Introduction

Nowadays, heavy metal ions cause various detrimental effects on both human health and the ecological system because they are non-biodegradable, toxic, and they tend to bio-accumulate in the food chain (Lu et al. 2023). These heavy metals, including lead, zinc, cadmium, copper, nickel, and chromium, are highly toxic even at low concentration. Among the heavy metals, lead (Pb) possesses a severe impact on human health. Pb (II) may lead to different forms of cancer, cardiovascular diseases, impaired renal functions, and neurological defects. Water decontamination, therefore, is an essential process in providing safe water for drinking and other practices. Different approaches have been developed for water treatment, such as flocculation, coagulation, biological methods, photocatalytic methods, and adsorption, to name just a few. The adsorption approach is the most popular and has several merits such as low cost, simplicity in design, high efficiency, and easy regeneration without yielding toxic products (Ren et al. 2023a, b).

Montmorillonite (MMT) is water-containing layered smectite clay, which has been increasingly used for the adsorption process due to its appealing features, such as large surface area, high porosity, ion exchangeability, and negatively charged surface (Maleki et al. 2023). Nevertheless, the separation of MMT from the liquid phase after adsorption is often complicated and time-consuming (Khodosova et al. 2022; Fan et al. 2023). Therefore, various modification systems are needed to deal with the separation problem. In this regard, spinel ferrites have been regarded as a promising adsorbent due to their physiochemical properties such as large surface area, strong magnetism, and high adsorption performance (Fan et al. 2017). Because of these advantages, CoFe2O4 nanoparticles have been widely used to remove Pb (II) from aqueous solution. For example, (Ren et al. 2023a, b) synthesized CoFe2O4-graphene oxide nanocomposite for removal of Pb(II), and magnetic separation was applied, leading to lower cost and easier operation.

In recent years, it is more effective to use artificial intelligence to enhance modeling and optimization of process parameters (dosage, analyte concentration, solution pH, and contact time) than traditional methods. In most cases, the response surface methodology (RSM) comprises a central composite design (CCD), which has been commonly used to investigate the interactive effect of independent variables and then to boost the removal efficiency (Rasoulzadeh et al. 2020a, b). On the other hand, artificial neural networks (ANN) learn from experimental data and are capable of analyzing multivariate nonlinear and complex systems. As a result, it is more accurate and can replace the polynomial regression-based RSM modeling approach (Foroughi and Azqhandi 2020; Saber et al. 2021). Furthermore, many studies have been reported on the ANN-predicted model of Pb (II) removal from aqueous solution using functionalized bentonite (Lingamdinne et al. 2023a, b), thiosemicarbazone modified chitosan (Zaferani et al. 2019), polyamine-polyurea polymer modified with pyromellitic dianhydride(Kiraz et al. 2019), multiwall carbon nanotubes (Al-Jadir et al. 2023), and polyacrylamide cellulose supported Fe3O4 magnetic nanocomposites (Rasoulzadeh et al. 2020a, b). The results show that the ANN model was the most fitted model that accurately predicts Pb (II) removal efficiency.

To our knowledge, there have been no prior experimental report assessed synchronously by the comparison of RSM and ANN to optimize Pb (II) adsorption on MMT/CF composite. Hence, the main objective of this study is to evaluate the adsorption capacity of MMT/CF composite as eco-friendly and low-cost adsorbent in Pb (II) removal with the aid of the RSM and ANN application. Moreover, the adsorbent was synthesized by straight forward approach followed by characterization using XRD, SEM–EDX, FTIR, BET, and TGA-DTG.

Materials and methods

Reagents

Analytical grade chemicals such as cobalt (II) chloride hexahydrate (CoCl2.6H2O, 97%), ferric (III) chloride hexahydrate (FeCl3·6H2O, 99%), hydrochloric acid (HCl, 37%), sodium hydroxide (NaOH, 99%), and lead nitrate (Pb(NO3)2, 99%) were obtained from Materials Science and Engineering Laboratories, Jimma Institute of Technology, Jimma, Ethiopia. Raw montmorillonite clay was collected from the Afar region, Ethiopia. Distilled water was used for all experiments.

