Membrane fractioning of pre-treated waste activated sludge for the recovery of valuable biocompounds

Journal of Water Process Engineering 55 (2023) 104086 Available online 4 August 2023 2214-7144/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Membrane fractioning of pre-treated waste activated sludge for the recovery of valuable biocompounds Daniel Nú˜ nez a , Paula Oulego a , Mahdi Nikbakht Fini b , Jens Muff b , Sergio Collado a , Francisco A. Riera a , Mario Díaz a , * a Department of Chemical and Environmental Engineering, University of Oviedo, c/Juli´ an Clavería 8, 33006 Oviedo, Spain b Department of Chemistry and Bioscience, Aalborg University Esbjerg, Niels Bohrs Vej 8, Esbjerg 6700, Denmark ARTICLE INFO Keywords: Biomolecules Modelisation Ultrafiltration Waste valorisation Wet oxidation ABSTRACT The recovery of added-value biocompounds from waste activated sludge (WAS) is a promising alternative to its current management. In this study, WAS was partially wet oxidised producing a highly complex stream mainly composed of humic acids, proteins and carbohydrates. This stream was ultrafiltered to assess the influence of membrane material and molecular weight cut-off (MWCO) on its fractioning based on the different sizes of the biomolecules contained in the oxidised WAS. Flat-sheet polyethersulfone (PES), permanently hydrophilic polyethersulfone (PESH), and polyacrylonitrile (PAN) membranes with a MWCO of 50 KDa were evaluated. The best performance was obtained with PES membrane, achieving high retention values (around 70 % for proteins and carbohydrates, and 47 % for humic acids) and high differences in selectivity between proteins and carbohydrates with humic acids (around 23 %). For the size fractioning experiments, PES membranes of 10 and 3 kDa (PES10 and PES3) were used, obtaining the best results when filtering the PES50 membrane permeate with the PES3 membrane, retaining 83 % of carbohydrates, 87 % of proteins and 69 % of humic acids. These results open the possibility of separating carbohydrates, proteins and humic acids through an integrated membrane process. Besides, membranes were characterized by atomic force microscopy, infrared spectroscopy, and contact angle measurements. Multiple fouling models were assessed, and the main fouling in PAN membrane, and to a lesser extent in PES, was reversible. Conversely, PESH membrane fouling had a strongly irreversible character. Cake filtration can be considered the main fouling mechanism in all experiments. 1. Introduction The biorefining of biowastes has been proposed as a sustainable means of waste valorisation, obtaining energy and biochemical resources while the volume of final waste is reduced [1]. Although it is yet at its conceptual phase [2], the steady rising trend of the biorefinery market value, expected to show an annual growth of 2.2 % until reaching a total value of USD 52680 million by 2027 [3], is a reliable marker for its potential. Waste activated sludge (WAS) is a promising raw matter for biorefinery, as it is a source of biomolecules (proteins, lipids, carbohydrates, humic acids and enzymes), phosphorus, bioplastics, bio-pesticides and also energy [4–10]. The current global market size of just the biomolecules present in WAS is of almost USD 200 billion, turning its recovery economically appealing [11]. WAS mainly consists of flocs of bacterial cells, which are formed by weak physical chemical interactions with extracellular polymeric substances [12] (a complex mixture of polymers generated by bacterial excretion, released after cell lysis or present in the incoming wastewater [13]). Thus, in order to recover the above-mentioned biocompounds, a prior solubilisation of the sludge is needed in order to break both the structure formed by the extracellular polymeric substances and the cell walls, thus releasing their intracellular content to the liquid medium. Several solubilisation methods have been tested for WAS, including ultra-sonication [14,15], cavitation [16], alkali treatment [15,17], ozonolysis [18], wet oxidation (WO) and thermal hydrolysis [19]. Among these techniques, WO is particularly interesting, due to its feasibility to be used at an industrial scale for sludge stabilisation [20]. It should be noted that this solubilisation results in a complex matrix, * Corresponding author. E-mail address: [email protected] (M. Díaz). Contents lists available at ScienceDirect Journal of Water Process Engineering journal homepage: www.elsevier.com/locate/jwpe https://doi.org/10.1016/j.jwpe.2023.104086 Received 19 December 2022; Received in revised form 3 July 2023; Accepted 22 July 2023 Journal of Water Process Engineering 55 (2023) 104086 2 where interactions between the different released biomolecules can difficult their purification. For instance, electrostatic interactions between proteins and humic acids occur while binding to heavy metals such as Cu 2+ , Zn 2+ , and Cd 2+ [21]; and even aggregates are formed when complexed with Cu 2+ [22], hindering a suitable