Synthesis and Characterization of g-C3N4/Ag3PO4/TiO2/PVDF Membrane with Remarkable Self-Cleaning Properties for Rhodamine B Removal

g-C3N4/Ag3PO4/TiO2 nanocomposite materials were loaded onto a polyvinylidene fluoride (PVDF) membrane using a phase inversion method to obtain a photocatalytic flat membrane for dye removal. The morphology, structure, and photocatalytic activity of the g-C3N4/Ag3PO4/TiO2 nanoparticles and composite membrane were evaluated. The g-C3N4/Ag3PO4/TiO2/PVDF membrane exhibited superior morphology, hydrophilic properties, and antifouling performance compared with the raw PVDF membrane. Four-stage filtration was performed to evaluate the self-cleaning and antifouling capacity of the g-C3N4/Ag3PO4/TiO2/PVDF membrane. Upon irradiating the composite membrane with visible light for 30 min, its irreversible fouling resistance (Rir) was low (9%), and its flux recovery rate (FRR) was high (71.0%) after five filtration cycles. The removal rate of rhodamine B (RhB) from the composite membrane under visible light irradiation reached 98.1% owing to the high photocatalytic activity of the membrane, which was superior to that of raw PVDF membrane (42.5%). A mechanism of photocatalytic composite membranes for RhB degradation was proposed. Therefore, this study is expected to broaden prospects in the field of membrane filtration technology.


Introduction
Rhodamine B (RhB) is one of the most popular dyes for printing, textile, and leather industries, which could cause pollution to the environment if not treated appropriately. Traditional physical methods, such as physical adsorption, coagulation, and chemical oxidation, and biological methods have been unable to achieve the efficient treatment of dye wastewater. Membranes with good selectivity, high efficiency, and excellent stability are widely used to treat municipal drinking water and wastewater. However, membrane fouling and further treatment of concentrated contaminant solutions are bottlenecks that have hindered the development of membrane separation technologies [1,2]. Surface modifications, which increase the hydrophilicity and improve the self-cleaning ability of membranes, are effective methods for enhancing the antifouling properties of membranes. Polyvinylidene fluoride (PVDF) has been widely employed in many areas because it involves a simple preparation process, can be easily modified, presents high strength, and is inexpensive [3,4]. However, PVDF membranes should be modified to confer on them hydrophilic properties and improve their anti-pollution performance. Among the various modification methods studied, the introduction of inorganic nanocomponents into PVDF membranes is an effective approach because nanoparticles present surface and interface effects; moreover, a small amount of dopant can confer on membranes many new functions. Recently, doping PVDF membranes with nanoparticles with photocatalytic properties, such as Fe(NO) 3 [5], Rb x WO 3 [6,7], C 3 N 4 [8][9][10], ZrO 2 [11,12], ZnO [13,14], Ag 3 PO 4 [15,16], TiO 2 [17,18], has become a popular research topic, because the fabricated photocatalyst-loaded membranes effectively integrate photocatalytic oxidation degradation of pollutants and separation within a single device. Since photocatalysts can effectively degrade contaminants in solutions and on the membrane surface, photocatalyst-loaded membranes present outstanding removal efficiency and remarkable anti-pollution performance [19,20].
Graphitic carbon nitride (g-C 3 N 4 ), which primarily comprises C and N, presents remarkable application potential owing to its excellent photocatalytic performance and chemical stability. However, the photocatalytic activity of pure g-C 3 N 4 has the disadvantages of a high carrier recombination rate, sparse adsorption sites, and active sites above 450 nm, which result in a volatile photocatalytic efficiency [21,22]. The recombination rate of photogenerated carriers in composites fabricated using two or more semiconductors (or heterojunction structures) is lower than that of their individual components; therefore, the photocatalytic performance of composites is superior to that of their individual components. Numerous semiconductor materials, such as MoO 3 [23], TiO 2 [24], ZnO [25], SnO 2 [26], ZrO 2 [27], BIOI [28], Ag 3 PO 4 [29][30][31], and graphene oxide (GO) [8,32], were coupled with g-C 3 N 4 to accelerate the separation of photogenerated electron-hole pairs and increase the bandgap energy. Among them, Ag 3 PO 4 has been applied as a highly effective compound for the visible-light-driven photodegradation of various organic pollutants in an aqueous solution; however, it is prone to photocorrosion in practical applications. The electric field that typically forms upon coupling g-C 3 N 4 with Ag 3 PO 4 can significantly increase the transfer rate of photogenerated charge carriers [29]. The simultaneous addition of TiO 2 and Ag 3 PO 4 to g-C 3 N 4 inhibits the photocorrosion of Ag 3 PO 4 because the band positions of TiO 2 and Ag 3 PO 4 match. Furthermore, other methods aimed at improving photocatalytic activity by combining g-C 3 N 4 with Ag 3 PO 4 and other Ag-based compounds or TiO 2 to form heterojunctions have been successfully developed. Abbasi-Asl et al. [30] prepared a TiO 2 /Ag 3 PO 4 /g-C 3 N 4 semiconductor composite and used it for the degradation of metronidazole (MNZ). The photodegradation efficiency of the composite nanoparticles for MNZ reached 97.18% under optimal experimental conditions. The comparison experiment showed that the photocatalytic efficiency of ternary composites (TiO 2 /Ag 3 PO 4 /g-C 3 N 4 ) was far better than that of pure and binary samples. Cui et al. [22] fabricated g-C 3 N 4 /Ag 3 PO 4 /PVDF photocatalytic porous membranes using a phase inversion method and utilized them for RhB removal. The removal efficiency of the g-C 3 N 4 /Ag 3 PO 4 /PVDF membrane for RhB reached 97%, which was superior to those of the g-C 3 N 4 /PVDF and pure PVDF membranes (85% and 41%, respectively).
In this study, a novel type of hydrophilic PVDF membrane loaded with g-C 3 N 4 /Ag 3 PO 4 / TiO 2 photocatalytic nanocomposite was prepared via blending and phase inversion. The morphology, structure, and photocatalytic activity of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 nanoparticles and composite membrane were characterized and evaluated. A four-stage filtration experiment was used to evaluate the membrane performance for the separation of bovine serum albumin (BSA) and the membrane self-cleaning mechanism. Additionally, the photocatalytic degradation and rejection rate of the composite membrane for RhB in solution were evaluated. The renewability of the composite membrane over repeated RhB photodegradation and adsorption cycles was also evaluated. A mechanism of photocatalytic composite membranes for RhB degradation was proposed.

