A review on microfiltration membranes: fabrication, physical morphology, and fouling characterization techniques

1 Introduction

Microfiltration (MF) is a commonly used pressure-driven separation process with various applications, including particle removal, clarification, water treatment, and biological separation (Anis et al., 2019). Several studies have highlighted the strong and complex relationship between the morphology and fouling behavior of MF membranes (Ho and Zydney, 1996b; Ho and Zydney, 1999a; Hwang and Liao, 2012; Xiao et al., 2014), which remains a significant challenge in their application. Thus, characterizing parameters such as pore size, pore size distribution, pore connectivity, tortuosity, and pore shape is crucial for guiding membrane fabrication to achieve the desired structure for optimal performance. In addition to morphology, characterizing fouling is essential for developing strategies to minimize its impact and for creating predictive models that anticipate when fouling is likely to occur. This review aims to focus on physical morphology and fouling characterization methods for MF membranes.

Over the past three decades, reviews have explored the morphological characterization of porous polymeric membranes, employing various direct microscopic and indirect porosimetric techniques (Nakao, 1994; Sakai, 1994; Khulbe and Matsuura, 2000; Zhao et al., 2000; Chen V. et al., 2004; Yin et al., 2005; Morison, 2008; Jena and Gupta, 2010; Tung et al., 2014; Khulbe and Matsuura, 2017; Johnson et al., 2018; Tanis-Kanbur et al., 2021; Widakdo et al., 2023; Rudolph-Schöpping et al., 2024). Among these techniques, gas liquid porometry is the most frequently used one for characterizing MF membrane morphology owing to its speed and affordability. However, there exists no review covering mathematical models for processing raw data based on the structural features of membranes. Only one review, conducted by Morison (2008) has examined equations for evaluating liquid-liquid porosimetry data. MF membranes exhibit distinct structural features due to alternative manufacturing methods beyond phase inversion. Neglecting these structural features in mathematical models can result in significant errors in determining porosity, pore size, and pore size distribution (PSD) (Islam and Ulbricht, 2019). Thus, the review of the models for the analysis of gas liquid displacement data deserves special attention. Regarding fouling characterization techniques for MF membranes, only a few reviews have addressed this topic comprehensively (Chen V. et al., 2004; Ding et al., 2017; Tummons et al., 2020; Tanudjaja et al., 2022; Huisman et al., 2024), and none have thoroughly covered MF membrane fabrication techniques. This review stands out by bridging these gaps in the literature. It not only provides an in-depth exploration of physical morphology and fouling characterization techniques but also offers insights into fabrication methods for MF membranes. By comparing equations for evaluating gas-liquid porometry data, it contributes to the existing literature.

The review is organized into four sections: 1) fabrication techniques, 2) physical characterization, 3) fouling characterization techniques for MF membranes, and 4) future directions and concluding remarks. The physical characterization section places particular emphasis on the gas-liquid displacement (GLD) method, accompanied by an in-depth review of the mathematical models employed to determine the pore size distribution (PSD) from GLD data.

2 Preparation techniques

Polymeric MF membranes are classified into two categories: “tortuous-pore” structure, where the pores are interconnected through a torturous path, and “capillary-pore” structure, where the pores follow a straight-through path (Baker, 2012). Different preparation techniques including stretching, track etching and phase inversion, are used in the manufacture of MF membranes.

2.1 Stretching

The stretching method involves temperature-controlled uniaxial deformation of the crystalline structure and is thus limited to semicrystalline polymers such as polyethylene (PE), polypropylene (PP) and polytetrafluoroethylene (PTFE). The process begins by creating a precursor film with the desired lamellar morphology through melt extrusion. The film is then annealed at high temperatures to enhance its crystalline structure. Following this, cold-stretching, either uniaxially or biaxially, is performed, usually at room temperature, to induce pore formation through the separation of lamellae. Finally, hot-stretching is conducted at elevated temperatures to further enlarge the pores (Wang et al., 2017). The method controls the porosity, pore size, and mechanical strength of the membranes by adjusting the stretching rate and temperature, as well as by incorporating lubricant during stretching. In the 1980s, Asahi Kasei Corporation introduced the wet process, involving the mixing of ultrahigh molecular weight polyethylene with a hydrocarbon liquid. The mixture is then extruded into a sheet, biaxially oriented, and finally, the liquid is extracted using a solvent (Doi et al., 1975; Doi et al., 1977). Because the wet process involves the use of a solvent and requires post-extraction procedures, leading to higher operating costs, the dry process emerges as a more technologically convenient alternative. Figure 1 demonstrates the morphological changes in the wet and dry processed membranes following biaxial and uniaxial stretching. The dry cast film, resulting from the extrusion of polypropylene (PP) melt with a high draw ratio and rapid cooling, exhibited shish-kebab crystals (Figure 1A). Subsequently, after annealing and uniaxial stretching, continuous slit-like pores developed throughout the bulk structure due to the tearing of lamellar interfaces during stretching (Figure 1B). Figure 1C shows the structure of the ultra high molecular weight polyethylene (UHMWPE) film obtained through extrusion after mixing with plasticizer oil, then cooled within the range from 20 to 40 °C, facilitating solid liquid separation. Consequently, the structure contains UHMWPE crystals and uniformly distributed oil droplets within the cast film.

2.2 Track-etching

In the track etching method, the polymer film is exposed to highly energetic particles to damage its structure. Subsequently, the damaged film is immersed in an etching solution, which dissolves the damaged molecules, thus forming pores in the structure. This method is capable of producing membranes with a very narrow pore size distribution and low porosity. However, it is restricted to polymers with high thermal stability, such as polycarbonate (PC), PP, polyvinylidene fluoride (PVDF) and polyimide (PI) (Apel, 2001). The distinctive feature of the trach etching method is the possibility of adjusting the pore size and number of pores independently. The pore diameter can change from 100 to 1,000 nm’s (Podgolin et al., 2021) (Figure 2). The pore diameter can be adjusted by changing the etchant concentration, temperature, etching time, addition of surfactants and different solvents in the etching solution to enhance the homogeneity and etching rate (Apel, 2001; Kiy et al., 2021; Froehlich et al., 2022).

For track etched membranes fouling was reported to be inverse function of porosity. To enhance the surface porosity and average pore area of track etched membranes while decreasing fouling tendency, a combination of two methods was employed. Essentially, the track etched membrane undergoes uniaxial stretching (Hanks et al., 2007; Worrel et al., 2007).

