3.1. Characterization of the Material
Figure 2 shows the FTIR spectra of Fe
3O
4 nanoparticles (NPs) that depict a high-intensity band at 581 cm
−1. This band is associated with the Fe–O stretching vibration from the magnetic nanoparticles [
42], and the bands at 1637 and 3423 cm
−1 are associated with the O–H stretching modes and bending vibration of water, respectively. In the FTIR spectra of PAM, the absorbance band at 3448 cm
−1 is due to amidogen. The absorbance peaks at 2923 cm
−1 and 1659–1119 cm
−1 are associated with the –CH
2 stretching vibrations, C=O stretching vibrations, N–H bending vibration, and C–N bond, respectively [
43]. The FTIR spectra of
[email protected]3O
4 reveals that the band at 3396 cm
−1 is due to the –OH stretching vibration of cellulose, PAM, and Fe
3O
4.
The other bands at 2915 cm
−1 are attributed to the C–H asymmetric and symmetric stretching vibrations of the cellulosic ring and PAM. The peak at 1659 cm
−1 is associated with the –C=O bond from PAM. The loop at 1324 cm
−1 is a direct result of the OH bending vibrations. The absorption peaks at 1167–899 cm
−1 are related to the –CH
2 parallel scissoring in the cellulosic pyranoid ring, C–O anti-symmetric bridge stretching, the crystal absorption peak of the cellulose C–O–C pyranoid ring skeletal vibrations, and β-glycosidic linkages [
44]. It was determined that the peak at 574 cm
−1 is associated with the Fe–O bond. The FTIR spectra of
[email protected]3O
4 after adsorption of Pb(II) indicates a shift in the frequency at certain band points, suggesting the involvement of the –OH and –NH
2 groups for adsorption in the encapsulation process.
Figure S2 represents the XRD spectra of cellulose and
[email protected]3O
4. The characteristic peaks of cellulose are present at 2θ values of 22.27° and 34.28° [
45]. After uniting and fusing Fe
3O
4 MNPs in the matrix of PAM-g-Cell, the crystalline structure transitioned to a semi-crystalline structure and there appeared to be specific crests in the XRD spectra of
[email protected]3O
4 at 42.43°, 56.62°, and 62.38°. These are characteristic peaks of Fe
3O
4 MNPs.
The morphology of cellulose, PAM, and
[email protected]3O
4 nanocomposite, in addition to the EDX images before and after adsorption of Pb(II), are presented in
Figure 3a–f. The structure of cellulose appears to be sinewy and smooth in the SEM image, while the PAM has an apparently flaky surface. Nonetheless, after fusion with PAM and support of the Fe
3O
4 MNPs, the sinewy structure transforms into a permeable material. After adsorption of Pb(II), the permeable structure transforms into irregular flakes because of adsorption of water molecules on the surface of the material, as shown in
Figure 3e. The SEM images are the basis of the inference that cellulose was effectively united with PAM with the reinforcement of Fe
3O
4 MNPs nanoparticles in the hybrid copolymer framework.
EDX (
Table 1) images of
[email protected]3O
4 before and after treatment were acquired to confirm the adherence of metal ions onto the exterior of
[email protected]3O
4, as shown in
Figure 3d,f. The appearance of the Pb(II) bar in the EDX spectrum of Pb(II) adsorbed on
[email protected]3O
4 indicates that there is adherence between the Pb(II) ions and the surface of the material.
It is difficult to analyze the distribution of Fe
3O
4 MNPs in the polymer matrix because of the low magnification in the SEM micrographs. Thus, an appropriate technique for the analysis of the polymer framework is the utilization of TEM. Based on the TEM micrographs shown in
Figure 4,
[email protected]3O
4 in association with PAM forms a nanocomposite, in which the nanoparticles are distributed throughout the polymer lattice. It is evident that the Fe
3O
4 MNPs were consistently covered by the PAM-g-Cell matrix. The average size of the Fe
3O
4 MNPs was 20.5 nm.
Thermogravimetric techniques are effective strategies for illustrating the temperature response of a material to heat.