separation of these molecules by immobilised metal affinity chromatography. Besides, a selective precipitation is also not possible from this complex matrix, as proteins, carbohydrates and humic acids co-precipitate with several precipitation methods [19]. These molecules have important industrial applications separately: proteins are used in cosmetics, food industry, pharmaceuticals and animal feed; humic acids can be applied in agriculture, pharmaceuticals or ecological remediation, among other uses; and carbohydrates are often used in the food industry [11]. Thus, their separation is of great interest for the incorporation of WAS as a valuable raw material in a context of circular economy. Besides, lipid recovery for its use as biofuel from WAS faces difficulties during its purification with solvent extraction, as other lipidic contaminants such as wax esters, terpenoids and polycyclic aromatic hydrocarbons are extracted together with the desired lipids [4]. For these reasons, fractioning the solubilised WAS would improve the efficiency of further separation and purification steps. To that end, membrane filtration is a suitable technology for this purpose due to its advantages, such as high selectivity, low energy consumption, low cost, and mild operating conditions [23]. Nevertheless, the performance of the membrane filtration is affected by several factors, especially the choice of membrane material and its molecular weight cut-off (MWCO) [24–26]. Most membranes are polymeric, and the choice of this polymer is critical for the efficiency of the operation, since it affects the permeability rate, the separation ability or the fouling process, key parameters in the filtration process [23]. Despite their utter importance, polymeric materials have not yet been studied for the filtration of solubilised WAS, and the effect of MWCO has been scarcely studied. Hence, only Li et al. have tested the use of 1, 10, 30, and 50 kDa polysulphone membranes to concentrate humic acids [27,28]. Therefore, the objective of this work was to study the influence of the membrane material and MWCO on the fractionation of hydrothermally solubilised WAS in order to separate carbohydrates, proteins, and humic acids based on their size differences aiming to obtain partially purified streams of these compounds. In this sense, 3 different polymeric materials: polyethersulphone, hydrophylic polyethersulphone and polyacrylonitrile, with a MWCO from 3 KDa to 50 KDa, were evaluated, paying special attention to fouling modelling. Hydrophilic polyethersulfone and polyacrylonitrile have a hydrophilic character, unlike polyethersulfone, which is a hydrophobic material. Besides, an integrated membrane process for the recovery of biomolecules from WAS was also proposed. 2. Experimental 2.1. Oxidised waste activated sludge Waste activated sludge was collected from the thickening unit of a wastewater treatment plant located in northern Spain (Baí˜ na, Asturias). The collection was performed by trained plant personnel to ensure the representativeness of the samples. WAS was immediately solubilised by a partial WO at 160 ◦C and 40 bar for 80 min. These oxidation conditions were selected to maximise the production of the target molecules: if the intensity of the treatment is too high, the target molecules get oxidised or mineralised; if the intensity is too low, the sludge does not completely solubilise [29]. Additionally, as the oxidation intensity increases, the particle size decreases, reducing the retention capabilities of the membranes. A constant flow of 1200 mL/min of O 2 saturated with steam was maintained during the entire reaction. The content of the reactor was kept stirring at 150 rpm. A more detailed description of the reactor can be found in [30]. After the reaction, the oxidised WAS was centrifuged in order to work with the liquid phase. Sodium azide 0.1 % (w/v) was added to the oxidised WAS in order to prevent microbiological growth. The oxidised sludge was stored at 4 ◦C for 15 days and then replaced with fresh oxidised sludge. 2.2. Membrane filtration 2.2.1. Membranes MQ (Synder Filtration) polyethersulfone (PES), UH050 (Microdyn Nadir) hydrophylic polyethersulfone (PESH), and MW (Suez) polyacrylonitrile (PAN) flat-sheet membranes with MWCO of 50 kDa (named as PES50, PESH50, and PAN50, respectively) were employed to perform the material screening experiments. Additionally, ST (Synder Filtration) and VT (Synder Filtration) PES flat-sheet membranes with MWCO of 3 and 10 kDa, respectively (named as PES3 and PES10), were used for the cut-off size screening experiments. All membranes were cut to a circular shape of 9 cm of diameter, and a filtration area of 63.62 cm 2 . Membrane hydrophilicity was characterized by contact angle measurements. The images were obtained with a CAM 200 optical contact angle meter (KSV Instruments Ltd., Finland). Sessile water droplets were dropped on the clean and fouled membrane surfaces using a syringe and let to spread freely. Images of the droplets were taken by a highresolution CCD camera at 40 ms intervals for the first 0.36 s, and at 1 s intervals for the subsequent 19 s. Equilibrium sessile drop contact angles were determined from the steady-state angles using the KSV CAM 200 software by measuring the angle between the baseline of a liquid drop and the tangent at the solid–liquid boundary. All contact angle measurements were performed in triplicate using three different membrane samples. 