2.3 Phase inversion

Phase inversion stands as the most commonly utilized technique in the preparation of MF membranes. In this technique, an initially homogeneous polymer solution is forced to separate into polymer lean and polymer-rich phases. Phase inversion can be induced by 4 factors: 1) Coagulation in a nonsolvent bath (NIPS) 2) Evaporation (EIPI) 3) Coagulation in vapor phase (VIPS) and 4) Cooling or heating below/above the critical temperature of polymer solution (TIPS) (Mulder, 1996). Among these methods, in EIPS, polymer precipitation is initiated by solvent evaporation, which delays the polymer solidification process.

Different morphologies are produced by varying the demixing rate. In the case of instantaneous demixing during NIPS process, the polymer precipitates and a solid film is formed very rapidly after immersion in the nonsolvent bath. Instantaneous demixing results in a highly porous substructure characterized by fingerlike macrovoids and finely porous, thin skin layers (Figure 3). When demixing is delayed, the membrane structure becomes spongy. The choice of solvent and nonsolvent significantly influences the morphology of the resulting membranes. Higher miscibility between solvent and nonsolvent leads to instantaneous demixing and a more porous membrane. Additionally, adding solvent to the coagulation bath prevents instantaneous demixing. Furthermore, variations in the composition of the casting solution and coagulation bath temperature can be varied to control the demixing rate and morphology of the membrane.

In the VIPS method, the casted membrane is first exposed to nonsolvent vapor and the diffusion of nonsolvent into the membrane and simultaneous evaporation of solvent induces phase separation. In the following step, the membrane is immersed into a coagulation bath to form the final membrane structure. The most important advantage of VIPS method is the slow phase separation allowing to achieve desirable cellular-like or bicontinuous, uniform pore structure. Evaporation temperature and exposure time to vapor phase are two important parameters affecting the morphology. Figure 4 shows that the asymmetric structure of the PVDF membrane changes from the long fingerlike porous to macrovoid-like structure when the evaporation temperature was increased from 50 C to 80 C. Further increase to 110 C resulted in cellular-like porous structure. All the membranes have large open pores at the top surface and cellularlike porous structure adjacent to the bottom surface with the pore size increasing with the increased evaporation temperature. Increasing vapor exposure time changes the cross section morphology from long fingerlike porous structure to the graded density cellularlike porous structure (Figure 5).

Most polymers, such as polypropylene, polyethylene, and PVDF, processed via TIPS, exhibit a semicrystalline nature. Once a stable dope solution is formed, as indicated by the starting point in Figure 6, the temperature is gradually lowered at a controlled rate, leading to either solid-liquid or liquid-liquid phase separation. At high polymer concentrations, phase separation occurs predominantly through solid-liquid (S-L) separation, wherein the polymer crystallizes directly from the casting solution. In contrast, at lower polymer concentrations, liquid-liquid (L-L) demixing occurs after passing through a metastable region. As depicted in Figure 6, L-L demixing yields a cellular morphology, while spherulitic morphology is attained at the conclusion of S-L demixing.

While phase inversion followed by track etching and stretching methods dominate the production of MF membranes, there are also other methods such as aperture array lithography (Han et al., 2005), selective-etching from the polymer blend (Ilyin et al., 2020; Remanan et al., 2021), phase inversion microfabrication (Gençal et al., 2015), a combination of dip-coating and breath figure approach (Mansouri, et al., 2013) and sacrificial-layer approach (Ji et al., 2010). Table 1 summarizes the methods used for manufacturing MF membranes, while Table 2 lists the polymers utilized in MF membrane preparation via phase inversion, track etching and stretching methods.

TABLE 1. Different methods used for microfiltration membrane preparation.

TABLE 2. Polymers used in the preparation of mf membranes by phase inversion, stretching and track-etching techniques.

3 Physical characterization techniques

Several physical characterization methods have been utilized to determine morphology-related parameters of MF membranes, including pore size, PSD, membrane thickness, and porosity (Zeman and Zydney, 1996). These methods can be classified into two categories: direct and indirect methods. Direct methods include microscopy techniques ranging from SEM, field emission scanning electron microscopy (FESEM), atomic force microscopy (AFM), to transmission electron microscopy (TEM) (Johnson et al., 2018). While direct methods offer advantages like real-time observation, they also come with limitations such as restricted observation area or the need for coating with a conductive material, particularly in the case of SEM and FESEM (Tung et al., 2014). To address these limitations, indirect characterization methods, which rely on theoretical principles to convert measurements into PSDs (Tanis-Kanbur et al., 2021) are widely utilized for membrane characterization. Table 3 summarizes polymeric MF membranes analyzed using indirect physical methods.

TABLE 3. Polymeric MF membranes characterized using indirect physical characterization methods.

The numbers in the parantheses represent the mean pore diameter of the membrane.

C. Alexowsky, M. Bojarska, M. Ulbricht, Porous poly (vinylidene fluoride) membranes with tailored properties by fast and scalable non-solvent vapor induced phase separation, J. Membr. Sci. 577 (2019) 69–78.

PA6: Polyamide6: PA: Polyamide; CA: Cellulose acetate; PES: Polyethersulfone; PVDF: Polyvinylidene fluoride; PET: Polyethylene terephthalate; PP: Polypropylene; MCE: Mixed cellulose ester; PC: Polycarbonate; PSF: Polysulfone; PEEK: Polyether ether ketone.

Among various indirect techniques, Mercury intrusion is commonly used to identify porous materials like catalyst supports and adsorbents, providing details such as specific pore volume and specific internal surface area. However, since Mercury intrusion typically requires very high pressures (Honold and Skau, 1954; Lowell and Shields, 1987), it is not a suitable method for evaluating MF membranes. Capillary condensation, based on the Kelvin equation, provides another means for pore characterization. The method relies on the significant condensation of inert gases like N₂ and He, which predominantly occurs within pores with very fine diameters (Dollimore and Heal, 1964). Thus, the primary application of the method is limited to evaluating the PSD of ultrafiltration and nanofiltration membranes. It is important to note that both Mercury intrusion and capillary condensation methods are based on equilibrium and they do not distinguish between through-pores and dead-end pores. Dead-end pores do not contribute to the transport properties of membranes. Consequently, both Mercury intrusion and capillary condensation methods cannot provide precise information about the PSD of membranes. Based on this fact, Jakobs and Koros (1997) suggest using dynamic methods for determining PSD of membranes.

