K. After degassing, the sample was transferred to the analysis system where it was cooled in liquid nitrogen. Chemical characterization of surface functional groups was undertaken using pressed potassium bromide (KBr) pellets containing 5% of carbon sample by a Fourier transform infrared (FTIR) spectrometer (FTIR-2000, Perkin Elmer). The FTIR spectra were analyzed between 4,000 and 400 cm⁻¹.

Adsorption batch tests were carried out in a thermostatic bath of Grant type with adjustable stirring speed and temperature. In this bath, 100 mL of boron solution was prepared in distilled water at 25 °C at different concentrations. One gram of activated carbon was added into the batch at a stirring speed and at a well-defined temperature. The pH of the solutions was natural (pH 5.5). Aqueous samples were taken from the solution and the concentrations were analyzed. The time range was 5–60 minutes. All samples were filtered prior to the analysis to minimize the interference of the carbon fines in the analysis. Boron concentration of all samples was determined by a UV-visible spectrophotometer (VWR UV-1600PC) at wavelength of 410 nm.

Numerous theories and models have been developed to interpret the different types of isotherms; the resulting equations can be used to predict quantities adsorbed from a minimum of experimentation.

The fixed-bed column studies were performed using a laboratory-scale glass column with an internal diameter of 1.2 cm and a length of 50 cm. The length of activated carbon bed was varied according to the amount of adsorbent. The boron solution of known concentrations was pumped upward through the column at a desired flow rate controlled by a peristaltic pump (Ismatec BVP-Z). The boron solutions at the outlet of the column were collected at regular time intervals and the concentration was measured using a UV-visible spectrophotometer at 410 nm. Each experiment used different volumes of inlet effluent depending on the saturation time at certain experimental condition. All the experiments were performed at 25 °C using deionized water at the native pH of the solution (5.5). The breakthrough curves were obtained by continuous monitoring. Deionized water at a flow rate of 3 mL min⁻¹ was introduced into the column, before each experiment, in order to humidify the porosity of the activated carbon.

The fractional factorial design is used to reduce a number of experiments when there are many factors included in the study. So, this design is a way to simplify the full factorial design because it reduces the number of experiments required (Anderson & Whitcomb 2005).

In this study the influence of three operation variables (factors) on a parameter (response), full factorial design (2³) (Zuorro 2015), was applied: initial boron concentration (C₀), weight (W) of adsorbent and volumetric feed flow rate (Q).

Four response factors were evaluated: breakthrough time (t_b), saturation time (t_s), treated volume (V_s) and fraction bed utilization (FBU). Breakthrough time defines the time to reach 5% of initial concentration. Saturation time presents the time to reach 95% of initial concentration of boron. Data were analysed using the software Statistica 8 (Bakkal et al. 2015).

Successful design of a column adsorption process requires prediction of the breakthrough curve for the effluent. The maximum adsorption capacity of an adsorbent is also needed in design. Mathematical breakthrough curve models were used to predict a fixed-bed column performance and to calculate kinetic constants and uptake capacities (Cruz-Olivares et al. 2013). Three analytical breakthrough curve models were employed to fit the experimental data: Thomas, Bohart–Adams, Yoon–Nelson and Yan models.

It is important to specify that the expression of the Yoon and Nelson model is mathematically analogous to the equation which represents the Thomas model.

The chemical surface groups of the raw material and activated carbon were studied from the FTIR spectra. The FTIR spectra showed that the intensity of all the peaks decreased significantly for the activated carbon compared to its precursor. This is associated with the destruction of intermolecular bonds during thermal and chemical activation. For the FTIR spectra of activated carbon, it has been reported that the band at around 3,384.24 cm⁻¹ can be attributed to the stretching vibration of O-H of hydroxyl functional groups (Torrellas et al. 2015); a reduced intensity for the prepared activated carbon was maybe due to the decrease in volatile matter and moisture content (Sulaiman et al. 2018). The bands at 2,353.14 and 2,328.51 cm⁻¹ have been allocated to the C≡C stretch (Islam et al. 2015). The peak at around 1,734 cm⁻¹ corresponds to the stretching vibration of C=O of the carbonyls for the raw material, and the absence of this peak for the activated carbon indicated numerous removals of aromatic and aliphatic bonds during the thermochemical process (Mahmood et al. 2017). The peak detected at 1,633.58 cm⁻¹ corresponded to stretching vibrations of C=C and a weakened peak for a reduction in the aromatic content contrasted with the precursor (Hsu et al. 2014). The peaks occurring at 1,358.17 and 1,060.28 cm⁻¹ were all attributed to oxygen functionalities such as C-O stretching vibrations and C-O bending vibrations of primary alcohols (Islam et al. 2015).