Pre-treatment of montmorillonite

The raw Montmorillonite was dried at 100 °C for 4 h followed by crushing into powder. The sample was then washed with distilled water to remove water-soluble impurities. Afterward, the sample was dried for 20 h at 100 °C in the oven followed by grinding. It was then sieved through a 75-mesh to get a fine powder.

Synthesis of CoFe2O4 and montmorillonite/CoFe2O4 nanocomposite

CoFe2O4 was prepared according to the previous work with little modification (Jiang et al. 2021). The MMT/CF composite was also synthesized based on the synthesis of CF. Initially, 1.0 g of MMT was added in 50 mL of ethylene glycol and sonicated for 30 min to distribute the powders uniformly. Then, the necessary amount of FeCl3·6H2O and CoCl2·6H2O in a 2:1 ratio was added to the MMT suspension, followed by stirring for 60 min at room temperature. After that, 2-M NaOH solution was dropped to the solution under constant stirring to adjust the pH around 10.5, then stirring for 20 min. The resulting mixture was placed in a 100 mL Teflon-lined stainless-steel autoclaves, kept at 200 °C for 9 h, and cooled down to room temperature. Afterward, the product was washed three times with distilled water and ethanol and dried at 100 °C for 12 h, and then calcined at 400 °C for 3 h in a muffle furnace.

Characterization

X-ray powder diffraction (XRD-700, Shimadzu, South Korea) was used to examine the phase crystallinity of MMT/CF nanocomposite. To study the surface morphology and elemental composition, a scanning electron microscope coupled with energy-dispersive X-ray analysis (SEM–EDX) (COXIEM-30, Shimadzu, South Korea) was used. Fourier-transform infrared spectroscopy (FT/IR-4000 series, JASCO) was used to detect functional groups. Brunauer–Emmett–Teller (BET) using a Quantachrome analyzer (Nova station C, version 11.0) was employed to estimate the surface area and porosity of the MMT/CF based on the principle of adsorption/desorption of nitrogen at 77.3 K and 60/60 s (ads/des) equilibrium time. All the characterization tests were performed in the Materials Science and Engineering Laboratories, Jimma Institute of Technology, Jimma, Ethiopia.

Lead adsorption experiments

For the adsorption test, a stock Pb (II) solution of 1000 mg/L was prepared by dissolving Pb(NO3)2 into distilled water. Then, a sample of 25-mL-volume adsorbate solution of specified initial concentrations (15–55 ppm) was stirred at 250 rpm at room temperature using different adsorbent dosage in a 100-mL conical flask. Finally, atomic absorption spectrometer (ZEEnit 700P, Atomic Flame Mode, Analytik Jena) was used to analyze the residual Pb (II) in the filtrates. The pH of the solution was controlled by adding 0.1-M NaOH and 0.1-M HCl solutions. Magnetically separated adsorbents were used for the regeneration after drying in an oven at 100 °C for 2 h.

The adsorption capacity and removal efficiency of the adsorbent were then calculated using Eqs. (1 and 2) as follows:

$$q_{e} = \frac{{V\left( {C_{o} - C_{e} } \right)}}{m}$$
(1)
$$\eta \left( \% \right) = \frac{{(C_{o} - C_{t} )}}{{C_{o} }} \times 100$$
(2)

where Co and Ce (mg/L) are initial and residual concentration of Pb(II), Ct is the concentration at a given time, V (L) is the volume of the adsorbate, m (g) is mass of the adsorbent, qe(mg/g) is the equilibrium adsorption capacity, and \(\eta\) (removal efficiency).