2.2.2. Equipment and filtration conditions Filtration experiments were carried out in duplicate using an FT17 Cross-flow Filtration Unit (Armfield Ltd., United Kingdom), which allows to perform tangential flow filtrations with flat sheet membranes. Prior to conducting the experiments, all polymeric membranes were preconditioned by running the equipment with no pressure for 30 min using distilled water. Subsequently, water was filtered under the operational conditions (indicated below) for an additional 30 min. The flux obtained in this sted was considered as the flux at t =0. The permeate flux of the clean membranes was measured during this step for further fouling modelling. All the material screening filtration experiments were performed under the following conditions: temperature of 50.0 ±0.4 ◦C, transmembrane pressure (TMP) of 4.0 ±0.2 bar and crossflow velocity (CFV) of 3.00 m/s. The oxidised WAS was filtrated without permeate recirculation until a volume concentration rate (VCR) of 2.5 was reached. During the MWCO screening experiments, the oxidised WAS was filtered with the PES50, PES10 and PES3 membranes. Besides, in order to assess the viability of the fractioning of the oxidised WAS, the permeate obtained after the filtration with PES50 was subsequently filtered with the PES10 or with the PES3 membranes, naming these permeates as PES50-10 and PES50-3, respectively. The experiments were performed under the same conditions than those used in the material screening ones. Only for obtaining the PES50-3 permeate, pressure was set at 30 bar and a VCR value of 1.25 was achieved. Permeate flow was determined by gravimetric measurements of the permeate, which were collected automatically by the FT17 Cross-flow Filtration Unit software. Flux was calculated by the following equation (Eq. (1)): J=QP AM (1) Where J is the permeate flux (m⋅s −1 ), Q P is the permeate flow (m 3 ⋅s −1 ), and A M is the membrane surface area (m 2 ). In addition, samples of the permeate and retentate were collected periodically and kept at 4 ◦C for further analysis. D. Nú˜ nez et al. Journal of Water Process Engineering 55 (2023) 104086 3 After the filtrations, the fouled membrane was rinsed with distilled water until a constant flux was obtained, and its permeability was measured for further fouling modelling. 2.2.3. Fouling modelling Resistance-in-series, Hermia’s, and Mentha’s fouling models were employed to characterise both the reversibility and main mechanism of membrane fouling occurred during the different filtration experiments. Resistance-in-series model expresses the total hydraulic resistance of the membrane (R T , m −1 ) as the sum of different resistances caused by reversible fouling (R rev , m −1 ), irreversible fouling (R irrev , m −1 ), or by the membrane itself (R m , m −1 ) (Eq. (2)). Hydraulic resistance can be calculated as shown in Eq. (3): RT=Rm+Rrev +Rirrev (2) R=TMP μ J(3) Where μ is the dynamic viscosity of the WAS at 50 ◦C (kg⋅m⋅s −1 ). By adding or subtracting the resistances obtained with the clean, fouled, or rinsed membrane fluxes, R m , R rev , and R irrev can be easily calculated. A more detailed explanation of these calculations can be found in the Appendix of [31]). The main fouling mechanism occurred on each membrane during ultrafiltration was determined through Hermia’s model [32] (Eq. (4)): dJ dt = − Kj⋅(J−J0)⋅J2−n(4) Where t is time (min), K j is the model constant that depends on the fouling phenomenon, J 0 is the limiting flux (m⋅s −1 ), and n is a constant that varies for the fouling mechanism: complete pore blocking (CPB) (n =2, K b in min −1 ), where the active membrane area is blocked by particles larger than the pore size; internal pore blocking (IPB) (n =1.5, K i in m −1 ), where membrane pores are blinded by either adsorption or deposition of particles smaller than the pore size; particle pore blocking (PPB) (n =1, K p in m −1 ), where particles might seal a pore over time, or bridge it and not block it completely; and cake filtration (CF) (n =0, K c in min⋅m −2 ), where a cake of particles that does not enter the pores is formed on the membrane surface [33]. The K j for the four models were calculated by minimizing the difference between the predicted values and the experimental data, calculated as the sum of squared residuals (SSR). The model with the lowest SSR was chosen as the most suitable one for each set of experimental data. In addition, flux was also modelled using the Mehta’s model [34], which takes into account the two flux decline domains that take place during membrane filtration: domain 1, where a rapid flux decline occurs during the early stage of filtration; and domain 2, where the flux decline decreases until the flux remains quasi-stable [35]. It can be expressed as follows (Eq. (5)): J=J0−J∞1⋅exp− α t+ (J∞1 −J∞2)⋅exp−βt+J∞2 (5) Where J∞1 is the flux at the end of domain 1 (m⋅s −1 ); J∞2 is the flux at the end of domain 2 (i.e., at the end of the experiment) (m⋅s −1 ); and α (min −1)and β(min −1)are two constants determined experimentally that describe the rate of flux decline associated with the membrane fouling and the concentration polarization and gel layer formation, respectively. 