The elemental analysis of activated carbon was carried out using GmbH Elementar. Results show that the activated carbon is composed of 67.58% carbon, 2.16% hydrogen, 29.01% oxygen and 1.22% nitrogen.

In order to evaluate the surface charge of activated carbon, the pH at the point of zero charge (pH_pzc) was determined and the results showed that the pH_pzc value was 7.2. This value indicates the neutral character of the surface. The pH_pzc value of activated carbon suggests that its surface can have a predominance of negative charges, since for pH_solution > pH_pzc the surface tends to be predominantly positive, while for pH_solution < pH_pzc the surface tends to be predominantly negative.

The nitrogen adsorption–desorption isotherm of activated carbon corresponded to a type I isotherm, which indicates the presence of the microporous structure. The average pore diameter of the sample is 21. 29 Å, confirming the existence of a microporous structure with a surface area of 836.44 m² g⁻¹. The large specific surface area is helpful for the adsorbate molecules approach to the adsorbent-active sites, which can enhance the adsorption efficiency.

To study the mechanism of boron adsorption by activated carbon, three most frequently used kinetic models, namely, pseudo-first-order (Equation (1)), pseudo-second-order (Equation (2)) and intraparticle diffusion model (Equation (3)), were used to assess the adsorption of boron. The derived model parameters from fittings of the data in Figure 1, including the kinetic constant (k), square of correlation coefficient (R²), and equilibrium adsorption capacity (q_e), are shown in Table 1.

The R² value of the pseudo-second-order model (0.954) was closer to 1.0 and greater than that of the pseudo-first-order model (0.899). On the other hand, the q_e values calculated from the pseudo-second-order model were nearer to the observed experimental ones (q_exp). These results indicated that the adsorption of boron on activated carbon was well fitted to the pseudo-second-order kinetic model compared with the first-order model. It can be concluded that the pseudo-second-order kinetic model gives a good correlation for the adsorption of boron on activated carbon.

The intraparticle diffusion model was used to determine the rate-limiting step of the adsorption kinetics (Georgieva et al. 2015). This model accepts that the adsorption mechanism happens through the diffusion of adsorbate molecules into the pores of adsorbent material. The plots for q_t versus t^1/2 are shown in Figure 2.

If the plot of q_t versus t^1/2 gives a straight line, then the adsorption process was controlled by intraparticle diffusion. But, if the data show multi-linear plots, then two or more steps affect the adsorption process.

As can be seen in Figure 2, plots were separated into two linear regions. The first stage indicates a curve generally ascribed to boundary layer diffusion effects or external mass transfer effects. The second stage is the gradual adsorption period with control of the intraparticle diffusion. The deviation of straight lines from the origin may be due to difference in rate of mass transfer in the initial and final stages of adsorption. Additionally, the deviation of the straight line from the origin shows that the pore diffusion was not the only rate-controlling step for the boron adsorption process (Kavitha 2016).

Adsorption isotherm studies play an important role in the predictive modeling procedure for the analysis and design of an adsorption process. In this study, five adsorption isotherm models (Langmuir (Equation (4)), Freundlich (Equation (5)), Dubinin–Radushkevich (Equation (6)), Temkin (Equation (7)) and Sips (Equation (8))) were applied to the experimental data by the software OriginPro8.6. Figure 3 shows the non-linear plots of isotherm models fitted to these experimental data. The values of isotherm parameters are given in Table 2.

From the correlation coefficients obtained for the different models studied, it can be noticed that the Langmuir and Sips models were those that can give a good fit to the experimental data obtained.

