Optimization of ammonia nitrogen removal by SO₄²⁻ intercalated hydrotalcite using response surface methodology

Abstract

SO₄²⁻ intercalated Mg–Al hydrotalcite (S-LDH) was prepared under microwave irradiation and characterized by powder X-ray diffraction (XRD), Fourier-transform infrared spectroscopy (FT-IR) and scanning electron microscopy (SEM). The S-LDH had a typical layered structure containing SO₄²⁻ anions, and intercalation of SO₄²⁻ into the interlayer spaces caused reduction of crystallinity and crystal sizes compared to the Mg–Al hydrotalcite precursor. Furthermore, the Box–Behnken design, an experimental design for response surface methodology (RSM), was used to study the NH₃–N removal by S-LDH under microwave irradiation. The parameters including S-LDH amount, irradiation time, temperature and initial pH were optimized by RSM. From the analysis of variance, it was found that RSM could be used effectively to model and improve the NH₃–N removal efficiency. The optimum conditions were 1.0 g L⁻¹ of the S-LDH, 9.0 min, 70 °C and initial pH 10.0 to achieve 90.4% of the NH₃–N removal rate. Overall, the S-LDH showed a high performance under moderate operating conditions for NH₃–N removal.

---

1 Introduction

Ammonia nitrogen contaminated wastewater has posed a great threat to the safety of water resources. Nitrogen compounds in wastewaters exist usually in the form of the ammonium ion (NH₄⁺).^1,2 Biological and physicochemical treatments are commonly used because these techniques are the most economical and applicable. Due to the presence of many refractory and toxic compounds like cyanides and polynuclear aromatic hydrocarbons, a biological treatment process is not efficient for the removal of NH₃–N.^1,3,4 And physicochemical treatment processes introduce other critical problems such as membrane fouling, high cost or the generation of toxic by-products.^5,6 To overcome such issues, microwave (MW) technology is employed for NH₃–N removal from wastewaters.^7,8

Recently, microwave (MW) radiation has gained a great deal of attention, because of the molecular-level heating and ease of operation. It also has been successfully employed in environmental remediation, especially in wastewater treatment.^1,9,10 MW alone is found to be useful treatment process for NH₃–N removal, which can decrease NH₃–N concentration from 5000 to 350 mg L⁻¹ at 750 W.¹ In our previous work, the efficiency of MW in NH₃–N treatments can be amplified by coupling MW with the SO₄²⁻ intercalated hydrotalcite.¹¹ Additionally, some interesting reports had appeared on the application of the hydrotalcite materials for nitrate removal, where the hydrotalcite can significantly adsorb nitrate anions, and then effectively reduce nitrites into nitrogen.^12,13 The effectively synchronized removal of cyanide and metal ions in the Zn/CN electroplating wastewater and Ni wastewater can be implement by Zn/Al hydrotalcite, where high removal efficiency is contributed to Ni(CN)₄²⁻.¹⁴ At the same time, the organo/MgAl hydrotalcite is used as an amendment in contaminated soils to reduce the mobility of polycyclic aromatic hydrocarbons due to its high irreversibility of the adsorption process.¹⁵

As the process of NH₃–N removal under MW irradiation is not a simple and single chemical process, it is indispensable to understand the optimal working conditions to get maximum the NH₃–N removal efficiency.¹⁶ The traditional ‘‘one-factor-at-a-time approach” is an operation frequently used in optimization to obtain high yields of the desired products. But that method disregards the complex interactions among various physicochemical parameters. The response surface methodology (RSM) is a powerful statistical technique for optimizing experimental conditions and investigation of critical processes which can evaluate multiple parameters and their interactions by reducing the number of experimental trials, and also predict their behavior under given sets of conditions.¹⁷

In this work, the SO₄²⁻ intercalated hydrotalcite was prepared, and employed to remove the NH₃–N in synthetic wastewater. The operating conditions (temperature, pH, hydrotalcite amount and irradiation time) for the removal of NH₃–N by the SO₄²⁻ intercalated hydrotalcite under MW irradiation were optimized by RSM.

2 Experimental

2.1 Materials

All the reagents/reactants were of analytical grade and solutions were prepared with deionized water. 0.1 mol L⁻¹ NaOH and HCl solutions were used for pH adjustment. A pH electrode (Mettler Toledo S40K) was used for pH measurements.

