Constructing Positively Charged Thin-Film Nanocomposite Nanofiltration Membranes with Enhanced Performance

Using polyethylenimine (PEI) as the aqueous reactive monomers, a positively charged thin-film nanocomposite (TFN) nanofiltration (NF) membrane with enhanced performance was developed by successfully incorporating graphene oxide (GO) into the active layer. The effects of GO concentrations on the surface roughness, water contact angle, water flux, salt rejection, heavy metal removals, antifouling property, and chlorine resistance of the TFN membranes were evaluated in depth. The addition of 20 ppm GO facilitated the formation of thin, smooth, and hydrophilic nanocomposite active layers. Thus, the TFN-PEI-GO-20 membrane showed the optimal water flux of 70.3 L·m−2·h−1 without a loss of salt rejection, which was 36.8% higher than the thin-film composite (TFC) blank membrane. More importantly, owing to the positively charged surfaces, both the TFC-PEI-blank and TFN-PEI-GO membranes exhibited excellent rejections toward various heavy metal ions including Zn2+, Cd2+, Cu2+, Ni2+, and Pb2+. Additionally, compared with the negatively charged polypiperazine amide NF membrane, both the TFC-PEI-blank and TFN-PEI-GO-20 membranes demonstrated superior antifouling performance toward the cationic surfactants and basic protein due to their hydrophilic, smooth, and positively charged surface. Moreover, the TFN-PEI-GO membranes presented the improved chlorine resistances with the increasing GO concentration.


Introduction
Nowadays, many industries such as mining [1], metallurgy [2], electroplating [3], and microelectronic manufacturing [4] discharge a large amount of industrial wastewater contaminated by heavy metals. It is becoming a huge environmental and public risk due to the persistent and high toxicity of heavy metals to human beings and the environment. Therefore, it is urgent to develop efficient and practical solutions to remove heavy metals from industrial wastewater.

Preparation of NF Membranes
Polysulfone (PSU) UF membrane was lab-made and used as the supporting layer in our previous report [39]. The PSU support was fixed on a homemade polypropylene frame. GO was added into the aqueous solution consisting of 1.0 wt% PEI, 2.0 wt% TEA, and 0.08 wt% DMAP, which was treated by ultrasonic dispersion for 1 h. The obtained aqueous solution was poured on top of the PSU UF membrane and drained off after 1 h. Subsequently, 0.2 w/v% TMC organic solution was further poured on top of the PEI-saturated PSU support and removed after 4 min. Finally, the composite membrane was heated at 60 • C for 20 min. The obtained TFN membranes with incorporating GO were named TFN-PEI-GO-x, where x denotes the aqueous solution containing x ppm of GO in the aqueous solution. The control PEI-based TFC membrane using PEI monomers without addition of GO was named TFC-PEI-blank. The control PIP-based TFC membrane using PIP monomers without addition of GO in our previous study [43] was named TFC-PIP-blank. The membrane ID and the corresponding IP reaction conditions are listed in Table 1. Table 1. Preparation conditions of composite NF membranes.

Characterization of NF Membranes
The cross-sectional and surface morphologies of the fabricated membranes were observed by SEM (LEO-1530, Jena, Germany). The surface roughness was recorded by atomic force microscopy (AFM, MI5500, Agilent, Santa Clara, CA, USA). The chemical groups of the membrane surfaces were characterized by attenuated total reflectance-Fourier transform infrared spectroscopy (ATR-FTIR, Bruker Vertex 70, Fällanden, Switzerland). The chemical compositions of the membrane surfaces were

Performance of NF Membranes
The membrane separation performance was tested using a laboratory cross-flow membrane separation device with 70 cm 2 effective area (FlowMem-0021-HP, FMT, Xiamen, China). The tested membranes were prepressured for 30 min under the transmembrane pressure of 0.5 MPa, the temperature of 25 • C, and the tangential velocity of 1.8 m/s. The salt rejections were first evaluated using four kinds of 2000 ppm salt solutions (Na 2 SO 4 , MgSO 4 , NaCl, and MgCl 2 ). The removals of heavy metal ions were further evaluated using five kinds of 500 ppm heavy metal salt solutions (ZnCl 2 , CdCl 2 , CuCl 2 , NiCl 2 , and Pb(NO 3 ) 2 ). The salt rejections were measured under the full batch mode (both the concentration and permeate stream were recycled into the feed tank) to keep the feed concentration constant. The key parameters of the related ions are listed in Table 2. Table 2. Key parameters of the related ions [46].