Figure 5 shows the TGA spectra of cellulose and
[email protected]3O
4. From this figure, it is evident that the deterioration of pristine cellulose occurs in two stages. The initial step at 100.57 °C occurs because of the loss of hydrated and constituent water. The second stage involves weight reduction that occurs at approximately 159.80 to 594.55 °C. This is generally associated with the deterioration of cellulose (natural carbon) under oxidizing conditions [
46]. Disintegration past 594.55 °C occurs because of oxidation of the decayed products of cellulose. In addition,
[email protected]3O
4 also undergoes a two-stage deterioration, as the first one indicated by a peak at 134.31 °C is due to the loss of water molecules. The second step occurs at approximately 188.62 to 297.37 °C, which can be inferred from cellulose deterioration. Weight reduction at more than 297.37 °C decreases because of the exceedingly stable spinel-organized Fe
3O
4 nanoparticles in the framework of the biopolymer. For the correlation of TGA thermograms of cellulose and
[email protected]3O
4 nanocomposites, it was determined that on average, 26% of the Fe
3O
4 nanoparticles are available in the polymer framework.
The surface area and porosity of the material can be evaluated using the N
2 adsorption–desorption curve at low temperature. The assembled isotherm and pore size distribution controlled by using the BJH model is illustrated in
Figure 6. The data reveals that the material can be considered as a mesoporous solid with a type IV isotherm based on to the IUPAC nomenclature, with pore closeness of various breadths [
47]. From the literature, the surface territory and porosity of neat Fe
3O
4 nanoparticles were observed to be 286.9 m
2/g and 0.6928 cm
3/g, respectively [
48]. The normal hysteresis circle for this case is type H1, which is related to the permeable materials that display a limited dispersion of moderately uniform round and hollow pores [
49]. The BET surface region of the nanocomposite produced by the N
2 adsorption–desorption technique was determined to be 65.89 m
2/g, with a BJH porosity of 0.073 cm
3/g. The surface zone and explicit pore volume of Fe
3O
4 diminished in the
[email protected]3O
4 nanocomposite because of the disappearance of the nanoporosity of Fe
3O
4 nanoparticles due to functionalization with the PAM-g-Cell polymer lattice.
The magnetic property of the material was analyzed by using a vibrating sample magnetometer (VSM) PAR 155 (Quantum Design, Manchester, UK) in the range of −10000 to +10000 Oe. The magnetic hysteresis (M–H) loop indicates that the material is showing ferromagnetic behavior at room temperature [
50]. The saturation magnetization (
ms) value of Fe
3O
4 MNPs depends upon the synthesis protocol and type of non-magnetic groups attached to their surfaces [
51]. From the literature, it was found that the
ms value for bare Fe
3O
4 MNPs was 56.8 emu g
−1 [
52], 75 emu g
−1 [
53], 71.2 emu g
−1 [
54], and 56 emu g
−1 [
55], based on the difference in synthesis protocol. While looking at the
Figure 7, the
ms value for the
[email protected]3O
4 was found to be 40.80 emu g
−1. The decrement in the
ms value of bare Fe
3O
4 magnetic nanoparticles (MNPs) suggests the functionalization of non-magnetic polyacrylamide-g-cellulose matrix on the surface. Thus, based on the VSM studies, it was concluded that there is a correlation between Fe
3O
4 MNPs and the polymer matrix, with a negative synergistic effect on magnetic properties due to the presence of non-magnetic mass.
Figure S3 shows the zeta potential curve for
[email protected]3O
4 as a function of pH, using 0.1 M KCl solution for a pH range of 1–10. Examination of this result revealed that the isoelectric point (IEP) of the material had a pH of 4.32. Hence, the surface of the nano-sorbent is positive below this pH value and negative above this value. The maximum and minimum values of the zeta potential were determined to be +35.68 mV at pH 1 and –38.67 at pH 10.
3.3. Adsorption Isotherms
According to the Langmuir model, a reversible chemical equilibrium exists between the surface of the material and the bulk solution, which is only limited for a finite number of active sites. The non-linear plot (
Figure 9a) for the Langmuir model was selected for Pb(II) based on non-linear regression. It can be observed from
Table 2 that the highest monolayer adsorption capacities (
qm) controlled by the Langmuir model are 314.47, 239.34, and 100.79 mg g
−1 for Pb(II) at 323, 313, and 303 K, respectively.