Experimental design and statistical analysis using RSM

The effect of independent parameters such as adsorbent dosage, initial concentration, solution pH, and contact time were optimized using CCD-RSM. The range of the process variables was wisely chosen for maximizing the removal efficiency. An increase in the adsorbent dosage led to increase in the amount of heavy metal ions adsorption on MMT/CF, and the adsorption removal decreased with increase in the initial Pb(II), which corresponded to ordinary rules of adsorption (Manzoor et al. 2024). Thus, the adsorption of Pb (II) onto MMT/CF was studied by varying the adsorbent dosage (0.15–0.55 g/L), with a dose of 0.55 g/L being possible to allow for all Pb (II) to be removed by the adsorbent. The pH range was chosen to be (3.5–7.5) because at low pH values, there is competition between Pb (II) and H+, while at high pH values the Pb(II) ions could be converted to Pb(OH) (Kochar et al. 2023). A short range of contact time (20–60 min.) could also allow a more suitable basis for experiments, ensuring that the impact of duration could be well addressed.

Table 1 shows the levels of the four independent variables for Pb(II) removal.

Table 1 Level of independent variables
Full size table

ANN modelling

Nowadays, artificial neural network (ANN) is considered as a well-known deep learning model that simulates the functions of human neurons through a network of interconnected neurons.

Figure 1 shows the proposed ANN architecture which consists of four input nodes in the input layer, a hidden layer with nine neurons, and a single output node. The input variables applied in this investigation include dose (g/L), concentration (mg/L), pH, and duration (min), whereas the corresponding outputs are represented by the response variable R (%). The model uses KerasTuner for hyperparameter optimization, to determine the optimal configuration of the model.

Fig. 1
figure 1

General layout of ANN Model

Full size image

Results and discussion

XRD analysis

The powder XRD patterns of MMT, CoFe2O4, and MMT/ CoFe2O4 nanocomposite samples are presented in Fig. 2a–c, respectively. For MMT/ CoFe2O4 nanocomposite, characteristics peaks at a 2θ of 7°, 20.4°, 36.1°, 54.5°, and 62.1°, correspond to the crystallographic planes of (001), (100), (105), (210), and (300), could be ascribed to MMT (the JCPDS pattern 09–0432) (Moja et al. 2020). These peaks are the same as those observed in Fig. 2a, with minor impurities of quartz, kaolinite, and dolomite. However, the diffraction peaks at a 2θ of 18.3°,30.15, 35.5°, 37.1°, 43.2°, 53.6°, 57.1°, and 62.65° belong to the crystal indexes of (111), (220), (311), (222), (400), (422), (511), and (440), respectively. These results are consistent with standard XRD data of spinel CoFe2O4 (JCPDS: 01-077-0426), indicating the presence of CoFe2O4 in the composite (Wang et al. 2018). Such results are also similar to those displayed in Fig. 2b. Moreover, the composite exhibited a lower d001 values, which confirms the successful loading of the MMT surface (Thamer, Aldalbahi et al. 2019). The approximate crystallite size of the composite was also calculated to be 22.6 nm according to the Debay–Scherer equation (Basak et al. 2022), D = K/βcosθ, where the constant k is 0.89, λ is the X-ray wavelength, β is the full-width half maximum, and θ is the Bragg’s angle.

Fig. 2
figure 2

XRD pattern of MMT (a), CF (b), and MMT/CF composite (c)

Full size image

SEM–EDX analysis

The morphological properties MMT and MMT/CF composite are presented in Fig. 3a, b, respectively. The as-synthesized MMT (Fig. 3a) depicted a microstructure with smooth surface in which the MMT particles are distributed with less aggregations (Chu et al. 2020). However, the surface morphology of MMT/CF nanocomposite revealed a rough morphology with high agglomeration and uneven particle distribution (Fig. 3b). This is due to the magnetic property of CF (Yuan et al. 2009). Hence, the as-prepared MMT/CF composite could provide an efficient adsorption process to the targeted Pb (II) due to the available surface-active sites and its porous nature.

Fig.3
figure 3

SEM micrograph of a MMT and b MMT/CF

Full size image

Energy-dispersive X-ray (EDX) analysis was done to further examine the elemental composition of the samples (Fig. 4a and b). In Fig. 4a, the EDX spectrum of MMT shows the existence of Na, Mg, Al, Si, and O without any more peaks. Also, The EDX spectrum of MMT/CF composite indicated the presence of Na, Mg, Al, Si, O, Co, and Fe in the structure (Fig. 4b).