Synthesis of g-C 3 N 4
g-C 3 N 4 was prepared via melamine pyrolysis using a one-step synthesis method. Melamine (10 g) was added to a porcelain crucible with a lid and calcinated at 550 • C under a heating rate of 5 • C/min in a muffle furnace for 2 h, and then it was cooled to 25 • C. The obtained milky yellow powder was purified with a 1.0 M HCl solution at 150 • C under a heating rate of 3 • C/min for 4 h. Eventually, the fabricated solid was collected through centrifugation, followed by drying, and ground into a uniform powder using an agate mortar and pestle prior to further use [31,33].

Synthesis of g-C 3 N 4 /Ag 3 PO 4 /TiO 2 Nanocomposites
Ag 3 PO 4 was synthesized via precipitation. g-C 3 N 4 /Ag 3 PO 4 /TiO 2 composite nanoparticles were prepared simultaneously as follows: 0.4 g of g-C 3 N 4 , 1.0 g of PVP, and 0.4 g of TiO 2 were added to 80 mL of distilled water (suspension 1). After 15 min of intermittent ultrasonication, the suspension was subjected to magnetic stirring for 30 min. Next, 5.1 g AgNO 3 was added to suspension 1 (mixture 2). Thereafter, 100 mL of a 0.1 mol/L Na 2 HPO 4 solution was added dropwise to mixture 2 using a burette. After the mixture was stirred for 3 h in the dark, a yellow sediment was separated via centrifugation, followed by rinsing three times with water and ethanol, alternately. Lastly, the residue was dried to a constant weight in a vacuum oven at 50 • C.

Preparation of Photocatalytic Composite Membranes
PVDF-based photocatalytic membranes using PVP as the pore former were synthesized by blending g-C 3 N 4 /TiO 2 /Ag 3 PO 4 nanoparticles with PVDF using phase inversion technique. In brief, 1.5 g of g-C 3 N 4 /TiO 2 /Ag 3 PO 4 nanoparticles, 2.0 g of PVP, and 40.0 g of DMAc were added to a 250 mL beaker placed in a 60 • C oil bath, and the reactants were stirred at a constant speed until the mixture became transparent. Thereafter, 7.5 g of PVDF powder was added to the beaker step by step with continuous stirring for 6 h at 60 • C to form a homogeneous and transparent casting mixture. Next, the mixture was placed in a vacuum-drying oven for 12 h at 60 • C to remove air bubbles. After degassing at 20 • C and 70% humidity, the dopant mixture was uniformly spread on a clean glass plate using a coater with 200 µm gap. The glass plate was allowed to rest in ambient air for 30 s, and then it was placed in a coagulation bath at 20 • C. Lastly, the prepared membrane was placed in ultrapure water for 24 h. Figure 1 shows the schematics of the preparation of the nanoparticles and the photocatalytic composite membranes.

Characterization of the g-C3N4/Ag3PO4/TiO2 Nanocomposites
The crystalline structure and phase purity of the photocatalyst powders we lyzed using an Ultima4 (Rigaku) X-ray diffraction (XRD) instrument. The morpho the photocatalyst powders was evaluated using a JSM 7001F (JEOL) SEM appara F-4500 (Hitachi Corp.) photoluminescence (PL) spectrometer was used to investig  The crystalline structure and phase purity of the photocatalyst powders were analyzed using an Ultima4 (Rigaku) X-ray diffraction (XRD) instrument. The morphology of the photocatalyst powders was evaluated using a JSM 7001F (JEOL) SEM apparatus. An F-4500 (Hitachi Corp.) photoluminescence (PL) spectrometer was used to investigate the photochemical properties of the samples.

