3.1. Characteristics of the Adsorbent (Wool Fibers)
Figure 1 displays the histograms of fiber diameter distribution (
Figure 1a,b), as well as an example of an SEM micrograph of wool fibers given as the inset. These histograms were built by counting 300 fiber diameters from different SEM images of wool samples. The inset image in
Figure 1a shows cuticle-like structures on the top surfaces, which are specific to the wool fibers [
2]. As given in the histogram from
Figure 1a, the wool fiber diameters varied from 19 μm to 128 μm, following a normal distribution. The histogram analysis disclosed an average fiber diameter equal to 63 μm. The cumulative histogram (
Figure 1b) indicated that 50% of the inspecting fibers bestowed diameters greater than 57 μm. According to the classification mentioned by Rajabinejad et al. [
1], wool of a very coarse grade has a maximum fiber diameter of over 36.2 μm. Hence, the wool explored in this study is of a very coarse grade, as evidenced by the histogram analysis (
Figure 1).
The EDX spectrum of the wool fiber (
Figure 2) disclosed the presence of all expected chemical elements, i.e., C, O, N, S (Pt is due to fiber metallization). The EDX characterization technique provides the qualitative elemental analysis, which is surface specific (i.e., the estimation has a local character). The weight and atomic percentages reported in the inset table (in
Figure 2) represent the locally averaged values. These are in reasonable agreement with the general chemical composition of wool reported in the literature [
2].
Dynamic vapors sorption (DVS) measurements for the wool sample, in the relative humidity (RH) interval of 0 to 90%, are reported in
Figure 3, as the adsorption and desorption branches of the water vapors. The special structure of wool fibers (the exterior of the fiber is hydrophobic, while the interior of the fiber is hygroscopic) determines its behavior in the presence of moisture. The DVS measurements were completed in order to assess the specific surface area of the wool fibers explored in this study. Next, the BET method (Brunauer–Emmett–Teller) was employed to estimate the surface area, based on the water vapor desorption in the dynamic conditions, and for the relative humidity range of 10–40%. Thus, the DVS measurements and BET method indicated a specific surface area for wool fibers equal to 275 m
2/g.
The mechanical tests on coarse wool fibers revealed an elongation at breaks (or tensile strain) of 28.7% and 39.9% for dry wool fiber (RH = 0%) and wet wool fiber (RH = 55%), respectively. Hence, the wool fiber can be stretched more in conditions of greater relative humidity (RH). More details regarding the mechanical characterization are reported in the
Supplementary Materials (SM) in Figure S1 and Table S1.
In addition, Fourier transform infrared spectroscopy (FTIR) was employed to gain the structural information on the investigated coarse wool fiber. The infrared (IR) spectrum of the coarse wool fiber is reported in
Figure S2 (Supplementary Materials). The identified main absorption bands, characteristic of the wool fiber, are mentioned next. These broadband signals correspond to various molecular vibrations: O–H stretching vibrations (centered at 3438 cm
−1), which overlap the absorption bands of amide A (3252 cm
−1) and B (3077 cm
−1); C–H stretching vibrations (three bands at 2962, 2922 and 2853 cm
−1); a shoulder at 1748 cm
−1, characteristic of carboxylic groups (–COO
−) that are specific to aspartic and glutamic amino acid residues form keratin; the C=O band associated with amide I (1647 cm
−1); the C-NH band of amide II (1539 cm
−1); two bands (1448 and 1386 cm
−1) attributed to protonated groups –NH
3+, emerging from the ionic bonds established between the structural units of the substituted peptides with carboxylic groups of aspartic acid and amino groups of lysine; and the band specific to the vibrations of amide III (1262 cm
−1). Likewise, other vibrational bands were identified: broad bands centered at 1161, 1117, and 1040 cm
−1 (corresponding to the cysteine groups indicating the existence of oxidized cystine in raw wool) [
18]; bands at 900, 840, 768, and 700 cm
−1, corresponding to out-of-plane C–H bending [
19]; the last relevant band 620 cm
−1, corresponding to the C–S vibrations; and the low-intensity bands at 525 cm
−1 and 462 cm
−1, attributed to S–S stretching vibrations [
19]. Additional details regarding the attributions of IR peaks for wool-based materials may be found elsewhere [
10].
