*Article* **Valorization of Goat Cheese Whey through an Integrated Process of Ultrafiltration and Nanofiltration**

**Antónia Macedo 1,2, \*, David Azedo 1 , Elizabeth Duarte <sup>2</sup> and Carlos Pereira 3**


**Abstract:** Goat cheese whey is a co-product that comes from goat cheese manufacture. Due to its high organic load, adequate treatment is necessary before its disposal. Additionally, the recent growing interest in caprine products, attributed to their specific nutritional and nutraceutical characteristics, such as the lower allergenicity of their proteins and higher content of oligosaccharides, compared with bovine products, made the recovery of goat cheese whey a challenge. In this study, an integrated process for the recovery of sweet goat whey components was carried out. It includes filtration, centrifugation and pasteurization, followed by sequential membrane processes, ultrafiltration/dilution, nanofiltration of ultrafiltration permeates in dilution mode and the concentration/dilution of nanofiltration retentates. Ultrafiltration was performed with membranes of 10 and 1 kDa. Membranes of 10 kDa have higher permeate fluxes and, in a single stage of dilution, allowed for better protein retention and higher lactose purity, with a separation factor of 14. The concentration of lactose by nanofiltration/dilution led to the retention of almost all the lactose in retentates and to a final permeate, whose application in cheese dairy plants will allow for the total recovery of whey. The application of this integrated process in small- or medium-sized goat cheese dairies can represent an important contribution to their sustainability.

**Keywords:** goat cheese whey; ultrafiltration; nanofiltration; dilution mode

#### **1. Introduction**

Goat cheese whey is a liquid co-product of goat cheese manufacture. It retains about 55% of the nutrients found in milk, including lactose, soluble proteins, bioactive peptides, lipids, minerals and vitamins [1]. In comparison with bovine and sheep cheese whey, goat cheese whey contains a higher concentration of oligosaccharides, namely sialic acid, that seems to promote the development of infants' brain [2]. Besides, it is rich in nonprotein nitrogen compounds, namely nucleotides and free amino acids, making it suitable for baby food or children with a cow's milk allergy [3,4]. Therefore, there is new and growing interest in producing caprine products due to the nutraceutical and hypoallergenic properties of caprine milk compared to cow's milk [5], which contributes to the increasing volumes of goat whey produced. Despite its nutritional and nutraceutical value, goat cheese whey is usually treated as waste, deposited in septic tanks or partially mixed with the wastewaters coming from cheese washing and the cleaning operations of equipment and from the cheese dairy; it is then delivered to wastewater plants. Its high values of chemical oxygen demand (COD) and biological oxygen demand (BOD5), about 50–120 g L <sup>−</sup><sup>1</sup> and 27–60 g L −1 , respectively [6,7], can lead to the decline in treatment efficiency, transforming this co-product into one of the main environmental problems that the dairy industry has faced for decades. So, its reuse has the advantage of generating value-added products while mitigating its negative impacts on the environment.

**Citation:** Macedo, A.; Azedo, D.; Duarte, E.; Pereira, C. Valorization of Goat Cheese Whey through an Integrated Process of Ultrafiltration and Nanofiltration. *Membranes* **2021**, *11*, 477. https://doi.org/10.3390/ membranes11070477

Academic Editor: Mohammad Peydayesh

Received: 20 May 2021 Accepted: 23 June 2021 Published: 28 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Membrane technologies are the most common separation processes used for recovering valuable fractions from cheese whey. Ultrafiltration (UF) is well-established in the food industry to produce whey protein concentrates (WPC) from bovine cheese whey [8], using membranes with a cut-off equal to or higher than 10 kDa [9,10]. These membranes allow the retention of the predominant whey proteins (β-lactoglobulin (β-Lg), α-lactalbumin (α-La), immunoglobulin (Ig), serum albumin (SA)) and other minor proteins, such as lactoferrin (LF) and lactoperoxidase (LP), while lysozyme, glycomacropeptide (GMP) or bioactive peptides, with lower molecular weights or sizes [11], can mostly permeate through the membranes, together with lactose, other sugars and minerals. To increase the permeation of these smaller compounds, thus improving the separation of protein and lactose fractions, the dilution mode of UF retentates is also applied to obtain purified streams [12]. The recovery of permeates of ultrafiltration can be carried out by nanofiltration (NF), followed by the dilution mode of retentates [13]. NF, due to its specific characteristics, especially for the separation of the smaller solutes present in UF permeates, can be a valuable tool, retaining mostly lactose, the main component responsible for the environmental damage of whey [14]; bioactive peptides; oligosaccharides, free amino acids; and bivalent ions, originating a final permeate with a very low organic load [15,16]. The use of NF in dilution mode can enhance the permeation of salts, especially sodium and chloride, which can be high in UF permeates because of salt addition to milk during cheese manufacture. Therefore, the application of NF to UF permeates, in dilution mode, contributes to the reduction in osmotic pressure, a major drawback in these processes [17]. In addition, the purification of retentates may allow its application in food or pharmaceutical industries, due to their nutritional and nutraceutical characteristics, as cited above. This investigation aims to contribute to improvement in the sustainability of the artisanal production of goat cheese, carried out mainly in small- and medium-sized cheese dairy plants. This study has an innovative character in that it presents an integrated membrane process for the total recovery of a co-product with a very complex composition, such as goat cheese whey. Therefore, the main objectives are:


#### **2. Materials and Methods**

#### *2.1. Sampling and Pretreatment of Goat Cheese Whey*

Six samples of goat cheese whey (GCW) were collected in the same cheese dairy, located close to Beja, Portugal. The GCW was produced during the cheese-making process, which involved the following steps: milk pasteurization; filtration of the milk; addition of salt to the milk cheese; rennet coagulation; syneresis, where whey was released; the new addition of salt to the curd; pressing and packing, before expedition. A volume of about 10 L of each sample was collected and carried out to our laboratory, keeping them refrigerated in ice during transportation. After arriving, samples were filtrated two times through cotton cloths, like those used in a traditional cheese dairy, to remove suspended solids and casein fines. After that, samples were skimmed in an Elecrem SAS, Fresnes, France, centrifuge, at a temperature of about 35 ◦C to remove most of the lipids and some minor residues of casein and bacteria. Since these samples have a high concentration of lipids, the reduction of their concentration is crucial to avoid membrane fouling. Finally, whey samples were subjected to a low pasteurization process, at 65 ◦C, for 30 min. When it

was not possible to process the samples in the same day, they were immediately preserved at about 3 ◦C until the next day.

#### *2.2. Permeation Experiments*

All the permeation experiments were carried out in a plane-and-frame module, Lab Unit M20, from Alfa Laval, Navskov, Denmark. This filtration rig is a commercial installation with a membrane area ranging from a minimum of 0.036 m<sup>2</sup> , which corresponds to two membrane sheets, to 0.72 m<sup>2</sup> , the maximum area. The membranes were grouped in pairs, resting on the top and bottom of the same support plate. The plates were separated by spacers that acted as feed chambers 0.5 mm in height and that were divided into 30 channels to increase tangential velocity and thus, promote mass transfer in the adjacent layer near the membrane surface. In each support plate, there were individual collectors for the permeate. Support plates and spacers were made of polysulfone, and the module frame was made of stainless steel. The unit had a hydraulic system that allowed the flat plate module to be compressed, making it perfectly watertight.

These experiments included the UF of the pretreated goat cheese whey to remove residual protein and fat and to obtain a protein-containing retentate and a lactose-rich permeate. For a better recovery of lactose in permeates, dilution mode in ultrafiltration (DUF) of the retentates was also carried out. After that, all of the permeates resulting from UF and DUF were mixed and subjected to NF to recover the lactose fraction. For purifying this lactose-rich retentate, which may increase the possibilities for its use in various industries, dilution mode in NF was also applied to the retentates of the NF process (Figure 1).

**Figure 1.** Experimental scheme.

− Before each permeation test, the hydraulic permeability to pure water was determined by measuring the permeate fluxes at different transmembrane pressures at a feed circulation velocity of 0.94 ms−<sup>1</sup> and using Equation (1). The hydraulic permeability of pure water is the slope of the linear regression obtained from the experimental water fluxes and corresponding transmembrane pressures.

$$J\_w = \left(\frac{L\_p}{\mu}\right) \Delta P \tag{1}$$

where *J<sup>w</sup>* is the water permeate flux (ms−<sup>1</sup> ); *Lp*/*µ* is the hydraulic permeability to pure water (ms−1Pa−<sup>1</sup> ); *L<sup>p</sup>* is the intrinsic permeability of the membrane (m), related with its morphological characteristics; *µ* is the water viscosity (Pa·s), and ∆*P* is the applied transmembrane pressure (Pa).

After the tests, a cleaning and disinfection cycle was performed, according to the procedure shown in Table 1. To ensure that membrane's permeability characteristics were kept, the hydraulic permeability to pure water was again determined and, if it was at least 95% of the initial value, the same membranes were used in the following tests.

.


**Table 1.** Process of cleaning and disinfection of membranes <sup>a</sup>

a Into tolerate pH limits of membranes; Na-EDTA, ethylenediaminetetra-acetic acid, sodium salt.

#### 2.2.1. Ultrafiltration Experiments

Ultrafiltration experiments were carried out with two different kinds of membranes, one with an active layer made of regenerated cellulose acetate and a molecular weight cut-off of 10 kDa, designated as RC70PP, and another one with an active layer made of a composite fluoropolymer and a molecular weight cut-off of 1 kDa, named ETNA01PP. Both materials used to manufacture the membranes are hydrophilic in nature, which minimizes the effects of fouling by organic matter, particularly proteins, as described elsewhere [18]. The best operating conditions of transmembrane pressure (0.2 MPa) and feed circulation velocity (0.94 ms−<sup>1</sup> ) were selected, based on the results obtained in total recirculation UF experiments, carried out in the range of transmembrane pressure between 0.1–0.4 MPa, with the same membranes used in this study and established in previous works [18,19]. The highest permeate flux and relative flux (*Jp*/*Jw*) and the best separation between protein and lactose fractions were the criteria used for selection. The temperature varied from 16 ◦C to 22 ◦C and, to correct for the different viscosities of the permeates, all the permeate fluxes were converted to 25 ◦C [20]. The pressure drop along the module was about 0.1 MPa.

The first set of ultrafiltration experiments was done in three steps: preconcentration until the volume concentration factor, VFC = 2.0; dilution mode by adding deionized water; and postconcentration. This procedure allows the achievement of the concentration/purification of the protein fraction in the retentate and a better recovery of lactose in the permeate, thus contributing to improvements in the separation of these components.

Starting from an initial volume of 8.75 L of each sample, three UF experiments with GCW were performed in concentration mode until a volume concentration factor (VCF) of about two was reached with each of the membranes (10 kDa and 1 kDa), using a membrane area of 0.072 m<sup>2</sup> . After concentration, the dilution (DF) of the final retentates was realized in three stages in a discontinuous mode. In each of them, a volume of deionized water, equal to the observed volume of the retentate in the tank, was added. After homogenization and stabilization at the same operating conditions of transmembrane pressure, feed circulation velocity and temperature, a new concentration process took place until the same volume of permeate was collected, thus maintaining the volume of the retentate. Samples of raw and pretreated GCW, retentates and permeates of UF and retentates and permeates of DF were taken for analyses.

The results obtained from this first set of experiments, with both membranes, were analyzed in terms of the following parameters: productivity, measured by volumetric permeate fluxes (*Jp*) and their evolution along the concentration processes, and the separation factor, *α*, for lactose and protein, which should be greater than 1; the higher it is, the better the separation between the two solutes.

The volumetric permeate fluxes were determined according to Equation (2):

$$J\_p = \frac{\Delta V}{A\_m \times \Delta t} \tag{2}$$

where *J<sup>p</sup>* is the volumetric permeate flux; ∆*V* is the volume (m<sup>3</sup> ) of permeate collected during an interval of time ∆*t* (s), and *A<sup>m</sup>* is the total membrane area (m<sup>2</sup> ).

The separation factor, *α*, is defined as [21]:

$$\alpha = \frac{\mathbb{S}\_{microsolute}}{\mathbb{S}\_{macrosolute}} \tag{3}$$

where *Smicrosolute* is the sieving coefficient for the microsolute (lactose), and *Smacrosolute* is the sieving coefficient for the macrosolute (protein). *S<sup>i</sup>* , the sieving coefficient for a component *i*, is given by: *S<sup>i</sup>* = *cp*/*c<sup>r</sup>* , in which *c<sup>r</sup>* is the concentration of a solute in the bulk retentate, and *cp*, the concentration of the solute in the bulk permeate.

Based on the results obtained in this first set of experiments, the best membrane in terms of productivity and the separation of protein and lactose fractions was selected to carry out NF experiments.

#### 2.2.2. Nanofiltration Experiments

Nanofiltration experiments were realized with the mixture of permeates resulting from the UF (PUF+PDFU1+PDFU2+PDFU3) experiments (Figure 1). Permeates were mixed, homogenized and subjected to NF. NF experiments were performed with membrane NFT50 (NF), commercialized by Alfa Laval, Navskov, Denmark. These membranes have an active layer made of polyamide semi-aromatic (polipiperazine). The preconcentration of the feed was carried out until a VCF of about 2.0, at a transmembrane pressure of 2 MPa, a feed circulation velocity of 0.94 ms−<sup>1</sup> and a membrane area of 0.072 m<sup>2</sup> . The diafiltration of the final retentates of NF was performed in three stages by adding a volume of deionized water equal to that of the retentate in the tank, and, afterwards, the concentration process proceeded until an identical volume of permeate was collected. The experimental conditions used in this process were the same as those used for preconcentration by NF.

The performance of the process of concentration by NF, in dilution mode, followed by the diafiltration of the retentates obtained, was determined in terms of permeate fluxes (productivity), *Jp*; efficiency of the removal of salts, *µremoval*; and the evaluation of the quality of the final permeate for possible further application in cheese dairy plants.