2.3. Atomic force microscopy The roughness of the fouled and clean membranes was analysed by atomic force microscopy (AFM). All AFM measurements were performed at room temperature (20 ◦C) using a Nanoscale scanning tunnelling microscope (Nanotec Cervantes FullMode SPM), working in contact mode in air medium with gold coated silicon nitride tips. Membrane samples were fixed to the sample holder of the microscope with highvacuum silicone grease. Roughness parameters were determined from the collected data using the WSxM 5.0 software [36]. Membrane roughness was compared in terms of mean roughness (R a [nm/ μ m]), root mean square roughness (rms), peak-to-peak distance (nm/ μ m), and surface skewness and kurtosis. R a is the mean value of the surface relative to the centre plane; rms is the standard deviation of the heights for all the pixels in the image from the arithmetic mean [37]; and skewness and kurtosis describe the shape of a probability distribution, reflecting the obliquity and the flatness of the curve, respectively [38]. 2.4. Infrared spectroscopy Infrared spectra (FTIR) of the clean and fouled membranes were taken in the range from 600 to 4000 cm −1 using Varian 670-IR FTIR spectrometer equipped with a Golden Gate horizontal attenuated total reflectance (ATR) accessory. Experimental conditions were 32 scans, 4 cm −1 resolution and aperture open. 2.5. Analytical methods Proteins, humic acids, carbohydrates, colour number (CN) and chemical oxygen demand (COD) were measured by colorimetric methods. Proteins and humic acids were measured following the modified Lowry method described by Frølund et al. [39], using bovine serum albumin and commercial humic acid as standards. Carbohydrates were measured according to the Dubois method [40] using D-glucose as standard. Spectral absorbance coefficients (SAC [cm −1 ]) were measured at 436, 525 and 620 nm and used to calculate the CN value (cm-1) according to Eq. (6): CN =SAC2 436 +SAC2 525 +SAC2 620 SAC436 +SAC525 +SAC620 (6) The absorbances of proteins, humic acids, carbohydrates and SAC were measured with a Helios Alpha UV–Vis spectrophotometer (Thermo Scientific, USA). Density was measured at 50 ◦C and 1 atm with a pycnometer. Kinematic viscosity was measured at 50 ◦C and 1 atm with a CannonFenske inversed-flow viscometer (Proton, UK). Dynamic viscosity was calculated by multiplying the kinematic viscosity by the density. pH was measured with a Basic 20 pH meter (Crison, Spain). COD values were determined by the potassium dichromate method [41], and the absorbance at 600 nm was measured with a HACH DR/2500 spectrophotometer (Hach Company, USA). Total organic carbon (TOC) was determined with a Shimadzu TOC-V CSH TOC analyser (Shimadzu, Japan). Rejection coefficients (RC i ) were calculated according to the Eq. (7): RCi=1−CPm,i CRt,i (7) Where CPm,i and CRt,i the concentration of the compound “i” in the permeate and the retentate (g⋅L −1 ), respectively. All analytical measurements were conducted at least three times. 3. Results and discussion 3.1. Oxidised waste activated sludge The oxidised WAS was slightly acid and presented a deep brown colour. Its main physical-chemical characteristics are shown in Table 1. D. Nú˜ nez et al. Journal of Water Process Engineering 55 (2023) 104086 4 3.2. Membrane material screening 3.2.1. Contact angle measurements Measured contact angles of water on the polymeric membranes used in the ultrafiltration of the oxidised WAS are shown in Table 2. A selection of the pictures of the sessile drops, from which the contact angles were calculated, can be found in Fig. A1. A surface is considered hydrophilic if the contact angle is lower than 90◦[42]. Thus, PES50 could be considered hydrophobic, while PESH50 and PAN50 were found to be hydrophilic. This was in accordance with the results obtained by other authors related to fouling resistance of ultrafiltration membranes [43]. After filtering the oxidised WAS, the fouled PES50 and PESH50 turned more hydrophilic than the pristine ones, while PAN50 became less hydrophilic after being fouled, which showed the different nature of the foulant-membrane interactions depending on the membrane material: it seems that PES50 and PESH50 were coated with more hydrophilic foulants, while PAN50 interacted with foulants less hydrophilic than itself. 