3.2. Adsorption in Batch Mode
Batch-mode preliminary adsorption tests on coarse wool fibers were carried out. First, the coarse wool investigated herein was tested for the adsorption of various cationic and anionic dyes from aqueous solutions (see
Figure S3, Supplementary Materials). Results revealed that the coarse wool fibers showed a better performance for the retention of cationic dyes, the adsorption of the cationic dye BB9 being the most promising in terms of both the adsorption capacity and removal efficiency (
Figure S3). Second, the influence of the initial pH of the aqueous solution on the efficiency of BB9 adsorption onto wool fibers revealed better performance at naturally occurring levels of pH 6.5 (
Figure S4).
To further assess the effectiveness of coarse wool fibers for removing the cationic organic pollutant (BB9) from wastewater, batch adsorption tests were carried out to reveal the kinetics and isotherms. These adsorption assays enabled us to evaluate the contact time when the equilibrium was reached (from kinetics), as well as the maximum adsorption capacity (from isotherms). The results of adsorption kinetics and isotherms for the retention of BB9 onto the wool adsorbent are shown in
Figure 4. As depicted in
Figure 4a, the adsorption equilibrium was attained at a contact time equal to 90 min, indicating an adsorption capacity of 7.97 mg/g. Greater contact times (
t > 90 min) did not significantly change the adsorption capacity, since the stationary state (equilibrium) was achieved. The pseudo-first-order (PFO) and the pseudo-second-order (PSO) kinetic equations were applied to model the experimental data. Likewise, adsorption isotherms were recorded at two values of temperature: 300 K and 320 K (
Figure 4b). According to
Figure 4b, as the equilibrium concentration (
Ce) increases, the adsorption capacity (
qe) increases as well, reaching a stable plateau for a given level of temperature. The observed maximal values of adsorption capacity at equilibrium were 22.50 mg/g at 300 K and 33.09 mg/g at 320 K. It turns out that the equilibrium adsorption capacity (
qe) increased with temperature (
Figure 4b). The experimental data regarding the adsorption at equilibrium were interpolated to the Freundlich, Langmuir, and Langmuir-revised isotherm models, using the nonlinear regression analysis. Note that the Langmuir-revised equation, proposed by Azizian and co-workers [
16], takes also into account the solubility of the adsorbate (pollutant) in aqueous solutions. For instance, the saturation concentration of BB9 in aqueous solutions is about
CS = 4.36 × 10
4 mg/L (or 0.1363 mol/L). An advantage of the revised Langmuir model is that it provides a dimensionless constant (
KML) that can be applied directly for the computation of thermodynamic parameters [
16]. The values of parameters of the considered mathematical models (for both kinetics and isotherms) are summarized in
Table 1, along with the indicator of the goodness-of-fit (Chi-square test, χ
2). It should be mentioned that the less the Chi-square value (χ
2), the better the fit between the experimental data and calculations.
According to the data shown in
Figure 4a, and by inspecting chi-square values (
χ2) in
Table 1, it may be inferred that the kinetics of adsorption obeys the trend sketched by the PSO equation, rather than by the PFO one. Similarly, by analyzing the isotherms in
Figure 4b and the corresponding chi-square values in
Table 1, it can be claimed that the Langmuir models better predict the behavior of the adsorption system at equilibrium, when compared to the Freundlich model. It seems that the adsorption of the BB9 dye onto the surface of the wool fiber led to the formation of the molecular monolayer. Regarding the comparison of Langmuir and Langmuir-revised models, both predict similar trends (
Figure 4b) with almost identical chi-square tests (
Table 1).
