To compare the internal microstructures of PVC, PVC-BT and PVC-LaBT, cross sections of these membranes are shown in Figure 2.

As can be seen from Figures 2a1–c1, the thickness of PVC membrane is about 50 μm, while that of PVC-BT and PVC-LaBT were 100 μm and 116 μm respectively. All three membranes were prepared by 200 μm applicator, but the thickness of the finished membranes is less than 200 μm. This is because the membranes shrunk during the process of phase conversion. During the experiment, we also optimized the thickness of the membrane. When the thickness was 100 or 150 μm, the membrane was prone to breakage during the casting process, and small holes were likely to appear on the membrane surface. When the thickness exceeded 200 μm, due to the high fluidity of the casting solution, the scraper was unable to evenly smooth the surface of the membrane.

Furthermore, the addition of adsorbent increased the viscosity of PVC solution, which reduced the contraction space of the membrane during phase conversion process. Therefore, the thickness of PVC-BT and PVC-LaBT is larger than that of pure PVC membrane.

By observing the structure of the membrane sections, it can be seen that there are some channels in PVC membrane, but the number and size are both small. The cross section of PVC-BT shows a mesh structure, but the pore size is small. There are numerous of pores with large size distributes in the internal of PVC-LaBT.

To compare the internal microstructures of PVC, PVC-BT and PVC-LaBT, cross sections of the three membranes were visualized by 5,000×, 100,00× and 2,000× under a SEM. As can be seen from Figure 2a2,a3, there are few pores with small sizes in pure PVC membranes. These pores are irregular in shape and have thick walls. According to Figure 2a4, when the inner wall of the pores was observed by 200,00×, only a small part of the inner wall has pores with small size, while there are nearly no pores can be found to distribute over most of the inner wall. As can be seen from Figures 2b2–b4, the interior of PVC-BT membrane has a porous mesh structure. These pores are relatively numerous, with small size and irregular shape. Figures 2c2–c4 shows that the interior of PVC-LaBT shows a loose porous structure. The pores have a irregular shape. Compared to the PVC and PVC-BT membranes, the PVC-LaBT membrane has a large internal aperture, wide and dense distribution, and very thin pore walls.

The comparison of SEM images shows that PVC-LaBT has the most internal pores with large size, which indicates that the specific surface area of PVC-LaBT is the largest. Compared to PVC-LaBT, the internal pore size of PVC-BT is smaller. This is because after the modification of bentonite, La³⁺ enters the interlayer of bentonite through ion exchange, increasing the interlayer spacing and improving the specific surface area, resulting in larger membrane pore size. Compared with the composite membrane, the pores of pure PVC membrane without adding matrix are the least, indicating that the addition of bentonite can improve the specific surface area of the PVC membrane.

2D and 3D atomic force microscope (AFM) test images were obtained to observe and compare the surface roughness and morphology of PVC, PVC-BT and PVC-LaBT membranes. The data were analyzed to obtain the parameters of different membrane surface such as average roughness (Ra), maximum height roughness (Rmax) and root mean square roughness (Rq).

According to Figures 3a1, a2, the pure PVC membrane has a smooth surface, with small and uniform surface particles. After PVC AFM data analysis, we obtained its Ra of 8.5 nm, Rq of 10.7 nm and Rmax of 90.3 nm. As can be seen from Figures 3b1, b2, the surface of PVC-BT is relatively rough. Compared with PVC, the surface particles distribute more unevenly. After PVC-BT AFM data analysis, we obtained its Ra of 9.7 nm, Rq of 12.6 nm and Rmax of 122.0 nm. Figures 3c1, c2 shows that the surface of PVC-LaBT is the roughest, which is due to the wide disparity between surface particle sizes. Furthermore, the distribution of surface particles is the most uneven. The AFM data of PVC-LaBT analyzed Ra of 12.2 nm, Rq of 15.7 nm, and Rmax of 170.0 nm. The surface roughness information of PVC, PVC-BT and PVC-LaBT membranes were filled in Table 1 for comparison.

In conclusion, PVC-LaBT has the largest surface roughness, followed by PVC-BT and PVC membranes. According to the comparison of the Ra value of the three membranes, the roughness of PVC-BT is only 14.5% larger than that of pure PVC membrane, while that of PVC-LaBT is 29% higher than that of pure PVC membrane. From this information, it can be inferred that compared to PVC and PVC-BT membranes, PVC-LaBT has the largest specific surface area and the best adsorption performance.

To compare the performance of different membranes, 1 mg·L⁻¹ of phosphate cross flowed through PVC, PVC-BT and PVC-LaBT. The phosphate concentration (C _t) was tested every hour for 8 h. The experimental results are shown in Figure 5.

