Enhanced oil removal from a real polymer production plant by cellulose nanocrystals-serine incorporated polyethersulfone ultrafiltration membrane

As discharging oily wastewater from industries to the environment is a potential threat for the aquatic ecosystem, in this research, oil removal from a real case of Kermanshah polymer production plant wastewater was investigated. The focus of this study was on improving the oil rejection performance of polyethersulfone (PES) ultrafiltration membrane due to adding cellulose nanocrystals (CNC) and modified CNC with serine amino acid (CNC-Ser) in PES mix matrix. From the results, the membranes embedded with CNC-Ser showed better performance in terms of water flux, flux recovery ratio, and oil rejection (higher than 97%) compared to the modified membranes with CNC. The lowest water contact angle (41.37°), smoother surface, and higher negative surface potential (− 24 mV) were achieved for the optimum loading of CNC-Ser. Besides, long-term performance of the membranes with optimum loading of CNC and CNC-Ser were compared in both dead-end and cross-flow setups.


Introduction
With developing industrialization, the treatment of industrial wastewater has become a main concern in all over the world (Guan et al. 2021). The treatment and disposal of oily wastewaters is a major challenge for some industries like petrochemical, food, and pharmaceutical. Petrochemical industry inevitably generates large volumes of oily wastewater over the chemical process; for instance the production of high density polyethylene generates waxes as a byproduct (Huang et al. 2018;Li et al. 2021). Therefore, petrochemical industry must separate oil from wastewater before releasing wastewater to the environment. Also, seawater is being polluted by oil leakage and discharging effluents, nowadays (Elshorafa et al. 2020). Various technologies have been applied for treating oily wastewaters, i.e., air flotation (Santos et al. 2021), ultrasonic irradiation (Sadatshojaie et al. 2021), bioremediation (Mai et al. 2021), microwave irradiation (Garcia-Costa et al. 2021), and pyrolysis (Singh et al. 2021). It has been proved that the mentioned methods are not qualified enough as a result of their disadvantages like being costly, inefficient in low oil concentration (≤ 100 ppm), and generate second pollutions in some cases.
To present more details of the mentioned methods, the generated microorganisms as known excess sludge should be disposed in bioremediation technique, and also the extracted oil in air flotation and ultrasonic irradiation should be disposed. Air flotation is an expensive method, mainly because it requires pressurized air, and also ultrasonic and microwave irradiations consume a lot of electric energy. Likewise, pyrolysis is a thermal decomposition of materials producing volatile products and char as a leftover which requires lots of energy (Pauzan et al. 2021;Wang et al. 2020;Yan et al. 2019). Membrane technology has been noticed by over recent decades resulted from its impressive separation even in low ranges of oil concentration; however, the concentrated rejected oil should be disposed, and also fouling phenomenon over separation occurs and needs to be cleaned frequently which could be a restriction for membrane technology (Eggensperger et al. 2020;Kamali et al. 2019;Yusuf et al. 2020).
Ultrafiltration has been used frequently for oil separation from oily wastewaters. In order to reduce the fouling phenomenon and increase the performance of ultrafiltration, the polymeric membranes have been modified by different additives and fillers to obtain more hydrophilic membranes, e.g., TiO 2 , carbon nanotubes (CNT) , covalent organic framework (COF) (Fan et al. 2018), mesoporous , C 3 N 4 (Shi et al. 2019), metal-organic framework (MOF) (Gholami et al. 2018), and cellulose nanocrystals (Balcik-Canbolat & Van der Bruggen 2020).
In 2019, Liu et al. reported mesoporous hybrid PAV/SiO 2 nanoparticle in PVDF membrane for simultaneous oil and heavy metal rejection. From the reported results, an acceptable heavy metal rejection beside of adequate antifouling ability and high water flux have been observed ). Gholami and his coworkers studied antifouling property and oily wastewater rejection by adding Zn-based MOF (TMU-5) to PES polymer membrane (Gholami et al. 2018). Due to hydrophilic nature of nanoparticles in membrane matrix, excellent hydrophilic and antifouling properties have been achieved. In another study, a direct osmosis TFC membrane including poly [3-(N-2-methacryloylxyethyl-N, N-dimethyl)-ammonatopropanesulfonate] (PMAPS) has been used for oil rejection from oily wastewater which high flux, low-fouling phenomenon, and 95% of oil removal efficiency have been reported (Lee et al. 2019). Also, Zarghami and others used polymerization of dopamine on PES membrane in which dopamine improved the water flux with good self-cleaning and antifouling properties (efficiency above 98%) (Zarghami et al. 2019). Besides, ZnO microsphere/carbon nanotube as a filler in PES membrane has been applied which high rejection, good flux and long-term stability have been obtained (Saththasivam et al. 2019). In another research, ZNG-g-PVDF-based membrane has been exhibited high antifouling quality and high oil rejection for very low concentration of oil pollution (13 ppm) (Zhang et al. 2018).
The application of nano-biomaterials as filler in polymer matrix has been noticed by researchers because of their favorite features like being cost-effective, biocompatible, and nontoxic. Cellulose nanocrystals (CNCs) as renewable nanoscaled biomaterials have been applied in different researches to enhance and modify polymer characteristics due to its high hydrophilicity, good dispersion in water, and polar solvents. Besides, the hydrophilic characteristics of CNC could be improved by conjugating with hydrophilic groups such as amino acids. Hence, amino acid with − COOH and − NH 2 groups provides superior hydrophilic nano-biomaterial compared to CNC alone which has only − COOH group (Daraei et al. 2017;Wang et al. 2021).
From the literature, there are no any reports about using cellulose nanocrystals (CNC) modified with serine amino acids to improve hydrophilicity of PES membranes. CNC are derived from cellulose showing high surface area which have been applied for absorption and catalyst field. In spite of high stability and hydrophilicity of CNC, it has not been used broadly to improve the features of membranes for wastewater treatment. As a notice, it can be used to construct a novel membrane by the phase inversion method. In the current study, a nanocomposite UF membrane based on PES and CNC modified with serine amino acid including − NH 2 and − COOH functional groups is prepared via a phase inversion technique for purification of oily wastewaters. The CNC-serine is firstly synthesized and characterized and then embedded into the PES membrane. In order to identify membrane characteristics, contact angle, SEM, AFM, zeta potential, and antifouling tests are used. It should be mentioned that the optimal membranes have been used to reject oil from Kermanshah polymer production plant wastewater.