Fig. 4
figure 4

EDX analysis of a MMT and b MMT/CF

Full size image

FTIR analysis

In order to identify the presence of functional groups on the surface of prepared adsorbent, it was determined by FTIR analysis (Fig. 5a and b). In the structure of MMT and MMT/CF composite, the peaks at 3445 and 1636 cm−1 are attributed to the existence of O–H and H–OH stretch vibrations in these materials (Chang et al. 2023). The intensity of the peaks was reduced for MMT/CF composite as compared to MMT, suggesting that the composite has less bonded H2O molecules. The FT-IR spectrum of pristine MMT exhibited characteristic absorption peaks at 1037 cm−1 (Si–O or Al–O) (Tayyebi et al. 2015). The absorption peak of 614 cm−1 displayed that MMT/CF corresponds to the valence vibrational bond of metal–oxygen bond (Co–O and Fe–O) (Idehara et al. 2023). It should be noted that the area of the peak and the intensity of the peaks found in the as-synthesized composite changed significantly, which may be attributed to the interaction of CF with MMT layer (Foroutan et al. 2020).

Fig. 5
figure 5

FTIR spectra of a MMT and b MMT/CF composite

Full size image

BET analysis

The BET surface area, pore volume, and pore size of the prepared MMT/CF are listed in Table 2. The calculated surface area, pore volume, and pore diameter were found to be 243.56 m2/g, 0.298 cm3/g, and 8.91 nm, respectively. This high surface area with high porosity is the main factor behind adsorbing Pb (II) ions on the surface of MMT/CF.

Table 2 BET analysis of MMT/CF composite
Full size table

Thermal property analysis

The thermal analysis for MMT clay and magnetic composites of MMT/CF are shown from TGA and DTA curves. As shown in Fig. 6a, b, the weight of MMT and MMT/CF was significantly reduced in three stages. The first stage of weight loss for MMT and MMT/CF composite was 8.816%, 9.618%, respectively. The primary cause for this weight loss is the evaporation of the water molecule from their structure (Alekseeva et al. 2019). The second weight loss was determined to be 1.668% and 3.475% for MMT and MMT/CF, respectively. This is due to dihydroxylation, clay decomposition, and decarburization of cobalt ferrite particles. The third weight loss reaches up to 800 °C due to the release of OH-groups from the structure. Overall, the sum of weight loss of the MMT (13.846%) < MMT/CF (15.761%). The results suggest that the MMT is more stable than the magnetic composite (Larraza et al. 2012).

Fig. 6
figure 6

TGA-DTA analysis of a MMT and b MMT/CF composite

Full size image

Modeling of Pb (II) removal by RSM-CCD

With the aid of the RSM-CCD method, the combined effect of four independent factors on the process response was investigated to set the optimal values of the process parameters. In this work, a total of 30 experimental runs were carried out, 24 = 16 cube points, six replications at the center point and eight axial points The experimental removal efficiency of Pb (II) relative to the software (Design-Expert 13) prediction and their residuals are presented in Table 3.

Table 3 The relationship between the actual values and results predicted by RSM and ANN
Full size table

As shown in Table 4, the quadratic model is the most fitting model as supported by the largest values of R2 (0.9953), adjusted R2 (0.9909), and predicted R2 (0.9853) with a small standard deviation (1.91). Maximized value of the predicted R2 and the adjusted R2 implies that the model is significant [36]. Furthermore, the P value of the quadratic model is lower than 0.0001, indicating the significance of the model.

Table 4 lack of fit test
Full size table

Accordingly, the quadratic model correlates the four independent variables to the response (R) can be described by the following Eq. (3).

$$R=+89.67+9.26A-10.14B+7.14C+6.02D-2.15AB+1.21AC-1.46BC-7.26{A}^{2}-5.19{B}^{2}-10.18{C}^{2}-7.67{D}^{2}$$
(3)

where R (%) is the removal efficiency of Pb (II), A is the dosage of MMT/CF, B is the initial concentration, C is the pH, and D is the contact time.