Liquid-liquid displacement (LLD) is one of the dynamic techniques for membrane pore characterization. Initially, all the pores in the membrane under examination are filled with a wetting liquid, facilitated by capillary forces (Kamide and Manabe, 1980; Kesting, 1985). Subsequently, another liquid is introduced into the pores to gradually displace the initial one. By monitoring the dependence of transmembrane pressure on the flux of the displacing liquid, the PSD can be determined. Peinador et al. (2020) analyzed six commercial cartridges made from PES along with the small membrane coupons detached from these cartridges using LLD and GLD, respectively. The results for the mean pore size obtained from both methods was found satisfactory. However, the LLD and GLD methods yielded different results for the number-based PSD. Implementing LLD presents significant challenges, particularly in selecting two appropriate completely immiscible liquids or liquid compositions to replace each other under incremental pressure increase for a given membrane. Water and butanol are commonly used as a liquid pair in LLD. The interfacial tension between water and butanol is 1.7 mJ/m², making it suitable for characterizing mesopores ranging from 2 to 50 nm in size. However, in the case of MF membranes with pore sizes exceeding 50 nm, typical liquid pairs exhibit insufficient interfacial tensions to generate the wide range of bursting pressures necessary for progressive liquid-liquid displacement within the pores. The surface tensions of gas-liquid systems (17–73 mJ/m²) significantly exceed the typical interfacial tensions of liquid-liquid systems (1–3 mJ/m²). Utilizing an inert gas, such as nitrogen or air, for displacing the liquid within membrane pores can effectively overcome the limitation for LLD method. Over the past few decades, the GLD has become the standard protocol, appearing in various versions of the ASTM standard methodology (ASTM F 3 16-86) and has been utilized to characterize numerous commercial MF membranes (Reichelt, 1991; Zeman, 1996; Jakobs and Koros, 1997). The upcoming sections will first exclusively focus on reviewing the GLD method as the only indirect characterization method, due to its wide acceptance within industry standards. Following this, direct microscopic techniques will be discussed.

3.1 Gas/liquid displacement (GLD) permporometry

GLD method measures the Laplace pressure required to force a wetting liquid out of a capillary and relates this pressure to the pore size corresponding to the largest pore. During a typical measurement, liquid A is allowed to swell the membrane and gas pressure on one side of the membrane is incrementally increased until the first bubble or measurable gas flow on the opposite side is observed. A pressure differential, ∆P=P _*1* -P _*2*, develops at the interface and Young–Laplace equation relates this pressure difference to surface tension of the liquid filling the pores (γ) and pore diameter (d _*p*).

Δ P = 4 γ d p Cos θ (1)

Equation 1 is derived from the mechanical force balance at the curved surface, assuming that a) a single hemispherical meniscus exists at the discharge of liquid/gas from the membrane b) the pore shape is circular c) the contact angle corresponds to the static contact angle b) kinetic effects during discharging of liquid from the pores are negligible. However, the validity of these assumptions has been criticized in various studies. For instance, Adamson (1976) explored a more complex scenario involving a non-spherical meniscus. Chatterjee (2008) showed that relying on a single meniscus to estimate the capillary radius from the measured liquid extrusion pressure can result in significant errors, thus, a correction factor is necessary to account for two menisci. GLD is a dynamic process (Blake and Haynes, 1969), and Lavi et al. (2008) demonstrated that neglecting inertia and dynamic contact angle effects could result in significantly inaccurate estimation for the capillary radius. In many real membranes, pores are not always circular and can be better approximated by other geometries. Zeman (1996) demonstrated that kinetic effects are far from negligible and should be taken into account in the analysis of GLD porosimetry data. He rederived the expression for bubble point pressure to take into account kinetic effects as follows:

Δ P = 4 γ d p + 8 η A + η B 2 0.5 L ρ 0.5 d p (2)

Equation 2 was initially proposed by Bechhold et al. (1931) and considers the effects of commonly ignored parameters, the membrane thickness (L), presure ramp ρ = d P d t, and the viscosities of the wetting fluid (η_A) and gas (η_B). Notably, the second term in Eq. 2 represents the kinetic effect, which can only be eliminated by extrapolating the GLD porosimetry data to ρ = 0. In the literature, either the Young–Laplace equation or its simplified form for complete wetting of the pores (Cantor equation) is frequently utilized to characterize MF membranes (Bottino et al., 1991; Miller and Tyomkin, 1994; Calvo et al., 1995; Rebenfeld and Miller, 1995; Hernández et al., 1996; Lee et al., 1997; Li et al., 2006; Sanz et al., 2006; Yu et al., 2010; Gribble et al., 2011). However, these equations can only provide information about the maximum pore size within the pore distribution, which correlates with the minimum pressure required to initiate the first observed air bubble. The PSD of the membrane is determined by measuring gas flux through a dry membrane first, followed by a wet membrane, as a function of increasing pressure (Islam, and Ulbricht, 2019). The accuracy of the measurement depends on the experimental conditions and theoretical approach employed to convert raw data into the PSD. The pressure ramp applied during experiments should be carefully adjusted, and efforts should be made to eliminate nonequilibrium effects while collecting the experimental data. Zeman (1996) reported the impact of pressure ramp on the PSD of the approximately 0.2 µm-rated MF membranes, including an isotropic cellulose acetate (CA) membrane, and an anisotropic PVDF membranes. Pressure ramp rates applied for the GLD were ρ = 0.1, 0.2, 0.5, and 1.0 psi/s. He reported that the PSD strongly depend on the pressure ramp for both membranes and the sample orientation for the anisotropic PVDF membrane. The ramp rate effect was attributed to the distribution of tortuous pore lengths, the relatively extended duration needed for liquid expulsion from the pores and great difference in the viscosities of the two fluids.

Commercial capillary porometers measure gas fluxes through both dry and wet membranes. The software within the porometer estimates various parameters, including mean flow pore pressure and pore diameter, bubble point pressure and bubble point pore diameter, and illustrates integral and differential flow distribution versus pore diameter curves. However, it is unable to provide the PSD of the membrane due to the requirement for reliable mathematical models. The GLD method is an indirect method that relies on combining the raw data with a mathematical model, typically containing adjustable parameters. Due to various pore structures found in MF membranes, ranging from capillary pores with straight-through paths to capillaries of different length, diameter, shape, roughness and regularity, interpreting the data accurately poses significant challenges. If the model lacks sufficient information about the critical characteristics of the pore structure, it may fail to yield meaningful and precise insights into the material’s morphology.

The subsequent section provides a review of the models utilized for determining PSD from GLD porometer data.