2.2 Preparation of SO₄²⁻ intercalated hydrotalcite

Mg/Al–CO₃²⁻ hydrotalcite (LDH) with Mg/Al molar ratio of 3.0 were prepared by urea hydrolysis ([urea]/[NO₃⁻] molar ratio of 3.0) according to our previous works.¹⁸ Mg(NO₃)₂·6H₂O (0.12 mol L⁻¹) and Al(NO₃)₃·9H₂O (0.04 mol L⁻¹) were dissolved in deionized water, and then added into a three-neck flask. The solution was maintained at 105 °C for 12 h under stirring, and then aged statically at 80 °C for 12 h. The formed solid was collected by filtration and washed to neutral using deionized water, and subsequently dried at 80 °C for 18 h, which was denoted as LDH. Part of the precursor LDH was calcined at 500 °C for 4 h in a muffle furnace, which was denoted as LDO.

The SO₄²⁻ intercalated hydrotalcite was prepared by intercalation–reconstruction. 2.0 g LDO sample was put into 1000 mL CO₂-free solution containing 550 mg L⁻¹ Na₂SO₄ at initial pH 4. Under a N₂ atmosphere, the mixture was maintained at 60 °C and 150 rpm magnetic stirrer for 80 min in a XH-100A microwave chemical reactor (500 W) (XiangHu Technology Co., Ltd., Beijing, China). After cooling, the resulting slurry was centrifuged at 8000 rpm for 10 min. The residuum was washed with deionized water to neutrality, and subsequently dried at 80 °C for 10 h, which was denoted as S-LDH.

2.3 Removal of ammonia nitrogen

The ammonia nitrogen (NH₃–N) synthetic wastewater containing 100 mg L⁻¹ NH₄Cl was prepared by dissolving NH₄Cl in deionized water. According to the experimental design, different amount of the S-LDH was added into the synthetic wastewater at a certain initial pH. Under a N₂ atmosphere, the reaction was conducted at a special temperature and 150 rpm magnetic stirrer in a microwave field for a given time. After reaction, the reaction solution was cooled and centrifuged at 7000 rpm for 3 min, and the supernatant was kept for NH₃–N measurement. The precipitate was washed three times, and the washing solutions were also stored for NH₃–N determination. The residual concentration of NH₃–N was measured by the Nessler's reagent spectrophotometry method (Hitachi U-2910) at 425 nm based on Chinese State Environmental Protection Standards HJ-535-2009. All the experiments were carried out in triplicate under the same condition and average values were reported. The NH₃–N removal ratio (η%) was calculated as follows,where C₀ and C_t were the initial and final values of ammonia nitrogen, respectively.

After approximation of the best conditions by ‘‘one-factor-at-a-time” method in our preliminary experiments, the response surface methodology (RSM) was used to test the effect of S-LDH amount, irradiation time, temperature and initial pH on the NH₃–N removal rate. Box–Behnken Design (BBD) was used to design experiment and optimize one response variable namely NH₃–N removal rate (η). Each independent variable was coded at three levels between −1 and +1, where the variables S-LDH amount (S), irradiation time (t), temperature (T) and initial pH (P) were changed in the ranges shown in Table 1. Twenty-nine experiments were augmented with three replications at the design center to evaluate the pure error, and carried in randomized order. After reaction, the response η was measured.

Table 1 Experimental range and levels of the independent variables

Independent variables	Symbols	Units	Code levels
			−1	0	+1
S-LDH amount	S	g L^−1	0.5	1.0	1.5
Irradiation time	t	min	5.0	10.0	15.0
Temperature	T	°C	45.0	60.0	75.0
Initial pH	P		8.0	10.0	12.0

---

The statistical software package Design Expert software (version 8.0.5) was used for regression analysis of the experimental data and to plot response surface. The model generated during RSM implementation was validated by conducting experiment (three replications) on given optimal setting. The second-order polynomial model was applied to predict the response variable (η) as shown below,

where η was the response value (NH₃–N removal rate), β₀, β_i, β_ii and β_ij were the regression coefficients for interception, linear, quadratic and interaction terms, respectively. X_i and X_j were the independent variables.