Analysis of residuals was also done to validate the significance of the selected model. The difference between the experimental value and the predicted one is called Residual (Simsek et al. 2022). Lower values of the residuals imply the significance of the suggested model by CCD.

The variation of normal probability against externally studentized residuals is shown in Fig. 7a. This profile shows that all the externally studentized residual points are distributed along the line indicating the significance of the selected model. This was further supported by the plot of a random scatter of residuals versus the expected values (Abu-Hamdeh et al. 2021) (Fig. 7b). The plot revealed that the points follow a straight line, confirming the model is significant.

Fig. 7
figure 7

Normal probability plot of externally standardized residuals (a) and the plot of residuals versus run number for removal of Pb (II) (b)

Full size image

ANOVA examination

Analysis of variance (ANOVA) table was generated to confirm the accuracy of the equation Table 5. This evaluation is performed by the model F statistic (F value) and the probability value (P value). A model is said to be significant, if F value obtained is high (Sultana et al. 2020). Herein, the model F value of 226.72 together with the low lack-of-fit F value (0.3210) implies the significance of the model. Furthermore, P values less than 0.0500 indicate model terms are significant.

Table 5 ANOVA result for Pb (II) removal from CCD
Full size table

The coefficient of determination R2 was calculated to be 0.9953, indicating that 99.53% of the variability in the response can be explained by the quadratic model, and therefore, only 0.47% of the observed variability is not explained by the model. The high value of adjusted R2 (0.9909) further confirms that the model is favorable. The predicted R2 of 0.9853 is in reasonable agreement with the adjusted R2 representing the effectiveness of the model.

The signal-to-noise ratio (48.186) was evaluated with adequate precision, implying that the model can be used to navigate the space of design. The low value of the coefficient of variance (C·V = 2.92%) indicates that the deviation between the actual and prediction values is low.

The reasonably low value of sum of squares of prediction error (PRESS = 171.36) suggests the chance of error during the experiment, and thus the predicted values are low.

Three-dimensional responses and contour plots

The 3D response surfaces plots help to investigate the effect of the independent variables and their interactions on the dependent one (Fig. 8a–c).

Fig. 8
figure 8

Three-dimensional and contour plots showing the effects of a analyte concentration and adsorbent dosage, b pH and adsorbent dosage, and c pH and adsorbent dosage on the removal efficiency of MMT/CF toward Pb (II)

Full size image

The adsorbent dosage value is one of the most important variables in the removal of heavy metal ion (Weyrich et al. 2023). As can be seen from Fig. 8a, b, increasing the adsorbent dose increases the removal efficiency of the adsorbent, because it increases the number of active sites, which could adsorb more Pb (II) ions; thus, increases the removal efficiency (Cheblaoui et al. 2023). As shown in Fig. 8a, the increase in the adsorbent dosage from 0.25 to 0.45 g increases the removal efficiency from 76.1% to 98.8%. Figure 8a, c shows that increasing the initial concentration decreases the removal of Pb (II) ions, because at higher initial concentrations, the Pb (II) ions forms an aggregation and overlap on the surface of the composite, reducing the surface area on the adsorbent and as a result, the adsorption efficiency decreases, which is consistent with the results of a previous study. The maximum removal of Pb (II) ions at a concentration of 25 mg/L was 76.34%. However, the lowest removal of Pb (II) ions at a concentration of 45 mg/L was 58.70%.

The pH has also significant effect on removal efficiency as shown in Fig. 8c. It can be seen that the maximum removal efficiency of Pb (II) was attained at pH 6.0. This is because at low pH values, there is competition between Pb (II) and H+ to occupy the active sites on the adsorbent surface, as a result the removal efficiency decreases. Nevertheless, at high pH values, the adsorbent surface becomes negatively charged, therefore, Pb(II) ions are converted to Pb(OH)2 (Huang et al. 2017).

In general, the optimum values of the independent variables for the maximum Pb(II) removal efficiency were found to be 0.45 g adsorbent dosage, 26.94 mg/L initial Pb (II) concentration, 6.01 solution pH, and 49.2 min. time. Considering these optimum conditions, the predicted and experimental removal efficiencies of Pb(II) were 99.08 and 97.88%, respectively.