Membrane Characterization
The roughness and morphology of the membrane surfaces were evaluated using the SEM JSM 7001F (JEOL, Tokyo, Japan) and a Dimension Icon (Bruker, Madison, WI, USA) AFM device, respectively. TGA experiments were conducted using a Q5000IR (TA Instruments, New Castle, DE, USA) system. The interfacial interactions and crystallinity in the membranes were analyzed using an IS10 (Thermo Fisher, Waltham, MA, USA) FTIR spectrometer. The mechanical performance of the membranes was measured using a CMT4204 electronic universal testing machine (XinSanSi, Shanghai, China) at room temperature. The hydrophilic performance of the membranes was estimated by testing their water contact angles with an OCA 40 (Dataphysics, San Jose, CA, USA) goniometer. The sessile drop technique with dynamic contact angle measurement was used, and contact angles were measured for 30 s [34].

Permeability Measurements
The prepared membranes were placed in a dead-end stirred ultrafiltration cup to evaluate their permeability ( Figure 2). The effective filtration area of the ultrafiltration cup was 50.2 cm 2 . Permeability measurements were performed under an operating pressure of 0.1 MPa, as follows: The membrane was pre-pressured with deionized water until the permeate flux and operating pressure were stabilized. The volume of deionized water passing through the membrane was determined within 5 min. The permeate flux was calculated as follows: where J 0 is initial permeate flux of deionized water (L/(m 2 h)), V, A, and t are the volume of deionized water collected within 5 min (L), effective filtration area (m 2 ), and time employed during filtration (h), respectively. The pore size and porosity of the membrane were measured by a weighing method. The operation process and calculation formula were similar to the existing reports [35]. The porosity (ε) was calculated as follows: The pore size and porosity of the membrane were measured by a weighing method. The operation process and calculation formula were similar to the existing reports [35]. The porosity (ε) was calculated as follows: where m d and m w are the weight of the membrane before and after immersion in n-butanol, ρ b is the density of n-butanol, and V d is the dry membrane volume.2.7.2. Membrane reusability After the membranes were pre-pressurized with deionized water, the feed solution was changed to a 0.2 g/L BSA solution, and the stable flux values (J P ) were calculated using Equation (1). The concentration of BSA in the permeate was determined after 30 min, and the rejection of BSA (R) was calculated as follows: where C 0 (mg/L) and C (mg/L) are related to the contaminant concentrations before and after filtration, respectively. Subsequently, the fouled membrane was rinsed repeatedly with deionized water, and its permeate flux (J rw ) was recorded after 10 min. Lastly, the membrane was illuminated using a Xe lamp, and the water flux (J w ) was determined after 30 min. This process was repeated five times. The flux recovery ratio (FRR), total fouling ratio (R t ), reversible fouling ratio (R r ), and irreversible fouling ratio (R ir ) of the membranes were calculated as follows:

Removal of RhB Using the Photocatalytic Membrane
A 10 mg/L solution of RhB, which was used to simulate wastewater, was filtered using an ultrafiltration cup and the photocatalytic membrane under the illumination of a 300 W Xe lamp. During this process, a piece of 420 nm cutoff filter was applied to cover the top of the ultrafiltration cup. The system was kept in dark for 30 min to reach adsorption equilibrium before light illumination. During the visible light illumination, 3 mL of permeate solution was collected from the ultrafiltration cup every 7.5 min and analyzed using an ultraviolet (UV)-visible (Vis) spectrophotometer at 554 nm to determine the concentration of residual RhB. The rejection of RhB was calculated using Equation (3). The time evolution of the concentration of RhB was fitted using the following pseudo-first-order kinetic equation: where k (min −1 ) represents the rate constant and t is the time employed during filtration.

Characterization of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 Nanocomposites
SEM images of g-C 3 N 4 , Ag 3 PO 4 /TiO 2 , and g-C 3 N 4 /Ag 3 PO 4 /TiO 2 are shown in Figure 3. A two-dimensional, micrometer-sized, solid agglomerated structure of g-C 3 N 4 was homogeneously formed (Figure 3a). g-C 3 N 4 presented a typical thin lamellar structure of graphite-phase C 3 N 4 with distinct folds, which conferred on it its large specific surface area. Ag 3 PO 4 comprised 300-500 nm spheres, which connected tightly with the ultra-thin g-C 3 N 4 layers ( Figure 3b). The Ag 3 PO 4 /TiO 2 nanoparticles presented a regular spherical shape similar to that of the pure Ag 3 PO 4 particles that attached to the g-C 3 N 4 sheets (Figure 3c,d).
where k (min −1 ) represents the rate constant and t is the time employed during filtration.