2.4 Characterization of samples

Powder X-ray diffraction (XRD) patterns were collected on a Japan Rigaku D/max 2550 PC (λ = 1.5405 Å) with Cu Kα radiation. The step size was 0.02° (2θ) at 0.5 s step time with a filament intensity of 30 mA and a voltage of 40 kV. Fourier transform infrared (FT-IR) spectrum was recorded on Perkin-Elmer Spectrum One B instrument using KBr pellet technique. Scanning electron microscopy (SEM, JEOL JSM-6700F) was used to characterize the surface morphology.

3 Results and discussion

3.1 Characterization of the materials

3.1.1 XRD analyses. Powder XRD patterns of the LDH and S-LDH samples are shown in Fig. 1. The two samples had the typical hydrotalcite layered structure with sharp and intense (003), (006), (009), (110) and (113) reflections and broadened (015) and (018) reflections.^18,19 Further analyses revealed some differences in the cell parameters among the two samples. The interlayer space (d₀₀₃ about 0.771) of the LDH was typical of a CO₃²⁻ anions intercalated hydrotalcite, while the interlayer space of the S-LDH was 0.879 nm, possibly due to the intercalated SO₄²⁻ anions.^20,21 No other crystalline phase was observed in the LDH precursor, indicating that the sample was a highly crystalline hydroxide structure. Conversely, MgO as an impure phase was observed in the S-LDH. Comparing with the S-LDH, the LDH precursor possessed the higher crystallinity due to a more stable baseline with sharper and higher diffraction peaks.^20,22 The parameter a, average cation–cation distance in the brucite sheet, was calculated from the (110) XRD reflection (Table 2). The similarity in a values among the LDH and S-LDH indicated that intercalation of SO₄²⁻ anions did not change the microstructure of the brucite sheets. The crystal sizes of the samples from rough calculation using the Debye–Scherrer equation were predicted. The crystal sizes in a direction (d_a) of the S-LDH were smaller than those of the LDH, and the crystallite sizes in c direction (d_c) followed a similar trend, demonstrating that the S-LDH had smaller crystallite sizes. The results revealed that the crystal sizes decreased after the SO₄²⁻ intercalation reaction, which could be further confirmed by SEM analyses.

Fig. 1 XRD patterns of the LDH and S-LDH samples.

Table 2 Crystallographic parameters of the samples^a

Parameter	LDH	S-LDH
a FW: half-width of diffraction peak, da: crystallite size in a axis direction, dc: crystallite size in c axis direction.
d003 (nm)	0.771	0.879
d110 (nm)	0.153	0.152
FW003 (rad)	0.538	1.062
FW110 (rad)	0.483	0.889
a (nm)	0.304	0.304
dc (nm)	14.68	7.43
da (nm)	18.80	10.24

---

3.1.2 FT-IR analyses. The FT-IR spectra of the LDH and S-LDH samples in the region 4000–400 cm⁻¹ are displayed in Fig. 2, which were typical of hydrotalcites. The peaks at around 3446 cm⁻¹ (structural –OH vibrations) and 1630 cm⁻¹ (water bending vibrations) were observed in the two samples. The bands below 1000 cm⁻¹ were attributed to the lattice vibration of M–O and M–O–M (M: Mg, Al).^23,24 The peaks at 1353 cm⁻¹ and 681 cm⁻¹ were associated with CO₃²⁻ stretching vibration in the interlayer spaces.¹⁹ For the S-LDH, the peak at 1353 cm⁻¹ disappeared, while the peak at 681 cm⁻¹ shrank. It indicated that the amount of CO₃²⁻ in the S-LDH was significantly decreased, but only traces of CO₃²⁻ still existed in the interlayer space of the S-LDH. There were new peaks appeared at 1115 and 618 cm⁻¹, which were due to the asymmetric and symmetric stretching vibration of S=O and S–O, respectively.^20,25,26 The results indicated that SO₄²⁻ as a replacement for CO₃²⁻ were intercalated into the interlayer spaces by intercalation–reconstruction process and trace amounts of CO₃²⁻ still remained in the interlayer spaces of the S-LDH.

Fig. 2 FT-IR spectra of the LDH and S-LDH samples.