Ions
Hydrated Radius (nm) The water flux and rejection were calculated by Equations (1) and (2), respectively: where J is the water flux (L·m −2 ·h −1 , abbreviated as LMH), V is the volume of the permeate (L), A is the effective area of the membrane (m 2 ), and ∆t is the filtration time (h).
where C p and C c refer to the solute concentration in the permeate and in the concentrate, respectively. The salt concentration was calculated according to the standard curve between the concentration and the conductivity measured by a conductivity meter (Mettler Toledo S3, Schweiz, Switzerland). Two kinds of cationic surfactants (CTAC, DTAC) and the basic protein (lysozyme) were used as the model foulants to evaluate the antifouling performance of the selected NF membranes. First, the pure water flux was recorded for 60 min at 0.5 MPa. Then, the pure water was replaced with a 2000 ppm solution containing the model foulant, and the flux of the pollutant solution was recorded for an additional 60 min. After water washing, both the water filtration procedure and pollutants' filtration procedure were repeated.
After immersion in a 2000-ppm NaClO solution, the selected NF membranes were remeasured in terms of water flux and salt rejection. The chlorine resistances were evaluated based on the flux variation and the rejection variation.

Membrane Morphologies
The surface and cross-section morphologies of the fabricated NF membranes were characterized by SEM, as shown in Figure 1. While using the small molecular PIP as aqueous reactive monomers, the TFC-PIP-blank membrane presented a dense surface with the typically discrete nodular structures [43]. While using the macromolecular PEI as aqueous reactive monomers, both the TFC-PEI (Figure 1(a1)) and TFN-PEI-GO membranes (Figure 1(b1,c1)) exhibited the dense and smooth surfaces without nodular structures. Although PEI had a large number of reactive groups (primary amine groups), its reactivity was still lower than PIP due to its large steric hindrance effect and low diffusion rate. PEI (Figure 1(a1)) and TFN-PEI-GO membranes (Figure 1(b1,c1)) exhibited the dense and smooth surfaces without nodular structures. Although PEI had a large number of reactive groups (primary amine groups), its reactivity was still lower than PIP due to its large steric hindrance effect and low diffusion rate.
According to the cross-sectional SEM images (Figure 1(a2,b2,c2)), the thickness of the active layers decreased with the increasing GO concentration. The addition of GO into aqueous solution slowed the PEI diffusion toward the organic phase, resulting from the large steric hindrance of GO nanosheets and the hydrogen bonding interaction between GO and PEI. Namely, the IP reactivity between PEI and TMC was reduced by the introduced GO, which accordingly decreased the thickness of the active layer. The thickness of the active layer slightly decreased from 256.8 nm (TFC-PEI-blank, Figure 1(a2)) to 223.3 nm (TFN-PEI-GO-40, Figure 1(c2)). In this study, we used two approaches to promote more macromolecular PEI monomers to take part in the IP reaction. On the one hand, the impregnated time of aqueous solution was extended to 1 h and the IP reaction time was extended to 4 min. On the other hand, DMAP as an effective phase transfer catalyst was added into aqueous solutions to eliminate the steric hindrance of acyl transfer reaction and promote the IP reactions [47]. The introduction of DMAP into the aqueous solution significantly increased the degree of the cross-linking reaction between highly steric-hindered PEI and TMC [48]. As a result, both the TFC-PEI-blank and the TFN-PEI-GO membranes showed the thicker active layer compared with the TFC-PIP-blank membrane [43]. The surface roughness of the fabricated NF membranes was further characterized by AFM. The key parameters of surface roughness include the mean roughness (Ra), the root mean square of Z data (Rq), and the mean difference between the five highest peaks and five lowest valleys (Rz). According to Figure 2 and Table 3, the surface roughness of the