Results and Discussion
3.1. Characterization of the g-C3N4/Ag3PO4/TiO2 Nanocomposites SEM images of g-C3N4, Ag3PO4/TiO2, and g-C3N4/Ag3PO4/TiO2 are shown in Figur 3. A two-dimensional, micrometer-sized, solid agglomerated structure of g-C3N4 was ho mogeneously formed ( Figure 3a). g-C3N4 presented a typical thin lamellar structure o graphite-phase C3N4 with distinct folds, which conferred on it its large specific surfac area. Ag3PO4 comprised 300-500 nm spheres, which connected tightly with the ultra-thin g-C3N4 layers ( Figure 3b). The Ag3PO4/TiO2 nanoparticles presented a regular spherica shape similar to that of the pure Ag3PO4 particles that attached to the g-C3N4 sheets (Figur 3c,d).  Figure 4a shows the XRD patterns of pure g-C3N4, Ag3PO4/TiO2, and g C3N4/Ag3PO4/TiO2. The characteristic peaks in the XRD pattern of pure g-C3N4 at 2θ = 27.8  Figure 4a shows the XRD patterns of pure g-C 3 N 4 , Ag 3 PO 4 /TiO 2 , and g-C 3 N 4 /Ag 3 PO 4 / TiO 2 . The characteristic peaks in the XRD pattern of pure g-C 3 N 4 at 2θ = 27.8 • (strong) and 12.6 • (weak) were attributed to the (002) and (100) planes of g-C 3 N 4 , respectively [36].  (322), respectively, of cubic-phase Ag 3 PO 4 . No impurity peaks were present in the XRD patterns of the samples, and the sharp peaks indicate good crystallization of the samples [33]. It is noteworthy that in the ternary catalyst, the (002) and (100) planes of g-C 3 N 4 disappeared, while the diffraction peak of Ag 3 PO 4 was more prominent, which indicated that the crystal of Ag 3 PO 4 grew. As shown in Figure 4b, it was clear that the PL peak intensity of C 3 N 4 was the highest, that is, the photogenerated electron-hole recombination of its monomer was the easiest. However, the spectral PL intensity of g-C 3 N 4 /Ag 3 PO 4 /TiO 2 composites was significantly suppressed in comparison to g-C 3 N 4 or Ag 3 PO 4 /TiO 2 , which showed that the ternary composite catalyst had the highest photogenerated current-carrying separation efficiency and the lowest recombination rate of the photogenerated charge carriers [37]. That is to say, the composite of g-C 3 N 4 /Ag 3 PO 4 /TiO 2 can effectively improve the diffusion rate and charge mobility of photogenerated charge. posites was significantly suppressed in comparison to g-C3N4 or Ag3PO4/TiO2, which showed that the ternary composite catalyst had the highest photogenerated current-carrying separation efficiency and the lowest recombination rate of the photogenerated charge carriers [37]. That is to say, the composite of g-C3N4/Ag3PO4/TiO2 can effectively improve the diffusion rate and charge mobility of photogenerated charge.  Figure 5a,b show the SEM images of the surface and cross-section of the raw PVDF membrane, respectively, whereas Figure 5c,d show those of the g-C3N4/Ag3PO4/TiO2/PVDF membrane. Surface and cross-sectional images were used to study the dispersion of nanoparticles on the PVDF membrane and probe the thickness and homogeneity of the coating layer. As shown in Figure 5a,c, the raw PVDF membrane displayed a cavity-like structure, and the composite membrane presented a dense top layer supported by a porous layer in the surface image. All cross-sections of membranes showed an asymmetric microscopic porous structure and spongy layers surrounding a finger-like macroscopic cavity. By contrast, the finger-like pore wall of the composite membrane appeared rougher and more porous. The addition of nanoparticles to PVDF promoted the demixing rate by increasing the thermodynamic instability of the membrane. During phase inversion, g-C3N4/Ag3PO4/TiO2 nanoparticles strongly adsorbed water molecules, which provided abundant sites for the infiltration of water molecules and led to pore formation in the surface layer of the membrane.  Figure 5a,c, the raw PVDF membrane displayed a cavitylike structure, and the composite membrane presented a dense top layer supported by a porous layer in the surface image. All cross-sections of membranes showed an asymmetric microscopic porous structure and spongy layers surrounding a finger-like macroscopic cavity. By contrast, the finger-like pore wall of the composite membrane appeared rougher and more porous. The addition of nanoparticles to PVDF promoted the demixing rate by increasing the thermodynamic instability of the membrane. During phase inversion, g-C 3 N 4 /Ag 3 PO 4 /TiO 2 nanoparticles strongly adsorbed water molecules, which provided abundant sites for the infiltration of water molecules and led to pore formation in the surface layer of the membrane.  Figure 6 shows the AFM images of the raw PVDF and g-C3N4/Ag3PO4/TiO2/PVDF membranes. Unlike the PVDF membrane, the g-C3N4/Ag3PO4/TiO2/PVDF membrane presented distinct ridges and valleys on its surface; furthermore, more holes were obtained on the surface of the composite membrane than on the surface of the raw PVDF mem-  Figure 6 shows the AFM images of the raw PVDF and g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membranes. Unlike the PVDF membrane, the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane presented distinct ridges and valleys on its surface; furthermore, more holes were obtained on the surface of the composite membrane than on the surface of the raw PVDF membrane. The higher surface roughness of the composite membrane was attributed to the nodular structures that formed on the top layer of the membrane owing to the aggregation of the photocatalytic material. The R a values of the raw PVDF and g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membranes were 93 and 309 nm, respectively, and the R q values of the raw and composite membranes were 113 and 395 nm, respectively. These data indicated that the surface of the composite membrane was rougher than that of the raw membrane, which provided optimal conditions for adequate contact between photocatalytic materials and contaminants in solutions. This was attributed to the nodular structure generated via the aggregation of photocatalytic material on the membrane surface [8,38]. It was worth noting the results did not match with reference [15] which indicated that the addition of a ternary composite of Ag 3 PO 4 /GO/APTES decreased the roughness of PVDF membranes.