3.1.3 SEM analyses. In order to investigate the morphology of the LDH and S-LDH, the two samples are observed by SEM (Fig. 3). As shown in Fig. 3, all the crystal particles of the LDH showed well-developed plates and were made up of individual platelets with a narrow size distribution (2.0–3.0 μm), where the LDH had more well-developed layered structure than the S-LDH. The S-LDH had thin flat crystals with various edges indicting the existence of layered structure, where there was some tendency for platelets to aggregate in clumpy manner. The crystal sizes of the S-LDH with a broad size distribution (0.5–2.5 μm) were significantly smaller than those of the LDH precursor. The results revealed that the intercalation–reconstruction method could rebuild the layered structure, decrease crystallinity and crystal sizes, which supported the speculation from the XRD analyses.

Fig. 3 SEM images of the LDH and S-LDH samples, ×10 000.

3.2 Optimization of the NH₃–N removal by RSM

3.2.1 Regression analysis of BBD. Based on Box–Behnken Design (BBD), the values of independent variables and the response (η, NH₃–N removal rate) obtained in the multivariate study for each experiment were represented in Table 3. Each result was expressed as arithmetic mean of three replications. Regression analysis is the general approach to fit the empirical model with the collected response variable data. By using regression analysis, the responses obtained in Table 4 were correlated with the four independent factors using the polynomial equation as in eqn (2).

Table 3 Experimental Box–Behnken design matrix and its response and predicted value

Run	Experimental variables				Response η (%)
	S (g L^−1)	t (min)	T (°C)	P	Expt.	Predicted
1	0.5	5	60	10	55.9	55.2
2	0.5	10	75	10	74.1	75.9
3	1	10	60	10	86.9	86.1
4	1	10	60	10	86.1	86.1
5	1	10	45	12	72.9	73.5
6	1	5	45	10	51.7	51.8
7	1.5	10	60	12	90.3	90.0
8	0.5	15	60	10	72.5	71.8
9	1	5	75	10	78.7	78.6
10	1	10	75	8	79.2	77.4
11	1.5	5	60	10	75.2	74.7
12	1.5	10	45	10	74.8	74.4
13	1	10	60	10	86.5	86.1
14	1	10	45	8	55.3	54.1
15	1	10	60	10	85.8	86.1
16	1	5	60	12	76.6	77.1
17	1	15	45	10	74.2	74.1
18	1	10	75	12	90.6	90.7
19	1	15	60	8	75.7	76.6
20	1	15	60	12	86.8	87.5
21	1	5	60	8	54.7	55.4
22	1.5	10	75	10	89.2	89.6
23	1	10	60	10	85.1	86.1
24	1.5	10	60	8	74.8	76.1
25	0.5	10	45	10	49.5	50.6
26	1	15	75	10	88.3	87.9
27	0.5	10	60	8	54.8	54.9
28	1.5	15	60	10	90.2	89.7
29	0.5	10	60	12	75.3	73.7

---

Table 4 ANOVA analysis for response surface quadratic model (α = 0.05)

Source	Sum of squares	DF	Mean square	F	p-Value
Model	4625.45	14	330.39	233.82	<0.0001
S	1052.81	1	1052.81	745.08	<0.0001
t	750.50	1	750.50	531.14	<0.0001
T	1234.24	1	1234.24	873.48	<0.0001
P	800.33	1	800.33	566.40	<0.0001
St	0.64	1	0.64	0.45	0.5119
ST	26.01	1	26.01	18.41	0.0007
SP	6.25	1	6.25	4.42	0.0540
tT	41.60	1	41.60	29.44	<0.0001
tP	29.16	1	29.16	20.64	0.0005
TP	9.61	1	9.61	6.80	0.0207
S^2	304.58	1	304.58	215.56	<0.0001
t^2	262.79	1	262.79	185.98	<0.0001
T^2	283.84	1	283.84	200.87	<0.0001
P^2	199.98	1	199.98	141.53	<0.0001
Residual	19.78	14	1.41
Lack of fit	17.89	10	1.79	3.79	0.1054
Pure error	1.89	4	0.47
Cor total	4645.23	28

---

R^2	0.9957
Adjusted R^2	0.9915
Predicted R^2	0.9772
Adeq precision	46.925
CV	1.57%