CNC modification
Serine and the carboxyl groups of CNC were conjugated by 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC)/Nhydroxysuccinimide (NHS) coupling to synthesize aminefunctionalized CNC. Dried CNC was dispersed in ethanol by sonication for 45 min. EDC (54.3 mg, 0.4 mmol) and NHS (50.6 mg, 0.4 mmol) were added to the CNC solution (0.5 g. ml −1 , 0.5 g). Then, the serine was added to the CNC solution and stirred for 1 day at room temperature. The resulting CNC-Ser solution was centrifuged and washed in ethanol to remove unreacted serine (Fereiro et al. 2018).

Preparation of mixed matrix membrane
Phase inversion method has been applied for preparing flat-sheet symmetric porous membranes modified with CNC-based additives. The composition of the used different mixed solutions is shown in Table 1. An appropriate amount of additive was mixed with DMAc and sonicated (DT 102 H Bandelin ultrasonic, Germany) for 25 min. Then, PVP and PES were added to the solution and mixed for 24 h (400 rpm). After obtaining homogeneous casting solution, a film applicator and several neat glassy plates were utilized to fabricate membranes with thickness of 150 µm. In next step, whole glass was immersed into distilled water (room temperature) to form the solid polymeric membranes. Then, the flat-sheet membranes were immersed into fresh distilled water for 24 h. In order to dry the obtained membranes, they were kept between two filter papers for another 24 h (Gholami et al. 2020a(Gholami et al. , 2020b.