Basic Properties of Membranes
Some basic properties of two types of membranes are shown in Table 1 and Figure 7. The result of the porosity test was consistent with that of the SEM. The thickness of the composite membrane was higher than that of the raw membrane owing to the presence of nanoparticles. Tensile strength measurements were used to investigate the mechanical properties of raw and composite membranes. As shown in Table 1, the tensile stress of the g-C3N4/Ag3PO4/TiO2/PVDF membrane was lower than that of the raw PVDF membrane. That can be explained by the fact that more porous structures could decrease the membrane's mechanical performance. Likewise, agglomeration resulting from excessive incorporation of g-C3N4/Ag3PO4/TiO2 made the dispersion of particles in the polymer matrix nonuniform. When the membrane worked under the action of an external force, the internal stress concentration polarization took place, resulting in the decreased mechanical performance [5].
As shown in Figure 7a, there are no obvious differences in the FTIR spectra between the raw and composite membranes. The obtained results show that relative overlapping

Basic Properties of Membranes
Some basic properties of two types of membranes are shown in Table 1 and Figure 7. The result of the porosity test was consistent with that of the SEM. The thickness of the composite membrane was higher than that of the raw membrane owing to the presence of nanoparticles. Tensile strength measurements were used to investigate the mechanical properties of raw and composite membranes. As shown in Table 1, the tensile stress of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane was lower than that of the raw PVDF membrane. That can be explained by the fact that more porous structures could decrease the membrane's mechanical performance. Likewise, agglomeration resulting from excessive incorporation of g-C 3 N 4 /Ag 3 PO 4 /TiO 2 made the dispersion of particles in the polymer matrix nonuniform. When the membrane worked under the action of an external force, the internal stress concentration polarization took place, resulting in the decreased mechanical performance [5].  Two types of membranes were heated at 15°C/min at temperatures ranging from 25°C to 700 °C. The TGA curves of the raw PVDF and g-C3N4/Ag3PO4/TiO2/PVDF membranes, which were used to evaluate the thermal stability of the membranes (Figure 7b), showed that the TGA curves of the raw and composite membranes were comparable, which was attributed to the decomposition and evaporation of small polymer fragments at high temperatures. The significant mass loss (72.3%) in the range of 306-473 °C was attributed to the thermal decomposition of the PVDF backbone. A distinct weight loss was exhibited in the TGA curve of g-C3N4/Ag3PO4/TiO2/PVDF as the temperature was increased from 317 to 540 °C, and the weight loss rate was ~60.9% at 700°C. The weight loss rate and quantity of the g-C3N4/Ag3PO4/TiO2/PVDF membrane were lower than those of the raw PVDF membrane. This was attributed to the addition of composite particles forming hydrogen or coordination bonds between organic macromolecular interactions, hindering the thermal motion of molecules and the movement of macromolecular chains. Consequently, the energy needed to break the bonds increased, and the thermal stability of the membrane increased [35]. Cacho-Bailo [40] obtained similar results by adding a metal-organic skeleton to polysulfone membranes.
Surface hydrophilicity is a critical factor for the anti-pollution performance of membranes. Generally, the lower the contact angle, the higher the membrane's hydrophilic performance; in other words, the greater the resistance against fouling. The contact angle of the raw PVDF membrane (83.15°) was higher than that of the composite membrane (70.32°) (Figure 7c). This was attributed to the nanoparticles loaded on the top layer of the PVDF membrane. Overall, the addition of g-C3N4/Ag3PO4/TiO2 to PVDF improved membrane hydrophilicity. Generally, the more hydrophilic composites added, the stabler and denser the hydration layer which could rule out more contaminant molecules was As shown in Figure 7a, there are no obvious differences in the FTIR spectra between the raw and composite membranes. The obtained results show that relative overlapping is the dominant phenomenon in these spectra. The distinct peaks at 1440 and 1170 cm −1 in the FTIR spectra of the raw PVDF and g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membranes were attributed to the C-H deformation vibration and the C-F vibration, respectively. Furthermore, the absence of the characteristic absorption peaks of Ag 3 PO 4 in the FTIR spectra of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane was attributed to the relatively low amount of Ag 3 PO 4 or overlap with other peaks [39].
Two types of membranes were heated at 15 • C/min at temperatures ranging from 25 • C to 700 • C. The TGA curves of the raw PVDF and g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membranes, which were used to evaluate the thermal stability of the membranes (Figure 7b), showed that the TGA curves of the raw and composite membranes were comparable, which was attributed to the decomposition and evaporation of small polymer fragments at high temperatures. The significant mass loss (72.3%) in the range of 306-473 • C was attributed to the thermal decomposition of the PVDF backbone. A distinct weight loss was exhibited in the TGA curve of g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF as the temperature was increased from 317 to 540 • C, and the weight loss rate was~60.9% at 700 • C. The weight loss rate and quantity of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane were lower than those of the raw PVDF membrane. This was attributed to the addition of composite particles forming hydrogen or coordination bonds between organic macromolecular interactions, hindering the thermal motion of molecules and the movement of macromolecular chains. Consequently, the energy needed to break the bonds increased, and the thermal stability of the membrane increased [35]. Cacho-Bailo [40] obtained similar results by adding a metal-organic skeleton to polysulfone membranes.
Surface hydrophilicity is a critical factor for the anti-pollution performance of membranes. Generally, the lower the contact angle, the higher the membrane's hydrophilic performance; in other words, the greater the resistance against fouling. The contact angle of the raw PVDF membrane (83.15 • ) was higher than that of the composite membrane (70.32 • ) (Figure 7c). This was attributed to the nanoparticles loaded on the top layer of the PVDF membrane. Overall, the addition of g-C 3 N 4 /Ag 3 PO 4 /TiO 2 to PVDF improved membrane hydrophilicity. Generally, the more hydrophilic composites added, the stabler and denser the hydration layer which could rule out more contaminant molecules was obtained. At the same time, hydrogen bonds formed between g-C 3 N 4 and water molecules, which promoted the diffusion of water molecules in the membrane and weakened the adsorption of contaminants on the membrane surface [41,42]. The hydrophilicity improvement of the membrane also resulted from the hydroxyl radical produced by the composite nanoparticles. In addition, according to previous studies [43,44], surface roughness can improve membrane wettability owing to the formation of composite nanoparticle layers. A few contaminant molecules could accumulate in the "valleys" on the surface accordingly. These results were consistent with the aforementioned SEM and AFM results.