The efficiency of removal of a certain solute is given by:

$$
\mu\_{removal} = \frac{c\_{ri} - c\_{rf}}{c\_{ri}} \times 100\tag{4}
$$

where *cri* is the concentration of a solute in a retentate *i*, before a stage of diafiltration, and *crf* is its concentration in the retentate after that stage.

#### *2.3. Cleaning and Disinfection Cycle*

After the permeation experiments, samples were removed from the installation, and three flushes were carried out with water to ensure that no residues were present. The cleaning and disinfection cycle realized and the operating recommended conditions of the manufacturer for the membranes in this study are shown in Table 1. The cleaning procedure included four steps, in each one a different chemical was added, performed under recirculating conditions, which means that both permeate and retentate were recycled to the feed/retentate tank. A transmembrane pressure of 0.1 MPa, a feed circulation velocity of 0.92 ms−<sup>1</sup> and a temperature of 25 ◦C were used during this operation. For NF membranes, a transmembrane pressure of 1 MPa was applied, maintaining the same values of feed circulation and temperature. Between each two cleaning solutions, water was permeated to remove the previous reagent, checking if the pH was already restored. After cleaning, a final disinfection step was carried out, as presented in Table 1, using the same transmembrane pressure, feed circulation velocity and temperature.

#### *2.4. Physicochemical Characterization of the Samples*

The samples (feed, retentates and permeates) were analyzed for: pH (by potentiometry); lactose, by determination of reducing sugars [22]; total solids, by gravimetry [23]; total nitrogen, by the Kjeldahl reference method; crude protein, obtained from total nitrogen multiplied by the factor 6.38 [24] and adapted for cheese whey; the fat content, determined by infrared spectroscopy using the equipment Milkoscan134B, previously calibrated for cheese whey with the standard method of Rose-Gottlied for milk and dairy products; sodium and potassium, by emission flame photometry, according to the procedure described in [25]; calcium and magnesium by atomic absorption spectrophotometry with air–acetylene flame [25]; chloride, by volumetric precipitation, according to the method of Charpentier-Volhard [26]; and phosphates, by the spectrophotometric method of ammonium molybdate [27].

#### **3. Results and Discussion**

*3.1. Physicochemical Characterization of Raw and Pretreated Goat Cheese Whey*

The average composition of raw and pretreated goat cheese whey is shown in Table 2.

.



<sup>a</sup> n (number of samples) = 6.

The goat cheese whey used in this study is classified as a sweet cheese whey because its pH is around 6.0 and is produced from milk coagulated by the enzymatic hydrolysis of casein through chymosin action, at a pH not lower than 5.6 [28]. Apart from water (around 90.8% *w*/*w*), the main components are lactose, followed by minerals, lipids and nitrogen compounds. Lactose, lipids, and nitrogen compounds are, in order of importance, primarily responsible for the high organic loading of these co-products, which are translated into high levels of COD and BOD, as stated in Section 1.

The pretreatment realized (filtration, centrifugation, pasteurization) allowed for a removal of about 58% of lipids, 5% of nitrogen compounds and 15% of ash, leading to a decrease of 6% of the total solids. These results suggest that a part of the organic matter present in the raw goat cheese whey, mainly related to its lipid content, was quickly removed during the pretreatment. This co-product, rich in lipids, after pasteurization, may eventually be reused in cheese dairies, added to milk cream to increase the yield of the manufacture process of goat butter and/or other types of spreads, and will be the subject of further study. However, lactose, most of the nitrogen fraction and around 42% of fat is still present in the pretreated goat cheese whey, thus contributing to its high content of organic matter.

Regarding the mineral composition, the most salient aspect is the very high concentrations of chloride and sodium, which are in contrast with goat milk, where the dominant minerals are potassium, chloride, calcium and phosphate [4]. This resulted from the addition of sodium chloride to the cheese milk, during the manufacture of goat cheese.

#### *3.2. Permeation Experiments*

#### 3.2.1. Characteristics of Membranes

Before permeation experiments, the hydraulic permeability of membranes to pure water was determined (Table 3), according to the procedure described in Section 2.2. In Table 3 is displayed the hydraulic permeability of membranes, the intrinsic permeability and the MWCO of membranes, furnished by the supplier and determined, for NF membranes, according to the procedure described elsewhere [16].

**Table 3.** Hydraulic permeability of membranes (±95% confidence interval), intrinsic permeability and MWCO of membranes.


(1) The dynamic viscosity of pure water, at a temperature of 25 ◦C, used to calculate the intrinsic permeability of membranes, was 8.91 × 10−<sup>4</sup> Pa·s; (2) In accord with [16], for the same set of membranes.

#### 3.2.2. Performance of Ultrafiltration Experiments

#### • Permeate fluxes

The evolution of permeate fluxes along the process of concentration by UF, followed by dilution in UF mode with three stages for both types of membranes (RC70PP and ETNA01PP), is displayed in Figures 2 and 3, respectively. The horizontal line, in both figures, represents the water fluxes at the transmembrane pressure of 0.2 MPa, at which permeation experiments were carried out. As can be observed, until a VCF of about 2.5, the permeate fluxes obtained with samples and with water are close, which indicates that, in the experimental conditions used, fouling is negligible for both membranes. More experiments will be realized in the future on the highest VCFs to study flux behavior when protein concentration is increased. The effect of MWCO, and the corresponding mean pore radius, is evident, because permeate fluxes obtained with membranes of 1 kDa were around 50% of those produced with membranes of 10 kDa. Therefore, the use of membranes with higher MWCO allowed for higher permeate fluxes, as expected, probably because both membranes are made from hydrophilic materials, which are less susceptible to fouling by proteins, the component most responsible for this phenomena in the UF of cheese whey.

1.73 × 10ିହ <sup>−</sup>

1.05 × 10ିହ <sup>−</sup>

**Figure 2.** Variation of average permeate fluxes (±standard deviation) with time, during the process of UF/DF, obtained with membranes RC70PP (n = 3 experiments), at a transmembrane pressure of 0.2 MPa.

**Figure 3.** Variation of permeate fluxes (±standard deviation) with time, during the process of UF/DF, obtained with membranes ETNA01PP (n = 3 experiments), at a transmembrane pressure of 0.2 MPa.

In the case of UF experiments with the membrane RC70PP, at the very beginning, a decline in permeate fluxes was observed, probably due to polarization-concentration phenomena, which is more important when permeate fluxes are higher, due to the rapid accumulation of retained compounds near the membrane surface. However, after that, an average constant flux of about 1.73 × 10−<sup>5</sup> ms−<sup>1</sup> was reached. With membranes ETNA01PP (Figure 3), the initial decline in permeate fluxes was much less pronounced, because the lower average permeate flux, around 1.05 × 10−<sup>5</sup> ms−<sup>1</sup> , minimized the effect of the intensity of concentration-polarization phenomena [29].

ࢻ

−

ߙ

ሺߙሻ

ߙ

Relative to the DF process carried out in three stages, we can observe in Figure 2 that the intensity of average permeate fluxes along the dilution processes were slightly higher than those observed during preconcentration by UF, because are the lower species, such lactose and minerals that are preferentially permeating UF membranes. During each stage, permeate fluxes were kept almost constant. The range of permeate fluxes during the dilution process realized with membranes RC70PP was between 1.79 × 10−<sup>5</sup> ms−<sup>1</sup> for DF1 and 2.00 × 10−<sup>5</sup> ms−<sup>1</sup> at the third stage (DF3). The same was true for the dilution process performed with membranes ETNA01PP (Figure 3), the permeate fluxes ranging from 1.08 × 10−<sup>5</sup> ms−<sup>1</sup> (DF1) up to 1.17 × 10−<sup>5</sup> ms−<sup>1</sup> (DF3). These results show that permeate fluxes, during the dilution processes, were not affected by the permeation of the lower species across UF membranes.

• Analysis of separation factors

The concentrations of lactose and protein in retentates and permeates, as well as the corresponding separation factors between these two components, along UF/DF processes, are shown in Tables 4 and 5 for membranes RC70PP and ETNA01PP, respectively.

**Table 4.** Separation factors (1) (*α*) between lactose and protein, along with UF/DF processes for membrane RC70PP.


(1) *α* was determined at a transmembrane pressure of 0.2 MPa, feed circulation velocity of 0.92 ms−<sup>1</sup> and temperature of 25 ◦C.

**Table 5.** Separation factors (1) (*α*) between lactose and protein, along with UF/DF processes for membrane ETNA01PP.


(1) *α* was determined at a transmembrane pressure of 0.2 MPa, feed circulation velocity of 0.92 ms−<sup>1</sup> and temperature of 25 ◦C.

The observation of the data displayed in Table 4 allows the conclusion that the use of dilution in ultrafiltration mode for UF retentates led to a large increase in the separation factor between lactose and protein, from 6 to 14, right after the first stage of dilution. This is mainly due to the permeation of lactose into permeate streams, as can be confirmed by the decrease in its concentration in UF retentates. However, after the first stage (DF1), the separation factor between those components remained or even declined. Therefore, since dilution mode involves water consumption, which should be minimized for economic and environmental reasons, the use of a second and third stage in dilution mode will be dependent on the desired purification of the final protein fraction.

For preconcentration by UF, the membranes of lower MWCO (1 kDa) led to a better separation between lactose and protein, because *α* is around 10 and, for the other membranes, it is 6 (Table 5). This is likely due to the greater accumulation of the protein fraction in retentates and, consequently, the lower loss of protein into the permeates, probably the lower-molecular-weight whey proteins, such as glycomacropeptide (GMP), with a molecular weight of 6.80 kDa, and bioactive peptides, as described elsewhere [9]. However, unlike what was observed with membranes of 10 kDa, during the dilution process in three stages, the separation factor decreased to about 9 and was kept constant until the end of this process. This may be due to the higher accumulation of the protein fraction in retentates that may have hampered the removal of lactose into the permeate stream, leaving it retained within the protein fraction, which can be confirmed by its higher concentration

in retentates. Therefore, relative to the separation factor, despite the fact that membranes of lower MWCO allowed, in a single UF operation, the obtention of a better separation between protein and lactose fractions, the final decision as to which of the membranes should be selected will depend on the intended application of protein retentates.

3.2.3. Performance of Nanofiltration Experiments

• Variation of permeate fluxes with VCF

Nanofiltration experiments were carried out with the permeates resulting from the UF/DF of membranes with the higher MWCO because their separation between the protein and lactose fractions was better. During the process of dilution in the nanofiltration mode of UF permeates, it was observed a sharp decline in the average permeate flux of about 45%, ranging from 1.59 × 10−<sup>5</sup> ms−<sup>1</sup> (57.24 Lh−1m−<sup>2</sup> ) to 8.33 × 10−<sup>6</sup> ms−<sup>1</sup> (29.99 Lh−1m−<sup>2</sup> ), and until the VCF of 2.34 was attained. A similar pattern was observed during the DF of the NF retentates performed in three stages. Permeate fluxes were only slightly higher than those measured during the preconcentration process by NF, ranging from 1.74 × 10−<sup>5</sup> to 8.72 × 10−<sup>6</sup> ms−<sup>1</sup> , and the decline of permeate fluxes varied from 39 to 47%. This behavior is explained by the fact that UF permeates are mainly composed of the smaller solutes of cheese whey, like lactose and minerals, especially sodium, chloride and potassium, that mostly contribute to its higher osmotic pressure [11], leading to a decrease in effective membrane pressure and thus, to the decrease of permeate fluxes. One possibility to overcome this disadvantage will be the reduction of the amount of sodium chloride added to the milk and curd during the manufacturing process, which will also be beneficial for human health. Another factor that can also contribute to the decrease in NF productivity is the possible formation of insoluble salts of calcium or magnesium phosphates near the membrane surface due to their high retention by NF membranes.

• Physicochemical characterization of NF and DF/NF samples and removal efficiency

Table 6 shows the physicochemical characterization of the following samples (Figure 1): final UF permeate, which is the feed for nanofiltration; CNF, the concentrate of nanofiltration; PNF, the permeate of nanofiltration; CDNF3, the concentrates of 3rd stage of diafiltration; and PDNF3, the corresponding permeates.


**Table 6.** Physicochemical characterization of NF and DF/NF samples.

(1) Not detectable.

In Table 6, it can be observed that the main components of the feed of the DF/NF process (mixture of all the permeates from ultrafiltration) are, apart from water, lactose and minerals, nitrogen compounds at the lowest concentration (0.023% *w*/*w*). During the previous ultrafiltration, most nitrogen compounds were retained by UF membranes, in accord with the sieving coefficient of about 8% (Table 4), which corresponds to a membrane rejection around 92%.

Lipids and nitrogen compounds were preferentially retained by NF membranes, probably through steric hindrance and non-electrostatic membrane–solute interactions, the main mechanisms responsible for the retention of uncharged molecules in nanofiltration membranes [30,31].

Relative to minerals, chloride and sodium are predominant due to their preferential permeation through UF membranes, as expected when negligible fouling problems occur. The distribution of ions between retentates and permeates in nanofiltration can result both from steric hindrance and electrostatic interactions between ions and surface charge, based on the Donnan exclusion mechanism [31]. Since NF membranes used in this work have an isoelectric point, pH<sup>i</sup> = 4.2, this means that, at the pH of our samples, they carried a negative charge [30]. Then, the counter-ions, especially calcium and magnesium due to their higher density charge, were adsorbed at the membrane surface by electrostatic interactions, and the co-ions, such as chloride, were mainly repulsed by the membrane surface to satisfy the electroneutrality condition. The chloride ion even had a negative rejection during the concentration process by NF, probably because of its higher density charge.