3.1.1 Hagen-Poiseuille model

The Hagen-Poiseuille model, shown by Eq. 3, is commonly used to predict flux (J) through porous membranes. The main assumptions utilized in deriving this equation are as follows: 1) All pores are cylindrical and parallel to each other, following a straight path. 2) Fluid is Newtonian and flow is in Laminar regime. 3) There is no slipping at the pore wall, i.e., the fluid velocity at the pore wall is zero.where n _*i* is the pore-number density of the ith class of pores with diameter d _*i*. In the case of non-parallel pores, the equation is modified by multiplying the L with the tortuosity (τ) of the membrane (L e f f = L τ.

J = π ∑ n i d i 4 128 η − Δ P L (3)

Zeman (1996) justified the assumption of laminar flow for modeling flow through MF membrane pores. For a capillary with a diameter of 0.7 µm and a length of 370 μm, he determined the flow velocity to be 0.6 cm/s and the corresponding Reynolds numbers for the flow of isopropanol (η = 2.43 × 10⁻² P) and air (η = 1.8 × 10⁻⁴ P) to be 7 × 10⁻⁴ and 1.4 × 10⁻⁴, respectively. These values are typical for GLD experiments and for MF membranes. For the fastest, dry membrane air flow, the Reynolds number is on the order of 5 × 10⁻², which is still low and flow is Laminar (Zeman, 1996).

3.1.2 Mixed Poisseuille Knudsen flow model

Gas transport through a capillary depends on the ratio of the pore size to the mean free path of the gas used as the displacing fluid in GLD experiments. When the mean free path of the gas molecules is greater than the pore diameter, the Hagen-Poiseuille equation loses its validity. In this scenario, it is not reasonable to assume zero velocity at the capillary wall. The flow under this condition is known as Knudsen flow and is governed by the collisions of the gas molecules with the wall (Mulder, 1996).

According to gas kinetic theory (Shao et al., 2004), the mean free path of a gas, λ , is expressed by Eq. 4.where p and p ₀ are gas pressures indicated by gauge and the atmospheric pressure, respectively; k is the Boltzmann constant. The extent of Knudsen diffusion is influenced by the surface tension of the pore-filling liquid. When water (with a surface tension of 72.8 mJ/m² at 20°C) is employed as the pore-filling liquid and nitrogen as the displacing gas, the contribution of Knudsen diffusion was estimated to be approximately 5%. This percentage rises to about 15% when wetting liquids with a surface tension value of around 18 mJ/m² are utilized (Shao et al., 2004). The calculations demonstrate that pure Knudsen diffusion alone cannot control gas flow through MF membranes. Instead, both Hagen Poiseille flow and Knudsen diffusion contribute to the gas transport. For this mixed flow regime, a correction term was added to the Hagen Poiseille equation, as utilized in the study of Islam et al. (2020): where u ¯ m is the average molecular velocity of the gas, P ¯, is the average pressure of the gas at the inlet and exit of the capillary, M _*w* is the molecular mass of the gas, R is the universal gas constant. The coeffiicent S is estimated by the expression.where ν = π / 4 , 3 π / 16 , 1.

λ = 1 π 2 k T p + p o 1 d 2 (4)

J = u ¯ m P a π ∑ S i n i d i 3 12 + π ∑ n i d i 4 128 η p − Δ P L (5)

u ¯ m = 8 R T π M w p = P ¯ P a

S = ν + K n 1 + K n K n = λ d p

Hernández and his group (Hernández et al., 1996) determined the pore size distribution of track-etched PC MF membranes using a combination of Poiseuille and Knudsen flow, assuming an S value of unity in Eq. 5. To validate the accuracy of the model, they calculated the pure water permeabilities based on the determined PSD as follows:

L p , w = π ∑ n i d i 4 128 η L (6)

They found that the PSD determined using the mixed Poiseuille and Knudsen flow accurately predicted permeability values that closely matched the experimental data.

Islam et al. (2020) rearranged Eq. 5 as and defined gas permeability as ξ = J − Δ P / L

J = S a v g u ¯ m P a π ∑ n i d i 3 12 + π ∑ n i d i 4 128 η p − Δ P L (7)

S a v g = ∑ S i n i d i 3 ∑ n i d i 3

They tested the validity of the mixed Poiseuille and Knudsen flow model by first fitting the gas permeability data for PES, PP, PVDF, and PTFE membranes to the model to adjust the S_avg value in Eq. 7. The mixed Poiseuille Knudsen flow model accurately predicted the permeability data of the PTFE membrane, while its predictions deviated from the experimental data for the other three membranes. The deviation was attributed to the assuming a constant average S value. Given the known S_avg value and the PSD, flux through dry and wet membranes were predicted using 2 approaches. In the first approach, the viscosity, η, was replaced by air viscosity, η _*a*, and the corresponding value of K _*D;g* was estimated K D , g = π ∑ n i d i 4 128. In the second option, the Darcy coefficient, K _*D;g*, describing gas permeability was equated to the Darcy coefficients K _*D;l* obtained from the data collected with the wet membrane. The predictions demonstrated that the wet and dry flux data for the PTFE membrane were accurately predicted with the mixed Poiseuille Knudsen flow model. However, for the PVDF data, the predictions deviated significantly from the experimental data. This difference can be attributed to the distinct morphological features present in these membranes. While the PTFE membrane exhibits a network structure where fine fibrils are interconnected to form nodes, the PVDF membrane has a sponge-like structure.

3.1.3 Klinkenberg model

Klinkenberg model describes the gas flux through a dry membrane, J _*d*, with one adjustable parameter, A, and constant viscosity as follows:

J d = π 128 ∑ n i d i 4 η k P ¯ P a + A P ¯ λ P a ∑ n i d i 3 η k − Δ P L (8)

The first term in Eq. 8 represents viscous contribution while the second term accounts for the slippage effect. The parameter A is determined by minimizing the difference between experimental and predicted gas permeabilities ξ = J d / − Δ P / L π / 128 = ∑ n i d i 4 η k P ¯ P a + A P ¯ λ P a ∑ n i d i 3 η k. Islam et al. (2017) applied this model to estimate the gas permeabilities for the PET track-etched membranes, but the fitting was not satisfactory leading to deviations between the experimental and predicted dry and wet flow curves. The disadvantage of the Klinkerg model is the presence of an adjustable parameter, A, hence, determining the PSD requires an iterative solution. The wet and dry flow curves were generated with the fitted A value and the estimated value of the specific pore numbers, n_i/η_k, for given diameter d _*i*.