Antifouling Performance of Membranes
The ratios of reversible and irreversible fouling (R ir and R r , respectively) to the total fouling (R t ) of the membranes are critical parameters for evaluating the antifouling performance of membranes (Figure 8a,b). R ir represents the fouling molecules that are tightly anchored to the membrane surface and trapped within its pores, whereas R r primarily represents the foulant that is loosely bound to the membrane surface. Before visible light irradiation, the R ir values of the membranes were higher than the corresponding R r values (Figure 8a,b), indicating that irreversible fouling was the predominant process. The R ir value of the composite membrane decreased considerably, from 61.1% to 8.9%, whereas the R r value increased significantly, from 33.7% to 85.3%. For the composite membrane, a fraction of irreversible fouling was converted into reversible fouling. Notably, the R r values of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane were remarkably greater than those of the raw PVDF membrane under both experimental conditions. Therefore, the addition of photocatalytic nanomaterials to PVDF improved the antifouling ability of the raw PVDF membrane.
In practice, it is necessary to evaluate the antifouling properties of membranes. Therefore, we subjected both membranes to four-stage filtration experiments for five cycles. The FRR values of both membranes decreased gradually after each cycle (Figure 8c), which was attributed to the accumulation of irreversible sums and the adsorption of dirt in the membrane pores. However, the composite nanoparticles promoted the self-cleaning ability of the membrane and lowered the FRR value of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane from 90.1% to 71.0%. In contrast, the FRR value of the raw PVDF membrane decreased from 78.9% to 50.3% after five filtration cycles. This was attributed to the photocatalytic performance and high hydrophilicity of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 nanoparticles on the membrane surface, which could prevent the accumulation of foulant on the membrane surface during filtration. The BSA rejection of the raw and composited membranes was higher than 90% during the five filtration cycles. The addition of g-C 3 N 4 /Ag 3 PO 4 /TiO 2 to PVDF increased the BSA rejection and flux of the PVDF membrane.  The amount of RhB decreased only by 42.5% after 120 min of filtration using the ra PVDF membrane (Figure 9a). RhB removal by the raw PVDF membrane occurred excl sively via absorption. In contrast, the degradation efficiency of the C3N4/Ag3PO4/TiO2/PVDF membrane for RhB was significantly higher (98.1%) under vis ble light irradiation. The kinetic equation −ln(C/C0) = kt was used to evaluate the phot degradation of RhB (Figure 9b). The k value of the raw PVDF membrane (0.00549 min was negligible as it was approximately seven times lower than that of the composite mem brane (0.03618 min −1 ) under the same experimental conditions. These results indicate th the g-C3N4/Ag3PO4/TiO2 coating improved the photocatalytic activity of the PVDF mem brane. The UV-Vis absorption spectra of an RhB solution subjected to photodegradatio using the g-C3N4/Ag3PO4/TiO2/PVDF membrane under visible light irradiation for 90 m are shown in Figure 9c. The maximum absorption peaks at 554 nm declined significantl consistent with the results shown in Figure 9a. The membranes were used to remove RhB for five cycles, and the experimental data provided information about the reusability and stability of the g-C3N4/Ag3PO4/TiO2/PVDF membrane. The removal rates of RhB using the composite membrane and a dead-end stirred ultrafiltration cup under visible light irradiation for five cycles were 98.1%, 97.0%, 96.5%, 95.6%, and 95.3%, given chronologically (Figure 10a,b). After five cycles, the pho- The amount of RhB decreased only by 42.5% after 120 min of filtration using the raw PVDF membrane (Figure 9a). RhB removal by the raw PVDF membrane occurred exclusively via absorption. In contrast, the degradation efficiency of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane for RhB was significantly higher (98.1%) under visible light irradiation. The kinetic equation −ln(C/C 0 ) = kt was used to evaluate the photodegradation of RhB (Figure 9b). The k value of the raw PVDF membrane (0.00549 min −1 ) was negligible as it was approximately seven times lower than that of the composite membrane (0.03618 min −1 ) under the same experimental conditions. These results indicate that the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 coating improved the photocatalytic activity of the PVDF membrane. The UV-Vis absorption spectra of an RhB solution subjected to photodegradation using the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane under visible light irradiation for 90 min are shown in Figure 9c. The maximum absorption peaks at 554 nm declined significantly, consistent with the results shown in Figure 9a.