3.2.2. Permeability tests The fluxes obtained with PES50, PESH50 and PAN50 are shown in Fig. 1. Final fluxes approximately 3 times higher were obtained with PESH50 (67.2 ±0.9 LMH) and PAN50 (74 ±3 LMH) compared to the one achieved by PES50 (25 ±2 LMH), due to their hydrophilic character. In this sense, membrane hydrophilicity prevented fouling, in accordance to what was reported by other authors [44,45]. The resistance-in-series models (Fig. 2) confirmed the aforementioned about hydrophobicity and its higher tendency to fouling. In this sense, all the resistances (membrane, reversible and irreversible) measured for PES50, which added up a total hydraulic resistance of 1.07⋅10 14 m −1 , were higher than those of PESH50 (13.3, 2.5 and 1.1 times higher, respectively) and PAN50 (3.4, 2.4 and 6.3 times higher, respectively). On the other hand, the different behaviour between fluxes in PESH50 and PAN50 can be explained based on the values of irreversible fouling for each membrane. Thus, the resistance-in-series modelling showed that the higher tendency to fouling observed in PESH50 is due to irreversible fouling, since its R irrev accounted for the 49.6 % of its total hydraulic resistance, while R irrev observed in PAN50 only represented 12.4 % of the total hydraulic resistance. The total R irrev also seemed to be correlated with the hydrophobicity of the membrane, as PES50 showed the highest R irrev ([2.8 ±0.4]⋅10 13 m −1 ), followed by PESH50 ([2.45 ± 0.01]⋅10 13 m −1 ) and PAN50 ([4.3 ±0.6]⋅10 12 m −1 ). Besides, reversible fouling was found to be the main fouling in PES50 and PAN50 membranes, corresponding to a 53.4 % and to a 68.9 % of the total fouling for PES50 and PAN50, respectively. The fluxes obtained in this study were in the same order of magnitude than those obtained by other authors when PES membranes were used during the filtration of milk [46], refinery and petrochemical wastewater [47] and oil-in-water emulsion [48,49]; and with PAN membranes when tap water [50] and oil-in-water emulsion [49] were filtered. Besides, lower initial fluxes were obtained when PESH membranes were employed for the filtration of molasses [51], while similar fluxes were attained for olive oil washing wastewater [52]. It should be noted that the higher fluxes obtained with PESH50 and PAN50 come along with lower rejection coefficients and lower selectivities between proteins and humic acids (Table 3). In particular, the lowest rejection coefficients for CN, TOC, and the three measured biocompounds were those corresponding to PESH50. On the other hand, the highest rejections were obtained with PES50, also achieving the highest rejection differences between proteins and humic acids (22 % difference vs 7 % difference obtained with PAN50, and 10 % difference obtained with PESH50); and between carbohydrates and humic acids (23 % difference vs 13 % obtained with PAN50 and 17 % difference obtained with PESH50). These higher retentions may be due to the formation of a thicker cake layer on top of the membrane, which would act as a secondary filtration mesh, increasing the selectivity of the membrane [53]. The formation of this thicker fouling cake could be observed through the resistance-inseries model (Fig. 2), where the R rev , mainly associated with the formation of the fouling cake [54], was more than two-fold higher after filtering with PES50 ([5.7 ±0.8]⋅10 13 m −1 ) than with PESH50 ([2.3 ± 0.2]⋅10 13 m −1 ) or PAN50 ([2.41 ±0.07]⋅10 13 m −1 ). Rejection differences between PESH50 and PAN50 are coherent with this explanation, as the R rev of PESH50 is slightly lower than that of PAN50. As both higher rejection coefficients and higher rejection differences between humic acids and the other biomolecules (proteins and carbohydrates) were achieved with PES50, the fractionation tests with membranes of different MWCO (MWCO screening experiments) was carried out with PES membranes. 3.2.3. Flux modelling As it can be seen in Fig. 3 and Table 4 (SSR), CF was the best-fitting Hermia’s model in all three cases, although the fittings indicated that none of the Hermia’s models fully explain the fouling mechanism, thus indicating several fouling mechanisms may have occurred throughout the filtration. Indeed, the fact that different fouling mechanisms occur at different stages of the filtration is well documented in the literature [55,56] and it was taken into account by Mehta’s model [34]. Thus, CF was the main fouling mechanism overall, although irreversible pore blocking also occurred; especially during the filtration with PESH50, where the IPB model showed better fitting than in the filtrations with PES50 and PAN50, reflecting the above-mentioned more irreversible nature of the PESH50 fouling. Besides, CF has also been described by other authors as the main fouling mechanism of natural organic matter during ultrafiltration with PES membranes [57,58]. However, it should be noted that the best fitting of the experimental data was achieved with Mehta’s model, which provides information about the effect of membrane fouling (parameter α ) and