3.1.4 Forchmeier model

The Forchheimer model considers local turbulence effects and gas compressibility, with viscosity assumed to be constant. The steady-state gas flux through a dry membrane is described by the following equations:

J d = − η K D + η K D 2 + 4 P ¯ m P a − Δ P L β ρ a / 2 β ρ a (9)

Like Klinkerberg model, the Forchheimer model also contains one adjustable parameter, β . To fit this parameter, experimental and predicted flux data were plotted by rearranging Eq. 9 as follows:

Π F = η K D J d + β ρ a J d 2

Π F = P ¯ P − Δ P L

K D = π 128 ∑ i = 1 n n i D i 4

The Forchheimer model provided a better fit of the gas flux data for all three PET membranes compared to the Klinkenberg model (Islam et al., 2017). However, the predicted dry flux curve deviated from the experimental data.

3.1.5 Variable viscosity Poiseuille (VVPM) model

In purely viscous or mixed flow models, fluid is assumed to be Newtonian., i.e., the viscosity is constant. This assumption is valid at low pressures (Bird et al., 2001), however, during transport in confined spaces such as capillaries, the viscosity becomes pressure dependent even if the pressure is low (Creutz, 1974). Islam et al. (2017) proposed following equation for steady dry gas flow with a pressure dependent viscosity.where η_App represents the apperant viscosity at the applied pressure. They determined the pressure dependence of viscosity by evaluating the value of K _*D* /η _*App* for all the applied pressure during flow through dry membrane. Assuming that apparent viscosity is the same for both the dry and wet flow at any pressure, the gas flux through wet membrane, J _*w*, is given by the following relation:

J d = K D η A p p P ¯ P a − Δ P L (10)

J w = K w η A p p P ¯ P a − Δ P L K w = π 128 ∑ n i d i 4 (11)

The viscous flow modified with variable viscosity accurately predicted the experimental dry and wet flux data of track-etched PET membrane without any deviation (Islam et al., 2017). The VVPM model also accurately reproduced wet and dry fluxes of PES and PP membranes produced by phase inversion technique. However, the estimated porosities exceeded unity (Islam and Ulbricht, 2019). To resolve this issue, the VVPM model was modified by including 2 additional parameters, non-uniformity and tortuosity coefficient into the definition of K _*D* as shown below (Islam and Ulbricht, 2019).

K D = π 128 ∑ n i d i 4 ω + 1 4 τ (12)

The term ω + 1 τ characterizes irregularity of the pores where ω = 0 for uniform circular pores, while τ = 1 for straight-through pores. With the modified VVPM model, the predicted porosities for the PES and PP membranes perfectly matched with the experimental values.

3.2 Microscopic characterization techniques

Microscopy techniques provide direct visual information about membrane morphology, including surface pore shape and size, their distributions, pore density, surface porosity, and cross-sectional structure through image processing of micrographs.

3.2.1 The scanning electron microscopy

The SEM easily scans the membrane surface, enabling imaging of wide areas within a magnification range of up to 100,000. The resolution limit is about 0.1 μm for regular SEM instruments and 2.5 nm for the most advanced high-resolution instruments employing a field-emission gun (field-emission scanning electron microscopy, FESEM). The main limitation of the SEM is the requirement for membranes to be conductive since nonconductive samples cause charge accumulation, leading to scanning faults and image artifacts. To prevent overcharging effect, the membrane should be coated with an ultrathin layer of an electrically conductive material. However, this coating can hide fine sample features when observed at very high magnifications (Zhang et al., 2014). Except the usage of environmental SEM, which allows imaging in wet conditions, the SEM requires dry samples. Both coating and drying of the samples can change the pore size and PSD. Image processing software allows quantification of parameters such as membrane thickness, and selective layer thickness in two dimensions (2D) (Masselin et al., 2001; She et al., 2008). However, 2D imaging cannot quantify blind and closed pores, nor can it determine parameters such as inner specific surface area, 3-D pore size distributions, tortuosity, spatial pore arrangement, and connectivity of pores which can only be accurately assessed using 3D data.

3.2.2 Focused ion beam-scanning electron microscopy (FIB-SEM)

Uchic et al. (2011) highlighted the benefits of automated serial sectioning instruments to obtain 3-D data from SEM images. The first step in the analysis is to perform sectioning by a variety and/or combination of methods, such as cutting, grinding, polishing, laser ablation, and ion sputtering. In the second step, microstructural data from each section is collected and finally in the last step, computer software programs are used to reconstruct a 3-D data from the image analysis. Up to now, only two studies have utilized this technique for characterizing MF membranes. The primary challenge in the FIB-SEM method is the presence of the shine-through artifact in 3D reconstructions. To evaluate the impact of this artifact, Roberge et al. (2022) conducted the analysis using both an empty membrane and an embedded membrane, the latter prepared by filling the empty pores of the membrane with a resin. 3D data for the PES membrane with an average nominal pore diameter of 0.1 µm (Koch membrane systems, United States) were generated by first obtaining 8 large volumes and then extracting three smaller volumes of identical sizes, referred to as triplicates, from each volume, focusing on areas near the surface. Pore sizes, volume porosity, pore tortuosity and connectivity of the membrane as a function of depth were determined by analyzing triplicates. Each triplicate was subdivided into smaller volume layers, with their top and bottom surfaces parallel to the membrane surface. Each layer had a thickness of 300 nm chosen to be at least three times larger than the nominal pore size provided by the manufacturer (100 nm). The layers overlapped with the next by 50%, ensuring full coverage of the entire membrane structure and the measurements spaced apart by 150 nm. Finally, the mean pore diameter, tortuosity and connectivity of the pores were determined from automatic pore segmentation. The mean pore sizes and porosities for the embedded PES membrane (d_p = 105 ± 20 nm and = 22 ± 2 % ) were estimated to be higher than the values determined from 2D images (d_p = 90 ± 10 nm and ε = 6 ± 2 %). In contrast, for the empty membrane, predicted values from 3D images were lower (d_p = 60 ± 7 nm and ε = 5 ± 2 %). The reliability of the parameters determined from 3D analysis was evaluated by comparing the permeabilities (K) calculated from the Hagen-Poiseuille equation K = ε d p 2 32 τ with experimental values. For empty membrane, the permeability was overestimated by a factor of 50 (Measured flux: 250 L/m². h and calculated flux: 13.500 L/m².h). Significant difference between the permeabilities was attributed to the insufficiency of the Hagen-Poiseille model in predicting the morphology of the PES membrane. The authors concluded that microscopic data obtained from FIB-SEM images can initiate developing more accurate models for describing transport through porous membranes. The most important challenge for the FIB-SEM method is long time required for generating optimum 3D images and small focused volume of the analyzed membrane (∼1–10 μm³). The method is particularly challenging for polymeric membranes due to their poor conductivity, and sensitivity to ion and electron beam damage. In addition, the method is expensive and issues such as charging, cross-sectioning artifacts, and shadowing can complicate the accurate analysis of data. Recently, Chamani et al. (2023) developed a novel approach based on data science tools to reconstruct the 3-D structure of an ultrafiltration membrane using a single 2-D cross-sectional image from a 3-D tomographic data set. The difference between experimental and predicted properties such as mean pore radius, mean throat radius, coordination number and tortuosity was less than 15%. The approach seems to be promising but needs to be validated for MF membranes as well. Prochukhan et al. (2024) analyzed polymethyl methacrylate (PMMA) membrane, but they did not employ resin embedding into the structure due to low tensile properties of the membrane. Figure 7 outlines the initial image preparation steps along with the pore size distribution. From the images, the porosity of sample volume (16,682.4 μm³) was measured as 70.50% ± 4.50%. by setting intensity to 0 (black) for the pores and 255 for the membrane structure (white). The PSD determined from the images yielded an average pore diameter of 1.68 μm with a sample deviation of 1.10 μm. On the other hand, Mercury intrusion porosimeter (MIP) simulation resulted in a mean diameter of 1.10 μm. Additionally, 3D images allowed for the deteremination of pore connectivity and tortuosity of the membrane.