According to the Freundlich model, adsorption on the surface of a solid always follows the principle of heterogeneity (i.e., the formation of a multilayer of guest molecules on the surface). The values of K
F and
n were obtained by applying the equilibrium data to the non-linear regression (
Figure 9b). The interpretation of the type of interactions that occur between the metal ions and the sorbent is possible based on the value of the Freundlich constant
n. If the value of
n is greater than 1, then there will be strong interactions between Pb(II) and the sorbent (i.e., favorable adsorption). However, if
n is less than 1, this does not favor adsorption. It is evident from
Table 2 that for all the temperature runs, the estimation of
n is higher than 1, which indicates that the adsorption of Pb(II) is favorable under the given optimized conditions. The values of
R2 and
χ2 were determined to be (0.99, 2.59) at 323 K, (0.98, 3.69) at 313 K, and (0.98, 0.57) at 303 K. The value of K
F was determined to be 13.14 mg g
−1(dm
3/g)
n at 323 K, 4.48 mg g
−1(dm
3/g)
n at 313 K, and 2.22 mg g
−1(dm
3/g)
n at 303 K.
Temkin isotherms take into account the impact of the heat of adsorption, which diminishes with the interactions of the adsorbate and the adsorbent. The non-linear plot (
Figure 9c) of
qe versus
Ce facilitates the selection of the values of A
T and b
T. The Temkin constants incorporated in
Table 2 clearly suggest that the adsorption proceeds via chemisorption of the Pb(II) rather than the ion exchange mechanism. The values for the binding constant A
T, given in
Table 2 as 1.39, 0.42 L, and 0.28 L mg
−1, account for the high proclivity of Pb(II) towards the adsorbent surface at 323, 313, and 303 K. The b
T values are 131.96 J mol
−1 at 323 K, 105.32 J mol
−1 at 313 K, and 118.23 J mol
−1 at 303 K. The
R2 and
χ2 values were determined to be (0.98, 3.72) at 323 K, (0.97, 5.64) at 313 K, and (0.97, 4.53) at 303 K. The value of g (0.96, 5.31, and 5.81) calculated using the R–P model (
Figure 9d) correspond to unity, indicating that the Langmuir model is the best model for analysis of the equilibrium data. Therefore, based on the higher value of the regression coefficient and the lower value of
χ2 (i.e., for Pb(II)), the values are 0.99, 0.99, and 0.99 and 0.10, 0.31, and 0.47 at 323, 313, and 303 K, respectively. This indicates that the Langmuir model is the best fitting model for fundamental experimental information in all temperature ranges.
3.4. Adsorption Kinetics
Figure 10a–c was utilized to fit the equilibrium information obtained from the adsorption of Pb(II) (100 mg L
−1) on
[email protected]3O
4 at an ideal pH and at temperatures of 323, 313, and 303 K. The values of the dynamic variables
R2 and
χ2, which were determined using the non-linear method based on dynamic equations, are recorded in
Table 3. It is evident that the
R2 values (0.79, 0.82, and 0.89) are sufficiently large in combination with the low values of
χ2 (2.04, 2.06, and 1.17) for the application of the pseudo-second-order model (
Figure 9b), followed by the pseudo-first-order model (
R2 = 0.43, 0.46, and 0.56;
χ2 = 5.75, 6.30, and 4.67) (
Figure 9a) at 323, 313, and 303 K. In addition, the
qe,cal values (60.80, 57.85, and 53.89 mg g
−1) selected from the pseudo-second interest condition were determined to potentially concur with the
qe,exp values (61.34, 58.34, and 54.22 mg g
−1). The
qe,cal values (59.69, 56.64, and 52.74 mg g
−1) evaluated from pseudo-first-order dynamic conditions have a base synchronization with the fundamental values,
qe,exp (61.34, 58.34, and 54.22 mg g
−1). Therefore, the adsorption framework for Pb(II) onto
[email protected]3O
4 can be best depicted as a pseudo-second-order pathway, and the rate-determining step is possibly chemisorption at 323, 313, and 303 K. The higher estimation of k
2 of 0.019 g mg
−1 min
−1 at 323 K, 0.017 g mg
−1 min
−1 at 313 K, and 0.016 g mg
−1 min
−1 at 303 K indicates higher transport of Pb(II) from bulk to
[email protected]3O
4 surfaces at a high temperature.