Characterization of CNC-based membrane morphology
Scanning electron microscope (SEM, Philips-XL30, the Netherlands) was applied with acceleration voltage of 20 kV. Small sample of membrane was frozen by liquid nitrogen and dried, and then by sputtering, a thin layer of gold was coated on the membrane surface to make conduction. Atomic force microscopy (AFM) was used to study the surface of the prepared membranes. In this method, specific area of the membrane sample (2 × 2 cm) was fixed on specimen holder; then a high-resolution microscope (Nanosurf® Mobile S scanning probe-optical microscope, Switzerland) physically scanned the level, three dimensionally (4 × 4 µm). The difference between peaks and valleys is defined as surface roughness. Smoother surface has less difference in these parameters. The AFM images can be observed as Sa (average roughness), Sq (root of two data), and Sz (the average data between lowest valley and highest peak). The membrane water tendency was quantified by water contact angle (WCA) measurement. Less contact angle between membrane surface and water droplet indicates more hydrophilic nature. For this purpose, on the proper size of clean membrane, a small droplet of distilled water was injected (4 µl). After waiting for 10 s, to face with stable conditions, a digital microscope (Contact Angle Meter XCA-50) captured the cross-sectional images, and the WCAs were calculated. More importantly to make the obtained results more reliable, 5 random places on the membrane surface were measured, and average data was reported.

Porosity measurement of the prepared membranes
The membrane porosity was calculated based on gravimetric method. At first, 4 cm 2 of all membranes was cut and weighted precisely and then submerged into distilled water for 24 h, and finally it was weighted again. Based on Eq. 1, the membrane porosity was calculated: where ω 1 and ω 2 are the wet and dry weight of membranes. A, L, and dW are effective surface, membrane thickness, and water density (998 kg/m 3 ), respectively.
Besides, mean pore radius (r m ) was calculated according to Guerout-Elford-Ferry equation (Eq. 2): where η is the water viscosity (8.9 × 10 −4 Pa s), Q is the volume of the permeate pure water per unit time (m 3 /s), and ΔP is the operating pressure (0.5 M Pa) (Samari et al. 2021).
The surface free energy (ΔG s ) of the membranes was calculated using the Young Dupre equation (Eq. 3) (Hurwitz et al. 2010): where θ is the water contact angle value and γ L is the surface tension of water (72.8 mJ/m 2 ).

Membrane setup and performance
A stainless steel dead-end setup with volume capacity of 150 ml and effective membrane surface of 12.56 cm 2 was applied to evaluate membrane performance. The applied driving force to pass feed solution through the dead-end setup was the pressure of nitrogen gas (3 bar). (1) (3) −ΔG s = (1 + cos ) L Also, to reduce feed polarization, a stirrer was provided in setup. The weights of obtained samples passed through setup was determined by a digital balance, and the pure water flux (PWF) was calculated based on Eq. 4: where M, A, and ∆T are permeating weight (kg), effective membrane area (m 2 ), and filtration time (h), respectively (Gholami et al. 2020a(Gholami et al. , 2020b.
Also, the long-term performance of the optimum membranes in a cross-flow setup with volumetric flow rate of 250 l/min and operating pressure of 3 bar was evaluated.

Fouling effect
Resistance of the membranes was examined by passing milk powder solution (1000 ppm) through dead-end setup as fouling agent. At first, distilled water was passed through deadend setup (60 min, 3 bar), in the second step, milk powder solution was filtered for 90 min, and then the membrane was washed and immersed into distilled water (20 min), and finally, the distilled water was filtered again (60 min). After all, flux recovery ratio (FRR) and the membrane resistance are calculated according to Eq. 5 in which FRR was considered membrane resistivity against fouling: where j w.1 and j w.2 were pure water flux (PWF) of the first step and second step, respectively. Different types of fouling including total fouling ratio (R t ), reversible fouling ratio (R r ), and irreversible fouling ratio (R ir ) were also calculated:

Oil rejection and characteristics of Kermanshah polymer production plant wastewater
The simulated oily wastewater with 100, 300, and 500 ppm was utilized to investigate rejection performance, and finally the long performance of the optimum membrane in removing oil from a real petroleum wastewater was examined in both dead-end and cross-flow setups. In dead-end setup, oil-water emulsion (diesel oil, SAF 40) has been provided by heating up the mixture at 50 °C and stirring (400 rpm) for 100 min without emulsifier. Also, with applying stirrer and sonication, the oil droplet size distribution was kept constant. Also, Kermanshah polymer production plant wastewater, Kermanshah, Iran, as a real oily wastewater is used to evaluate the membrane performance in this study. The characteristics of the polymer production plant wastewater are as follows: COD of 500 mg/l, TSS of 395 mg/l, and pH of 5.5-7.