---

The statistical significance and fitness of the model equation as well as the effects of significant individual terms and their interactions on the chosen response was evaluated. According to the ANOVA results (Table 4), the model was high significant with a p-value less than 0.0001 to predict the response value, indicating the model was significant for 95% confidence intervals. In terms of the significant coefficients, the linear terms of S (S-LDH amount), t (irradiation time), T (temperature) and P (initial pH) were the very important factors due to the p-values less than 0.0001. The results clearly indicated that all the variables were the important factors affecting the response (η). For interaction terms, the quadratic term of the tT (p < 0.0001) was high significant term, and tP, ST and TP were significant terms (p ≤ 0.05), but St and SP were insignificant (p > 0.05). All the quadratic terms of the S², t², T² and P² were high significant terms (p < 0.0001). Insignificant terms could be manually removed from the model to improve the regression model and optimization results, and the quadratic model was given in eqn (3), as shown below:

η = 86.08 + 9.37S + 7.91t + 10.14T + 8.17P − 2.55ST − 3.23tT − 2.70tP − 1.55TP − 6.85S² − 6.36t² − 6.62T² − 5.55P² (3)

To test the fit of the model, the regression equation and determination coefficient (R²) were evaluated. A high coefficient of determination (R² 0.9957) value indicated that the response variation of 99.57% for the NH₃–N removal was attributed to the independent variables and only 0.43% of the total variations were not explained by the model, namely the obtained model gave a good response estimate within the studied range. The predicted R² is a measure of how good the model predicts the values for the response, and the adjusted R² verifies the experimental data and the model precision.¹⁷ The predicted R² and adjusted R² were close to 1, which indicated adequacy of the model. Adequate precision value gives a measure of signal to noise ratio, which the ratio greater than 4 is desirable. The ratio of 46.925 indicated adequate model discrimination.

Lack of fit, which is the weighted sum of squared deviations between the mean response at each factor level and the corresponding fitted value, was insignificant relative to the pure error for the response with a p-value of 0.1054. Insignificant lack of fit was good, implying that the model could be used to navigate the design space. If the CV is not greater than 10%, a model can be considered reasonably reproducible.²⁷ A relatively low value of coefficient of the variation (CV = 1.57%) clearly indicated a very high degree of precision and reliability of the experiments carried out.²⁸ ANOVA results of the quadratic model indicated that the quadratic mode could be used to navigate the design space.

Three-dimensional response surfaces were plotted on the basis of graphical representations of the regression equation in order to investigate the interaction among the variables. The model suggested that there were significant interactions between temperature and irradiation time, between irradiation time and initial pH, between S-LDH amount and temperature, and between temperature and initial pH. Therefore, it was of great interest to further characterize the interactions in the range of the process variables. Fig. 4–7 represented the effect of two variables on the NH₃–N removal rate, while the other two variables were held at zero level.

3.2.2 Interactions between the variables. The combined effect of temperature and irradiation time on the NH₃–N removal rate at constant S-LDH amount of 1.0 g L⁻¹ and initial pH 10 is presented in Fig. 4. The interaction of the two variables indicated a significant impact on the NH₃–N removal rate, where higher NH₃–N removal rate could be attained at higher irradiation time and temperature. At low temperature there was a moderate increase in the NH₃–N removal rate with irradiation time, while a rapid increase of the NH₃–N removal rate with time at high temperature. The NH₃–N removal rate increased from 52.2 to 73.9% when time increased from 5.0 to 14.0 min at 45 °C and then reached equilibrium. At 75 °C, the NH₃–N removal rate increment corresponded to the extension of irradiation time from 5.0 to 12.0 min, and subsequently became flat. The maximum NH₃–N removal rate (89.3%) of the S-LDH was obtained at 66.3 °C for 12.0 min.

Fig. 4 Interaction effect between the temperature and irradiation time towards NH₃–N removal rate.

Fig. 5 shows the NH₃–N removal rate with varying irradiation time and initial pH at constant S-LDH amount (1.0 g L⁻¹) and temperature (60 °C). The response surface plot revealed that the initial pH had an interaction with the irradiation time, where the conjugate effect of initial pH and irradiation time was significant with the contour curve of oval shape. The initial pH displayed an effect on the response from 8.0 to 10.8, where the NH₃–N removal rate increased with initial pH, and then reached equilibrium when the initial pH was further raised. Similarly, the NH₃–N removal rate also increased to a maximum with the increasing irradiation time and then leveled off with the further extension of irradiation time (>12.3 min). The best results (above 89.8% removal rate) were observed for higher initial pH (>10.8) and irradiation times higher than 12.0 min.