Modeling of Pb (II) removal by ANN

The ANN architecture with the sigmoidal function now becomes the most widely used technique for modeling the removal process of Pb (II) by several adsorbents (Lingamdinne et al. 2023a, b). The ANN was trained by the RSM matrix and their respective experimental response. The network has four input variables (adsorbent dosage, initial Pb (II) concentration, solution pH, and time) and one output variable (Pb (II) removal). Training and validation can reduce the errors associated with the ANN model (Vázquez-Sánchez et al. 2023). The network was trained until the R2 attained its maximum value and RMSE, MAD and ADE reached the lowest values. Additionally, each experimental run along with the predicted values based on the ANN model, was assessed and tabulated as compared with the experimental and predicted RSM values in Table 8. Furthermore, the residuals are found to be very low at all runs, implying that the ANN can best describe the experimental value.

Figure 9 illustrates the experimental and ANN-predicted values. The model's ability to predict the response variable is confirmed by high R2 value and decreased RMSE value.

Fig. 9
figure 9

The experimental and predicted values by ANN

Full size image

RSM versus ANN models

In this work, a comparative study was investigated between RSM and ANN models using on several statistical parameters, including the coefficient of determination (R2), root mean square error (RMSE) and absolute average deviation (AAD) (Oskui et al. 2019; Onu et al. 2021, Prabhu et al. 2023).

The R2 was calculated using Eq. (4):

$$R^{2} = \frac{{\mathop \sum \nolimits_{i = 1}^{N} \left( {Y_{{i,{\text{exp}}}} - \overline{Y}_{{{\text{exp}}}} } \right)\left( {Y_{{i,{\text{pred}}}} - \overline{Y}_{{{\text{pred}}}} } \right)}}{{\left( {\sqrt {\mathop \sum \nolimits_{i = 1}^{N} \left( {Y_{{i,{\text{exp}}}} - \overline{Y}_{{{\text{exp}}}} } \right)} } \right)\left( {\sqrt {\mathop \sum \nolimits_{i = 1}^{N} \left( {Y_{{i,{\text{pred}}}} - \overline{Y}_{{{\text{pred}}}} } \right.} } \right)}}$$
(4)

where N is the number of experimental data, \({Y}_{i,{\text{exp}}}\) is the ith data point of experimental output, \(\stackrel{\_\_}{{Y}_{{\text{exp}}}}\) is mean value of experimental output, \({Y}_{i,{\text{pred}}}\) is the ith data point of predicted output, and \({\overline{Y} }_{\mathrm{ pred}}\) is the mean value of predicted output.

The RMSE and AAD was calculated using Eqs. (5) and (6), respectively.

$${\text{RMSE}} = \sqrt {\frac{{\mathop \sum \nolimits_{i = 1}^{n} \left( {Y_{{{\text{pre}}}} - Y_{\exp } } \right)^{2} }}{n}}$$
(5)

The AAD was calculated using Eq. (6)

$${\text{AAD}} = \left[ {\frac{{\mathop \sum \nolimits_{i = 1}^{n} \left( {\left| {Y_{{{\text{pre}}}} - Y_{\exp } } \right|/Y_{\exp } } \right)}}{n}} \right] \times 100$$
(6)

Ypre is the predicted response, Yexp is the experimental response, and n is the number of experiments (n = 30).

The statistical parameters that measure and compare the accuracy of both models were estimated based on (Table 6). Generally, the high value of R2 and lower values of the RMSE and AAD determines the modeling ability of a given model (Saber et al. 2021). RMSE is 1.3587 and 0.851 for RSM and ANN, respectively, which illustrates that the predicted value using ANN has less deviation compared to RSM. The obtained R2 values for CCD and ANN are 0.9953 and 0.998, respectively. Furthermore, the AAD value of CCD-RSM (1.683) was lower than ANN (0.505). Hence, ANN has more prediction ability in comparison to RSM. Because ANN can best describe any kind of nonlinearity with no need for a standard experimental design (Huang et al. 2017). Conversely, RSM is preferable to ANN because it gives information about the effect of each parameters and their interactive effects (Ahmadi et al. 2021). Thus, ANN model is more suitable and quite trusted.