concentration polarization and gel layer formation (parameter β) on the flux decline. The optimised values for these two constants are shown in Table 4. Higher α and β values represent faster initial membrane fouling and faster stabilisation of the flux by the establishment of the concentration polarization gradient and gel layer formation, respectively. The α values obtained for the filtrations with PESH50 and PAN50 were 10-fold higher than those of PES50 (3.88), while the β values were 4-fold lower than those obtained for PES50 (5.20⋅10 −2 ). This indicates that a strong initial membrane fouling occurred after starting the filtration of the oxidised Table 1 Main physical-chemical characteristics of the oxidised waste activated sludge. Parameter Value pH 5.04 ±0.03 COD a (g O 2 L −1 ) 20.5 ±0.5 TOC a (g L −1 ) 8.00 ±0.01 CN a (cm −1 ) 3.9 ±0.2 Proteins (g L −1 ) 3.4 ±0.3 Humic acids (g L −1 ) 8.4 ±0.2 Carbohydrates (g L −1 ) 2.75 ±0.03 a COD: chemical oxygen demand; TOC: total organic carbon; CN: colour number. Table 2 Water surface contact angles on the studied membranes. Membrane size Membrane material Membrane state Contact angle 50 kDa PES Clean 90 ±3 Fouled 70 ±2 PESH Clean 72 ±1 Fouled 54 ±11 PAN Clean 39 ±4 Fouled 57 ±10 D. Nú˜ nez et al. Journal of Water Process Engineering 55 (2023) 104086 5 WAS with PESH50 and PAN50, causing a rapid decrease in the flux. After this initial drop, the stabilisation of the flux by concentration polarization occurred more slowly. Regarding the behaviour during the filtration with PES50, it was opposite to that of the other polymeric membranes: the initial drop caused by membrane fouling was less drastic, and the equilibrium in concentration polarization was reached faster. This result is in accordance with the literature, and can be explained considering the polarization sieving model [59], based on the differences of hydrophilicity between the membranes: Thus, in hydrophilic membranes, an initial irreversible adsorption layer is formed, regardless of the solute concentration, and subsequent 0 200 400 600 800 1000 1200 1400 1600 0 0.2 0.4 0.6 0.8 1 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 J (LMH) J/J0 VCR b) 0 20 40 60 80 100 0 0.2 0.4 0.6 0.8 1 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 J (LMH) J/J0 VCR 0 60 120 180 240 300 360 420 0 0.2 0.4 0.6 0.8 1 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 J (LMH) J/J0 VCR c) a) Fig. 1. Normalised flux variation over VCR for the oxidised WAS filtration with (a) PES50 (J 0 =102.4 ±0.2 L/m 2 h), (b) PESH50 (J 0 =1590 ±40 L/m 2 h) and (c) PAN50 (J 0 =420 ±40 L/m 2 h). Fig. 2. R m ( ), R rev ( ) and R irrev ( ) after filtration with PES50, PESH50 and PAN50. D. Nú˜ nez et al. Journal of Water Process Engineering 55 (2023) 104086 6 fouling will appear in the form of a gel-polarization layer. On the other hand, in hydrophobic membranes, the size of the irreversible adsorption layer increases until its thickness protects the hydrophobic surface from the adsorbed molecules, which generates higher values of irreversible fouling than in hydrophilic surfaces, and only after this limit is reached, the gel-polarization layer starts to form [59]. According to this, the behaviour of PES50 can be explained by its hydrophobic character, since more time was required for the initial membrane fouling to be fully occur, and the final concentration polarization layer needed less time to stabilise, thus starting to form at higher feed concentrations. 3.2.4. Atomic force microscopy AFM images were taken from clean and fouled membranes in order to analyse the surface morphology (Fig. 4). The vertical profile of the membrane surface was represented by the colour intensity, with lighter colours indicating higher regions, and darker colours indicating depressions. The surface of all the three clean membranes was clearly arranged in a “crest and valley” or “nodule and valley” pattern, originated by the random orientation and overlapping of the fibre structure [60]. This surface arrangement has also been observed by other authors when PES, PAN and polyamide membranes with flat-sheet, tubular and hollowfibre geometries were used in ultrafiltration and nanofiltration processes [60–63]. The presence of valley-like formations is highly related to fouling, as foulant particles tend to deposit in these formations [64]. This fact was supported by the images of fouled membranes, where crest-like formations could no longer be seen, indicating that the valleylike regions had been clogged by foulants. In addition to AFM imaging, membrane roughness was compared in terms of mean roughness (R a [nm/ μ m]), root mean square roughness (rms), peak-to-peak distance (nm/ μ m), and surface skewness and Table 3 Rejection coefficients (RC) obtained with the 50 kDa polymeric membranes. PES50* PESH50* PAN50* RC CN* 0.80 ±0.02 0.69 ±0.03 0.73 ±0.04 RC TOC* 0.47 ±0.03 0.34 ±0.04 0.38 ±0.05 RC COD* 0.48 ±0.08 0.41 ±0.03 0.4 ±0.1 RC CH* 0.70 ±0.02 0.57 ±0.07 0.60 ±0.06 RC Prot* 0.69 ±0.06 0.50 ±0.05 0.54 ±0.09 RC HA* 0.47 ±0.04 0.40 ±0.03 0.47 ±0.03 *CN: colour number; TOC: total organic carbon; COD: chemical oxygen demand; CH: carbohydrates; PROT: proteins; HA: humic acids; PES50: polyethersulphone, 50 kDa; PESH50: permanently hydrophilic polyethersulphone, 50 kDa; PAN: polyacrylonitrile, 50 kDa. 0 20 40 60 80 100 120 140 0 5 10 15 20 25 30 35 40 J (LMH) t (min) 0 200 400 600 800 1000 1200 1400 1600 1800 0 5 10 15 20 25 30 35 40 J (LMH) t (min) 0 50 100 150 200 250 300 350 400 450 0 5 10 15 20 25 30 35 40 J (LMH) t (min) a) b) c) Fig. 3. Hermia’s (complete pore blocking [ ], intermediate pore blocking [ ], partial pore blocking [ ] and cake formation []) and Mehta’s ( ) flux models for PES50 (a), PESH50 (b) and PAN50 (c) experimental fluxes (●). Table 4 Fitting parameters for the adjusted models. PES50 PESH50 PAN50 Hermia’s models CPB K b (min −1 ) 3.31⋅10 −2 5.29 9.43⋅10 −2 SSR 172,041.83 617,253.55 628,747.63 IPB K i (m −1 ) 4.58⋅10 −3 1.37⋅10 −1 7.31⋅10 −3 SSR 116,241.52 590,551.29 431,129.86 PPB K p (m −1 ) 6.54⋅10 −4 1.40⋅10 −3 5.97⋅10 −4 SSR 77,545.23 852,391.39 286,513.46 CF K c (min⋅m −2 ) 1.45⋅10 −5 4.42⋅10 −6 3.86⋅10 −6 SSR 32,776.51 255,336.20 129,005.00 Mehta’s model J∞1 (LMH) 46.05 114.16 116.08 α (min −1)3.88 30.70 30.70 β(min −1)5.20⋅10 −2 1.33⋅10 −2 1.26⋅10 −2 SSR 700.38 1401.89 783.73 D. Nú˜ nez et al. Journal of Water Process Engineering 55 (2023) 104086 7 kurtosis in order to analyse its relationship with fouling (Table 5). The loss of normalised flux was inversely related to R a and rms values. In this sense, PES50, whose normalised flux decreased the least (77 %) during the ultrafiltration of the oxidised WAS, also showed the lowest R a , rms and peak-to-peak distance values. Furthermore, the highest values were obtained with PESH50, which lost the highest proportion of normalised flux (96 %). The difference between normalised fluxes of PES50 and PAN50 was less marked than the roughness values may suggest, which could be explained based on their different hydrophilic properties (fouling allegedly increases with hydrophobicity [61]). The relationship between loss of normalised flux and R a and rms values is in accordance with the existing literature, indicating that surface roughness plays a key role in flux loss [61,64–66]. As it was previously commented, in the initial stages of filtration, the particles tend to deposit in the “valley-like” formations of the membranes, clogging these depressed regions. Membranes with lower surface roughness present fewer “valley-like” formations on their surface, so the attachment of solute molecules is restricted [64]. The effect of this fewer presence of valley-like regions can be attended contrasting the roughness values from Table 5 with the images in Fig. 4: valley formations were more evident in PAN50 and PESH50 than in PES50, and the depth of the valleys was lower in the latter membrane, as it can be confirmed by the peak-to-peak distance. As proved by the permeability tests and the AFM measurements, flux loss is a complex phenomenon, which depends on several factors, among which, the nature of the membrane surface (hydrophilic surfaces prevent fouling) and its rugosity (the higher the rugosity, the more space the solute molecules have for depositing) can be considered determinant. 3.3. Membrane molecular weight cut-off screening 3.3.1. Permeability tests As PES was selected as the most suitable material for WAS fractioning (Section 3.2.2), the MWCO screening experiments were carried out with membranes made of this polymeric material (PES10 and PES3). The flux obtained with PES3 was extremely low (around 0.5 LMH on average), making this filtration unfeasible, and thus no data from this experiment are depicted. The fluxes obtained with PES10, PES50-10 and PES50-3 are shown in Fig. 5. The highest flux obtained during the MWCO screening was achieved with PES50-10. However, the low rejection coefficients observed made this option unfeasible (Table 6). The highest rejections in terms of CN, TOC and COD were achieved with PES50-3. Nevertheless, in the case of the biomolecules, slightly differences were observed in rejections and rejection differences between PES10 and PES50-3. Based on the rejection coefficients and the observed decrease in flux during the ultrafiltration of the oxidised WAS using PES10, as well as the permeates obtained from PES50 with PES10 and PES3, it can be concluded that the majority of molecules retained by the PES10 membrane can also be retained by the PES50 membrane. Therefore, coupling these two membranes would be redundant and unnecessary. This behaviour was in accordance with the work by Urrea et al. [67], where the effect of WO on the different molecular weight fractions of WAS was studied. They reported that, after a WO treatment at 190 ◦C and 90 min, the molecular weight of the majority of the present molecules was comprised in the ranges between 0 and 35 kDa (referred by Urrea et al. as low molecular weight molecules) and 35–150 kDa (medium molecular