In NF/DF stages, the concentration of monovalent ions clearly decreased due to their removal into the permeate streams. The removal efficiencies of salts were calculated based on their concentrations in CNF and CDNF3 (Table 6) and using Equation (4). Calcium and magnesium were preferentially retained (with negative removals); phosphates were slightly removed (circa 2.5%); potassium, sodium and chloride had removal efficiencies of 44%, 54% and 78%, respectively, after the three-stage DF process. It is possible that the high retention of calcium and magnesium contributed to a small reduction in the surface charge of the membrane, and therefore, to a lower removal of the chloride anion present in the retentates.

The predominant components of NF permeate are water, chloride and sodium, and thus, they can be used in cheese dairy plants as washing waters for cheese or during cheese processing.

#### **4. Conclusions**

The fractionation of goat cheese whey using the sequential membrane processes proposed in this study allowed a good separation between protein and lactose fractions through ultrafiltration followed by diafiltration. The recovery of lactose by the nanofiltration of permeates contributes to minimizing the environmental impact of this co-product of goat cheese manufacture and, at the same time, allows for possible applications of the separated fractions.

**Author Contributions:** A.M. and D.A. developed the experimental work. A.M., E.D. and C.P. contributed to the Results and Discussion. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by partnership PORTUGAL2020-PDR, within the scope of the PDR2020-101-030768 project: LACTIES-Innovation, Eco-Efficiency and Safety in Dairy Industry.

**Data Availability Statement:** All the data of this study are presented along the text.

**Acknowledgments:** The authors would like to thank to project LACTIES, to Polytechnic Institute of Beja and to LEAF—Linking Landscape, Environment, Agriculture, and Food, Instituto Superior de Agronomia, University of Lisbon, Tapada da Ajuda, 1349-017 Lisboa, Portugal.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Review* **Mixed-Matrix Membrane Fabrication for Water Treatment**

**Tawsif Siddique † , Naba K. Dutta \* and Namita Roy Choudhury \***

> Chemical and Environmental Engineering, School of Engineering, RMIT University, Melbourne, VIC 3000, Australia; s3642366@student.rmit.edu.au or tawsif@uctc.edu.bd

**\*** Correspondence: naba.dutta@rmit.edu.au (N.K.D.); namita.choudhury@rmit.edu.au (N.R.C.)

† Currently on leave from Department of Mechanical Engineering, University of Creative Technology-Chittagong (UCTC), Chattogram 4212, Bangladesh.

**Abstract:** In recent years, technology for the fabrication of mixed-matrix membranes has received significant research interest due to the widespread use of mixed-matrix membranes (MMMs) for various separation processes, as well as biomedical applications. MMMs possess a wide range of properties, including selectivity, good permeability of desired liquid or gas, antifouling behavior, and desired mechanical strength, which makes them preferable for research nowadays. However, these properties of MMMs are due to their tailored and designed structure, which is possible due to a fabrication process with controlled fabrication parameters and a choice of appropriate materials, such as a polymer matrix with dispersed nanoparticulates based on a typical application. Therefore, several conventional fabrication methods such as a phase-inversion process, interfacial polymerization, cocasting, coating, electrospinning, etc., have been implemented for MMM preparation, and there is a drive for continuous modification of advanced, easy, and economic MMM fabrication technology for industrial-, small-, and bulk-scale production. This review focuses on different MMM fabrication processes and the importance of various parameter controls and membrane efficiency, as well as tackling membrane fouling with the use of nanomaterials in MMMs. Finally, future challenges and outlooks are highlighted.

**Keywords:** membrane; mixed-matrix membranes; MMMs; fabrication; membrane fouling; nanomaterials; phase-inversion process; interfacial polymerization; electrospinning

#### **1. Introduction**

Membranes can be described as films that act as selective barriers between two adjacent phases that allow the transportation of substances from one compartment to another [1]. Membranes play a vital role in separation technology, as well as in energy applications. Membranes are mostly polymer-based, which is adjusted by their synthesis process for the separation of specific substances, and results in efficient cost-effective separation technology with high performance. However, polymer-based membranes have some limitations due to their unavoidable built-in disadvantages, such as poor chemical and physical resilience.

Mixed-matrix membranes (MMMs) are an important class of organic–inorganic nanocomposite membranes with dispersed nanoparticles in polymeric films. Mixed-matrix membranes are based on either classical porous fillers such as zeolites, porous silica and carbon molecular sieves, or nonporous fillers such as graphene oxide, which has the ability to modify the free volume of a polymer by altering the molecular packing of the polymer chains in the membrane. The typical features of nanoparticles, such as stability, surfacearea-to-volume ratio, surface charge, etc. [2], make them excellent candidates for inclusion in polymers for biomedical and environmental applications, including conventional watertreatment processes [3].

In the field of functional membranes, the use of a wide range of nanoparticles and the combination of them with other engineered novel materials gives great scope for engineering the shape and structure of the membranes with the desired performance. As a result, the

**Citation:** Siddique, T.; Dutta, N.K.; Choudhury, N.R. Mixed-Matrix Membrane Fabrication for Water Treatment. *Membranes* **2021**, *11*, 557. https://doi.org/10.3390/membranes 11080557

Academic Editors: Mohammad Peydayesh and Akihiko Tanioka

Received: 8 March 2021 Accepted: 16 July 2021 Published: 23 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

use of mixed-matrix membranes (MMMs) is under development, in which nanoparticles are used as the filler materials in the polymeric matrix of MMMs [4] for applications such as water filtration, gas separation, fuel-cell application, and pervaporation [5–7]. MMMs have been developed substantially as per their applications, and new types of applications of MMMs also have been introduced in the past decade by incorporating inorganic nanomaterials such as metal oxides, including zinc oxide (ZnO) [8], titania (TiO2) [9], iron oxides (Fe2O3, Fe3O4) [10], zeolite [5], silica [11], carbon nanotubes [4], graphene [12], graphene oxide (GO) [13], and metal–organic framework (MOF) [14] as fillers in the polymer matrix.

Currently, MMMs are fabricated using a wide range of fabrication processes based on the membrane materials and their applications. As the effective use of MMMs is increasing due to their various attractive properties, worldwide research on MMMs has experienced exponential growth, as indicated by the number of publications on MMMs in last 20 years (Figure 1).

**Figure 1.** The number of publications each year since 2001 based on the keyword "Mixed matrix membrane" in the Web of Science database (data collected on 7 October 2020).

Due to the significant roles of various fabrication processes on MMMs' properties, the central focus of this review is the fabrication strategy of mixed-matrix membranes for water purification. We begin our discussion with the crucial issue of membrane fouling and ageing and the use of nanomaterials in membrane technology to address the issue, and then describe various fabrication strategies of MMMs, along with parameters that control membrane fabrication.

#### **2. Membrane Fouling and Ageing: Major Challenges for Water-Separation Membranes**

In general, membrane fouling occurs when undesirable particles, macromolecules, colloids, or salts are deposited on the surface of the membrane or inside the membrane's pores. Membrane fouling can be subdivided into a few categories such as inorganic fouling, organic fouling, and colloidal/biocolloidal fouling, based on the membranes' separation processes and the foulants' chemical properties [15,16].

Inorganic fouling occurs due to the higher concentration of inorganic salts, such as sulfates, carbonates of sodium, calcium, etc., mainly when their presence in the solvents is beyond the solubility limits and results in precipitation on the membrane surface or into the pores of the membranes [17]. Organic fouling occurs when irreversible and strong foulants like humic substances, proteins, and polysaccharides are deposited on membrane surfaces [18]. In the case of surface-water, brackish-water, and seawater treatment, the main organic foulant is natural organic matter (NOM) [19]. When the membrane fouling occurs due to the deposition of colloids and the suspension of the nanoparticles or microparticles,

it is known as colloidal fouling. There are three types of colloids [20,21]: organic colloids such as natural organic matter, proteins, etc.; inorganic colloids such as SiO2, iron oxides, and hydroxides of iron and heavy metals; and biocolloids such as viruses, bacteria, and other types of microorganisms. Biocolloid-induced membrane fouling is also called biofouling [22], and is caused by a range of bacteria like Aeromonas, Corynebacterium, Bacillus, Flavobacterium, Pseudomonas, and Arthrobacter, and also by fungi like Trichoderma, Penicillium, and other eukaryote microorganisms [23]. As a result, this foulant layer affects the permeate flux in two different ways [24,25]: first by creating an additional hydraulic resistance that results in low water flux and membrane permeability at a fixed applied pressure, which can be overcome by applying higher pressure; and also by the formation of a porous cake layer inside the unstirred cake layer, resulting in a higher concentration polarization, which leads to higher solute concentration on the membrane surface, as well as an increase in the osmotic pressure of the membrane surface and a decrease in the membrane flux.

Thus, it is widely recognized that the adherence of organic compounds and biocolloids to the surface of the membrane is the key parameter for the fouling of the membrane, and this adherence ability of the foulants is influenced by hydrogen bonding, London–van der Waals attractions, and hydrophobic and electrostatic interactions [18,26]. From the above discussion, it is evident that the inhibition or minimization of the fouling process might be possible by preventing the adhesion interactions between the membrane and the foulant. This could be possible through the development of MMMs with appropriate physiochemical properties, which could combine an efficient separation process with lower membrane fouling.

Additionally, various hydraulic cleaning procedures have been introduced for reversing or reducing membrane fouling [27]. Membrane backwashing with clean water is a common practice for foulant removal. After repeated filtration and backwash cycles, some materials are adsorbed on the membrane surface and need to be washed by a cleaning agent like hypochlorite for ultrafiltration membranes, as they cannot be removed otherwise [27]. Long-term exposure to foulants and cleaning agents has been reported to irreversibly change the performance and characteristics of membranes; these irreversible changes are defined as membrane ageing [28]. The characteristics of membranes are mainly chemical composition, pore size, etc.; and membrane performance factors are fouling rate, clean membrane resistance, etc. The main limitation is that complete full-scale ageing studies need many years of observation and cannot be controlled rigorously [29].

#### *2.1. Effect of Membrane Surface Properties on Fouling and Ageing*

The interactions between a membrane and foulants are determined by the membrane's surface properties such as hydrophobicity or hydrophilicity, surface charge, and surface roughness [30,31].

#### 2.1.1. Hydrophilicity and Hydrophobicity of Membrane Surfaces

Usually, a membrane's hydrophilicity or hydrophobicity is evaluated with a wettability study using contact-angle measurement [32]. The commercial membranes are mostly fabricated from hydrophobic polymers with high thermal, chemical, and mechanical stability, including polysulfone (PSF), polyethersulfone (PES), polyvinylidenefluoride (PVDF), polyacrylonitrile (PAN), polypropylene (PP), polyethylene (PE), and polyamide (PA) [1]. These polymers exhibit a high contact angle, which leads to the adsorption of different solutes from the feed. It is established that a higher mass per unit area of hydrophobic solute is adsorbed by membranes with high contact angles than that by the membranes with a lower contact angle [33]. On the other hand, hydrophilic membranes attract fewer charged inorganic particles, microorganisms, and organic substances, and result in less fouling [34,35].

#### 2.1.2. Surface Charge

In the case of charged foulants, membrane fouling can be controlled by the electrostatic charge of membranes. Membranes possessing the same charge as that of the foulants will reduce membrane fouling due to electrostatic repulsion occurring between the foulant and the membrane, which prevents foulant deposition on the membrane (Figure 2) [36,37]. Therefore, fouling can be reduced by incorporating ionizable functional groups on the surface of the membrane. For example, in protein filtration, when the protein is negatively charged at neutral pH, a negatively charged membrane surface could be a better choice [1]. Similarly, for organic compounds with a positive charge, the positively charged membrane surface is the solution for low membrane fouling [38]. So, low-fouling membranes could be fabricated and developed by considering the potential foulant's charge on the membrane surface and inside the membrane pores from feed streams.

**Figure 2.** Schematic representation of various antifouling mechanisms with composite membranes: (**a**) thin layer of bounded water, (**b**) electrostatic repulsion, and (**c**) steric repulsion (adapted with permission from [39]).

#### 2.1.3. Surface Roughness

Membrane fouling and surface roughness are strongly related to each other in nanofiltration (NF) and reverse osmosis (RO) membranes. Smooth and hydrophilic cellulose acetate (CA) RO membranes have less tendency toward colloidal fouling than hydrophobic and rough PA membranes [38]. Table 1 shows the relationship between surface roughness and relative fluxes for filtration of a sodium chloride solution containing silica particles with commercial NF (Osmonics HL, Dow-FilmTec NF-70) and RO (Trisep X-20, Hydranautics LFC-1) membranes. From the tabulated data of their flux and surface-roughness values, it is clearly visible that the flux decreased with the increase of surface roughness of the membrane during the filtration process. The increase in membrane surface roughness also led to an increase in the total surface area, resulting in more foulant attachment on the surface, and a ridge–valley structure also favoring the accumulation of foulants at the membrane surface. Using atomic force microscopy (AFM), Vrijenhoek et al. [40] showed that colloidal particles mostly accumulate in between the valleys of rough membrane surfaces, which results in valley clogging and causes lower flux and permeability than the membranes with smooth surfaces.


**Table 1.** Correlation between the surface roughness of commercial RO/NF membranes and their relative fluxes for the filtration of a 0.05 M NaCl solution containing 200 mg/L silica particles (0.10 µm); pH = 6.8. Flux-decline values determined for 10 L of permeate volume filtered \*.

\* Adapted with permission from [40].

Considering all the above points, it is clear that the top membrane layer is the key area to control the fouling process, so the main goal could be the surface modification of the membrane to develop a low-fouling composite membrane by introducing polymer brushes and charged groups on the membrane's surface, as well as hydrophilization and creating smooth surfaces, which would minimize the undesirable interactions between the foulants and the membrane surface for low or zero fouling of the membrane.