The plot of lnK_D versus 1/T was linear with the slope and intercept giving the values of and respectively. The constants of the thermodynamics are shown in Table 3.

An increase in the value of depending on the temperature indicates that the adsorption process is more favorable at lower temperatures. The negative values for enthalpy change, ⁠, confirm the boron adsorption was an exothermic process. Negative entropy change, ⁠, values show that the solid–liquid interface decreases randomly during the adsorption.

A two-level full factorial experimental design had been conducted to evaluate the effects of three variables on four different response factors. The experimental data was analyzed using Statistica 8. Table 4 shows the experimental conditions used and the results obtained from each experiment. Analysis of variance (ANOVA) for the four response factors are shown in Table 5.

ANOVA presents successful evidences such as high determination coefficients for all response factors, with values up to 0.96 as stated in Table 5. The p-value <0.05 indicates if a factor or interaction is significant in the adsorption process (95% confidence level). All analysis for each response is presented in the following four subsections.

The breakthrough time (t_b) is the first response factor that was calculated and represents the time required to detect 5% of the contaminant at the column outlet. P-values <0.05 of the main effects (initial concentration, adsorbent weight and flow rate) indicate they are all significant (Table 5). The adsorbent weight shows the lowest p-value among the three variables. This result indicates breakthrough time is highly affected by feed adsorbent weight.

The high value of R² (0.946) indicates the regression model fits the experimental data well. Response surfaces of the breakthrough time were constructed to study the relationship between independent variables and their interactive effects on the response in three-dimensional planes as shown in Figure 4.

The breakthrough time increases with increasing adsorbent weight due to their positive effects in the regression model of t_b (Equation (17)). Shorter breakthrough times at lower adsorbent weight might be attributed to limited number of adsorption sites on the activated carbon leading to earlier t_b (García-Mateos et al. 2015; Mondal et al. 2016). It is evident from the figure also that breakthrough time decreased with an increase in the initial boron concentration. This happened because a large number of sorption sites get occupied with an increase in C₀. As the value of C₀ increases, mass transfer is apparent to control the resistance presented by the solution. The same was reported by Shaidan et al. (2012). On the other hand, an increment of the flow rate decreases the residence time in the column bed. Thus, shorter breakthrough times are observed at higher flow rates (Meng et al. 2013).

A high value for R², approximately 0.92181, indicates the model fits the experimental data well. Among the three variables of the study only adsorbent weight is significant in the statistical analysis (p-values <0.05, Table 5). However, initial boron concentration can be also significant due to its p-value closer to 0.05. Figure 5 shows the surface plots of the saturation time.

At high values of initial concentration, adsorption sites are filled more efficiently and the column saturation, 95% of C₀, is reached more quickly. This also demonstrates the gradient concentration has significant influence on the adsorption rate (Sotelo et al. 2013). Feed flow rate shows a negative effect on the saturation time. The saturation time decreased as the feed flow rate increased. At higher flow rates the residence time in the column is usually not long enough. In this case, the adsorbate molecule leaves the column before reaching the adsorption equilibrium (Chen et al. 2012). According to Figure 7(b), saturation time increased with increase in amount of adsorbent. At this condition, the number of binding sites is higher and the adsorbate has more time to diffuse through the pores of the adsorbent (Meng et al. 2013).

As can be observed in Table 5, the flow rate (Q) and initial boron concentration (C₀) are significant in the adsorption process, with p-values lower and closer to significance level of 0.05, respectively. The effects of these two variables are represented in the contour plot in Figure 6.

C₀ shows negative effect and Q shows positive effect on the volume of effluent treated. When flow rate decreases, less effluent is treated per gram of activated carbon. Lower inlet concentration causes slower transport of the boron molecules from the film layer to the adsorbent surface due to lower concentration gradient (de Franco et al. 2017).

The determination coefficient value of 0.96081 indicates this equation explains about 96% of the variability in the response. Figure 7 shows two surfaces plots of FBU.