TFN-PEI-GO membranes was According to the cross-sectional SEM images (Figure 1(a2,b2,c2)), the thickness of the active layers decreased with the increasing GO concentration. The addition of GO into aqueous solution slowed the PEI diffusion toward the organic phase, resulting from the large steric hindrance of GO nanosheets and the hydrogen bonding interaction between GO and PEI. Namely, the IP reactivity between PEI and TMC was reduced by the introduced GO, which accordingly decreased the thickness of the active layer. The thickness of the active layer slightly decreased from 256.8 nm (TFC-PEI-blank, Figure 1(a2)) to 223.3 nm (TFN-PEI-GO-40, Figure 1(c2)). In this study, we used two approaches to promote more macromolecular PEI monomers to take part in the IP reaction. On the one hand, the impregnated time of aqueous solution was extended to 1 h and the IP reaction time was extended to 4 min. On the other hand, DMAP as an effective phase transfer catalyst was added into aqueous solutions to eliminate the steric hindrance of acyl transfer reaction and promote the IP reactions [47]. The introduction of DMAP into the aqueous solution significantly increased the degree of the cross-linking reaction between highly steric-hindered PEI and TMC [48]. As a result, both the TFC-PEI-blank and the TFN-PEI-GO membranes showed the thicker active layer compared with the TFC-PIP-blank membrane [43].
The surface roughness of the fabricated NF membranes was further characterized by AFM. The key parameters of surface roughness include the mean roughness (Ra), the root mean square of Z data (Rq), and the mean difference between the five highest peaks and five lowest valleys (Rz). According to Figure 2 and Table 3, the surface roughness of the TFN-PEI-GO membranes was dramatically smaller than the TFC-PIP-blank and TFN-PIP membranes in our previous reports [43], which was consistent with the SEM results. Meanwhile, the surface roughness of the TFN-PEI-GO membranes gradually decreased with the increase of GO concentration, which was also ascribed to the reduced diffusion rate of PEI inhibited by the steric hindrances and hydrogen bonding interactions from GO on PEI. The retarded IP process led to the formation of a smoother active layer. In addition, it was found that the introduction of GO resulted in the formation of the wrinkled and patterned surfaces, which was probably caused by the reduced diffusion rate of PEI monomers. This result was similar to the findings from Zhu's research group [49]. However, they thought that the wrinkled and patterned membrane surfaces derived from the wrinkled GO nanosheets with the appropriate lateral size as the templates [49]. With the increase of the GO concentration, more and more wrinkled and patterned structures appeared on membrane surfaces, which was expected to increase the effective filtration area and, accordingly, improve membrane flux.
Polymers 2020, 12, x FOR PEER REVIEW 6 of 17 dramatically smaller than the TFC-PIP-blank and TFN-PIP membranes in our previous reports [43], which was consistent with the SEM results. Meanwhile, the surface roughness of the TFN-PEI-GO membranes gradually decreased with the increase of GO concentration, which was also ascribed to the reduced diffusion rate of PEI inhibited by the steric hindrances and hydrogen bonding interactions from GO on PEI. The retarded IP process led to the formation of a smoother active layer. In addition, it was found that the introduction of GO resulted in the formation of the wrinkled and patterned surfaces, which was probably caused by the reduced diffusion rate of PEI monomers. This result was similar to the findings from Zhu's research group [49]. However, they thought that the wrinkled and patterned membrane surfaces derived from the wrinkled GO nanosheets with the appropriate lateral size as the templates [49]. With the increase of the GO concentration, more and more wrinkled and patterned structures appeared on membrane surfaces, which was expected to increase the effective filtration area and, accordingly, improve membrane flux.