#### **3. Mixed-Matrix Membrane Materials**

#### *3.1. Polymers*

3.1.1. Glassy and Rubbery Polymers

In water-treatment processes, various polymers have been used in MMMs; some polymers employed are rubbery (e.g., polyethylene oxide) [41], but most are glassy (e.g., aromatic polyamides, cellulose acetate, and polysulfone). Classifying membranes for watertreatment processes as rubbery or glassy can be complex, since they are operated under hydrated conditions and can absorb substantial amounts of water (i.e., ~10–50 vol% water) [42–44].

Recently, ion and water transport in glassy hydrated polymers has been reported, and has become a topic of interest in the membrane field [45–49]. Xie et al. measured water and salt transport in a disulfonated poly (arylene ether sulfone) copolymer (i.e., BPS-32) [45]. BPS-32 was synthesized in the potassium counter-ion form (K) and acidified to the acid form (H), either in solid state or in solution, and subjected to various ion-exchange steps and thermal treatments. Due to its relatively high T<sup>g</sup> (278 ◦C), the membrane remained glassy upon hydration, and therefore its processing history had a profound impact on its water and salt transport properties.

More recently, Chang et al. prepared two chemically similar copolymers, rubbery 2 hydroxyethyl acrylate-co-ethyl acrylate (HEA-co-EA) and glassy 2-hydroxyethyl methacrylateco-methyl methacrylate (HEMA-co-MMA), to probe the impact of polymer backbone dynamics on ion and water transport properties [48,49]. Both had similar and relatively low water contents (~8% by mass). However, the rubbery membrane had salt permeability coefficients roughly 2–3 times higher than those of the glassy membrane. In a later study, Chang et al. reported water dynamics and tortuosity in the same membranes over several length scales [49]. Using pulsed-field gradient nuclear magnetic resonance (PFG nmR), they measured water diffusivity as a function of diffusion encoding time. The longer the diffusion encoding time, the greater the length scale over which diffusion was measured. Water-diffusion coefficients decreased with increasing encoding time, plateauing at long times as water-molecule diffusion became increasingly hindered by the polymer segmental obstructions on longer length scales. The long-duration plateau value of water diffusivity was regarded as equivalent to the value observed in measurements of bulk-transport properties [49]. Salt solubility and diffusivity were measured via equilibrium and kinetic desorption techniques, respectively. Equilibrium water solubility was also measured. Using the solution-diffusion model, water and salt permeabilities were calculated from these data. Water and salt diffusivity and permeability were lower in the glassy polymer than in the rubbery polymer. However, water/salt selectivity was enhanced in the glassy membranes,

corroborating the enhanced size sieving observed in their earlier study [48]. This result was mainly attributed to enhanced diffusivity selectivity in the glassy polymer, since salt solubility was similar in both polymers.

## 3.1.2. Modification of Polymers

#### Chemical Cross-Linking

In many cases, membrane materials have reactive functional groups that can be linked through covalent bonds by applying a suitable cross-linker, which gives a remarkable scope of membrane fabrication using the chemical cross-linking process and for modifying polymers [50–59]. This chemical cross-linking method is used for a membrane's mechanical strength enhancement or swelling reduction, as well as the increase of a specific solutes' selectivity with better solvent permeability depending on the applications [60–62]. The crosslinking medium, the cross-linker's concentration and molecular structure, and the reaction time/temperature mainly influence the cross-linking degree, as well as the charge density, which can be confirmed by Fourier transform infrared spectroscopy (FTIR) [50,58,63]. A polyvinyl chloride (PVC) membrane has been cross-linked with an activated-carbon loaded 4,4′ -oxidianiline to prepare the MMM for separation technology [60].

#### Chemical Grafting

Chemical grafting on a membrane surface can be performed by growing or grafting another polymer onto the surface. The hydrophilicity, selectivity, and antifouling property improve due to the grafted polymer. There are a few approaches to produce the active sites that can prompt the commencement of the graft polymerization; for example, plasma, UV, and ion-beam irradiation [64–66].

UV photo-grafting is performed on a polyimide membrane's active surface to modify it so it is suitable for wastewater-treatment applications. The outer active surface of a polysulfone UF hollow-fiber membrane was reported to be achieved by UV grafting, in which sodium p-styrene sulfonate (monomer), N,N′ -methylene bis acrylamide (crosslinker), and 4-hydroxybenzophenone (photo-initiator) were used. Figure 3 shows a UVphoto-grafting setup in which the support layers of hollow fibers are wetted by water and immersed in a monomer solution. At that point, the fibers pass through two UV polychromatic lamps [67].

Graft polymerization of a methacrylic acid monomer was reported to contribute to membrane hydrophilicity and negatively charge the membrane surface, as it could eliminate the disrupting endocrine chemicals and active pharmaceutical compounds [68]. Furthermore, the introduction of a redox reaction at the initial stage of surface grafting also offered hydrophilicity, and the redox reaction could be achieved in aqueous media at room temperature without any external activation [66]. Additionally, the concentration of the monomer needed to be higher due to the slow reaction kinetics of the redox initiation [69]. Commercial polysulfone (PSF) has been grafted by poly(polyethylene glycol) methyl ether methacrylate (PEG) side chains to improve the interfacial interaction with zeolitic imidazolate framework-8 (ZIF-8) nanoparticles to prepare the desired MMMs [70].

**Figure 3.** UV-photo-grafting setup for hollow-fiber membrane fabrication.

#### *3.2. Nanoparticles (NPs)*

Surface modifications of polymer membranes have led to various low-fouling membranes, and in some cases proved feasible for commercial purposes. However, the use of nanoparticles (NPs) in the membrane could be a better strategy for preparing low-fouling membranes in a simpler way with a long durability. The addition of a large variety of nanoparticles into the polymeric membrane has been extensively explored, leading to mitigation of membrane fouling with longer durability and high permeate flux [71,72]. The successful development of MMMs depends strongly on the polymer matrix selection, the inorganic filler, and the interfacial interaction between the two phases [73]. The selection of suitable types of inorganic filler and their surface modification dictates an MMM's overall performance. Various surface-modification strategies have been used to maximize the interfacial interactions. The superior permeability and selectivity of inorganic membranes with the processability of polymeric membranes are combined in MMMs to achieve synergistic performance, in which the rigid, porous-type inorganic NPs provide desirable properties, and the polymeric phase enables the ideal membrane formation, hence solving the issue of brittleness inherently obtained in the inorganic membranes [74].

#### 3.2.1. Metal Oxides

Amongst various metal oxide nanoparticles, titanium dioxide (TiO2) is very attractive due to features like ease of preparation, stability, and commercial availability, and membrane fouling could be significantly reduced by introducing TiO<sup>2</sup> into the polymer matrix of a membrane [9]. Additionally, the hydrophilicity and the free water fraction also increased with the deposition of TiO<sup>2</sup> nanoparticles on the polymer membrane surface. Studies on the effect of various sizes of TiO<sup>2</sup> nanoparticles in a hydrophobic polyvinylidenefluoride (PVDF) membrane revealed that the fouling activity of the PVDF membrane could be significantly improved using smaller nanoparticles [75,76], as this hydrophilic modification of PVDF membranes actually decreased the adsorption and deposition of hydrophobic organics on the membrane surface. For example, TiO<sup>2</sup> in polyvinyl acetate not only decreased the membrane-fouling activity, but also improved the thermal stability, which was determined by the increase in the glass transition temperature [77].

Silica nanoparticles also showed the same trend in polyester urethane and polyether urethane-based membranes [11]. Silica nanoparticles have shown performance enhancement of polydimethylsiloxane (PDMS) membranes in pervaporation [78], resulting in improved selectivity of the membranes in pervaporation as the polymer chains became more rigid, and the polymer-free volume was also decreased.

Zinc oxide (ZnO) is used as the filler material in membranes for photo-degradation of organic pollutants and dyes in water and wastewater, and provides antibacterial properties [8]. It has also good electrochemical activity [79].

#### 3.2.2. Magnetic Nanoparticles

Nowadays, magnetic nanoparticles are considered as potential candidates for MMMs [80]. Iron-based magnetic nanoparticles have been studied for a vast number of environmental applications, as they also have the ability of bacterial inactivation [81,82]. Fe3O<sup>4</sup> has been used as filler material in mixed-matrix membranes due to its attractive features for various applications such as oil–water separation [10], dye and magnetic-particle removal [83], etc.

#### 3.2.3. Carbon-Based Nanoparticles

Carbon-based nanomaterials are also considered as an efficient family of filler materials for MMMs due to their improved chemical and mechanical properties and costeffectiveness. Among them, graphene oxide (GO) has been explored extensively as a filler material in the polymer matrix for the fabrication of polymeric nanocomposite membranes [84–87]. GO is a two-dimensional material with one-atom thickness, resulting in ultrafast water transport across the GO nanocomposite membrane as it forms interconnected nanochannels [88]. The functional groups such as hydroxyl (—OH), carboxyl (—COOH), epoxide, and C=C on the GO surface offer excellent hydrophilic, antifouling, and antibacterial properties [12,89–91].

#### 3.2.4. Zeolites

Mixed-matrix membranes with zeolite fillers have attracted attention due to their excellent advantages, such as high permeability and improved selectivity [92]. Zeolite–MMMs could be considered ideal for the purification industry, since they combine the properties of a polymeric matrix and zeolite inorganic fillers [93]. Nevertheless, only a few studies have been performed on zeolite–MMMs for water treatment; it was determined that the size of zeolite should be designed to match the expected polyimide active film thickness, thereby providing a preferential flow path through the nanochannels of zeolites [94,95]. Natural zeolite can readily form a suspension to coat the membrane as a support [96]. In another study by Damayanti and coworkers, zeolite-based membranes demonstrated excellent performance and high efficiency for removal of micro-pollutants for laundry-wastewater treatment [97]. Membrane performance was measured based on the flux and rejection values. They studied the superior ability of zeolite membrane to treat laundry wastewater as determined by turbidity measurements and phosphate removal as the two significant parameters. More importantly, another advantage of zeolite-based nanomembranes is that such membranes show an enhanced hydrophilicity when zeolites are used, since they are hydrophilic in nature, which in turn contributes to the enhanced removal of pollutants from wastewater. In addition, zeolite membranes showed improved separation performance and antifouling properties, and the structure and surface properties of the membrane's thin-film layers were modified [98,99].

#### 3.2.5. Metal–Organic Frameworks (MOFs)

Metal–organic frameworks (MOFs) are a unique family of nanoparticles used with membranes for the enhancement of their separation performance, as well as in pervaporation to recover the bioalcohols [14]. MOFs decrease the ageing of the MMMs due to their good compatibility and interaction with the polymer matrix, which results in restrictions of chain mobility (one of the main causes of ageing) [100]. MOFs include ZIF-8 [101–103], HKUST-1 [103,104], and UiO-66 [100,105–107], mostly either as cast or modified [108]. MMMs with inorganic fillers or nanoparticles often have weak polymer−filler interfaces

due to the lack of compatibility between the two components, which can create an adverse effect. MOFs containing organic functionality in their bridging ligands can potentially interact favorably with the organic functionality in polymers. However, the organic functionality does not completely eliminate this compatibility issue due to the rigid, crystalline nature of MOFs. Therefore, strategies to improve interfacial interactions, such as chemical and physical interactions, pre- and post-synthetic modifications to MOF ligands, chemically functionalizing the polymer, and employing cross-linking-type reactions to tether the MOF frameworks to the polymer, have been pursued [77–85].

Nanoparticles are also incorporated in membranes for pervaporation applications. As an example, for ethanol dehydration, phosphotungstic acid (H3PW12O40) nanoparticles were added in a sodium alginate/poly(vinyl pyrrolidone) polymer blend [109]. Silica nanoparticles have shown performance enhancement of polydimethylsiloxane (PDMS) membranes in pervaporation [78], resulting in improved selectivity of the membranes in pervaporation as the polymer chains become more rigid, and the polymer-free volume was also decreased.

There is a large body of work using nanoparticles and their surface modification, leading to better properties in many fields, as the general trend of using nanoparticles is to improve and maintain the permeability of liquids and gases and to enhance the desired separation of the membranes. However, the mechanism behind these results was not studied extensively. It is noteworthy that the nanoparticles in the matrix influence the morphology and free volume of the membranes. So, the nanoparticles are used in the membrane for their performance enhancement, and the fabrication of mixed-matrix membranes with nanoparticles will be discussed in the following sections.

#### 3.2.6. Loading or Addition of Nanoparticles in a Polymer Solution

MMMs are the combination of two phases: the polymer matrix and the filler material, such as NPs. Therefore, the mixing of NPs in the polymer matrix is an important part of MMM fabrication, as the homogeneous dispersion of NPs in polymer matrix needs to be ensured for good-quality membrane fabrication. To obtain this, preparation of a homogeneous solution of NPs and polymer is required, which can be done using one of the three established processes described below.


Among these methods, the first and third methods are used for better distribution of inorganic particles because in a dilute suspension, the particles are prevented from agglomerating by a high shear rate during stirring, while the second method is commonly used for nanoparticle distribution in the polymer matrix [142].

#### **4. Fabrication Processes of MMMs**

Figure 4 shows the various membrane-fabrication processes that will be discussed in this review. The improvement in functional properties brought about by forming mixed-matrix membranes or nanocomposite membranes can be grouped in two categories: physical mixing and in situ synthesis [143]. The physical mixing method is very convenient to operate at a very low cost in large-scale production; as a consequence, it has been used extensively to fabricate nanocomposite MMMs. For any inorganic nanomaterials, the nanofillers and polymer dope typically are prepared independently and mixed using the solution, mechanical agitation, fusion, emulsion, etc. [144,145]. Inorganic particle deposition or direct coating onto the membrane surface could also be used to fabricate MMMs. Nonetheless, it is difficult to control the nanoparticles' distribution on or in the

polymer matrix during MMM fabrication through the direct mixing method of polymers and nanofillers. The interfacial adhesion of nanoparticles with the polymer can lead to larger aggregates during mixing, thus noticeably diminishing the advantages of the nano dimensions. In addition, polymer degradation upon melt compounding and phase separation of nanoparticles from the polymer phase is sometimes detrimental. The uniform dispersion of nanoparticles on or in the polymer matrix can be achieved by adjusting different processing parameters like shear force, time, and temperature, etc. [146], and the use of dispersing agents could be a promising way of obtaining a well-dispersed membrane [147].