An increase in the initial concentration leads to higher FBU values. This result can be related to the fact that, with more adsorbate molecules, the higher is the adsorption rate. On the other hand, the adsorbent weight shows negative effect on FBU. Even with higher amount of activated carbon, less of the bed is used for the adsorption process. Lower values of FBU are observed at higher amount of activated carbon due to the fact that not all of the adsorbent surface was accessible to molecules of the solute (Liao et al. 2013). It can be seen from Figure 7(b) that an increase in flow rate has a negative influence on the adsorption process. Higher values of FBU are observed at lower feed flow rates.

Traditional analytical models of Thomas, Bohart–Adams, Yoon–Nelson and Yan were employed to fit the breakthrough experimental data under different operation conditions of the experimental design. Model parameters and determination coefficients are shown in Table 6.

The Bohart–Adams model is suitable to reproduce the initial behavior of the breakthrough curve. In this case, the adjustment was carried out up to 50% of the initial concentration. The model was satisfactorily employed with an R² average of approximately 0.87. A good agreement of the experimental data to the Bohart–Adams model indicates that the surface diffusion is the rate-limiting step in the adsorption process (Calero et al. 2009). Both Thomas and Yan models were used to describe processes in which the external diffusion and intraparticle diffusion are not limiting stages of sorption (Chiavola et al. 2012). These models also consider the mass transfer at the solid surface as the limiting step. In addition, the adsorption equilibrium was represented by the Langmuir isotherm and has a reversible second-order kinetics (Mondal et al. 2016). The Yoon–Nelson model assumes that the rate of decrease in the probability of adsorption of the adsorbate molecules is directly proportional to the adsorbate molecule adsorption and the breakthrough of the adsorbent.

Among the four models tested, the Bohart–Adams model shows the lowest average of R², 0.672.

The breakthrough curve of boron at optimal point of the experimental design and the comparison between the adjustments to Thomas, Bohart–Adams, Yoon–Nelson and Yan models are shown in Figure 8.

The Yan model provides a better fit of the boron breakthrough curve compared to Thomas and Yoon–Nelson models due to its smoothly curved shape near the saturation zone. According to authors, the Yan model gives a better adjustment of experimental data by the fact this equation minimizes the mathematical errors of the Thomas model (Chiavola et al. 2012).

The Yoon–Nelson model also showed a good correlation (R² = 0.993) with the experimental results. However, this model can be successfully applied for the boron adsorption on activated carbon to predict the breakthrough curves with insignificant deviations between model and experimental values.

The Thomas model assumes a Langmuir-type adsorption–desorption, isothermal and isobaric operating conditions, a constant column void fraction and second-order reversible kinetics (Meng et al. 2013). The Yan model could overcome the drawback of the Thomas model and both have the same considerations. In the ‘Kinetic studies’ section, it is observed that boron adsorption onto activated carbon followed more closely the pseudo-second-order equation than the pseudo-first-order model.

Furthermore, the equilibrium adsorption is better described by the Langmuir model compared to Freundlich. Both of these results are considerations of Yan and Thomas models than can relate batch adsorption with the fixed-bed column behavior. The three models of Yan, Thomas and Yoon–Nelson were suitable for the description of boron adsorption.

An activated carbon from palm bark is prepared and used for the removal of boron from aqueous medium. The activated carbon is characterized by FTIR spectroscopy, Brunauer–Emmett–Teller analysis, elemental analysis, Raman spectroscopy, pH_pzc method and X-ray diffraction analysis techniques. The experimental data are better correlated by the pseudo-second-order kinetic model rather than the pseudo-first-order kinetics. The equilibrium data are better described by the Langmuir and Sips models. The maximum adsorption capacity is 2.377 mg g⁻¹. The thermodynamic results show the exothermic nature of adsorption of boron on activated carbon. Fixed-bed adsorption experiments reveal that initial boron concentration, adsorbent dosage and flow rate affected both breakthrough and saturation times. Mathematical models for the breakthrough curves are successfully applied. The Yan model fitted better the experimental data of all tests than other models (R²0.993) while the Thomas model is the best equation for the prediction of boron adsorption capacity (q_TH). Based on the results, it can be concluded that the activated carbon is an effective adsorbent for the removal of boron from aqueous medium.