Membrane Surface Properties
The surface chemical groups of the fabricated NF membranes were characterized by ATR-FTIR. According to Figure 3, the characteristic peaks at 1688 and 1394 cm −1 were assigned to the typical C=O stretching vibrations of the amide-I band and the stretching vibration of C-N in the amide [43,50], respectively. Meanwhile, the peaks at around 1222 cm −1 and 1320 cm −1 were assigned to the bending vibration of N-H [51]. This indicated the successful formation of the polyamide active layers on the PSU support. Additionally, the absorption peak intensities of the TFN-PEI-GO membranes were slightly lower than that of the TFC-PEI-blank membrane, which was probably due to the embedded GO nanosheets and the relatively thinner active layer. ers 2020, 12, x FOR PEER REVIEW The surface chemical groups of the fabricated NF membranes were characterized by ATRrding to Figure 3, the characteristic peaks at 1688 and 1394 cm −1 were assigned to the typica hing vibrations of the amide-I band and the stretching vibration of C-N in the amide [4 ctively. Meanwhile, the peaks at around 1222 cm −1 and 1320 cm −1 were assigned to the ben tion of N-H [51]. This indicated the successful formation of the polyamide active layers o support. Additionally, the absorption peak intensities of the TFN-PEI-GO membranes tly lower than that of the TFC-PEI-blank membrane, which was probably due to the embe anosheets and the relatively thinner active layer. XPS had higher detective sensitivity than ATR-FTIR, which was further employed to an hemical compositions of the membrane surfaces. As shown in Figure 4, the peaks at 284.5, 287.3 eV were assigned to C-C/C=C, C-O/C-N, and C=O functional groups [52], respect h also suggested that the polyamide active layer was successfully generated after IP pro over, two new peaks at 286.7 and 288.3 eV attributed to C-O-C and O-C=O were found fo -PEI-GO-20 and TFN-PEI-GO-40 membranes. These epoxy and carboxyl groups derived indicating the successful incorporation of GO into the active layer after the IP process. XPS had higher detective sensitivity than ATR-FTIR, which was further employed to analyze the chemical compositions of the membrane surfaces. As shown in Figure 4, the peaks at 284.5, 285.6, and 287.3 eV were assigned to C-C/C=C, C-O/C-N, and C=O functional groups [52], respectively, which also suggested that the polyamide active layer was successfully generated after IP process. Moreover, two new peaks at 286.7 and 288.3 eV attributed to C-O-C and O-C=O were found for the TFN-PEI-GO-20 and TFN-PEI-GO-40 membranes. These epoxy and carboxyl groups derived from GO, indicating the successful incorporation of GO into the active layer after the IP process. the chemical compositions of the membrane surfaces. As shown in Figure 4, the peaks at 284.5, 285.6, and 287.3 eV were assigned to C-C/C=C, C-O/C-N, and C=O functional groups [52], respectively, which also suggested that the polyamide active layer was successfully generated after IP process. Moreover, two new peaks at 286.7 and 288.3 eV attributed to C-O-C and O-C=O were found for the TFN-PEI-GO-20 and TFN-PEI-GO-40 membranes. These epoxy and carboxyl groups derived from GO, indicating the successful incorporation of GO into the active layer after the IP process. The water contact angle (WCA) is commonly used to evaluate membrane hydrophilicity. The relationship between the GO concentration and the WCA of the composite NF membranes is depicted in Figure 5. With the increase of the GO concentration, the WCA of the TFN-PEI-GO membranes gradually decreased. This indicated that the membrane hydrophilicity was enhanced with the increase of GO concentration. GO nanosheets contain a large number of oxygen-containing groups such as hydroxyl and carboxyl groups, which contributed to improving the membrane hydrophilicity. When the GO concentration increased from 0 to 40 ppm, the WCA of the fabricated NF membranes significantly decreased from 42.0° to 25.1°.

Water Flux and Salt Rejection
The water fluxes and salt rejections of composite NF membranes were measured and are shown in Figure 6. Notably, there was a critical GO concentration for water flux. With the addition of 20 ppm GO into aqueous solution, the TFN-PEI-GO-20 membrane exhibited the highest water flux of 70.3 LMH, which was 36.8% higher than the TFC-PEI-blank membrane. However, the water flux The water contact angle (WCA) is commonly used to evaluate membrane hydrophilicity. The relationship between the GO concentration and the WCA of the composite NF membranes is depicted in Figure 5. With the increase of the GO concentration, the WCA of the TFN-PEI-GO membranes gradually decreased. This indicated that the membrane hydrophilicity was enhanced with the increase of GO concentration. GO nanosheets contain a large number of oxygen-containing groups such as hydroxyl and carboxyl groups, which contributed to improving the membrane hydrophilicity. The water contact angle (WCA) is commonly used to evaluate membrane hydrophilicity. The relationship between the GO concentration and the WCA of the composite NF membranes is depicted in Figure 5. With the increase of the GO concentration, the WCA of the TFN-PEI-GO membranes gradually decreased. This indicated that the membrane hydrophilicity was enhanced with the increase of GO concentration. GO nanosheets contain a large number of oxygen-containing groups such as hydroxyl and carboxyl groups, which contributed to improving the membrane hydrophilicity. When the GO concentration increased from 0 to 40 ppm, the WCA of the fabricated NF membranes significantly decreased from 42.0° to 25.1°.