Photocatalytic Properties of Membranes
The membranes were used to remove RhB for five cycles, and the experimental data provided information about the reusability and stability of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane. The removal rates of RhB using the composite membrane and a dead-end stirred ultrafiltration cup under visible light irradiation for five cycles were 98.1%, 97.0%, 96.5%, 95.6%, and 95.3%, given chronologically (Figure 10a,b). After five cycles, the photocatalytic performance of the composite membrane was still as excellent as in its original state. The low attenuation rate after the five experimental cycles indicates the excellent stability and regeneration capacity of the composite membrane, for the reason that the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 nanoparticles are stably bound in the membrane without loss. The membranes were used to remove RhB for five cycles, and the experimental provided information about the reusability and stability of the g-C3N4/Ag3PO4/TiO2/P membrane. The removal rates of RhB using the composite membrane and a dead stirred ultrafiltration cup under visible light irradiation for five cycles were 98.1%, 97 96.5%, 95.6%, and 95.3%, given chronologically (Figure 10a,b). After five cycles, the p tocatalytic performance of the composite membrane was still as excellent as in its orig state. The low attenuation rate after the five experimental cycles indicates the exce stability and regeneration capacity of the composite membrane, for the reason that th C3N4/Ag3PO4/TiO2 nanoparticles are stably bound in the membrane without loss.

Mechanism of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF Membrane
The photodegradation mechanism of RhB by the C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane is shown in Figure 11. First, RhB molecules were electrostatically adsorbed onto the surface and within the pores of the membrane. Subsequently, the C 3 N 4 /Ag 3 PO 4 /TiO 2 nanoparticles became excited under visible light irradiation and photogenerated electrons and holes and migrated to the surface of the nanocomposite.
According to previous studies [30,45,46], the bandgap energies of C 3 N 4 , Ag 3 PO 4 , and TiO 2 are approximately 2.76, 2.31, and 3.20 eV, respectively. The conduction band (CB) of g-C 3 N 4 (-1.16 eV) is more negative than those of TiO 2 (−0.29 eV) and Ag 3 PO 4 (+0.299 eV); therefore, superoxide radicals (•O 2 − ) can be generated as the photoinduced electrons in the CB of g-C 3 N 4 reduce the absorbed O 2 . Similarly, the electrons in the CB of TiO 2 migrated to Ag 3 PO 4 . Furthermore, the valence band (VB) potential of Ag 3 PO 4 (+2.61 eV) was more positive than those of g-C 3 N 4 (+1.60eV) and the H 2 O/•OH pair (+2.38 eV); therefore, •OH active species can be formed via the oxidation of H 2 O by the holes in the VB of Ag 3 PO 4 [47]. Subsequently, the strong oxidizing radicals (•O 2 − and •OH) and photogenerated holes could thoroughly oxidize RhB into CO 2 and H 2 O. In addition, Ag + ions can be reduced to metallic Ag with the help of generating enough energy, which can be excited by Ag 3 PO 4 and g-C 3 N 4 under light irradiation. The photogenerated electrons in the CB of Ag 3 PO 4 can be transferred to metallic Ag via the constructed Ag bridge and bind to the holes in the VB of g-C 3 N 4 [47,48]. According to previously published papers [49][50][51], the hybrid C 3 N 4 /Ag 3 PO 4 /TiO 2 ternary composite photocatalyst had a higher electron-hole pair separation efficiency and a lower recombination rate than the binary composite photocatalyst. Figure 10. (a) Photocatalytic activity and (b) cycling test and reusability C3N4/Ag3PO4/TiO2/PVDF membranes under visible light irradiation for five cycles (initial R centration of 10 mg/L, pH = 6.4).