In addition, the mean free energy of sorption (also known as average adsorption potential energy) was calculated according to the Dubinin–Radushkevich (D-R) isotherm [
20,
21]. For the investigated system, the mean free energy of sorption was equal to
ES = 13.66 ± 0.08 kJ/mol, suggesting relevant electrostatic interactions between BB9 cation and the surface of fibers. The D-R isotherm equations [
21] and the estimated parameters are also given in
Table 1. Likewise, the changes in thermodynamic parameters (i.e., Gibbs free energy, enthalpy, and entropy) were assessed for the studied adsorption process. In this respect, the equilibrium constant was equated to the modified Langmuir constant (
KML), according to the modern approach as proposed by Azizian et al. [
16]. This parameter (
KML) represents a dimensionless constant, giving the ratio between the elementary rate constants of adsorption (
ka) and desorption (
kd):
KML =
ka/
kd. For this reason, the modified Langmuir constant (
KML) can be directly used in thermodynamic computations, without any supplemental mathematical transformations [
16,
17]. This approach was developed on a solid theoretical background [
16], thereby, enabling the ample assessment of the thermodynamic parameters for the adsorption processes from the liquid phase.
As a result of thermodynamics calculations, the following values of the mentioned parameters were established: (a) ΔGad = −20.23 ± 0.64 kJ/mol, (b) ΔHad = −0.508 kJ/mol, and (c) ΔSad = 63.630 J/(mol × K). These results pointed out a spontaneous adsorption process (ΔGad < 0), a slight exothermic effect, and the increment of the random collisions at the solid–liquid interface as the temperature increased. The positive entropy change might be attributed to (1) the formation of a less structured or more randomly oriented adsorbed layer on the wool fibers; (2) an increase in the number of possible arrangements of molecules in the adsorbed multilayers; and (3) an increase in the mobility or disorder of water molecules in the vicinity of the adsorption process.
After adsorption, the BB9-loaded fibers represent a spent adsorbent. Several micrographs of the spent adsorbent (loaded wool fibers) are given in
Figure S5 (Supplementary Materials) as an example. Additional tests in batch mode were carried out to assess the desorption efficiency of BB9 from the spent adsorbent into different liquid phases or eluents (see
Figure S6). Relevant desorption efficiencies of 65.18% and 42.29% were observed in acidic eluents, that is, in 1M solutions of citric acid and hydrochloric acid, respectively (
Figure S6). These outcomes suggested that the ion-exchange phenomena could be involved in the adsorption and desorption of the BB9 dye.
The ion exchange mechanism might be attributed to the keratin, which is the major component of wool. Keratin is recognized for its inclusion of carboxyl (–COO
−) and amino (–NH
3+) groups along the side chains, which play a role in electrostatic interactions, specifically in Coulomb forces [
2]. To bring some intrinsic details regarding the mechanism of the interaction between BB9 molecules (cationic form) and a keratin sequence (negatively charged), molecular docking simulations were performed. To this end, the docking computations were performed on a Dell Precision T7910 workstation using the AutoDock-VINA algorithm [
22] encompassed in the YASARA Structure v.20.8.23 program [
23]. As a receptor, an α-keratin with 93 residues was taken into account in order to simulate a sequence of keratin proteins retrieved from the RCSB Protein Data Bank (RCSB PDB, rcsb.org). The considered keratin sequence, for modeling purposes, contains 16 carboxylic groups –COO
− (residues: 15 GLU and 1 ASP) and 12 protonated groups –NH
3+/=NH2
+ (residues: 5 LYS and 7 ARG) on the side chain, bearing a total net charge of −4. The results of the molecular docking simulation revealed that the interaction between BB9 molecules (ligand) and α-keratin (receptor) took place with a binding affinity of −4.85 kcal/mol and a dissociation constant of 0.28 mM, being stabilized by the hydrophobic contacts. The best pose of the docked complex (BB9/α-keratin) is shown in
Figure S7 from Supplementary Materials. In addition, the analysis of the docked complex at the level of Yasara force-field enabled the estimation of the total intermolecular interaction energy, as well as its components, i.e., Van der Waals (VdW) and Coulomb energy interactions (see
Table S2, Supplementary Materials). These results disclosed a total interaction energy equal to −62.72 kcal/mol, from which, −26.13 kcal/mol was attributed to the VdW interactions, and −36.69 kcal/mol to Coloumb (electrostatic) interactions (
Table S2). These simulation outcomes underlined that the VdW and Coulomb interactions were comparable in magnitude. However, the electrostatic interactions were somewhat greater. Hence, the molecular docking simulation corroborated the contribution of the ion exchange phenomena, which relies on the electrostatic (Coulomb) interactions.