**Figure 4.** Various membrane-fabrication approaches.

**Figure 5.** In situ synthesis process of mixed-matrix membranes.


−

Table 2 shows a list of membranes prepared according to various fabrication processes and their basic properties, and Table 3 compiles the merits and disadvantages of various fabrication processes. In the next section, we will discuss the main membranefabrication processes.


**Table 2.** Basic properties of some membranes with their fabrication process.


#### **Table 2.** *Cont.*


#### *4.1. Phase Inversion Process*

Phase-inversion is the most popular method to form an asymmetric polymer membrane, and was first developed by Loeb Sourirajan in 1963 [175]. It offers several advantages over other membrane-fabrication methods such as material selection flexibility and the capability of making membranes with different pore sizes (between 1 and 10,000 nm) by varying the process parameters, solvent, and membrane material. The phase-inversion process is also called the phase-separation process, in which a homogeneous polymer solution is separated into two different phases, polymer-rich and polymer-poor, leading to two different layers of the porous structure. The mechanism of phase inversion primarily involves controlled transformation of a polymer solution to a solid state through liquid– liquid demixing, as shown in the ternary phase diagram of a polymer–solvent–nonsolvent system (Figure 6). Thermally induced phase separation (TIPS) and non-solvent-induced phase separation (NIPS) are the two approaches for the separation of a polymer solution. In TIPS, the polymer and solvent are mixed at a high temperature followed by cooling, which results in phase separation, whereas NIPS is a three-component process in which a non-solvent is used with the polymer and the solvent, and the main phase change occurs via the immersion of the polymer solution into the non-solvent [176]. During this immersion, the non-solvent is absorbed by the polymer solution and the volatile solvent is evaporated. An electrolyte membrane of PVDF and PAN polymers in which SiO<sup>2</sup> was used as a nanofiller has been fabricated by phase inversion for lithium-ion batteries [159].

**Figure 6.** Phase diagram for the phase-inversion process.

Membrane Fabrication through Immersion Precipitation

In the immersion precipitation method, a coagulation bath and a casting knife are used (Figure 7). The prepared homogeneous polymer solution is poured over a non-woven supporting mat, and then the dope is spread to a pre-defined thickness by using the casting knife. Afterward, the membrane is dipped into the bath. Before dipping in the bath, the casted dope is exposed to an ambient environment. The membrane property can be adjusted by controlling the temperature of the coagulant bath, as well as the exposure time in that bath, and the condition of the ambient environment. Although mostly water is used for the coagulant solvent, other non-solvents can also be used.

**Figure 7.** Illustration of membrane casting.

For hollow-fiber membranes [177,178], a bore fluid is required for hollow-fiber spinning as an internal coagulant. The process of hollow-membrane fabrication is complicated, as the phase separation occurs on both the inner and outer surfaces. A hollow-fiber fabrication process is illustrated in Figure 8. Extrusion of the bore fluid and the dope takes place simultaneously from the spinneret, and the pumps are used to control the flow rate. The developing fiber flows through an air gap and is finally immersed in the coagulant. A

rotating drum is used to collect the final fiber at a constant speed, but it should be equal to or higher than the speed of free-falling fibers to avoid the coiling of the fiber. Finally, the solidified fiber is collected from the bath, followed by water soaking to remove the remaining solvent. Then the membrane is dried by freeze-drying or solvent exchange to avoid pore collapse during drying [177,178].

**Figure 8.** Spinning process of hollow-fiber membranes.

#### *4.2. Interfacial Polymerization*

Polyamide membrane development by interfacial polymerization has been recognized as the most regularly utilized method to form superior RO-like and NF-like active layers. Interfacial polymerization uses two exceptionally responsive monomers at the interface of two solvents that are immiscible with each other, one of which should be organic, and other of which should be inorganic/aqueous. There are two types of interfacial polymerization: (1) for drug delivery applications, micro/nanocapsules or micro/nanospheres are produced by dispersing one phase into another as tiny droplets using high-speed stirring [179]; and (2) the common process of introducing a continuous layer on a support, leading to a thin film [180,181].

′ A few types of monomers and prepolymers; for example, piperazine, N,N′ -diaminopiperazine, and m-phenylenediamine for amine solution [182,183], and trimesoyl chloride, sebacoyl chloride, and iso-phthaloyl chloride for acyl halides solution [181] can be utilized for interfacial polymerization.

Mixed-matrix interfacial polymerization has been developed to insert nanoparticles throughout the polymer layer. The purpose is to improve the membrane's performance. Super-hydrophilic zeolite nanoparticles are utilized to improve the water permeability with high rejection of salts [156]. Aquaporin-based biomimetic membranes have been fabricated with a similar process, resulting in high separation performances [184].

#### *4.3. Multilayer Polyelectrolyte Deposition*

Polyelectrolyte is a polymer containing electrolyte(s) groups in its repeating units. Polyelectrolyte shows charge properties when it dissociates in an aqueous solution or water. The driving force of multi-layer polyelectrolyte deposition on the membrane surface is the electrostatic interaction between the oppositely charged molecules. Scheme 1 shows various polyanions and polycations used for layer-by-layer formation of a polyelectrolyte complex multilayer (PEM). Figure 9 shows such a process, in which it is clear that the deposition of an aqueous polyelectrolyte solution on a porous substrate in the desired sequence could be a facile method of membrane preparation [185].

**Scheme 1.** Commonly used polyanions and polycations for the development of active–selective layer: **Top:** polyanions, from left to right: poly(styrene sulfonate) (PSS) sodium salt, poly(acrylic acid) (PAA), sulfated chitosan (S-Ch); **Bottom:** polycations, from left to right: poly(diallyldimethyl ammonium chloride) (PDADMAC); chitosan(Ch); polyethylenimine (PEI).

**Figure 9.** Schematic drawing of multi-layer polyelectrolyte deposition on the outer surface of a hollow-fiber membrane (adapted with permission from [185]).

Multi-layer polyelectrolyte deposition is easy and adaptable for membrane preparation with thinner thickness and containing specific desired layers for high selectivity of the desired content. The function and structure of the layers can be different for specific applications based on the charge density of the polyelectrolytes and their molecular

structures. Polyelectrolyte membranes are used in different applications such as forward osmosis [185–187], NF [188], ion exchange [189], pervaporation [190,191], and gas separation [192,193].

#### Factors Affecting Multi-Layer Polyelectrolyte Deposition

The pH and ionic type of the polyelectrolyte multi-layer play a significant role in the development of a unique film of multi-layer polyelectrolytes. In the event that both permeable substrate and the polyelectrolyte are contrarily charged at a high pH and vice versa, the ideal pH utilized ought to be in the middle of the iso-electric point of the substrate and the polymer, as shown in Figure 10. Accordingly, inverse charges are conveyed by the substrate and the electrolyte [194]. An ultraviolet/ozone (UV/O3)-cleaned permeable alumina membrane with surface pore measurement of 0.02 µm is becoming attractive as a substrate for polyelectrolyte layer deposition because of its positive charges [195]. Plasma-treated/hydrolyzed polyacrylonitrile and cellulose acetic acid derivatization are negatively charged [186,187]. Furthermore, PES is also appealing as a supporting material in spite of the fact that it is neutral. Hence, the connection of polyelectrolyte layer depends on hydrophobic cooperation [185].

⚫ ○ **Figure 10.** Plot of pH and zeta potential of polyelectrolyte (—) and substrate (—); ate (---); ⚫, i ○ , iso-electric point of polyelectrolyte; and #, iso-electric point of substrate (adapted with permission from [196]).

Ionic strength of the polyelectrolyte solution can be expanded by including salts. At high ionic strength, the electrostatic repulsion of the polymer chain diminishes with the polymer coils becoming denser, with the deposited layer in collapsed form instead of a flat conformation. Subsequently, it builds the thickness of the individual layer [197]. In the climate of incredibly high salt concentration, just a limited quantity of polyelectrolyte can be absorbed by the substrate because of the opposition to the more modest charged particles from the salts [194]. To improve the density of the polyelectrolyte layers, crosslinking could be utilized to enhance the layers' stability. A cross-linking agent such as glutaraldehyde could be utilized in those cases [186,187]. Another parameter, the charge density of the polyelectrolytes, depends on the molecular structure and the degree of ionization of the polar groups. The charge density of the resultant multilayers, defined as the number of ionic groups per number of carbon atoms in the repeat unit of the polyelectrolyte complex/multi-layer [197,198] often guides the thickness. By adsorbing polyions from salt solutions of varying electrolyte concentrations, the layer thickness can be controlled over a wide range. In addition, consolidation of nanoparticles, such as silver on the active layer of a membrane, can also enhance the antifouling or antibacterial properties of the membrane. The layered structure of a multi-layer polyelectrolyte could improve the stability of the nanoparticles on the membrane surface [188].

#### *4.4. Dual-Layer Co-Extrusion/Co-Casting*

Improvement of composite dual-layer membranes is appealing, as beneficial properties of at least two polymeric materials can be consolidated for different applications. The material expense of the superior polymer can be decreased, and the polymer with extraordinary selectivity but poor mechanical strength can be reinforced, by consolidating them with an economical and strong polymer support layer [199,200].

Increasing uses of double-layer membranes include forward osmosis, gas separation, and NF membranes, which are made out of a thick selective layer supported by a porous polymer matrix [51,180,201–204]; and direct-contact membrane distillation, which requires an additional thin hydrophobic layer for wetting prevention and another hydrophilic layer for better water permeability [205].

Hollow-fiber [201,204] and flat-sheet [199,202] membranes are prepared by the duallayer co-casting method on the basis of same principles, which are the casting of two different polymer solutions or a single-step co-extrusion. Synchronous development of the double-layer structure should be possible by utilizing a triple-orifice spinneret for hollow-fiber membranes, and a double-blade casting machine could be used to prepare the flat-sheet membrane by a co-casting process (Figure 11) [201,202].

**Figure 11.** *Cont.*

**Figure 11.** (**a**) Schematic diagram of a dual-layer hollow-fiber spinning process; (**b**) cross section of a triple-orifice spinneret (adapted with permission from [201]); and (**c**) fabrication process of a dual-layer flat-sheet membrane using a double-blade casting machine (adapted with permission from [202]).

Complexity arises while fabricating dual-layer membranes in either a hollow-fiber or flat-sheet configuration because of the involvement of many parameters controlling the thermodynamic properties and the energy of the phase change. This parameter control results in uniform cross-sectional morphology, as well as better lamination between the two layers of the synthesized membrane. The fabrication parameters can be divided into two categories: the chemistry of the polymer solution and the operating conditions. The chemistry of the polymer solution relies upon the polymer concentration and type, the solvent's affinity to the polymer or coagulant, and the concentration and variety of nonsolvent additives (or pore formers) [201,202]. Working conditions incorporate an air gap for hollow-fiber spinning, the evaporation time for flat-sheet casting, the composition and temperature of the coagulant, the temperatures of the polymer solution, and the operating temperature [177,199,201,206,207].

#### *4.5. Dip-Coating*

In a dip-coating method, the membrane surface is coated by applying a polymer or organic materials. The polymer usually utilized as coating material should have some extraordinary properties; for example, it could be hydrophilic and negatively charged, and attach to the support layer easily. This group of polymers can be prepared by sulfonation; for example, sulfonated PES (SPES) and sulfonated poly(ether ether ketone) (SPEEK). The coating layer may upgrade the performance of the support layer; for example, by giving it a higher strength and better separation properties. Some basic properties should be taken into consideration while choosing the coating polymer; for example, the strength and stability of the polymer, layer-forming capabilities, easy solubility in solvents, cost, and cross-linking capability [208]. Three basic steps in the dip-coating process (Figure 12) are: (1) immersing a dry membrane in a coating solution, (2) permitting the coating material to interact with the substrate, and (3) drying the prepared membrane (Figure 12).

**Figure 12.** Hollow-fiber composite membrane fabrication by the dip-coating process.

SPES has been used as the selective layer of NF hollow-fiber membranes by dip coating due to its capacity of ion exchange (limit of 0.8 meq/g) and antifouling activities. SPES conveys negative charges due to the presence of a sulfonic acid group in the main chain. The significant disadvantage of this polymer is that it can swell in water easily. When the polymer is dried, the structure of the layer becomes brittle [199]. In addition, NF hollowfiber membranes have been prepared using PES as the substrate, followed by the dipcoating of SPEEK as the selective layer. The thickness of the coating layer generally relies upon the viscosity of the coating solution, which is impacted by temperature, grouping of the solution, and added substances. At a lower concentration, the viscosity of the solution is low, and as a result, the coating solution will infiltrate to the substrate pores [209].

#### *4.6. Electrospinning*

Nanofibrous membranes are in high demand nowadays because of their scaffold structure, larger surface area, and interconnected porosity. Among different fabrication methods, electrospinning is attractive in developing nanofibrous membranes because of its scalability, simple design, and low cost [210,211]. Figure 13 shows a typical electrospinning setup.