Fig. 5 Interaction effect between the irradiation time and initial pH towards NH₃–N removal rate.

The NH₃–N removal rate is shown in Fig. 6 as a function of S-LDH amount and temperature at constant initial pH 10 and irradiation time 10 min. Both S-LDH amount and temperature exerted the effect on the NH₃–N removal rate when the S-LDH amount and temperature were under 1.2 g L⁻¹ and 67.5 °C, respectively, and then achieved a balance with further increasing the S-LDH amount and temperature. At low S-LDH amount, there was a moderate increase in NH₃–N removal rate with irradiation temperature, while the increase of NH₃–N removal rate immediately rose with time at high S-LDH amount.

Fig. 6 Interaction effect between the S-LDH amount and temperature towards NH₃–N removal rate.

It is clear from Fig. 7 that the combined effect of temperature and initial pH on the NH₃–N removal rate was significant at constant irradiation time (10 min) and S-LDH amount of 1.0 g L⁻¹. At 45 °C, the NH₃–N removal rate increased to about 73.4% with an increase in initial pH from 8.0 to 11.2, and then reached equilibrium when initial pH was further increased. When temperature was increased to 75 °C, the NH₃–N removal rate also increased to 91.5% when the initial pH was adjusted from 8.0 to 11.0.

Fig. 7 Interaction effect between the temperature and initial pH towards NH₃–N removal rate.

3.2.3 Optimization analysis. Although the actual situation might be more complicated than what we reported, an attempt for the optimization of the NH₃–N removal was made by RSM. Numerical optimization method was used to optimize the desired response of the system, which the response was the NH₃–N removal rate and it should keep all the variables in the range of the experimental values. The optimal values of the selected variables for the NH₃–N removal rate were predicted using the optimization function of the Design Expert Software. The model predicted the maximum NH₃–N removal rate of 91.2% appeared under the conditions of S-LDH amount of 1.2 g L⁻¹, temperature 71.7 °C, initial pH 10.1 and irradiation time 9.3 min.

3.2.4 Verification of the optimal parameters. To validate the adequacy of the model equations, the additional experiments in five independent replicates were carried out under the obtained optimum operating conditions of S-LDH amount of 1.0 g L⁻¹, temperature 70 °C and initial pH 10 for 9 min. The mean experimental NH₃–N removal rate was 90.4 ± 0.18% and (n = 5), while the predicted NH₃–N removal rate was 91.2%. It could be seen that the error between the experimental and predicted values was less than 0.9%, indicating that the values of NH₃–N removal rate from the experiment were in good agreement with the predicted value. The model validation was determined by additional independent experiments demonstrated that the experimental response value was basically the same as the predicted response value, which confirmed the efficacy of the predicted model. Therefore, it concluded that the generated model had sufficient accuracy to predict the NH₃–N removal rate, and more future research work should be dedicated to the use of the S-LDH material.

4 Conclusions

Microwave irradiation was used not only to prepare the SO₄²⁻ intercalated Mg/Al hydrotalcite (S-LDH) by intercalation–reconstruction method, but also to remove NH₃–N in synthetic wastewater by the S-LDH. XRD, FT-IR and SEM analyses indicated that SO₄²⁻ anions were successfully intercalated into the interlayer spaces of the S-LDH. Response surface method (RSM) was employed to evaluate the effect of the S-LDH amount, irradiation time, temperature and initial pH on the NH₃–N removal rate of the S-LDH. A quadratic model was developed to correlate the reaction variables to the NH₃–N removal rate by a Box–Behnken design (BBD) approach. Based on the analysis of variance (ANOVA), the most influential factors on the experimental design response were identified, and the predicted response value after process optimization was found to fit well with the experimental value, indicating suitability of the model to optimize the conditions. The adequacy of the model was verified effectively by the validation of experimental data. Thus, application of RSM on NH₃–N removal by the S-LDH successfully increased the NH₃–N removal rate leading to improved economy of the process.

Acknowledgements

This work is supported by the Key Project of Hunan Provincial Natural Science Foundation of China (12JJ2008), and Xiangtan University Graduate Innovation Project (XJCX201405).