3.3. Adsorption in the Fixed-Bed Column: Optimization of Operating Parameters
Adsorption experiments in the dynamic regime were carried out at room temperature (296 K) by pumping the feed aqueous solutions that were loaded with the BB9 dye (
C0 = 5 mg/L) in the up-flow mode. Thus, the feed solution was pumped through the column (containing wool fibers as the fixed bed) at a controlled flow rate using a dispensing pump, as shown in
Figure 5.
The dynamic adsorption studies on the fixed-bed column were carried out by adopting the design of experiments (DoE) and response surface methodology (RSM). Generally, the design of experiments (DoE) allows the efficient sampling of the design space via defining and executing a set of experiments. For this application, a central composite design (CCD) was utilized for experimentations, that is, a face-centered CCD with three center points (
Table 2). The effects of two operating parameters (factors) were investigated through the experimental design. Namely, the height (
H, cm) of the adsorption column and the flow rate of the aqueous solution (
Fv, mL/min) were considered as key factors for this application. The output variable (response) tracked in this study was the removal efficiency of the BB9 dye during the dynamic adsorption. This was determined after 240 min of contact between the fibrous adsorbent and the contaminated solution in the column. Hence, for each experimental run in the design of the experiment (
Table 2), the process response
Y (%) was determined experimentally, as given by the following equation:
where
C0 denotes the initial concentration of the BB9 dye (5 mg/L) in the feed aqueous solution (influent), and
C is the residual concentration of the BB9 dye (pollutant) in the effluent, as determined after a contact time of
t = 240 min between the adsorbent and pollutant in the dynamic regime. According to the modeling protocol, the values of both operating parameters (
H and
Fv) were scaled to represent these factors as coded variables
x1 and
x2, whose values range between −1 (minimum level) and +1 (maximum level). Such a conversion scheme helps to mathematically explore the design space of the factors on the same dimensionless scale (e.g., in a model-based optimization procedure). The mathematical expressions employed for the coding of factors, as well as more details regarding RSM are given elsewhere [
24,
25]. As summarized in
Table 2, the operating parameters are reported as actual factors (
H and
Fv) and as coded factors (
x1 and
x2).
The experimental matrix summarized in
Table 2 implied 11 experimental runs (trials), where the factors were changed concurrently. Likewise, the response of the dynamic adsorption process (
Y, %) was determined experimentally for each run (set of conditions). The central assays (runs no. 9 to 11) were carried out to evaluate the reproducibility of the adsorption experiments in the fixed-bed column. It should be mentioned herein that for each condition (run) specified in
Table 2, the kinetic profile (relative concentration
Ct/
C0 versus time) was determined for a duration of 480 min. All these recorded kinetic profiles (breakthrough curves) are shown in
Figure 6. The response (color removal efficiency) observed in the middle of the time span (i.e., 240 min) was used as the output variable in developing the empirical model.
Hence, the mathematical empirical model (second-order polynomial with interaction term) was constructed based on the experimental matrix (
Table 2), and by using the multiple regression technique [
24,
25]. Thus, the estimated response (
) can be predicted in terms of coded factors (
x1 and
x2), with the aid of the following data-driven model:
subjected to −1 ≤
xj ≤ +1; (
j = 1, 2).
The resulting multiple-regression model (Equation (6)) was tested for significance using the analysis of variance (ANOVA) [
24]. The statistical descriptors assessed by the ANOVA method are summarized in
Supplementary Materials Table S3. The analysis of variance (ANOVA) disclosed a multiple correlation coefficient (
R2) equal to 0.978, suggesting that the constructed model can account for more than 97% of data variation.
The agreement between the observed data from the experiment and the estimated data from the model is shown in
Figure 7. The data points scattered in the vicinity of the bisector (45° straight line) suggest a good concordance between the model and the observations (
Figure 7a).
Moreover,
Figure 7b shows the normal plot of residuals. Note that the residual error means the difference between the observed response and the response estimated by the model. Thus, this graph (
Figure 7b) displays the departure of the residual errors from the normal distribution. As one can see from
Figure 7b, the residuals are distributed nearby and on both sights of the straight line, attesting a normal distribution.