Typically, the electrospinning system consists of a high-voltage power supply, syringe pump, syringe, needle, and a conductive collector where the fiber is gathered to make the membrane. Figure 13 represents a basic electrospinning system [212]. It can be classified as vertical and horizontal system based on the ordering of the spinneret. During the electrospinning process, the polymer solution is pumped at a suitable rate from the syringe to make small droplets at the tip of the spinneret. The voltage is supplied in the range of 1–50 kV from the high-voltage power supply, which results in charging of the droplet by the applied electric field, and eventually a solution jet is formed. The droplet is turned into a cone-shaped structure (Figure 14) to aim the solution jet toward the conductive collector. The threshold value of voltage causes the electrostatic force to overcome the surface tension of the droplets, which leads to the formation of jet from the cone's tip. However, an appropriate viscosity is required for a continuous jet of solution by avoiding the Rayleigh instability, which causes breakup into droplets [213]. This jet becomes thinner and dries before being deposited on the collector in fiber form [214].

**Figure 13.** Schematic diagram of electrospinning setup: (**a**) horizontal; (**b**) vertical (adapted with permission from [212]).

**Figure 14.** Formation of a Taylor cone with the increase of applied voltage.

In 1930, Formhals illustrated the principle of electrospinning first, though the first patent was obtained in United States earlier (1902) [215,216]. Nevertheless, the electrospinning process received attention after 1990, but it was recognized globally within a short time to prepare the polymer-based nanofibers of different diameters down to a few nanometers. In the last decade, the number of publications on electrospinning is notable (Figure 15).

**Figure 15.** The number of publications each year since 2001 based on the keyword "Electrospinning" in the Web of Science database (data collected on 20 October 2020).

At present, the electrospinning process is more advanced than before, which allows a more controlled property by adjusting the process parameters. Eventually, the electrospinning method will become preferable in different fields of study, such as energy storage, separation and membrane technology, drug delivery, tissue engineering, and so on [217–220]. There is a drive to apply the electrospinning method in large-scale applications. Fortunately, Donaldson and Freudenberg [220] have successfully implemented electrospinning technology in making a filtration membrane.

The most interesting property of the electrospinning technique is the controllability of the fiber diameter by monitoring the variables such as solution concentration, loading of filler material, voltage, flow rate, temperature, and humidity [214]. A wide range of fiber diameters, from micron-sized to a few nanometers, can be achieved. Figure 16 shows a non-woven nanofibrous membrane of polyacrylonitrile [221].

**Figure 16.** Scanning electron microscopy image of an electrospun polymer: a poly(acrylonitrile) non-woven nanofiber mat produced by electrospinning (adapted with permission from [221]).

The electrospinning method can be applied not only to polymers, but also to metals [222] and ceramics [223] for formation of micro- and nanofibers. However, polymers are mostly being studied, including mixed-matrix polymers containing polymer blends [224], drugs [225], and nanoparticles [226]. Although many polymers are being successfully electrospun into fiber, several polymer/solvent systems are very popular because of their suitable molecular weight, volatility, and conductivity of the solvent. This list includes polyamides [227], polyurethanes [228], polyester [229], poly(ethylene oxide) [230], polystyrene [231], poly(vinyl pyrrolidone) [232], poly(methylmethacrylate) [233], poly(vinyl alcohol) (PVA) [234], poly(lactic-co-glycolic acid) [235], polyacrylonitrile [236], and poly(caprolactone) [237], as well as bio-polymers such as chitosan [238], collagen [239], and gelatin [240].

In the next sections, the effect of properties of polymer solution and process parameter on the properties of electrospun membrane will be discussed. In addition, the roles of temperature and humidity are also mentioned.

#### 4.6.1. Effect of Intrinsic Properties of Polymer Solutions

The properties of a polymer solution largely control the structure of the nanofiber. Currently, a large number of studies have reported the role of solution viscosity, surface tension, concentration, and conductivity on nanofiber fabrication [241–245]. In the next section, the effect of these parameters will be described.

#### Polymer Concentration and Solution Viscosity

Several research reports showed that the structure and morphology of the electrospun membrane largely depend on the solution viscosity and concentration [214,246–248]. Polymer concentration profoundly influences the surface tension and viscosity of a solution, which eventually controls the development of nanofibers. The low-viscosity solution results in bead-on-string fibers. On the other hand, with increasing viscosity, the shape of beads is changed from globular to a spindle-like structure, which leads to the formation of a uniform fiber [229,249]. However, high viscosity also increases the diameter of the nanofiber. Therefore, it is required to optimize the threshold value to obtain a preferable fiber structure.

#### Electrical Conductivity

The spinnability of a polymer largely depends on the electrical conductivity of the dope solution, as the rheological behavior largely depends on it [250,251]. The category of solvent and polymer and the concentration of ionizable salts determine the electrical conductivity of the polymer solution [249]. Usually, a highly conductive solution forms a finer fiber and a wide range of fiber-diameter distribution [217,218]. In addition, increased electrical conductivity can help to form a stable Taylor cone that leads to producing a dense scaffold structure [252]. The conductivity can be enhanced by adding ions in the dope solution. Moreover, due to a higher charge density, the smaller ion can create a stronger elongation force on the jet [217,253–255]. Electrical conductivity can also be enhanced by adding a suitable acid with a higher dielectric constant, such as formic acid [256,257].

#### Surface Tension

Surface tension of the dope solution is an important parameter in tailoring the nanofiber structure. It can be adjusted by adding surfactants [254,258–260]. Lower surface tension forms a stable jet, and consequently, a uniform woven structure is formed. However, a higher amount of surfactant can cause other defects, such as a clustered structure.

#### Solvent

Solvent plays an important role in determining the morphologies of a nanofibrous membrane. During fabrication of a nanofiber, the solvent is continuously evaporated. Therefore, solvents with different evaporation and solubility rates can change the final

structure of the nanofibrous membrane [254,257]. It has been reported that a solvent with low solubility is suitable for electrospinning. Spinnability–solubility maps were used to select a suitable solvent for the polymer [261].

#### 4.6.2. Effect of Electrospinning Process Parameters

The process parameters of the electrospinning technique such as flow rate, applied voltage and collector-to-spinneret distance play important roles in determining the quality of the electrospun membrane. In the following section, the effect of electrospinning parameters on the final product will be discussed.

#### Applied Voltage

The applied voltage determines the electrostatic force between the spinneret and the collector, and the charge density in the droplets [258]. The fiber diameter decreases with increasing voltage [252]. However, it may cause increased bead structure on the polymer net [262].

#### Electrode Distance

The distance between the spinneret and the collector defines the intensity of the electric field and the duration of the jet touching the collector. The distance should be enough to allow sufficient time for fiber elongation [263].The fiber elongation and solvent evaporation can be decreased by decreasing the distance, which leads to formation of a thicker fiber [264]. However, reduced distance also helps to stabilize the solution jet [265], while an inappropriate distance causes formation of beads [263].

#### Solution Mass Flow Rate

The study of the impact of flow rate on the quality of nanofibers has not been studied extensively. However, Megelski et al. [231] noted that higher flow rates cause formation of thick nanofiber and beads. The fiber diameter is increased because of reduction of charge density of fiber jet [266]. A bead is formed as the unstable jet is formed by the removal of the higher solution from the tip [267].

#### Ambient Environment

The effect of temperature and humidity on the electrospinning process cannot be ruled out. A lower temperature decreases the evaporation rate of the solvent, and eventually fiber diameter is decreased, as there is more time to be elongated before solidification. On the other hand, at a higher temperature, the diameter of the fiber increases, as the solution jet solidifies faster [257,268]. Moreover, the relative humidity can also have an impact on the fiber properties. Higher humidity can form a finer membrane. On the other hand, a lower humidity increases the fiber diameter [252,257,258,262].

Although extensive research has been done to understand the effect, there is significant space for additional research to reach a better understanding of the possible cause of bead formation and control of the fiber diameter. More comprehensive study is required to control the solution properties in order to understand the effect on electrospinning.

#### **5. Future Directions**

Research on the fabrication process of mixed-matrix membranes is ongoing, as they have been found very useful in different applications. Among all the mixed-matrix membranes, nanofibrous-type MMMs are now more popular due to their properties and efficiency. Among all the spinning processes, electrospinning has some great features like high speed, capability, and low cost, resulting in a highly porous patterned nanofibrous polymer membrane [269,270]. The electrospinning process can fabricate a membrane with a larger specific area with smaller pores and fibers within a diameter of 10 to 1000 nm [271]. These unique properties of electrospun nanofibrous membranes make them desirable for a wide range of applications [272], such as SiO2-incorporated electrospun SPEEK, which

has been applied in a fuel cell [273]. Additionally, during the electrospinning process, it is easy to perform the ordering of the polymer, as well as the chain elongation. Considering all the mentioned characteristics of electrospinning, this process could be taken as the latest effective technology for the production of continuous, long-chain, mixed-matrix nanofibrous membranes using a combination of different polymers and nanomaterials for various applications on a large scale [246].

There is an opportunity for developing new technology combining 3D printing and electrospinning in the nanofiltration area, as has been done for biomedical applications (Figure 17). Recently, a 3D-printed mesh reinforcement on electrospun scaffolds was attempted, in which a poly (lactic acid) (PLA) mesh was 3D-printed into an electrospun poly(ε-caprolactone) (PCL) gelatin directly, resulting in better mechanical properties [274]. In Figure 18, the effect of 3D printing on the electrospun scaffold structure is clearly visible. ε

**Figure 17.** Fabrication of reinforced electrospun scaffolds. Electrospun scaffolds were produced from a 40:60 ratio of PCL:gelatin. The scaffolds were then placed in a 3D printer, and a PLA mesh was deposited onto one side of the scaffold. Two types of 3D-printed meshes were generated: one with a 6 mm distance between PLA struts, and the other with an 8 mm distance between struts (adapted with permission from [274]).

So, it can be concluded that this technique offers the same matrix-like structure with a higher mechanical strength of the electrospun membrane, and these modified and updated 3D-printed electrospun membranes could be used in a new range of membrane applications.

Nanocomposite materials; a combination of graphene, graphene oxides, or metal oxides such as ZnO, TiO2, etc.; and magnetic nanoparticles, etc., could be better alternatives as filler materials in mixed-matrix membranes for various applications such as heavy metal removal, wastewater treatment, desalination, etc., as some previous research has shown that these types of nanocomposite particles excellently combine the properties that they exhibit individually [275,276]. The synthesis route of these nanocomposite particles is simple as well, using methods such as chemical mixing, chemical precipitation, sol–gel techniques, etc. [275–277].

**Figure 18.** SEM images of reinforced electrospun scaffolds. (**A**–**C**) SEM images of the electrospun side of the reinforced scaffolds. The images show a uniform distribution of randomly oriented fibers. (**D**–**F**) SEM images of the 3D-printed side of the scaffolds. The high-magnification images (**F**) show that there is minimal damage to the electrospun fibers in the immediate vicinity of the 3D-printed PLA mesh. Yellow arrows depict the 3D-printed PLA. White arrowheads depict the PCL:gelatin scaffold (adapted with permission from [274]).

Finally, the demand from the end user based on the applications is the main driving force for obtaining a good market value and establishing a better position in the total global membrane market, including pharmaceutical and biomedical, water filtration and wastewater treatment, textile and metalworking industries, chemicals and petrochemicals, food and beverages, etc. So, the demand for such fabrication technology is also at its peak, and a cost-effective and easier fabrication technology is desirable for bulk and industrial production.

**Author Contributions:** Conceptualization, T.S. and N.R.C.; validation, T.S. and N.R.C.; writing original draft preparation, T.S.; writing—review and editing, N.R.C. and N.K.D.; supervision, N.K.D. and N.R.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Acknowledgments:** The authors gratefully acknowledge the financial support of RMIT University for providing a PhD scholarship to Tawsif Siddique (T.S.) to carry out this work. T.S. acknowledges the support of Umma Habiba for his PhD work.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Article* **Impact of Pre-Ozonation during Nanofiltration of MBR Effluent**

**Zoulkifli Amadou-Yacouba , Julie Mendret \* , Geoffroy Lesage , François Zaviska and Stephan Brosillon**

IEM (Institut Européen des Membranes), UMR 5635 (CNRS-ENSCM-UM2), Université de Montpellier, 34095 Montpellier, France; zoulkifli.amadou-yacouba@umontpellier.fr (Z.A.-Y.); geoffroy.lesage@umontpellier.fr (G.L.); francois.zaviska@umontpellier.fr (F.Z.); stephan.brosillon@umontpellier.fr (S.B.)

**\*** Correspondence: julie.mendret@umontpellier.fr; Tel.: +33-467-144-624

**Abstract:** This study aimed to investigate the impact of real MBR effluent pre-ozonation on nanofiltration performances. Nanofiltration experiments were separately run with non-ozonated real MBR effluent, ozonated real MBR effluent and synthetic ionic solution mimicking the ionic composition of the real MBR effluent. The specific UV absorbance and the chemical oxygen demand were monitored during ozonation of real effluent, and the mineralization rate was calculated through the quantitative analysis of dissolved organic carbon. The membrane structure was characterized using SEM on virgin and fouled membrane surfaces and after different cleaning steps. The results confirm the low effect of the ozonation process in terms of organic carbon mineralization. However, the chemical oxygen demand and the specific UV absorbance were decreased by 50% after ozonation, demonstrating the efficiency of ozonation in degrading a specific part of the organic matter fraction. A benefic effect of pre-ozonation was observed, as it limits both fouling and flux decrease. This study shows that the partial mineralization of dissolved and colloidal organic matter by ozonation could have a positive effect on inorganic scaling and decrease severe NF membrane fouling.

**Keywords:** wastewater reuse; organic matter; ozonation; nanofiltration; membrane fouling

#### **1. Introduction**

A promising solution to the challenge of water shortage is to consider urban and industrial wastewaters no longer as wastes but more as renewable resources of water, nutrients and energy. One of the most challenging limiting factors to wastewater reuse is the widespread occurrence of micropollutants in different environmental compartments. To overcome this issue, membrane processes have been demonstrated to remove well micropollutants [1,2]. Among the numerous available membrane processes, nanofiltration is widely recognized for the compromise it offers in terms of selectivity and flux permeability [3,4].