weight molecules). Moreover, hydrophobic molecules were also present due to the interactions with size-exclusion column. According to the results attained in this study, the sizes of the majority of the low molecular weight molecules were comprised between 0 and 10 kDa, and Fig. 4. AFM images of clean and fouled PES50 (A and B), PESH50 (C and D) and PAN50 (E and F) for membrane surface morphology analysis. Table 5 Membrane surface roughness parameters. Clean Fouled PES50 PESH50 PAN50 PES50 PESH50 PAN50 R a [nm/ μ m] 23.0 ± 0.6 74 ±5 67 ±8 56 ± 15 164 ± 37 92 ±6 rms 30.4 ± 0.9 94 ±7 89 ±5 41 ± 12 212 ± 47 119 ±7 peak-topeak distance [nm/ μ m] 142 ± 5 391 ± 32 365 ± 16 255 ± 59 918 ± 208 487 ± 26 Skewness −0.23 ±0.04 – −0.08 ±0.03 – – −0.05 ±0.03 Kurtosis 3.35 ± 0.03 2.72 ± 0.05 2.8 ± 0.3 3.4 ± 0.4 2.89 ± 0.01 2.7 ± 0.1 D. Nú˜ nez et al. Journal of Water Process Engineering 55 (2023) 104086 8 the ones of the medium molecular weight molecules were above 50 kDa. The resistance-in-series model (Fig. 6) showed that approximately half of the membrane fouling during the filtration with both PES10 and PES50-3 was irreversible, which contrast to the results obtained with PES50, where only 33 % of the fouling resistance was attributable to irreversible fouling. Moreover, a comparison between the values of the R rev and the R irrev obtained when filtering with PES50 and PES10 shows that no significant differences could be found between the values of reversible fouling, whereas the irreversible one of the PES10 membrane was around 2.5 times higher than that of the PES50 membrane. Thus, the additional flux decay observed between the filtration with PES10 and PES50 was exclusively due to an increase in irreversible fouling. It has been reported that irreversible fouling during the filtration of natural organic matter is primarily caused by the hydrophilic fraction of this organic matter [68–70]. Therefore, the additional irreversible fouling observed in this study is likely attributed to hydrophilic substances, presumably oxidised HA, since their rejection coefficients were lower than those of the other biomolecules, which suggests a lower average size. In this sense, several authors have reported that the oxidation of HA increased their hydrophilicity [71–74] by oxidizing benzene groups into different aldehydes and carboxylic acids [72,74]. After a subsequent reduction of the membrane MWCO to 3 kDa, the fouling profile remained similar (slightly more irreversible than reversible fouling), proving that the molecules retained by the PES50 membrane mainly caused reversible fouling. In this sense, Taniguchi et al. [57] compared the fouling of PES membranes with MWCOs from 10 to 1000 kDa during the UF of natural organic matter, and reported that membranes with lower MWCO (10 and 30 kDa) showed higher irreversible fouling, although fouling was mostly reversible in all cases. 3.3.2. Fouling modelling The fitting of the studied fouling models to the experimental data corresponding to the filtrations with PES10, PES50-10 and PES50-3 is shown in Fig. 7. Similarly to the filtration with the 50 kDa membranes, the best fitting Hermia’s model was CF in the three filtrations. This results were in accordance to those reported by Peeva et al. [75] related to the ultrafiltration of humic acid solutions with PES membranes. However, as it was discussed in the previous section, the fitting showed that this model alone could not adequately explain the observed fouling tendency. In Fig. 5. Flux variation over VCR for the oxidised WAS filtration with PES10 (a) (J 0 =101.5 ±0.9 L/m 2 h), PES50-10 (b) (J 0 =102 ±3 L/m 2 h) and PES50-3 (c) (J 0 = 13.72 ±0.01 L/m 2 h). Table 6 Rejection coefficients obtained with PES10, PES50-10 and PES50-3. PES10 PES50-10 PES50-3 RC CN 0.90 ±0.01 0.35 ±0.01 0.9942 ±0.0009 RC TOC 0.70 ±0.16 0.17 ±0.09 0.88 ±0.01 RC COD 0.60 ±0.04 0.09 ±0.04 0.66 ±0.04 RC CH* 0.79 ±0.05 0.14 ±0.05 0.83 ±0.05 RC Prot* 0.84 ±0.09 0.21 ±0.08 0.87 ±0.06 RC HA* 0.63 ±0.05 0.15 ±0.06 0.67 ±0.06 *CH: carbohydrates; PROT: proteins; HA: humic acids. D. Nú˜ nez et al. Journal of Water Process Engineering 55 (2023) 104086 9 this case, the ultrafiltration of both the oxidised WAS with PES10 and the permeates obtained from PES50 with PES10 and PES3 suffered a proportionately higher irreversible fouling compared with the ultrafiltration of oxidised WAS with the PES50 membrane, which in all cases could be mainly attributed to PPB, as it was the second-best fitting Hermia’s model (Table 7). The fact that PPB remained the main irreversible fouling mechanism in membranes with lower MWCO is in accordance with the literature. Thus, it has been reported that, when Fig. 6. R m ( ), R rev ( ) and R irrev ( ) after filtration with PES10 and PES50-3 membranes. Fig. 7. Hermia’s (complete pore blocking [ ], intermediate pore blocking [ ], partial pore blocking [ ] and cake formation []) and Mehta’s ( ) flux models for PES10 (a), PES50-10 (b) and PES50-3 (c) experimental fluxes (●). D. Nú˜ nez et al.