Characterization of CNC and CNC-Ser
To ensure that surface modification on CNC has taken place some characterized test like FT-IR, SEM and zeta potential were performed. Figure 1a and b show FT-IR spectra for CNC and CNC-Ser, respectively. Also, the functional groups related to the observed bands have been specified. Based on the figure, the bands of − NH 2 (around 3500 cm −1 ), − NH, and − COOH (around 1600 cm −1 ) appeared in the FT-IR spectrum of CNC-Ser; however, they are not observed in the spectrum of CNC. This outcome is a firm evidence to modify CNC by serine (Chen et al. 2018;Zangeneh et al. 2019).
SEM images of CNC and CNC-Ser are illustrated in Fig. 2a and b, respectively. From SEM images, the differences between the surface of CNC and CNC-Ser are obviously found out. The figure showed that the surface of CNC was rougher rather than CNC-Ser. It should be noted that the surface of CNC was become slicker by inserting serine amino acid on its surface (Azizi 2020).
Besides, zeta potential of CNC and CNC-Ser was assessed as another evidence for surface modification which are shown in Fig. 3a and b. Based on the figure, the location of zeta potential peak and also the intensity of zeta potential peak were varied with surface modification of CNC. It is observed that zeta potential peak was shifted from − 25 mV for CNC to − 50 mV for CNC-Ser and also the intensity of zeta potential peak was altered from 4000 total counts for CNC to 6000 total counts for CNC-Ser (Mohan et al. 2021).

Membrane hydrophilicity
It is accepted that the presence of hydrophilic additives in membrane matrix caused a reduction in water contact angle (WCA) due to enhancing in membrane hydrophilicity. According to WCA results presented in Table 2, WCA of the bare membrane (M1) was decreased with addition of CNC and CNC-Ser, verifying the effect of hydrophilic groups. In overall, the membranes embedded with CNC-Ser (M5-M7) showed lower WAC relative to the membranes embedded with CNC (M2-M4). This outcome is another proof for inserting serine amino acid with − NH 2 and − COOH groups on the surface of CNC which makes the surface more hydrophilic. The lowest value of WCA was reported for the lowest loading of CNC-Ser (M5 with 0.1 wt%) as a result of the proper dispersion of the additive into the polymeric matrix. The WAC for the lowest loading of CNC and CNC-Ser was found to be 41.37° and 63.52° for M5 and M2, respectively, in which this difference is a sign of the presence of hydrophilic functional groups of serine. It should be mentioned that an increase in the CNC-Ser loading (0.5 and 1.0 wt%) resulted an increase in WCA and a reduction in the hydrophilicity of the membranes which could be explained by an increasing trend of the viscosity casting solution in higher loading of CNC-Ser. Moreover, the aggregation of nanostructures in higher loading could be another reason for decreasing homogeneity and increasing WCA besides the viscosity increscent (Samari et al. 2020).
In contrast to CNC-Ser, a decreasing trend of WCA is continued with increasing loading of CNC from 0.1 to 1 wt%. PWF is another parameter to assess the presence of hydrophilic additives and porosity. According to the presented data for PWF in Table 2, the membranes with CNC-Ser represented higher values of porosity and PWF for the membrane embedded with CNC-Ser (M5, 91.56 kg/ m 2 .h) compared to the membrane with incorporated CNC (M3, 70.42 kg/m 2 .h). High hydrophilicity of M5 creates strong electrostatic interactions (e.g., hydrogen bonding) between the membrane surface functional groups (− COOH and − NH 2 ), and water molecules caused the formation of a hydration layer on the surface of the membranes and also the channel walls in the membrane matrix (based on the Hagen-Poiseuille flow equation). It should be noticed that for both series of membranes with incorporated CNC and CNC-Ser, higher loadings of additive agglomeration restricted porosity and PWF compared to lower loadings of additives. As an exception, PWF and porosity was not significantly decreased with increasing CNC loading from 0.1 to 0.5 wt% as a result of the loaded porous nanoparticles; however, a reduction in PWF was observed from M5 (0.1wt%) to M6 (0.5wt%) (with CNC-Ser) due to occupying membrane pours and channels by serine functional groups.