Nonetheless, the fouling propensity remains a very big challenge for a widespread usage of this process [5]. In fact, the nanofiltration membranes seem particularly vulnerable to severe fouling, which constitutes their main drawback. Their propensity to both organic and inorganic fouling was demonstrated by numerous previous studies [6,7]. The reduction of membrane permeability due to fouling causes a substantial increase in operational and maintenance costs and a decrease in effluent quality and membrane lifetime.

As a solution to fouling challenges, numerous authors have investigated the impacts of different types of pretreatment processes such as advanced oxidation. Ozonation, as pretreatment to mitigate the fouling propensity in NF process, is one of the most promising technologies [8–10]. Former studies have pointed out the increase in permeate flux, but very few studies have focused on fouling dynamics in the combined process of ozonation/nanofiltration [11,12]. In particular, there is a lack of data about the specific roles of inorganic salts and organic matter during pre-ozonation of real wastewater before nanofiltration.

**Citation:** Amadou-Yacouba, Z.; Mendret, J.; Lesage, G.; Zaviska, F.; Brosillon, S. Impact of Pre-Ozonation during Nanofiltration of MBR Effluent. *Membranes* **2022**, *12*, 341. https://doi.org/10.3390/ membranes12030341

Academic Editor: Mohammad Peydayesh

Received: 1 March 2022 Accepted: 16 March 2022 Published: 18 March 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

The aim of this study was to analyze the NF process applied to a real MBR secondary effluent. Specifically, it consisted in identifying the role of organic and inorganic matters in fouling mechanisms, monitoring the degradation rate of organic matter by an ozonation process and investigating the impact of pre-ozonation on performances of subsequent NF processes. In particular, the effect of the mineralization rate of organic matter on fouling mechanisms in NF was identified.

#### **2. Materials and Methods**

#### *2.1. Matrix Used for the Study*

#### 2.1.1. Real Secondary Effluent Matrix

An effluent from a full-scale domestic WWTP equipped with MBR, located close to Montpellier, France, was used as real matrix. The plant was designed to treat 13,000 m3/d of domestic wastewater. The MBR was equipped with KUBOTA Submerged Membrane Unit (SMU RW400) (KUBOTA, London, Englandflat-sheet microporous membranes made of chlorinated polyethylene (total surface of 16,240 m<sup>2</sup> ), with an average pore size of 0.2 µm. The characteristics of the MBR permeate are presented in Table 1. SUVA254 is the specific ultraviolet absorbance. The MBR effluent was immediately stored at nearly 4 ◦C after sampling in order to limit the variation of the composition and re-warmed at room temperature (20 ◦C ± 1 ◦C) before conducting the experiments.

**Table 1.** Characteristics of real MBR effluent (*n* = 5).


#### 2.1.2. Synthetic Ionic Solution Matrix Composition

In order to deeply investigate the impact of organic and ionic matter on fouling mechanisms, it was chosen to conduct experiments with matrix free of organic matter. Therefore, a synthetic ionic solution was prepared in ultra-pure water imitating the ionic composition of the real MBR effluent (Table 2).

**Table 2.** Ionic composition of real MBR effluent and synthetic ionic solution.


To prepare the solution with the aforementioned ions, different salts were used in the following concentrations (Table 3).


**Table 3.** Salts used to prepare the synthetic ionic solution.

#### *2.2. Nanofiltration Experiments*

#### 2.2.1. Membrane Selection and Characterization

The membrane used for this study is an NF-90 polyamide membrane from DOW Filmtec. It is considered as a "tight" NF membrane with an estimated MWCO around 150 Da. Before experiments, each membrane was firstly soaked in ultrapure water to remove preservative agent and then compacted at 18 bars for at least one hour or until stability of the permeate flux was reached. Thereafter, the membranes were fully characterized in terms of pure water permeability and sodium chloride rejection, with values corresponding to 8.4 ± 1.0 L h−<sup>1</sup> m−<sup>2</sup> bar−<sup>1</sup> and 88 ± 4%, respectively.

#### 2.2.2. Cross-Flow Nanofiltration Unit and Experimental Protocol

The filtration experiments were carried out with 140 cm<sup>2</sup> flat-sheet membrane samples in an Osmonics Sepa CF II cell (Sterlitech Corp., Auburn, WA, USA). The Sepa cell was fed by a pump Hydra-Cell, Wanner Engineering, Inc, Minneapolis, MN, USA) with the solution from a 16 L feed vessel (Figure 1). The wastewater temperature was kept constant (20 ± 1 ◦C) using a cryothermostat (F32, Julabo, Seelbach, Germany). The bench-scale NF experiments were performed at a cross-flow velocity (vT) of 0.5 m s−<sup>1</sup> with a medium foulant spacer, 47 Mil (1.194 mm). The transmembrane pressure (TMP) was set constant at 10 bars using a micrometric pressure control valve located on the retentate outlet. The membrane performances were monitored throughout the filtration experiment at ~0%, 15%, 40% and 60% until reaching 80% of water recovery (or the maximum water recovery rate reachable in case of earlier severe fouling). The flux was recorded throughout the experiment by measuring the permeate weight every 60 s. Retentate and permeate samples were collected for physico-chemical analysis. The volume of the collected sample for different analyses was considered in the apparent rejection determination. Considering that the NF system is made of stainless steel and all the tubing is in Teflon, it was assumed that compounds (organic and inorganic matter) adsorption was exclusively occurring on membrane material.

To evaluate the impact of organic and inorganic matters on membrane fouling mechanisms, three types of NF experiments were run: (1) non-ozonated real MBR effluent, (2) synthetic ionic solution and (3) ozonated real MBR effluent.

#### 2.2.3. Membrane Fouling Propensity Test

After each filtration experiment, the NF unit was cleaned first by ultrapure water cleaning, then recirculating caustic soda (NaOH, 2%) for 6 h and finally recirculating acid solution (HNO3, 2%) for 6 h. After each base and acid cleaning, the system was fully rinsed with deionized water until a conductivity of 50 µS cm−<sup>1</sup> and a neutral pH were reached in the NF permeate. Membrane fouling was characterized according to the flux recovery after effluent filtration and after different cleaning steps. Reversible fouling was estimated immediately after ultrapure water cleaning by comparison with water flux before the filtration, at the beginning of the experiment. Then, the irreversible fouling was determined using chemical cleaning. Two types of irreversible fouling were distinguished: organic irreversible fouling evaluated by the determination of flux recovery after NaOH cleaning and inorganic irreversible fouling (scaling) determined after acid cleaning (HCl). Flux was measured after these cleaning steps and compared to the initial flux so as to estimate the flux recovery proportion of each type of cleaning.

**Figure 1.** Experimental setup of nanofiltration bench-scale pilot. (**1**) Cryothermostat. (**2**) Mechanical stirrer. (**3**) Tank isolation valve. (**4**) Valve for sampling. (**5**) Pump. (**6**) And. (**8**) Pressure sensors. (**7**) Filtration unit. (**9**) Pressure control valve. (**10**) Conductivity meter. (**11**) Precision scale. (**12**) Data processing.

Membrane surface morphology, for virgin and fouled membranes and after each cleaning step, were characterized with a Scanning Electron Microscope (SEM, Hitachi Table top Microscope S-4800) interfaced with an Energy-Dispersive X-ray (EDX) spectroscopy system (Thermo-Fisher, Waltham, MA, USA). Membrane samples were coated with a thin layer of gold before SEM analysis. EDX measurements were performed at different locations on the membrane surface, in order to obtain a comprehensive elemental composition. SEM micrographs were obtained at an accelerating voltage of 2 kV and magnification of 25,000.

#### 2.2.4. Osmotic Pressure

The difference in osmotic pressure (∆π) between feed and permeate sides of the membrane was calculated using Equation (1) [13]:

$$
\Delta \pi = \left. \pi\_{\text{feed}} - \pi\_{\text{perm}} \right. \tag{1}
$$

with π feed representing osmotic pressure in the feed side and π perm representing osmotic pressure in permeate side.

The NF removal was high, and the ions concentrations (and consequently the induced osmotic pressure) at permeate side were negligible compared to that of feed side.

For each ion, the osmotic pressure is given by Equation (2) [13]:

$$
\boldsymbol{\pi} = \mathbf{C} \cdot \mathbf{R} \cdot \mathbf{T} \tag{2}
$$

For all the identified ions, Equation (3) enables estimation of π [13]:

$$\boldsymbol{\pi} = \mathbf{R} \cdot \mathbf{T} \sum\_{i=1}^{n} \mathbf{C} \mathbf{i} \tag{3}$$

with:

R: gas constant (= 8.314 J/mol K); T: temperature of solution (◦K); C: concentration of ion (mol/m<sup>3</sup> ); n: number of ions in the solution.

#### 2.2.5. Concentration Polarization

Due to concentration polarization, the osmotic pressure is not homogeneous in feed solution. In fact, the ions concentration and the induced osmotic pressure (π) are more important at membrane surface (πmemb) than in the bulk solution (πbulk). These values are linked by the relation given in Equation (4) [13]:

$$
\pi\_{\text{memb}} = \pi\_{\text{bulk}} \cdot e^{\frac{\text{l}^\text{g}}{\text{k}}} \tag{4}
$$

with:

k: diffusion coefficient (m2/s); Jp: flux (m3/s/m<sup>2</sup> ).

#### *2.3. Bench-Scale Ozonation System Setup*

Experiments were performed in a glass stirred batch reactor (Vreactor = 3 L) where the liquid solution is maintained at room temperature (20 ◦C) using a cryothermostat (Figure 2). The ozone was continuously produced from a lab-grade pure oxygen tank by an ozone generator (BMT 803 N). Before diffusion in the reactor, the ozone was diluted with oxygen at a gas flow of 60 L h−<sup>1</sup> and introduced through a porous diffuser at the bottom of the reactor. The gas ozone concentration ([O3]gas,in) was monitored after dehumidification by an ozone gas analyzer (BMT 964). The impact of pre-ozonation on NF process was investigated for 30 min reaction contact time, and the dissolved ozone dose (TOD) was determined using indigo method [14].

**Figure 2.** Experimental setup of ozonation bench-scale pilot.

The desired oxygen/ozone ratio was determined using two electro-valves connected to the monitoring software. The ozone dissolution rate was increased in the solution us-ing an agitator (400 rpm). The experiment consisted in applying an ozone gas concen-tration of 5 gO3/Nm<sup>3</sup> to determine the transferred ozone dose through Equation (5).

$$\text{TOD} = \frac{(\text{C}\_{\text{ge}} - \text{C}\_{\text{ges}}) \ast \text{Q}\_{\text{g}} \ast \text{t}}{\text{V}\_{\text{reactor}}} \tag{5}$$

with:

TOD: transferred ozone dose (gO3/m<sup>3</sup> ); Cge: gas-phase ozone inlet concentration (g/Nm<sup>3</sup> ); Cgs: gas-phase ozone outlet concentration (g/Nm<sup>3</sup> ); Qg: gas flow (m3/h); t: reaction time (h); Vreactor: reactor volume (m<sup>3</sup> ). Finally, the specific ozone dose [O3]specific was calculated with Equation (6):

$$\text{[O}\_3\text{]} \text{specific} = \text{TOD/TOC} \tag{6}$$

with:

TOC: total organic carbon (gC/m<sup>3</sup> )

#### *2.4. Chemical Analysis*

2.4.1. Ionic Chromatography

The concentrations of ionic compounds were determined in all samples by ionic chromatography:


#### 2.4.2. Global Indicators for Pollution Monitoring: TOC, UV254 and SUVA Analysis

The specific UV absorbance (SUVA254) corresponds to the ratio of UV absorbance at wavelength of 254 nm, measured in a 1 cm quartz cuvette using a UV–vis spectrophotometer (UV-2401PC, Shimadzu, Kyoto, Japan) and TOC value [15]. TOC analysis was performed using a TOC-VCSN Shimadzu analyzer (Shimadzu Japan).

#### 2.4.3. Scanning Electron Microscopy (SEM)

A Hitachi Microscope (Hitachi S4800 SEM) was used to inspect surfaces of the virgin and pre-fouled membranes. Small pieces were cut from the surfaces of membranes (postmortem analysis). Before analysis, the samples were dried in desiccator until measurement in order to remove residual moisture and then metalized with platinum. The surfaces of fouled and virgin membrane were magnified 5000–15,000 times.

#### **3. Results**

#### *3.1. Flux Evolution and Fouling Mechanisms during Nanofiltration*

One of the criteria to evaluate NF efficiency is the evolution of the permeate flux with the time of filtration. The recovery rate (Y) was calculated corresponding to the ratio between the extracted permeate volume and initial feed volume. In order to compare the flux evolution for different experiments, the relative flux corresponding to the ratio between the flux at any time (J) and the initial flux (J0) was considered. Figure 3 presents the normalized flux (J/J0) during nanofiltration of real MBR effluent under a TMP of 10 bars.

**Figure 3.** Flux evolution in NF experiment applied to MBR real effluent matrix: TMP = 10 bars, T ◦ = 20 ◦C, J<sup>0</sup> = 53 L·m−<sup>2</sup> ·h −1 , duration of the experiment = 24 h.

Figure 3 revealed a drop of almost 70% in the initial flux value when reaching the maximum conversion rate of 80%. As established in previous studies, the main fouling mechanism during MBR effluent filtration by NF is organic fouling [11]. According to some authors, the reason that could explain the flux drop is that the organic matters, particularly those with higher MW and hydrophobicity, corresponding to humic-like substances, deposited into the pores and onto the membrane [16–18]. The deposited organics enhance gel layer formation, which was related to the rapid flux decline at the first stage. Then, the slower flux decrease could come from gel layer compaction and interactions between inorganic salts and organic matter deposited on the membrane surface [19]. For instance, Lin et al. have studied the roles of organic, inorganic and biological fouling along with NF applied to raw effluent. The authors noticed that organic/inorganic binary fouling became dominant, contributing up to 39.7% of flux decline due to metal/organic complexation [7]. The third stage, corresponding to a more pronounced flux drop, could come from concentration polarization [5,17]. Nonetheless, to establish a clear distinction between the impacts of organic and inorganic contributions to flux decline, it is required to run NF experiments with OM-free matrix.