3.2.3 Transmission electron microscopy (TEM)

Transmission electron microscopy scans the surface in 2D with a resolution higher than SEM. However, the sample thickness cannot exceed 100 nm, thus, before the analysis, membrane should be embedded in a crosslinked resin to prepare thin cross sections (Sai et al., 2013). The choice of embedding medium is critical as it should have no influence on the membrane structure. Transmission electron microscopy tomography can provide 3D images, but the procedure is tedious and time-consuming (She et al., 2010). Moreover, it is not necessary since the resolution of the FIB-SEM method is sufficient for 3D imaging of MF membranes.

3.2.4 Atomic force microscopy (AFM)

Scanning probe techniques, such as atomic force microscopy (AFM), enable the examination of sample topology and surface porosity (Bowen et al., 1996; Hilal and Johnson, 2010; Kochkodan et al., 2014). One of the key advantages of AFM is its ability to operate without the need for vacuum conditions or sample coating. Additionally, measurements can be conducted in both air and liquid environments. By scanning the surface of a porous polymer membrane, the dimensions of individual pores can be measured, enabling the calculation of the mean pore size (Bessieres et al., 1996). If a sufficient number of pores are present, the pore size distribution can also be determined (Bowen et al., 1996). Otherwise, the method can yield erroneous results. Another limitation of AFM for the physical characterization of MF membranes is its inability to access underlying layers of the sample.

3.2.5 Confocal laser scanning microscopy (CLSM)

Confocal laser scanning microscopy (CLSM), frequently utilized in biological sciences, has also been used for membrane characterization (Charcosset and Bernengo, 2000a; 2000b; Charcosset et al., 2000; 2002; Marroquin et al., 2011). Images are obtained through optical sectioning from the bulk of the sample. Thus, compared to SEM and TEM which require mechanical sectioning, CLSM is a less invasive technique and provides reliable information about the internal morphology without introducing artifacts from sample sectioning. 3-D imaging requires staining the membrane with a dye, however, sample preparation is still simpler than SEM and TEM since staining is done only once for each membrane. The technical limitation of the CLSM is its resolution. In optimal conditions, CSLM can resolve features as small as 180 nm in the focal plane (x, y), and approximately 500–800 nm along the optic axis (z) (Hell, 2003). However, this resolution is enough for characterizing surface porosity of MF membranes. Green et al. (2006) used CLSM and 3D image reconstruction to visualize the connectivity of pores resulting from the bi-axial stretching of CaCO₃/polyolefin composite membranes and investigated the effect of low and overstretching on the structural features.

3.2.6 Tomographic X-ray imaging

Tomographic X-ray imaging is viewed as a promising microscopic characterization technique, offering insights in 2D/3D or 4D, on membranes, membrane modules, and membrane processes on a scale ranging from micro-to nanometers (Remigy et al., 2007; Schöpping et al., 2024). In tomographic imaging, the membrane sample is rotated over 180 or 360^◦, with respect to the X-ray beam propagation direction. Then, a 3D volume of the sample is reconstructed from the projection images (Figure 8).

The technique is non-invasive and non-destructive and gives a chance to image 3D internal structure of porous membranes in high resolution to evaluate important geometrical features such as porosity, tortuosity and pore size without need for physical sectioning. A major challenge in utilizing tomographic methods is the availability of X-ray sources for conducting tomographic experiments, whether in a laboratory setting or at a synchrotron light source. Additionally, after conducting the experiment, the projection images must undergo post-processing and reconstruction. Subsequently, the reconstructed tomographic slices of the sample require processing, analysis, quantification, and visualization. A recent review discusses opportunities, challenges, and limitations of tomographic X-ray imaging techniques for membrane characterization (Schöpping et al., 2024). A summary of literature findings regarding the characterization of MF membranes using microscopic techniques is provided in Table 4.

TABLE 4. Polymeric MF membranes characterized using microscopic methods.

4 Fouling characterization techniques

In early studies, membrane fouling was monitored through following permeate flux, feed channel pressure drop, and hydraulic resistance during filtration (Velasco et al., 2003; Buetehorn et al., 2010; Wu et al., 2019; Hube et al., 2021). However, these approaches cannot provide insights into the structural and mechanical properties of fouling layer and how these properties impact operational performance.

4.1 Scanning electron microscopy

Over the years, numerous ex-situ and in-situ invasive and non-invasive techniques have been developed for characterizing fouling in MF membranes. The SEM has been utilized for visualizing fouling and the technique can help determining fouling mechanism through measurements of porosity, thickness and pore size before and after filtration. The primary limitations of the technique include extensive sample preparation, which may risk damaging the foulant layer, and the small imaged area, which cannot offer an accurate representation of the fouling scenario.