Table 6 Comparison of the predictive abilities of RSM and ANN models
Full size table

Adsorption isotherm analysis

The equilibrium adsorption process was studied based on Freundlich and Langmuir isotherm models. The Langmuir isotherm shows monolayer formation of Pb (II) due to homogenous adsorption process on the surface of the adsorbent. Langmuir isotherm equation is given as Eq. (7) (Romero-González et al. 2006):

$$\frac{{C_{e} }}{{q_{e} }} = \frac{1}{{q_{\max } K_{L} }} + \frac{{C_{e} }}{{q_{\max } }}$$
(7)

where \({K}_{L}\) is the Langmuir isotherm coefficient and qmax is the maximum monolayer adsorption capacity.

Additionally, the separation factor (RL) may be used to determines the favorability of Langmuir adsorption model, expressed as follows (Eq. 8) (Zhao et al. 2021).

$${R}_{L}=\frac{1}{1+{{K}_{L}C}_{o}}$$
(8)

According to RL values, the isotherm is favorable (0 < RL < 1), unfavorable (RL > 1), irreversible (RL = 0), and linear adsorption (RL = 1) (Moustafa 2023).

On the other hand, Freundlich adsorption isotherm is for non-uniform distribution of adsorbate on heterogeneous surface. Freundlich adsorption isotherm equation is given as Eq. (9)

$$\ln q_{e} = \ln K_{F} + \frac{1}{n}\ln C_{e}$$
(9)

where KF is the Freundlich adsorption coefficient, and n is the Freundlich exponent (Velarde et al. 2023).

Table 7 and Fig. 10 give the parameters for the two isotherm models. The calculated values indicated that Langmuir isotherm is the best fit to the adsorption equilibrium data of Pb(II) removal. Furthermore, RL value was between 0 and 1, confirming that the Langmuir adsorption was significant. Therefore, the maximum monolayer adsorption capacity (\({q}_{{\text{max}}}\)) was found to be 101.01 mg/g.

Table 7 Isotherm models and their parameters
Full size table
Fig. 10
figure 10

Freundlich (a) and Langmuir (b) isotherm models

Full size image

Kinetic modeling

The pseudo-first-order and pseudo-second-order models were applied to study the rate of Pb (II) adsorption based on the obtained data (Fig. 11). The linear equation for the pseudo-first order and pseudo-second order for MMT/CF composite are calculated using Eqs. (10) and (11), respectively.

$$\log \left( {q_{e} - q_{t} } \right) = \log q_{e} - \frac{{K_{1} }}{2.303}t$$
(10)
$$\frac{t}{{q_{t} }} = \frac{1}{{K_{2} q_{e}^{2} }} + \frac{1}{{q_{e} }}t$$
(11)

where qt and qe(mg/g) are the amount of Pb (II) adsorbed (g) at time t (min) and at equilibrium, respectively. K1(min−1) and K2 (g mg−1 min−1) are the rate constants of pseudo-first-order and pseudo-second-order models, respectively. According to the calculated parameters in Table 8, the experimental data were favorably predicted by the pseudo-second-order kinetic model because of the high R2 values. Moreover, the calculated qe for the pseudo-second order agreed reasonably well to the actual qe values.

Fig. 11
figure 11

Pseudo-first-order and pseudo-second-order kinetics for adsorption of Pb (II)

Full size image
Table 8 Adsorption kinetic parameters of Pb(II) onto MMT/CFcomposite
Full size table

Comparison of MMT/CF with other adsorbents

The adsorption capacity of MMT/CF for Pb(II) removal was compared with other adsorbent materials and is presented in Table 9. It can be seen that MMT/CF composite was higher than previously reported works in terms of adsorption capacity (101.01 mg/g).