The final mathematical model with actual factors (operating parameters) was deduced using the substitution technique. Hence, the empirical model can be expressed in terms of actual factors as follows:
subjected to 10.0 ≤
H ≤ 14.0 (cm); 3.0 ≤
Fv ≤ 7.0 (mL/min).
The final model, relying on actual factors (Equation (7)), was employed to inspect (by simulation) the response surface, providing the 3D and 2D plots (
Figure 8). These graphs aim to disclose the synergetic effect of factors on the estimated response.
As one can see from
Figure 8, the main effect of feed flow rate
Fv is greater than the main effect of the adsorption bed height
H, but in the opposite sense. For instance, the increment of the feed flow rate (
Fv) in the system depresses the removal efficiency (
). In contrast, increasing the height of adsorption column
H gradually enhances the removal efficiency. However, owing to the interaction effect between these two factors (
Fv and
H), the influence of the
H factor is more evident at lower levels of flow rate (
Fv). At higher flow rates, the effect of adsorption bed height is not so obvious. According to
Figure 8, the optimal region emerges for the following domain of factors
Fv (3.0–3.5 mL/min) and
H (12.0–14.0 cm). To pinpoint the optimal conditions more precisely, numerical optimization (model-based) was carried out. To this end, the numerical optimization was completed using the simplex direct search algorithm [
26] encompassed in the Design-Expert software (v.10). The optimal conditions indicated by this procedure were
H = 13.5 cm (adsorption bed height) and
Fv = 3.0 mL/min (feed flow rate). Under these indicated conditions, the computed response variable was equal to
= 92.04% (estimated value), whereas the response confirmed experimentally (observed after a contact time of 240 min) was
Y = 92.56% (actual value). This difference, equal to 0.52%, represents the residual error between the model and experiment. Note that the value of the actual response (92.56%) in optimal conditions was the greatest when compared to any value reported in DoE (
Table 2). For the established best running conditions (
H = 13.5 cm and
Fv = 3.0 mL/min), the full kinetic profile of adsorption in the fixed-bed column (i.e., breakthrough curve) was determined for a longer time (
Figure 9). These collected data were further subjected to nonlinear regression analysis in order to establish the parameters of the particular kinetic models that were initially employed to assess the performance of the adsorption column. These models can estimate the kinetic profiles (breakthrough curves) for the fixed-bed adsorption systems. In the case of our application, the breakthrough curve recorded under the optimal conditions was modeled using Adams–Bohart, Thomas, Yoon–Nelson, Yan, and Clark kinetic models [
15,
27,
28,
29] in order to estimate the dynamic behavior of the system. Nonlinear regression analysis was performed (in Matlab v.9.9) to compute the main parameters of the models. To assess the accordance between the model predictions and experimental data from the breakthrough curve, the error function (
ε2) was calculated. This error function expresses the goodness-of-fit and represents (
ε2 = Σ(
C(
t)
exp −
C(
t)
calc)
2) the sum of squares of residual errors. Typically, the less the error function (
ε2) is, the better the prediction given by the model. The explicit mathematical expressions of the kinetic models describing the optimal breakthrough curve are summarized in
Table 3, along with the models’ parameters and the estimated error function (
ε2). The smallest values of the error function were found for the Thomas and Yoon–Nelson models, suggesting these models as the best ones in terms of goodness-of-fit. The predictions provided by other models (Clark and Yan) were also relevant. Instead, for the Adams–Bohart model, the highest value of the error function was determined, yet the predictions might be considered satisfactory for the experimentation region (
Figure 9).
The adsorption capacity in the dynamic regime (fixed-bed column) can be estimated from the breakthrough curves when applying the following equation [
28,
29]:
where
qx denotes the adsorption capacity (mg/g) attained at a given time
tx (min),
mS is the mass (g) of the adsorbent fixed bed, and
Fv is the flow rate (mL/min) of the feed solution that flows through the fixed bed. Consequently, the amount of pollutant (
λx, mg) retained into the fixed bed can be readily estimated (
λx =
qx ×
mS). In Equation (8), the variable
Ct can be substituted via the explicit relation given by a breakthrough model (i.e., Adams–Bohart, Thomas, Yoon–Nelson, Yan, or Clark, given in
Table 3).