As can be seen from Figure 5, the three kinds of membranes all have a high phosphate adsorption rate in the first 3 h, among which the adsorption rate of PVC-LaBT was the highest. After 3 h, the C _t of the solution treated by PVC, PVC-BT and PVC-LaBT were 0.75, 0.65 and 0.30 mg·L⁻¹ respectively. Since then, the adsorption rate has decreased, and after 7 h, the adsorption process basically entered the adsorption-desorption equilibrium stage. Finally, the C _t of the solution treated by PVC, PVC-BT and PVC-LaBT were 0.70, 0.50 and 0.10 mg·L⁻¹ respectively. This is partly because the solubility product constant (pK_sp) of La-phosphate is large, which enhances the chemical adsorption, and on the other hand the intercalation of La ions results in larger interlayer space and porosity, thus providing more attachment sites (He et al., 2022).

Because SA-LaBT has the best filtration-adsorption performance, it was selected for subsequent influencing factor, adsorption kinetic and isotherm experiments.

Phosphorus-containing wastewater mainly includes municipal sewage, acid pickling and phosphorization wastewater, and wastewater from the phosphorus chemical industry, etc. Their pH values are usually between 3 and 11. To find the effect of solution pH on the filtration-adsorption performance of phosphate by PVC-LaBT. The pH of 1 mg·L⁻¹ phosphate solution was adjusted with 1 mol·L⁻¹ HNO₃ solution or NaOH solution to 3, 5, 7, 9 and 11. The filtration-adsorption experiments were performed by using PVC-LaBT and phosphate solution with different pHs. The experimental results are shown in Figure 6a.

According to Figure 6a, the adsorption capacity of PVC-LaBT was excellent under acidic conditions. The C _t decreased to 0.3 mg·L⁻¹ and 0.1 mg·L⁻¹ after 4 h and 8 h, respectively. When solution pH ≥ 7.0, the adsorption capacity of PVC-LaBT weakened with the increment of pH. When the solution was neutral, the C _t decreased to 0.3 mg·L⁻¹ and 0.2 mg·L⁻¹ after 4 h and 8 h, respectively. The performance decline was not obvious. However, when pH was 9.0, the adsorption performance decreased rapidly. After 8 h, the C _t decreased only from 1.0 mg·L⁻¹ to 0.6 mg·L⁻¹, with the removal efficiency of only 40%. When pH was 11.0, the adsorption performance was completely lost. The reason for the drastic decrement in the performance of PVC-LaBT in the alkaline solution may be caused by the OH⁻ competing with PO₄ ³⁻ for the adsorption site (Zhang Y. et al., 2022).

To explore the effect of coexisting anions on the performance of phosphate filtration by PVC-LaBT, 0.5 mmol KCl, KNO₃, K₂SO₄ and K₂CO₃ were respectively added to the phosphate solution with a concentration of 1 mg·L⁻¹. At this time, the molar mass of the coexisting anions in the solution was approximately 70 times the molar mass of the phosphate. The prepared phosphate solution containing coexisting anions passed through PVC-LaBT by cycle cross-flow mode and the C _t was tested after 4 h and 8 h. The experimental results are shown in Figure 6b.

As can be seen from Figure 6b, the order of anion inhibiting phosphate adsorption was: CO₃ ²⁻>SO₄ ²⁻>Cl⁻>NO₃ ⁻. After 4 h, the C _t in the blank group decreased to 0.3 mg·L⁻¹, as was the C _t in the solution containing NO₃ ⁻. The C _t in the solutions containing Cl, SO₄ ²⁻ and CO₃ ²⁻ were 0.4 mg·L⁻¹, 0.5 mg·L⁻¹ and 0.5 mg·L⁻¹ respectively. At this point, the inhibitory effect of anions was not obvious. However, after cross flow filtration for 8 h, the C _t in the solutions containing Cl, NO₃ ⁻, SO₄ ²⁻ and CO₃ ²⁻ were 0.2 mg·L⁻¹, 0.1 mg·L⁻¹, 0.3 mg·L⁻¹ and 0.5 mg·L⁻¹ respectively. This shows that divalent anions inhibited phosphate adsorption more strongly, while monovalent anions were less inhibitory. This is because SO₄ ⁻ and CO₃ ²⁻ can complete for adsorption with the phosphate anions. However, the pK_sp between Cl⁻ or NO₃ ⁻ and La₃ ⁻ is much smaller than that of PO₄ ³⁻ and La₃ ⁻, so there is no significant inhibition on the adsorption of phosphate (Du et al., 2022).

In other studies, adsorbents have also been embedded into organic membranes for the treatment of phosphate. The relevant parameters are listed in Table 2. As can be seen from the table, PVC-LaBT has moderate water flux. It also has outstanding adsorption capacity for phosphate. In addition, PVC-LaBT has the advantages of simple preparation and low cost. Therefore, it is expected to be widely used.