4.2 Atomic force microscopy (AFM)

AFM can be operated noninvasively with the noncontact mode, its limited image size, similar to SEM imaging, restricts the amount of information that can be obtained from this technique. The roughness measurement for the clean and fouled membrane gives a quantitative measure of degree of fouling (Hilal et al., 2009). In addition, the pore size, surface porosity, and PSD of both clean and fouled membranes can be determined from a 3D map of the surface topography (Woo et al., 2017). A recent review discussed the in situ measurement capability of AFM for membrane fouling (Wei et al., 2024). Evapoporometry (EP) was used as another ex-situ technique to determine extent of fouling by characterizing the PSD of fouled membranes (Han et al., 2020).

4.3 The streaming potential

The streaming potential is a simple yet powerful method for in-situ monitoring of deposition on the membrane surface (He et al., 2017; Nakamura et al., 2012a; 2012b; Le Bolay and Ricard, 1995; Jia et al., 2015). Fouling alters the streaming potential of the membrane, enabling sensitive monitoring of membrane fouling. However, for the technique to effectively characterize fouling, there must be a substantial difference between the charges of the foulants and the membrane. Nakamura et al. (2012a) found that the change in the responses of zeta potential of membrane during filtration can indicate the location of pressure drop and differentiate between the cake filtration and the pore blocking filtration. Jia et al. (2015) monitored flux and zeta potential of the PVDF hollow fiber membrane during yeast filtration. They observed a rapid decline in zeta potential during the pore blocking process, followed by a gradual and linear decrease during cake formation. Conversely, it remained relatively stable during cake compression.

4.4 Electrical impedance spectroscopy (EIS)

Electrical impedance spectroscopy (EIS) is another noninvasive and in-situ technique used for monitoring fouling in MF membranes. The main disadvantage of the technique is the requirement to bond a metal film onto the membrane surface, which may alter the membrane structure. In addition, the method is limited to monitoring fouling at early stage (Jørgensen et al., 2023). Bannwarth et al. (2016) recorded electrical impedance during filtration of silica particles through flat sheet and hollow fiber polypropylene MF membranes. They determined the filter cake height and filter cake porosity by combining the measured impedance data with an equivalent circuit model.

4.5 The ultrasonic time domain reflectometry (UTDR)

The ultrasonic time domain reflectometry (UTDR) technique has also been utilized as an in-situ, non-invasive and real-time technique to observe fouling in/on MF membranes (Li et al., 2002; Li and Sanderson, 2002; Sikder et al., 2006; Silalahi et al., 2009; Xu et al., 2009; Kujundzic et al., 2010; Li et al., 2012; Lin et al., 2013; Li et al., 2014a; Li et al., 2014b; Jørgensen et al., 2016). The UTDR technique offers a unique advantage as one of the few feasible techniques for optimizing commercial-scale membrane modules and processes to mitigate membrane fouling. Furthermore, unlike optical techniques, it is capable of being used with opaque mediums (Rudolph et al., 2019). However, if the density of the fouling layer is comparable to or lower than that of the membrane, separating the fouling layer from the membrane becomes challenging. Morever, the propagation of acoustic waves in multilayer membrane structures should be understood to enhance the accuracy and utility of the complex ultrasonic spectra. Li et al. (2002) demonstrated the capability of the UTDR technique in distinguishing individual modes of fouling layer growth on nylon MF membranes. Silalahi et al. (2009) reported that it is possible to distinguish fouling due to cake layer formation, adsorption and compaction using the UTDR technique. Xu et al. (2009) transformed ultrasonic spectra into two dimensional (2D) color contour diagrams to obtain fouling layer thickness and amount of oil deposited at three different parts of the PES hollow fiber MF membrane.

4.6 The direct observation (DO) technique

The direct observation (DO) technique is another noninvasive method used for in-situ observation of foulant deposition under a microscope. To successfully apply this method, the membrane or feed solution should be sufficiently clear (Chen V. et al., 2004). The DO techniques are straightforward, affordable, and readily available for monitoring membrane fouling. However, they suffer from low resolution and therefore cannot detect small-sized foulants. Foulant deposition on the membrane surface can be detected by positioning the observation window on the permeate side of the membrane module, given that there is significant contrast between the deposition and the membrane. However, fluorescent labeling can enhance this contrast. With the use of DOTM, only the initial stages of deposition can be observed since the view becomes obstructed when a monolayer of particles on the membrane surface is formed (Ngene et al., 2010). The DO method can facilitate estimating cake height and investigating the effects of hydraulic conditions (Ye et al., 2011) chemical environment (Xu et al., 2015), particle surface charge (Lorenzen et al., 2016), and membrane structures (Çulfaz et al., 2011) on foulant layer growth, structure, and compression/relaxation.

4.7 Optical coherence tomography (OCT)

In recent years, optical coherence tomography (OCT) has received significant attention as an in-situ nondestructive technique for monitoring fouling (Hube et al., 2021; Lay et al., 2022). The technique has the capability to generate 2D and 3D scans rapidly, enabling observation of fouling on the milimeter scale without the need for staining. The OCT technique was initially used for in-situ monitoring of biofilm formation in a capillary cell (Haisch and Niessner, 2007; Derlon et al., 2012). Biofouling represents a significant form of fouling, particularly widespread on microfiltration (MF) membranes, especially within membrane bioreactors (MBRs). The process of biofilm formation begins with the accumulation of individual cells, which then aggregate to form initial cell clusters. These microscopic clusters further coalesce to form mesoscopic structures, ultimately leading to the development of macroscopic biofilm structures. Figure 9 illustrates various scales of biofilm formation along with corresponding techniques utilized for analyzing biofilm structure. Morgenroth and Milferstedt (2009) emphasized the significance of mesoscale structural properties in capturing ecological mechanisms within biofilm formation. In this regard, compared to CLSM, OCT offers greater advantages in understanding the structure-function relationship, as it provides detailed visualization of the mesoscale. OCT has been applied to characterize fouling of different commercial MF membranes (Hube et al., 2021; Lay et al., 2022). Trinh et al. (2018) developed an algorithm to use OCT for in-situ monitoring of internal fouling by oil emulsions non-invasively during the filtration process. Han et al. (2020) applied the OCT to compare the internal fouling behavior of 3 MF membranes, PVDF, PTFE and mixed cellulose ester (MCE) membranes, with the same average pore sizes. Although they were able to monitor internal fouling in real-time, the quantitative determination of fouling extent for the 3 membranes was not possible due to differences in membrane material and the complex boundaries between the liquid-foulant and membrane. Ranieri et al. (2024) monitored in-situ biofilm formation during 20 days of seawater filtration through PVDF MF membrane. They quantified the biofilm’s amount and evaluated its impact on hydraulic resistance using OCT scans. In the second phase of the treatment, filtration was stopped for 6 h to facilitate the back transport of foulants from the membrane surface, allowing biofilm relaxation. Additionally, in the absence of permeation, shear stress was applied through air scouring for 15 min to enhance biofilm removal. 2D OCT scans were obtained before and after relaxation and air scouring to determine changes in biofilm thickness and the extent of relaxation necessary for cleaning. Zhang et al. (2024) collected OCT scans at 5-s intervals to characterize fouling layer growth on the PVDF MF membrane. They determined the fouling layer thickness and estimated foulant adsorption and desorption by comparing digitalized cake layers sampled by OCT at two consecutive times. Lay et al. (2022) used OCT to investigate the effect of foulant shape on the fouling characteristics of PC membranes. Han et al. (2018) employed 3D OCT image analysis to investigate the impact of the surface charge of monodisperse latex particles on cake formation. They determined the fractions of fouled voxels for each layer and topology of the fouling layer by analyzing the number of adjacent fouled layers. A recent review presents a detailed discussion on the application of OCT for noninvasive monitoring of fouling in membrane processes (Huisman et al., 2024).