Water Flux and Salt Rejection
The water fluxes and salt rejections of composite NF membranes were measured and are shown in Figure 6. Notably, there was a critical GO concentration for water flux. With the addition of 20 ppm GO into aqueous solution, the TFN-PEI-GO-20 membrane exhibited the highest water flux of 70.3 LMH, which was 36.8% higher than the TFC-PEI-blank membrane. However, the water flux turned to decrease when the GO concentration was beyond the critical concentration. It was speculated that the GO agglomeration at high concentration led to the increased membrane filtration resistance and the decreased flux. With the help of DMAP catalyst, the TFN-PEI-GO membranes demonstrated the relatively high and stable Na 2 SO 4 rejections, above 91%, regardless of the variation of GO concentration. This suggested that the introduction of GO improved water flux without sacrificing the salt rejection. The increase of water flux benefitted from: (1) The introduction of GO generated a thinner active layer, which correspondingly alleviated the membrane filtration resistance; (2) the introduction of GO led to the formation of the wrinkled membrane surface, which increased the effective membrane filtration area; (3) the hydrophilicity of the active layer was enhanced with embedding GO; and (4) GO provided additional transmission channels with less mass transfer resistance to water molecules.  In order to explore in depth the separation mechanisms of the positively charged PEI-based NF membranes, the rejections of four kinds of salts were measured and are illustrated in Figure 7. The mass concentration (500 ppm) of different salts was equal and relatively low in this study. Thus, the influence of molar concentration variation on the salt rejection could be ignored for the dilute salt solutions. Because macromolecular PEI is a cationic polyelectrolyte with branched chains, containing a large number of primary amine groups, secondary amine groups, and tertiary amine groups, the fabricated PEI-based NF membranes presented the positively charged surfaces. The order of salt rejections was MgCl2 ≈ MgSO4 > Na2SO4 > NaCl, which matched very well with the characteristics of the positively charged NF membrane [33]. This order was completely different from the negatively charged PIP-based NF membranes. The salt rejections of the TFN-PEI-GO membranes were similar to those of the TFC-PEI-blank membrane, suggesting that the loading of GO at a low concentration did not change the characteristics of the TFN-PEI-GO membranes. In order to explore in depth the separation mechanisms of the positively charged PEI-based NF membranes, the rejections of four kinds of salts were measured and are illustrated in Figure 7. The mass concentration (500 ppm) of different salts was equal and relatively low in this study. Thus, th influence of molar concentration variation on the salt rejection could be ignored for the dilute salt solutions. Because macromolecular PEI is a cationic polyelectrolyte with branched chains, containing a large number of primary amine groups, secondary amine groups, and tertiary amine groups, the fabricated PEI-based NF membranes presented the positively charged surfaces. The order of salt rejections was MgCl 2 ≈ MgSO 4 > Na 2 SO 4 > NaCl, which matched very well with the characteristics of the positively charged NF membrane [33]. This order was completely different from the negatively charged PIP-based NF membranes. The salt rejections of the TFN-PEI-GO membranes were similar to those of the TFC-PEI-blank membrane, suggesting that the loading of GO at a low concentration did not change the characteristics of the TFN-PEI-GO membranes.