References

1. L. Lin, J. Chen, Z. Q. Xu, S. H. Yuan, M. H. Cao, H. C. Liu and X. H. Lu, J. Hazard. Mater., 2009, 168, 862–867.
2. M. Minhalma and M. N. de Pinho, J. Membr. Sci., 2004, 242, 87–95.
3. Y. M. Li, G. W. Gu, J. F. Zhao, H. Q. Yu, Y. L. Qiu and Y. Z. Peng, Chemosphere, 2003, 52, 997–1005.
4. X. Zhou, Y. Li and Y. Zhao, RSC Adv., 2014, 4, 15620–15629.
5. X. P. Zhu, J. R. Ni and P. Lai, Water Res., 2009, 43, 4347–4355.
6. W. P. Cao, RSC Adv., 2015, 5, 14042–14046.
7. T. L. Huang, J. M. MacInnes and K. R. Cliffe, Water Res., 2001, 35, 2113–2120.
8. K. W. Kim, Y. J. Kim, I. T. Kim, G. I. Park and E. H. Lee, Water Res., 2006, 40, 1431–1441.
9. N. Remya and J. G. Lin, Chem. Eng. J., 2011, 166, 797–813.
10. M. Zieliński, M. Zielińska and M. D. ębowski, Bioresour. Technol., 2013, 127, 223–230.
11. C. R. Chen, H. Y. Zeng, S. Xu, X. J. Liu, H. M. Xiao and H. Z. Duan, RSC Adv., 2016, 6, 12753–12760.
12. Y. Wang, J. Qu, H. Liu and C. Hu, Catal. Today, 2007, 126, 476–482.
13. E. Palomares, J. G. Prato, F. Rey and A. Corma, J. Catal., 2004, 221, 62–66.
14. Y. Xu, J. Zhang, Y. Liang, J. Zhou, J. Zhao, X. Ruan, Z. P. Xu and G. Qian, Sep. Purif. Technol., 2015, 145, 92–97.
15. F. Bruna, R. Celis, M. Real and J. Cornejo, J. Hazard. Mater., 2012, 225, 74–80.
16. J. N. Sahu, J. Acharya and B. C. Meikap, J. Hazard. Mater., 2009, 172, 818–825.
17. A. D. Pan, H. Y. Zeng, G. B. F. C. Alain and B. Feng, Food Chem., 2016, 201, 230–236.
18. H. Y. Zeng, X. Deng, Y. J. Wang and K. B. Liao, AIChE J., 2009, 55, 1229–1235.
19. S. Xu, M. C. Liao, H. Y. Zeng, Z. Q. Zhang, X. J. Liu and P. H. Zhu, Appl. Clay Sci., 2015, 108, 215–221.
20. S. Xu, M. C. Liao, H. Y. Zeng, X. J. Liu, J. Z. Du, P. X. Ding and W. Zhang, J. Taiwan Inst. Chem. Eng., 2015, 56, 174–180.
21. J. T. Kloprogge, D. Wharton, L. Hickey and R. L. Frost, Am. Mineral., 2002, 87, 623–629.
22. Q. J. Huang, H. Y. Zeng, W. Zhang, B. Feng, X. J. Liu, H. Z. Duan and P. X. Ding, Chem. Eng. J., 2016, 60, 525–531.
23. S. Xu, M. C. Liao, H. Y. Zeng, C. R. Chen, H. Z. Duan, X. J. Liu and J. Z. Du, Appl. Clay Sci., 2015, 115, 124–131.
24. T. T. Wang, Z. H. Yang, B. Yang, R. J. Wang and J. H. Huang, J. Power Sources, 2014, 257, 174–180.
25. A. V. Radha, P. V. Kamath and C. Shivakumara, J. Phys. Chem. B, 2007, 111, 3411–3418.
26. X. Ren, X. Hu, F. Zhang, J. Wang, J. Liang, W. Wu and J. Wang, Catal. Sci. Technol., 2015, 5, 4813–4820.
27. H. Y. Zeng, S. Xu, M. C. Liao, Z. Q. Zhang and C. Zhao, Appl. Clay Sci., 2014, 91, 16–24.
28. X. Z. Yuan, J. Liu, G. M. Zeng, J. G. Shi, J. Y. Tong and G. H. Huang, Renewable Energy, 2008, 33, 1678–1684.

---