#### *3.2. Influence of Ionic Matrix during Nanofiltration*

To evaluate the impact of organic matter on fouling, experiments were run with synthetic ionic solution (SIS) mimicking the ionic composition of the real MBR effluent. The flux was monitored along with permeate recovery rate and is presented in Figure 4 with that of real MBR effluent matrix.

The Figure 4 revealed a decline of 75% in permeate flux at 60% of recovery for SIS solution. The occurrence of the severe fouling may be linked to an inner fouling caused by ionic compounds. In fact, as the organic matter playing the role of competitor in ions adsorption is no longer present in solution, the ions are free to adsorb onto membrane surfaces and enhance membrane fouling while diffusing through membrane pores. Thus, it was not possible to reach such high conversion rates as with real effluent (Y = 60% instead of 80%). Teixeira and Rosa have studied the impact of the water inorganic matrix on the permeate flux and the natural organic matter (NOM) removal by nanofiltration [20]. They noticed a decrease in flux in the presence of calcium. According to the authors, the flux and rejection decreased further in the presence of 1 mM Ca2+, which reduced the membrane negative charge and sieving effects and increased chemical interactions. In fact, in the present study, all the detected ionic composition was mimicked by a synthetic ionic solution free of OM that could compete with the membrane in adsorbing the inorganic and mitigate the inorganic fouling.

**Figure 4.** Flux evolution during NF experiment applied to MBR real effluent (duration of the experiment = 24 h) and synthetic ionic solution matrixes SIS (duration of the experiment = 18 h). TMP = 10 bars, T◦ = 20 ◦C, J<sup>0</sup> -SIS = 64 L·m−<sup>2</sup> ·h −1 , J<sup>0</sup> -MBR = 53 L·m−<sup>2</sup> ·h −1 .

3.2.1. Influence of the Osmotic Pressure

During the nanofiltration experiments, the ionic compounds became more and more concentrated and induced an osmotic pressure, which is supposed to increase with permeate recovery. The osmotic pressure constitutes a resistance to physical pressure and should be overcome in order to get permeate flux through the membrane. The differential osmotic pressure between retentate and permeate streams was calculated for real MBR effluent and synthetic ionic solution and compared in Figure 5.

The monitoring of the osmotic pressure revealed that it increases with permeate recovery rate from around 1 bar at the beginning to 2.2 bars at 60% of recovery rate for SIS solution and up to 4 bars for the MBR effluent matrix. The Figure 5 clearly displays a similarity in the evolution of osmotic pressure for both real MBR effluent and the synthetic ionic solution mimicking the MBR ionic composition, even though the permeate flux drastically dropped in the case of SIS much earlier than in the MBR effluent case (Figure 4). This result confirms the suspected inner fouling due to inorganic scaling. As the solution is free of OM, which would adsorb the ions, they are free to interact with each other and with the membrane, enhancing the scaling [21,22].

**Figure 5.** Evolution of differential osmotic pressure in NF for real MBR effluent (duration of the experiment = 24 h) and SIS (duration of the experiment = 18 h). TMP = 10 bars, T◦ = 20 ◦C.

#### 3.2.2. Characterization of Membrane Fouling

At the end of each experiment, the membrane goes through different cleaning steps beginning with ultrapure water, followed by basic and acid-based cleanings, respectively. Scanning Electron Microscopy (SEM) analysis was applied to samples from membrane used for both real MBR effluent and SIS. Samples of virgin and fouled membrane and membrane after the different cleaning steps were used, and the results are presented in Figure 6.

**Figure 6.** SEM of membrane surfaces at different states: (**A**) For membrane used in MBR effluent experiment, (**B**) Membrane used in SIS experiment. 1. Virgin membrane, 2. Fouled after experiment, 3. UPW-cleaned membrane, 4. Base-cleaned and 5. Acid-cleaned.

Figure 6 visually illustrates the membrane surface state throughout the different steps. A mixture of inorganic and dissolved organic matter can be noticed on the fouled membrane used with the real MBR effluent (2.A), while the membrane fouled with SIS (2.B) displays disaggregated inorganic compounds only. The subsequent cleaning methods helped to identify the type of fouling that occurred during these experiments through foulants characterization [16,23]. In fact, for the membrane fouled by real MBR effluent, while the ultrapure-water-based cleaning likely removed part of the fouling matter (3.A), the sodium hydroxide cleaning significantly removed it, except for some inorganics (4.A) that were totally removed by hydrogen chloride acid washing (5.A). For the SIS-fouled membrane, on the other hand, the ultrapure-water-based cleaning was able to remove part of scaling (3.B). The sodium hydroxide cleaning was not able to remove the inorganics on the membrane surface (4.B). Only the acid cleaning totally recovered the fouled membrane surface to almost virgin state (5.B).

#### *3.3. Influence of Pre-Ozonation during Nanofiltration of Real MBR Effluent* 3.3.1. Monitoring of Organic Matter

To evaluate the impact of organic matter and its degradation by ozone on the performances of nanofiltration process, the mineralization rate of organic matter was monitored during ozonation process, and the results are given in Figure 7.

**Figure 7.** Monitoring of the mineralization rate of the real matrix during ozonation. T◦ = 20 ◦C, Vreactor = 3 L, Vstir = 400 rpm, [O<sup>3</sup> ]gas = 5 gO3/Nm3.

Ozonation, as revealed by some previous studies, is not sufficient to completely degrade organic matter [8]. This is confirmed by the current study, in which only a mineralization of 15% was achieved after 30 min (TOC around 8 mg/L). Even though the mineralization rate was relatively low, the ozonation engendered an important change in the organic matter. Indeed, even if the mineralization of the organic matter was moderate, chemical changes occur, and the efficiency of ozonation in terms of modification of organic matter structure was monitored through some common parameters. The chemical oxygen demand and the specific UV absorbance (SUVA254) are some of these indicators (Figure 8).

Both of the two parameters indicate the efficiency of ozonation process in oxidizing organic matter. The COD decreased from 33 mgO2/L to 23 mgO2/L after 3 min reaction time and to less than 20 mgO2/L at 30 min of reaction time (Figure 8a), corresponding to the introduction of oxygen in the chemical structure of the organic matter. This level of mineralization was already observed by Gong et al. and Justo et al. [15,24]. In addition, the ozonation decreased the SUVA by half after 30 min reaction time; this indicates the opening of the double bond mainly in the aromatic group. This parameter is a good indicator of the change in the chemical structure of the organic matter [25,26] (Figure 8b).

**Figure 8.** Evolution of global parameters of the real matrix during ozonation. (**a**): COD, (**b**): SUVA254, T ◦ = 20 ◦C, Vreactor = 3 L, Vstir = 400 rpm, [O<sup>3</sup> ]gas = 5 gO3/Nm<sup>3</sup> .

#### 3.3.2. Nanofiltration of Ozonated Real MBR Effluent

During the NF experiment applied to the ozonated real MBR effluent, the flux evolution was monitored, and the relative flux is displayed in Figure 9 in comparison with non-ozonated real MBR and SIS matrix.

**Figure 9.** Flux evolution in NF experiment applied to synthetic ionic solution (duration of the experiment = 18 h) and non-ozonated (duration of the experiment = 24 h) and ozonated MBR real effluent matrixes (duration of the experiment = 24 h). Ozonation reaction time = 30 min, T◦ = 20 ◦C, Vreactor = 3 L, Vstir = 400 rpm, [O<sup>3</sup> ]gas = 5 gO3/Nm<sup>3</sup> TMP = 10 bars, J<sup>0</sup> -MBR = 53 L·m−<sup>2</sup> ·h −1 , J0 -MBR + O<sup>3</sup> = 54 L·m−<sup>2</sup> ·h −1 , J<sup>0</sup> -SIS = 64 L·m−<sup>2</sup> ·h −1 .

Figure 9 reveals that when the nanofiltration experiment is run with ozonated real MBR effluent, the drop in flux trends is slightly slower, as around 10% of flux was re-covered by pre-ozonation. Even though the ozonation is not efficient in terms of mineralization, it changes the structure of the organic matter [12]. According to the cited authors, preozonation increases the hydrophilic fraction and anionic charge of organics and alters their size distribution [1,12]. In fact, the gel layer (coming from organic and inorganic complexation) was demonstrated to be responsible for membrane fouling. Therefore, the ozonation, by degrading part of this gel layer, leads to improve nanofiltration conditions by reducing the fouling celerity [11].

When the NF experiment was applied to non-ozonated real MBR effluent, the fouling essentially came from complexation of organic and inorganic matter [11]. When the nanofiltration experiment was run with SIS free of organic matter, the drop in flux trends was much more severe and occurred earlier (Figure 9). This demonstrated that during NF experiments, the propensity to inorganic fouling is much higher than that of organic fouling. After ozonating the real MBR, the trend of permeate flux in NF is improved due to delayed fouling, as the ozonation was insufficient to totally mineralize the effluent organic matter. These results demonstrate that the main drawback of the ozonation process, which is its limited mineralization rate, rather constitutes an advantage for a subsequent NF process: the residual organic matter prevents a severe inorganic fouling by competing with the membrane for adsorption of inorganics. For instance, Li et al. have studied the operational optimization and membrane fouling analysis of nanofiltration in municipal wastewater advanced treatment [16]. One of the main conclusions they came to is that inorganic fouling was mitigated because the inorganics were assumed to adsorb on the effluent organic matter.

#### 3.3.3. Cleaning and Nanofiltration Performances Recovery

Two other parameters used to characterize the fouling that occurred during NF experiments are the type of cleaning and the rate of flux that it allowed to be recovered. The values of flux at the beginning and the end of the experiments are recapitulated for all matrixes in Table 4.


**Table 4.** Values of permeate flux at the beginning and end of studied NF experiments.

Hence, after each nanofiltration experiment, ultrapure water was used to clean the membrane, and the permeability was measured. A basic cleaning using NaOH (0.1 N) and acid cleaning using HCl (0.1 N) were successively performed as well. It consisted in imbibing the membrane in the cleaning solution for 6 h for both chemical solutions. The membrane permeability recovery rates were determined for the three studied matrixes and are presented in Figure 10.

According to Figure 9, the SIS induced more severe fouling than the real MBR effluent. Then, ultrapure water cleaning enabled 53%, 69% and 15% flux recovery for MBR effluent, ozonated MBR effluent and SIS solution, respectively. Fouling corresponding to both MBR effluent and ozonated MBR effluent thus seem easier to remove, which is consistent with Figure 9. Sodium-hydroxide-based cleanings allowed non-negligible permeate flux recovery in the cases of real MBR effluent (35%) and ozonated MBR effluent (18%) and a significant flux recovery of SIS-fouled membrane (60%), for which chemicals are needed. The acid-based cleaning allowed the most important recovery for SIS solution, which is consistent with the inorganic nature of fouling in this case. According to Li et al., the water flushing samples after nanofiltration of wastewater were essentially composed of low MW with high intensity, which is typically related to humic substances, indicating that the humic substances could be removed easily by physical cleaning, which is in accordance

with the present result during ultrapure water cleaning [16]. In the case of severe inorganic fouling, the acid-based cleaning is required for flux recovery [17,23,27].

**Figure 10.** Flux recovery after pure water, NaOH and HCl cleanings in NF experiments applied to non-ozonated and ozonated real MBR effluent and synthetic ionic solution.

#### **4. Conclusions**

This research aimed to evaluate the impact of pre-ozonation on fouling propensity in nanofiltration. The fouling was mainly due to organics and inorganics complexation forming a gel layer (70% drop in flux at 80% of permeate recovery). When the NF experiment was run with an organics-free synthetic ionic solution, the fouling was more severe because of the high propensity of NF to inorganic fouling (75% drop in flux at 60% of permeate recovery). When the ozonated real MBR effluent was used for NF experiment, not only was the fouling delayed (62% drop in flux at 80% of permeate recovery), but the flux recovery was improved as well by a mere water cleaning. Therefore, pre-ozonating the effluent presents two advantages: it allows economical use of chemicals needed for chemical cleaning, and it contributes to improving the membrane lifetime by delaying chemical cleaning.

The SEM analysis confirmed that the acid cleaning was the most efficient to recover a virgin membrane state, even though the ultrapure water and basic cleanings can allow recovering an important part of flux, depending on the type of fouling linked to the nature of matrix used for the experiment.

The results demonstrate that the low mineralization rate of ozonation process is of high value to preventing a severe inorganic fouling. It mitigates the organic fouling by degrading partially and modifying the molecular structures of organic matter, which improves its hydrophilicity. On the other hand, the remaining organic matter, which resulted from the partial mineralization, prevented the membrane from a severe fouling, as a total mineralization would lead to occurrence of inorganic scaling. To sum up, ozonation might be the best AOP to couple with an NF process for better organic and inorganic fouling mitigation for wastewater reuse.

**Author Contributions:** Conceptualization, J.M. and G.L.; methodology, Z.A.-Y.; validation, S.B., formal analysis, F.Z.; investigation, Z.A.-Y.; resources, J.M.; data curation, Z.A.-Y.; writing—original draft preparation, Z.A.-Y.; writing—review and editing, J.M., G.L., S.B.; visualization, Z.A.-Y.; supervision, S.B.; project administration, J.M.; funding acquisition, J.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by ANR, grant number SAWARE ANR-16-CE04-0002-01.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**


#### **References**