Removal of Heavy Metals
The TFC-PEI-blank, TFN-PEI-GO-20, and TFN-PEI-GO-40 were selected to investigat aration performance of removing heavy metals. According to Figure 8, the select branes showed high rejections, above 91%, toward five kinds of heavy metal salts. The re b(NO3)2 was 91% and the rejections of the other heavy metal salts ranged from 94% to 97% icated that the positively charged NF membranes fabricated from PEI monomers were hel ance the removals of heavy metals. Because both the TFC-PEI-blank membrane and the TF membranes were positively charged, these membranes presented the same sequen oving heavy metals: ZnCl2 > CdCl2 ≈ CuCl2 > NiCl2 > Pb(NO3)2. This result was consisten hydration radius sequence of various heavy metal ions, as listed in Table 2: Zn 2+ > Cd 2+ > + > Pb 2+ . Five kinds of heavy metal ions belong to divalent cations. Thus, the rejections dep the hydration radius based on size exclusion effect. The larger hydration radius of heavy contributed to the high rejection of heavy metal ion. It is well known that the NF separation mechanisms include the Donnan exclusion effect and the size exclusion effect. Therefore, the positively charged TFN-PEI-GO membranes generated the stronger electrostatic repulsion toward the divalent Mg 2+ than the monovalent Na + . As a result, the rejection of divalent Mg 2+ was greater than that of monovalent Na + . For the salts having the same divalent Mg 2+ , the positively charged TFN-PEI-GO membranes also produced the stronger electrostatic attraction toward the divalent SO 4 2− than the monovalent Cl − . The strong adsorption of SO 4 2− significantly weakened the positively charged properties of membrane surfaces. As a result, the rejection of MgSO 4 was lower than that of MgCl 2 . However, for the salts having the same monovalent Na + , the Na 2 SO 4 rejection was much higher than NaCl rejection because SO 4 2− has a greater hydration radius than Cl − , which indicated that the size exclusion effect was more dominant than the Donnan exclusion effect.

Removal of Heavy Metals
The TFC-PEI-blank, TFN-PEI-GO-20, and TFN-PEI-GO-40 were selected to investigate their separation performance of removing heavy metals. According to Figure 8, the selected NF membranes showed high rejections, above 91%, toward five kinds of heavy metal salts. The rejection of Pb(NO 3 ) 2 was 91% and the rejections of the other heavy metal salts ranged from 94% to 97%. This indicated that the positively charged NF membranes fabricated from PEI monomers were helpful to enhance the removals of heavy metals. Because both the TFC-PEI-blank membrane and the TFN-PEI-GO membranes were positively charged, these membranes presented the same sequence for removing heavy metals: ZnCl 2 > CdCl 2 ≈ CuCl 2 > NiCl 2 > Pb(NO 3 ) 2 . This result was consistent with the hydration radius sequence of various heavy metal ions, as listed in Table 2: Zn 2+ > Cd 2+ > Cu 2+ > Ni 2+ > Pb 2+ . Five kinds of heavy metal ions belong to divalent cations. Thus, the rejections depended on the hydration radius based on size exclusion effect. The larger hydration radius of heavy metal ion contributed to the high rejection of heavy metal ion.
GO membranes were positively charged, these membranes presented the same sequence for removing heavy metals: ZnCl2 > CdCl2 ≈ CuCl2 > NiCl2 > Pb(NO3)2. This result was consistent with the hydration radius sequence of various heavy metal ions, as listed in Table 2: Zn 2+ > Cd 2+ > Cu 2+ > Ni 2+ > Pb 2+ . Five kinds of heavy metal ions belong to divalent cations. Thus, the rejections depended on the hydration radius based on size exclusion effect. The larger hydration radius of heavy metal ion contributed to the high rejection of heavy metal ion. The influences of different aqueous reactive monomers (PIP vs. PEI) on the copper removal of the TFC NF membranes were further investigated. As shown in Figure 9, regardless of CuSO4 or The influences of different aqueous reactive monomers (PIP vs. PEI) on the copper removal of the TFC NF membranes were further investigated. As shown in Figure 9, regardless of CuSO 4 or CuCl 2 , the Cu 2+ rejection of the positively charged TFC-PEI-blank membrane was remarkably higher than that of the negatively charged TFC-PIP-blank membrane. For the TFC-PEI-blank membrane, the rejections of both CuSO 4 and CuCl 2 were above 94%, and the CuSO 4 rejection was slightly lower than the CuCl 2 rejection because the positively charged TFC-PEI-blank membrane generated the stronger electrostatic attraction toward the divalent SO 4 2− than the monovalent Cl − . This suggested that the Donnan exclusion effect was predominant compared with the size exclusion effect. On the contrary, the negatively charged TFC-PIP-blank membrane presented the stronger electrostatic repulsion toward the divalent SO 4 2− than the monovalent Cl − . Thus, both the Donnan exclusion and size exclusion preferred to increase the CuSO 4 rejection rather than the CuCl 2 rejection for the TFC-PIP-blank membrane.
Polymers 2020, 12, x FOR PEER REVIEW 11 of 17 CuCl2, the Cu 2+ rejection of the positively charged TFC-PEI-blank membrane was remarkably higher than that of the negatively charged TFC-PIP-blank membrane. For the TFC-PEI-blank membrane, the rejections of both CuSO4 and CuCl2 were above 94%, and the CuSO4 rejection was slightly lower than the CuCl2 rejection because the positively charged TFC-PEI-blank membrane generated the stronger electrostatic attraction toward the divalent SO4 2− than the monovalent Cl − . This suggested that the Donnan exclusion effect was predominant compared with the size exclusion effect. On the contrary, the negatively charged TFC-PIP-blank membrane presented the stronger electrostatic repulsion toward the divalent SO4 2− than the monovalent Cl − . Thus, both the Donnan exclusion and size exclusion preferred to increase the CuSO4 rejection rather than the CuCl2 rejection for the TFC-PIPblank membrane.