Thus, the solving of Equation (8) can be performed by means of the numerical integration technique (e.g., the Simpson method). In our case, the adsorption capacity in the dynamic regime (
qx, mg/g) was determined with Equation (8) for different moments of time (
tx, min) in the breakthrough curve and applying different dynamic models. These results are summarized in
Table S4. The considered moments of time from kinetic profile were the breakthrough point (
tb = 90 min, when
Cb = 0.05 ×
C0), an intermediate point (
ti = 240 min), the saturation or exhausting point (
ts = 1080 min, when the condition
Cs = 0.95 ×
C0 is met), and the final point in the breakthrough curve (
tf = 1500 min). All these values are reported in
Table S4. For example, on the saturation point when the adsorption column began to be exhausted (
Cs = 0.95 ×
C0), the adsorption capacity was found to be
qs = 5.51 ± 0.04 mg/g, which corresponds to an amount of pollutant removal of
Λs = 10.64 ± 0.08 mg. Note that all dynamic models considered (Adams–Bohart, Thomas, Yoon–Nelson, Yan, and Clark) converged to these values. It should be emphasized herein that at the breakthrough point (
tb = 90 min), the observed color removal efficiency was equal to
Yb = 99.77%, and for the intermediate point (
t = 240 min), it was 92.56%. Up to a contact time of 280 min, the color removal efficiency was found to be still around 90%. Afterward, it started to decrease, for instance, it was 74.86% at
t = 480 min and 2.26% at the saturation point (
ts = 1080 min). Several photo snapshots of the fixed-bed column, at different contact times, are given in
Figure S8 from the Supplementary Materials.
Lastly, a comparative analysis was conducted, aiming to correlate the findings from this study with the outcomes reported in the literature (
Table 4).
The wool/dye adsorption systems reported in the literature are summarized in
Table 4; most of them referred to batch adsorption experiments. In our study involving coarse wool fibers loaded with BB9 dye, the results of adsorption performance under batch conditions fall somewhere in the middle when compared to findings from other studies listed in
Table 4.
Regarding fixed-bed adsorption, fibrous materials, other than wool, had been reported for the retention of dyes, such as amino-modified cotton fibers [
37], bagasse treated with tartaric acid [
38], and chemically modified kenaf core fibers [
39] (see
Table 4). For fibrous materials like modified cotton [
37], it was observed that the adsorption capacity was less for fixed-bed adsorption when compared to batch adsorption for the same system. In our study, working with coarse wool fibers, we observed a similar behavior. This might be attributed to the less intense stirring in the fixed-bed adsorption when compared to batch mode. Hence, in a column-mode adsorption, the dye molecules might not be distributed over the all-adsorptive sites of the material for the considered contact time [
37].
However, the utilization of wool fibers as a fixed-bed column and the analysis of the breakthrough curve are reported, to best of our knowledge, for the first time in this study. Thus, in the dynamic regime of adsorption, a removal efficiency of 92.56% was observed for 4 h of operability of the fixed-bed column (
H = 13.5 cm) at a feed flow rate of
Fv = 3.0 mL/min. This corresponded to passing a total volume of 0.72 L of contaminated solution ([BB9]
0 = 5 mg/L) through the column. Relaying on experimental evidence, it might be inferred that a single adsorption column (with wool as a fixed bed) can operate in a feasible regime (ensuring > 90% separation efficiency) for up to 280 min. For greater operating times (up to achieving the saturation point), the column might be included in a system of adsorption columns connected consecutively. It should be noted herein that for industrial applications the adsorbent column is usually replaced when the relative concentration
Ct/
C0 is equal to value 0.5, that is, 50% breakthrough is attained [
39]. For our application (wool fibers/BB9), the relative concentration of 0.5 (i.e., 50% breakthrough) was achieved after 780 min contact time, operating at the inlet flow rate of 3.0 mL/min. Above this point, the column might be still operational until attaining the value
Ct/
C0 = 0.9, which represents the operating limit of the adsorption column. For greater values (
Ct/
C0 > 0.9), the column becomes less operational, and for
Ct/
C0 > 0.95, it becomes completely exhausted. Ultimately, this study highlighted that coarse wool fibers have a relevant potential as low-cost sorbents for the uptake of cationic dyes from wastewater, in both batch modes and in fixed-bed adsorption systems.