Mechanism of the g-C3N4/Ag3PO4/TiO2/PVDF Membrane
The photodegradation mechanism of RhB by the C3N4/Ag3PO4/TiO2/PVDF brane is shown in Figure 11. First, RhB molecules were electrostatically adsorbed o surface and within the pores of the membrane. Subsequently, the C3N4/Ag3PO4/T noparticles became excited under visible light irradiation and photogenerated el and holes and migrated to the surface of the nanocomposite. Figure 11. Photocatalytic degradation process in the g-C3N4/Ag3PO4/TiO2/PVDF membrane According to previous studies [30,45,46], the bandgap energies of C3N4, Ag3P TiO2 are approximately 2.76, 2.31, and 3.20 eV, respectively. The conduction band g-C3N4 (-1.16 eV) is more negative than those of TiO2 (−0.29 eV) and Ag3PO4 (+0.2 therefore, superoxide radicals (•O2 − ) can be generated as the photoinduced elect the CB of g-C3N4 reduce the absorbed O2. Similarly, the electrons in the CB of T grated to Ag3PO4. Furthermore, the valence band (VB) potential of Ag3PO4 (+2.61 e more positive than those of g-C3N4 (+1.60eV) and the H2O/•OH pair (+2.38 eV); th •OH active species can be formed via the oxidation of H2O by the holes in the Ag3PO4 [47]. Subsequently, the strong oxidizing radicals (•O2 − and •OH) and ph erated holes could thoroughly oxidize RhB into CO2 and H2O. In addition, Ag + i be reduced to metallic Ag with the help of generating enough energy, which can be by Ag3PO4 and g-C3N4 under light irradiation. The photogenerated electrons in th Ag3PO4 can be transferred to metallic Ag via the constructed Ag bridge and bind holes in the VB of g-C3N4 [47,48]. According to previously published papers [49-hybrid C3N4/Ag3PO4/TiO2 ternary composite photocatalyst had a higher electro pair separation efficiency and a lower recombination rate than the binary composi tocatalyst.

Conclusions
In summary, g-C 3 N 4 /Ag 3 PO 4 /TiO 2 nanocomposites were prepared and used as precursors to fabricate g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF photocatalyst membranes using the phase inversion method. The g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane presented excellent self-cleaning properties and high RhB removal performance during membrane separation and photodegradation. Therefore, our results demonstrated that the nanocomposites improved the antifouling and self-cleaning performance of the membrane. Furthermore, the visible-light-driven photocatalytic degradation of RhB on the membrane surface by the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 nanoparticles conferred the composite membrane's excellent self-cleaning ability and remarkable regeneration capacity.
The photocatalytic performance of the g-C 3 N 4 /Ag 3 PO 4 /TiO 2 /PVDF membrane has been improved compared to previous studies of TiO 2 -based photocatalytic membranes. Mericq et al. [52] demonstrated that the TiO 2 nanoparticles improved the structure of the membrane as well as its antifouling performance under UV irradiation. Zhang et al. [41] found that Ag/g-C 3 N 4 membranes degraded a maximum of 77% of methyl orange in 100 min. Cui et al. [22] fabricated g-C 3 N 4 /Ag 3 PO 4 /PVDF membranes for RhB removal (97%), which were superior to those of g-C 3 N 4 /PVDF and pure PVDF membranes (85% and 41%, respectively). It is worth noting that they did not take the mechanical properties of the membrane and the exudation of nanoparticles into account. In fact, the incorporation of g-C 3 N 4 /Ag 3 PO 4 /TiO 2 which resulted in the decreased mechanical performance of the membrane cannot be ignored. Hence, the further improvement of the mechanical strength of the membrane is one of the keys to the application of photocatalytic membranes in the future. Overall, this study can expand the application of g-C 3 N 4 /Ag 3 PO 4 /TiO 2 as an advanced membrane material and the use of composite membranes for wastewater treatment. Data Availability Statement: The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Conflicts of Interest:
The authors declare that they have no conflict of interest.