Antifouling Performance
Two kinds of cationic surfactant (CTAC and DTAC) and lysozyme (LYZ) were used to evaluate the antifouling performance of the fabricated NF membranes. According to Figure 10, regardless of CTAC, DTAC, or LYZ, the normalized water fluxes of the TFC-PEI-blank and TFN-PEI-GO-20 membranes were significantly higher than those of the TFC-PIP-blank membrane.
After fouling by CTAC and water washing, the normalized water fluxes of the TFC-PEI-blank and TFN-PEI-GO-20 membranes reached 91.1% and 89.7%, respectively. However, the normalized water flux of the TFC-PIP-blank membrane was only 76%. After fouling by DTAC, the normalized water fluxes of the TFC-PEI-blank and TFN-PEI-GO-20 membranes reached 85.9% and 85.3%, respectively. However, the normalized water flux of TFC-PIP-blank was only 75.4%. LYZ is positively charged under neutral testing conditions due to its isoelectric point of 10.7. After fouling by LYZ, the

Antifouling Performance
Two kinds of cationic surfactant (CTAC and DTAC) and lysozyme (LYZ) were used to evaluate the antifouling performance of the fabricated NF membranes. According to Figure 10, regardless of CTAC, DTAC, or LYZ, the normalized water fluxes of the TFC-PEI-blank and TFN-PEI-GO-20 membranes were significantly higher than those of the TFC-PIP-blank membrane. than the TFC-PEI-blank membrane. This was caused by the higher flux of the TFN-PEI-GO-20 membrane than that of the TFC-PEI-blank membrane. It means that more foulants were adsorbed on the membrane surface and the concentration polarization was intensified due to the high membrane flux, which would aggravate membrane fouling.  After fouling by CTAC and water washing, the normalized water fluxes of the TFC-PEI-blank and TFN-PEI-GO-20 membranes reached 91.1% and 89.7%, respectively. However, the normalized water flux of the TFC-PIP-blank membrane was only 76%. After fouling by DTAC, the normalized water fluxes of the TFC-PEI-blank and TFN-PEI-GO-20 membranes reached 85.9% and 85.3%, respectively. However, the normalized water flux of TFC-PIP-blank was only 75.4%. LYZ is positively charged under neutral testing conditions due to its isoelectric point of 10.7. After fouling by LYZ, the normalized water fluxes of the TFC-PEI-blank and TFN-PEI-GO-20 membranes reached 85.0% and 83.1%, respectively, while the normalized water flux of the TFC-PIP-blank membrane was only 62.8%. Therefore, for the positively charged foulants, the above results indicated that the positively charged TFC-PEI-blank and TFN-PEI-GO-20 membranes had superior antifouling performance over the negatively charged TFC-PIP-blank membrane.
On the one hand, both the TFC-PEI-blank and TFN-PEI-GO-20 membranes possessed a more hydrophilic and smoother surface than the TFC-PIP-blank membrane, which was helpful to reduce membrane fouling. On the other hand, the positively charged TFC-PEI-blank and TFN-PEI-GO-20 membranes generated the stronger electrostatic repulsion toward the cationic surfactants and the positively charged LYZ, which further alleviated the membrane fouling. On the contrary, the negatively charged TFC-PIP-blank membrane had the strong electrostatic attraction toward the cationic surfactants and the positively charged LYZ, which resulted in the severe membrane fouling.
Although the TFN-PEI-GO-20 membrane was more hydrophilic compared with the TFC-PEI-blank membrane, the TFN-PEI-GO-20 membrane presented a slightly more severe membrane fouling than the TFC-PEI-blank membrane. This was caused by the higher flux of the TFN-PEI-GO-20 membrane than that of the TFC-PEI-blank membrane. It means that more foulants were adsorbed on the membrane surface and the concentration polarization was intensified due to the high membrane flux, which would aggravate membrane fouling.