Table 9 Comparison of MMT/CF composite with other adsorbents for Pb(II) removal
Full size table

Proposed mechanism for the adsorption process

Surface complexation, ion exchange, and electrostatic interactions are key in the adsorption of heavy metal ions (Alcalde-Garcia et al. 2023). Surface complexation, which is the ability of heavy metals to form multiatomic structures when they interact with functional groups in the adsorbent material. As can be seen from FTIR graph (Fig. 5), there are abundant hydroxyl group on MMT/CF composite, indicating that Pb(II) can interacts with –OH to form hydroxy complexes (-O-Pb) (Jiang et al. 2018). The "ion-exchange" process refers to the exchange of cationic pollutants with the interlayer cations on the adsorbent and adsorption of Pb(II) could be ion-exchange controlled process. Moreover, the other means of interaction between the Pb(II) and the adsorbent is electrostatic interaction because Pb(II) has a strong affinity toward the negatively charged MMT and hydroxyl (OH−) ions. Overall, the adsorption of Pb(II) onto MMT/CF composite may be driven by a combination of several steps: transportation of the Pb(II) to the adsorbent surface from aqueous solution followed by adsorption onto the adsorbent surface, and transport within the adsorbent particle (Raji et al. 2023). However, it is noteworthy that different literatures might verify adsorption mechanisms in a different way. Figure 12 illustrates possible mechanisms of Pb(II) onto the MMT/CF composite.

Fig. 12
figure 12

Schematic diagram showing the possible adsorption mechanisms of Pb(II) by MMT/CF composite

Full size image

Reusability of the adsorbent

The reusability process was cost-effective and reasonable for the Pb(II)-loaded adsorbent. The adsorbent was regenerated under optimum conditions. Also, the reusability of MMT/CF composite to remove Pb(II) from aqueous solution was tested up to three cycles of adsorption–desorption. As can be seen form Fig. 13, as the adsorption–desorption steps increased, the removal efficiency of the adsorbent toward the adsorbate decreased (Elgarhy et al. 2022). The decrease in removal efficiency may be attributed to the incomplete desorption of Pb(II) from the surface of the adsorbent or due to the loss of active sites in the adsorbent (Jiang et al. 2018). The removal efficiency of the Pb(II) using the present adsorbent was 97.98, 97.94, and 96.8% in the first, second, and third cycles, respectively. Similar results have been reported for the regeneration of MMT-based adsorbent (Irawan et al. 2019; Wang et al. 2021). Overall, MMT/CF composite can be used repeatedly for three cycles with little reduction in the removal efficiency toward Pb(II) ions in aqueous solutions.

Fig. 13
figure 13

Removal efficiency at diferent adsorption–desorption cycles

Full size image

Conclusion

In this study, a composite of MMT and CF was effectively used to remove Pb(II) from aqueous solution. The adsorbent was characterized using XRD, SEM–EDX, FTIR, BET, and TGA-DTA. The CCD-based RSM was utilized to optimize several input variables, including adsorbent dosage, initial Pb (II) concentration, solution pH, and time. The RSM model predicted an optimal Pb (II) removal efficiency of 99.08% at 0.45 g/L and 26.94 mg/L and 6.02 and 49.2 min and validated experimentally as 97.88%. Determination coefficient (R2), Mean-Square Error (MSE), and Absolute average deviation (AAD) were used to evaluate the operational conditions for the removal of Pb(II) using the MMT/CF. Even though both models could reasonably examine the correlation of response variable with independent factors, the ANN is more precise than the RSM to describe the process data. The R2, MSE, and ADD values of ANN were found to be 0.998, 0.851, and 0.5049, respectively. As can be seen from the isotherm and kinetics study, the adsorption process followed Langmuir isotherm and pseudo-second-order kinetics. Also, the synthesized adsorbent was stable for three-round tests toward the Pb(II) removal. Therefore, according to the obtained results, the MMT/CF can be considered as a cost-effective and sustainable adsorbent for adsorptive removal of Pb(II) from contaminated waters. However, further studies are recommended to be conducted to investigate: (a) influences of other parameters on the removal efficiency of the adsorbent, (b) efficiency of the adsorbent in the removal of other heavy metal ions, and (c) feasibility of using the adsorbent at an industrial scale.