4.8 Confocal scanning laser microscopy

Confocal scanning laser microscopy has primarily been utilized for characterizing biofouling of MF membranes (Wu et al., 2019). The main advantage of this technique is its ability to distinguish between different species and visualize, in real-time, the adsorption-desorption processes occurring at various depths within the membrane (Ferrando et al., 2005). The technique can also enable the visualization of cake formation on the membrane’s surface and/or pore constriction resulting from the deposition or adsorption of foulant agents. Tow et al. (2022) illustrated the in-situ formation of biofilm within large pores beneath the membrane surface, a finding supported by ex situ SEM images. They observed distinct biofilm conformations through in situ and ex situ CLSM analysis, underscoring the significance of the analytical approach. Despite the capability of the CLSM for wet state imaging and locating and identifying the foulants within the sample, its primary limitation is the short penetration depth (Rudolph et al., 2019). Marroquin et al. (2014) addressed this limitation by developing a cross-sectional CLSM imaging protocol capable of generating defect-free images across the entire thickness of membranes. Using this approach, they were able to determine the individual foulant deposition throughout the cross section. CLSM was utilized under both in situ and ex situ conditions. Ferrando et al. (2005) obtained 20 images via optical sectioning of the PC MF membrane after filtration of BSA and ovalbumin mixture. Through this process, they determined protein penetration of up to 25% of the total thickness and locations of adsorbed proteins. Additionally, they provided quantitative data on the pore surfaces where proteins were detected along with the thickness of cake layer (3 µm) deposited on the identified surface porosity.

4.9 Phase-contrast X-ray imaging (XMI)

Phase-contrast X-ray imaging (XMI) with synchrotron radiation is another powerful in-situ technique for characterizing membrane morphology. Park et al. (2020) utilized this technique to analyze the adsorption of water droplets in oil-water emulsions on hydrophilic membranes and the oil film between oil and water phases based on visualized data. Yeo et al. (2005) observed the deposition of ferric hydroxide particles as a cake inside the lumen of the membrane and also deposition and fouling within the membrane structure. Although, the XMI is a powerful method for characterizing membrane morphology, the equipment required is expensive and evaluating the thickness of the cake presents a challenge. Table 5 lists the studies that utilized in-situ techniques for fouling characterization.

TABLE 5. Methods utilized for characterizing fouling of polymeric MF membranes.

5 Future directions and conlcusion

MF membranes are utilized in a wide range of applications that demand tunable morphological characteristics such as pore sizes, pore shapes, pore orientations, pore connectivity and surface-to-volume ratios. These structural features are determined by the manufacturing process and significantly influence the membrane’s performance and susceptibility to fouling (Figure 10). Therefore, a thorough characterization of morphology and fouling layer is crucial to guide manufacturing processes and enhance performance for the desired separation. This review focuses on manufacturing techniques, physical and fouling characterization methods for MF membranes.

Many MF applications such as sterilization and clarification, particle removal, cell harvesting require an isoporous structure. This structure is less prone to fouling by particulate plugging and provides enhanced flux through uniformly distributed pores. While the track etching method can produce isoporous membranes, it is expensive, limited to a few polymers with high thermal stability and typically results in low membrane porosity. Phase inversion is a commonly used technique for industrial production, allowing for the processing of various commercially available, affordable polymers. Future research efforts should focus on the combining different phase inversion techniques to produce isoporous membranes with high pore density.

Regarding physical morphology characterization of membranes, GLD method is the most commonly used one for MF membranes. A comprehensive characterization requires combining GLD data with liquid flux data to determine the PSD. The key issue for an accurate PSD estimation is to employ mathematical models that consider the structural features of the membrane, rather than solely relying on classical expressions valid for cylindirical, isolated, parallel pores.

While the ASTM standard outlines the protocol for determining the maximum pore size, there is a need for standardized protocols for PSD calculation to ensure comparability of characterization results across different research groups. In the future, researchers should utilize the benefits offered by machine learning approach for quantitative analysis of porous structures. Real 3D images of MF membranes obtained by direct methods such as high resolution X-ray computer tomography (XCT) can be used in visualizing structural features such as aniostropy, tortuosity, pore shape and pore connectivity. By utilizing experimental data as input, machine learning algorithms can rapidly generate structural models and predict key features. This approach not only accelerates analysis but also facilitates a deeper understanding of membrane morphology. A further step in future efforts for the physical characterization of MF membranes should involve the development of nondestructive methods applicable to large-sized membrane modules.

Various in-situ and ex-situ methods have been developed for characterizing fouling of MF membranes. With the integration of multiple techniques, the characterization of each foulant not only on the surface but also through the cross section of the MF membrane should be done to understand the fouling mechanism in detail. The main challenge in fouling characterization is to switch from research equipment to more robust field use. Therefore, techniques suitable for in-situ fouling observation in commercial modules over the long term, coupled with advanced signal processing algorithms, are necessary. Quick and facile evaluation of large datasets is crucial for long-term online monitoring. In this regard, utilizing deep-learned neural networks to process images allows going beyond traditional image processing for significantly improved results.

Author contributions

SA: Writing–review and editing, Writing–original draft, Visualization, Validation, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal Analysis, Data curation, Conceptualization.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Nomenclature

Table 6.