Chlorine Resistance
In view of the degradation mechanism of the PA TFC membranes, the PEI-based PA membranes are vulnerable to chlorine attack due to a large amount of end amine groups for N-chlorination [35]. Therefore, an important objective of the present study was to enhance the chlorine resistance of the PEI-based PA membranes by incorporating GO nanosheets into the active layer.
The normalized water flux and the normalized Na 2 SO 4 rejection were used to evaluate the chlorine resistances of the fabricated NF membranes. As shown in Figure 11, the normalized water flux significantly increased with the increasing chlorine exposure time, and the variation of the normalized water flux was TFC-PEI-blank > TFN-PEI-GO-20 > TFN-PEI-GO-40. Meanwhile, the normalized Na 2 SO 4 rejection decreased with the increasing chlorine exposure time and displayed the opposite order: TFC-PEI-blank < TFN-PEI-GO-20 < TFN-PEI-GO-40. These results indicated that the incorporation of GO in the active layer contributed to improving the chlorine resistance of the PEI-based PA NF membranes. On the one hand, the skeleton structure of GO could absorb chlorine radicals to form O-Cl bonds, accordingly reducing the chlorine radical attack toward the PA layer [53]. On the other hand, the embedded GO nanosheets provided additional protection for the underlying PA, ascribed to their large specific surface area and the hydrogen bonding between GO and PA [54]. Thus, the chlorine resistance of the PEI-based PA NF membrane increased with the increase of GO concentration. radicals to form O-Cl bonds, accordingly reducing the chlorine radical attack toward the PA layer [53]. On the other hand, the embedded GO nanosheets provided additional protection for the underlying PA, ascribed to their large specific surface area and the hydrogen bonding between GO and PA [54]. Thus, the chlorine resistance of the PEI-based PA NF membrane increased with the increase of GO concentration.

Conclusions
To construct the positively charged NF membrane with enhanced performance, the PEI-based TFN NF membrane using PEI as the aqueous reactive monomers was fabricated by embedding GO into the active layer via the IP process. After incorporating an appropriate amount of GO into the PEI-based PA active layer, the optimal TFN-PEI-GO-20 membrane showed the maximal water flux of 70.3 L·m −2 ·h −1 without sacrificing salt rejection, which was 36.8% higher than the TFC-PEI-blank membrane. Meanwhile, the positively charged TFC-PEI-blank and TFN-PEI-GO membranes presented superior rejections for heavy metal ions (Zn 2+ , Cd 2+ , Cu 2+ , Ni 2+ , and Pb 2+ ) due to the synergistic effects of Donnan exclusion and size exclusion. Furthermore, both the TFC-PEI-blank and TFN-PEI-GO-20 membranes exhibited excellent antifouling performance toward the cationic surfactants and basic protein (LYZ), owing to the hydrophilic, smooth, and positively charged surface. Finally, the TFN-PEI-GO membranes presented the enhanced chlorine resistances due to the extra protection from GO. Therefore, the positively charged PEI-based TFN NF membrane embedded with GO shows a promising potential for wastewater purification from heavy metal ions.