In order to explore the adsorption kinetic of PVC-LaBT, the phosphate solution of 1 mg·L⁻¹ was recycled through PVC-LaBT. The phosphate concentrations were determined at different times. To understand the mechanisms of adsorption kinetics, the experimental data were fitted with the following five kinetic models (Equations 4–8), (Simonin, 2016; Azizian, 2004; Wang and Guo, 2022):

Pseudo-first-order model: q t = q e − q e e − k 1 t (4)

Pseudo-second-order model: q t = k 2 q e 2 t 1 + k 2 q e t (5)

Richie n^th-order model: q t = q e 1 − 1 β + k n n − 1 t 1 n − 1 (6)

Elovich model: q t = 1 β ln α β t (7)

Intraparticle diffusion model: q t = k i d t 0.5 + C (8) where k ₁ (1·h⁻¹), k ₂ (m²·mg⁻¹·h⁻¹) and k _n (m²⁽ⁿ⁻¹⁾⋅mg¹⁻ⁿ⋅h⁻¹) are the pseudo-first-order equilibrium rate constant, pseudo-second-order equilibrium rate constant and Ritchie nth-order equilibrium rate constant, respectively; n is a fitting parameter, dimensionless and without special significance; α (mg⋅m⁻²⋅h⁻¹) is the adsorption rate at initial time, while β (m²⋅mg⁻¹) denotes the constant of desorption. k _id (mg⋅m⁻²⋅h^−0.5) is the rate constant of intraparticle diffusion and C (mg⋅m⁻²) is determined by the thickness of boundary layer.

The experimental data fitting results are shown in Figure 7, and the corresponding kinetic parameters are listed in Table 3. As can be seen from Figure 7, in the first 3 h, the phosphate adsorption capacity increased rapidly. In the following 3–6 h, the increment speed of phosphate adsorption capacity became slow. In the next 6–8 h, the adsorption capacity tended to stabilize, and the adsorption process basically reached equilibrium.

The parameters listed in Table 3 show that the Pseudo-first-order, Pseudo-second-order, Richie nth-order and Elovich models all have high R² values, among which the Richie nth-order model has the highest R² coefficient (0.9969), indicating that the adsorption of phosphate by PVC-LaBT is controlled by various mechanisms, including chemical adsorption, electrostatic attraction, hydrogen bonding and outer-sphere complex formation (Pan W. et al., 2022). Moreover, the theoretical adsorption capacity calculated by the Richie nth-order model (131.73 mg⋅m⁻²) basically agrees with the experimental adsorption capacity value (129.29 mg⋅m⁻²).

Furthermore, to elucidate the diffusion mechanism of phosphate during adsorption, the kinetic data were fitted to Intraparticle diffusion model with three linear regions. The data are shown in Figure 7b and Table 3. This plot exhibits a multilinear morphology that can be divided into three linear regions, suggesting that intra-particle diffusion is not the only rate-control step for the entire reaction (Luo et al., 2021). The k _id1 value was 55.7865 mg⋅m⁻²⋅h^−0.5, indicating that the adsorption rate is the highest in the first stage, during which the phosphate ions were adsorbed to the surface PVC-LaBT. The slope of the second linear region decreased, with k _id2 value being 32.7635 mg⋅m⁻²⋅h^−0.5, demonstrating that the adsorption rate at this stage was controlled by intra-particle or pore diffusion. The last linear region flattened out, exhibiting the last equilibrium phase.

The isotherm experiments were performed by putting PVC-LaBT in the triple high pressure plate membrane testing equipment and cycle the phosphate solution of different initial concentrations (1, 2, 3, 4 and 5 mg⋅L⁻¹). The phosphate concentrations were tested after 20 h, and thus the phosphate adsorption capacities were calculated. At last, the experimental data were fitted with Langmuir and Freundlich models using Equations 9, 10 respectively listed below, in order to analyze the adsorption mechanism and predict the maximum adsorption capacity (Wang and Guo, 2020; Foo and Hameed, 2010).

Langmuir model: q e = q m K L C e 1 + K L C e (9)

Freundlich model: q e = K F C e n (10) where K _L (L⋅mg⁻¹) and K _F (mg^(1−1/n)·L^1/n·m⁻²) are the Langmuir adsorption equilibrium constant and the affinity parameter of Freundlich model respectively; q _m (mg⋅m⁻²) indicates the Langmuir maximum adsorption capacity; n is the Freundlich linearity constant.

According to Figure 8, the phosphate adsorption capacity of PVC-LaBT is gradually increased as the initial phosphate concentration increased from 1 mg⋅L⁻¹ to 5 mg⋅L⁻¹. The obtained isotherm parameters are tabulated in Table 4. The fitting degree of the model is mainly based on the value of R². Generally, the higher the value of R² indicates the better a fit is. Table 4 shows that the Langmuir model gives a high correlation coefficient R² (0.9820), indicating that the adsorption system was mainly monolayer and the phosphate ions dispersed uniformly on the active sites of PVC-LaBT without interaction (Abukhadra and Mostafa, 2019). However, the Freundlich model (R² = 0.9493) is not negligible. This indicates that there is also a multi-layer adsorption phenomenon with heterogeneous surfaces in the PVC-LaBT. That is to say, a portion of the adsorption active sites on the PVC-LaBT surface are unevenly distributed, which can lead to the phenomenon of active site stacking (Wang L. et al., 2023). Furthermore, the maximum adsorption capacity of PVC-LaBT predicted by Langmuir model was 244.09 mg⋅m⁻².