Figure 9c demonstrates the intra-particle diffusion model, suggesting that the sequestration process is facilitated by a diffusion mechanism.
3.7. Impact of Other Competitive Ions
Figure 11a depicts the impact of various metal ions including Na
+, K
+, Mg
2+, Ca
2+, Cu
2+, Cd
2+, Ni
2+, and Cr
6+ on the % adsorption of Pb(II) by
[email protected]3O
4. It was discovered that in the absence of other co-ions, the adsorption of Pb(II) was approximately 99%. For 20 mL of 50 mg L
−1 solution of NaNO
3, the adsorption effectiveness of Pb(II) was observed to be 99%, indicating that the proximity of the sodium particles does not have a synergistic impact on Pb(II) adsorption. The proximity of 20 mL of 50 mg L
−1 solution of KNO
3 caused a decrease of the adsorption proficiency to 96% for Pb(II), demonstrating a less significant effect of K
+ ions on Pb(II) adsorption. The presence of Mg
2+ ions significantly impacts the adsorption of Pb(II) because of identical charge valence for a specific adsorption site on the surface of the material. The mg
2+ particles significantly reduce the adsorption capacity of Pb(II) from 99% to 90%. With a larger size and competing valency, Ca
2+ reduces the efficiency of the material towards Pb(II) from 99% to 85%. The presence of Cr
6+ ions show a very small effect on adsorption of Pb(II) by asserting a decline of only 4% on Pb(II) adsorption. This may be due to the application of high pH (5) condition, as the optimized pH for Cr
6+ adsorption lies in the range of 2.5–3.5 [
59], while the presence of Ni
2+, Cu
2+, and Cd
2+ show a minimal to significant effect on Pb(II) adsorption, with decreased values of 90%, 79%, and 70% from 99%. Again, the solution pH plays an important key role here, as the optimum pH for Ni
2+ adsorption is in the range of 6–7 [
60], while for Cu
2+ it is 5.5–6.5 and for Cd
2+ it is 6–8 [
61,
62]. Therefore, it was concluded that the presence of Ca
2+, Cu
2+, and Cd
2+ in the wastewater can significantly affect the Pb(II) on
[email protected]3O
4 based on the pH condition of the system.
3.9. Optimization of the ANN Structure
To train and test the neural network model, the ANN tool was utilized to compute the scavenging process based on the application of the experimental data at different operating conditions [
40]. The optimization of the module was assessed with the aim of minimizing the MSE value and maximizing the
R2 value of the testing set (1–20 neurons correspond to the hidden layer).
Figure 12a illustrates the relationship between the neuron number of the hidden layer and the MSE for the Levenberg–Marquardt algorithm for Pb(II) adsorption. It is evident from the figure that with the increase in the number of neurons in the hidden layer, there is a decrease in value of MSE.
Table 4 represents the variation of the number of neurons with respect to the
R2 and MSE values for the ANN model. The results suggest that the
R2 and MSE values are 0.9915 and 0.0010, respectively. As such, the model associated with 10 hidden neurons for the selected ANN module was preferred for the interpretation of the adsorption behavior of Pb(II) on the nano-sorbent.
Figure 12b represents MSE versus the number of the epoch curve. The results indicate that there was not a significant change in the performance of the method after epoch 3 for the proposed ANN module.
As seen in
Figure 13, there is a good agreement between the experimental data and the predicted data, which represents the predicted removal data for training and testing. The input and output data can be correlated to each other using an equation called the objective function, which is given as follows:
where
x(1),
x(2), and
x(3) represent the inputs,
w1 and
b1 are the weight and bias of the hidden layer, respectively, and
w2 and
b2 are the weight and bias of the output layers, respectively.
Table 5 shows the weight and bias values of each layer, which were determined from the optimum ANN structure.
Figure 13 represents the regression coefficient plot for the proposed ANN model. A value of 0.92 indicates a good correlation between the experimental and predicted data. The point data prediction by the ANN module was approximately equal to the values observed by the batch model (i.e., optimum time of 178.85 min, pH 4.89, adsorbent dose of 1.41 g L
−1, and Pb(II) concentration of 98.55 mg L
−1.