Morphology analysis
The tissue of the membranes was assessed by SEM crosssectional analysis. From Fig. 4a-g, the top layer of the bare membrane was dense, and the obvious channels could not be observed. With adding the additives into the casting solution, the channels have been appeared in the top layers, and also more porous top layer was obtained. As an explanation, CNC and CNC-Ser nanoparticles made hydrogen bonds with the polymer (PES) matrix which caused a reduction in displacement speed of PES toward membrane surface resulted in a top layer with more porous structure (see Fig. 4b, c, e). However, with increasing loading of the additives, a thick top layer with no noticeable channels was obtained (see Fig. 4d and g). This result could be elucidated by increasing viscosity with increasing loading of additives, so that the viscosity of the solution was increased from 412 cp for M5 to 489 cp for M7 which caused a thick and dense top layer with less porosity (Zhang et al. 2010). The mentioned results were in a good agreement with porosity and PWF (presented in Table 2).
As the top layer of the membranes at lower loading of both additives (M3 and M5) was more porous rather than the others (M4 and M7), an improvement in passing water through the membrane channels and porosity data was obtained (higher PWF). From Fig. 4, the presence of CNC and CNC-Ser creates obvious changes in the tissue of the membranes particularly at higher loading of additives which resulted in more texture deformation (M4 and M7) .
Besides, to evaluate the roughness of the surface, AFM images of the bare and modified membranes are represented in Fig. 5. The surface topography of the membrane's surface was investigated by mean roughness (Sa), root mean square of the Z data (Sq), and the difference between highest and lowest valleys (Sz) which are presented in Table 3. According to the table, the values of Sa, Sq, and Sz for the modified membranes were lower than the bare membrane, indicating more smooth surface of the modified membranes. This result proved the presence of hydrophilic additives which enhanced the surface smoothness. M3 and M5 showed the least roughness for both membrane series with CNC and CNC-Ser, respectively, as a result of their good dispersions of additives in the polymeric matrix creating smoother surface with less wrinkled. While, the surface roughness increased at higher additive loading. Less roughness of M5 (Sa = 2.44 nm) compared to M3 (Sa = 2.99 nm) could be explained by more interactions between functional groups of CNC-Ser (− NH 2 and − COOH) with the matrix of PES membrane exclusively over the phase inversion (Lee et al. 2019).

Fouling behavior
Fouling behavior was evaluated by applying milk powder solution with concentration of 1000 mg/l as foulant solution. In this mean, three frequent steps were assessed including distillated water-milk powder solution-distillated water. Figure 6 illustrates the flux data over each step for all the membranes. The trends of changing flux for all membranes are similar; however, their antifouling ability and flux recovery are different. From the figure, the lowest flux for all three steps was observed for the bare membrane (M1) because of its low porosity (47.45%, Table 2) and hydrophobicity feature (WCA of 78.65°). In general, all the modified membranes showed better performance rather than the bare PES membrane, verifying the presence of hydrophilic additive. The highest flux over all three steps was obtained for M5 which based on Table 2 showed the lowest WCA and the highest porosity. The higher hydrophilicity of M5 enables the membrane to reject the foulant, provides satisfied flux, and improves antifouling behavior and flux recovery. It should be noted that the higher performance of M5 is due to the presence of serine with polar functional groups (Vatanpour et al. 2021).
Also, as zeta potential is an effective criterion for membrane performance in rejecting foulants and flux recovery, the surface charge of the bare membrane (M1) and the optimum membrane for each series (M3 for CNC series and M5 for modified CNC) at pH range of 4-8 was measured and is presented in Fig. 7. According to the figure, all the mentioned membranes displayed negative values of zeta potential (from -6.66 to -25.29 mV), and also an increase in pH value from 3 to 9 caused an increase in negative zeta potential (from -6.66 to -16.33 mV for M1, from -9.4 to -22.74 mV for M3, from -12.8 to -25.29 mV for M5). In each studied pH value, M5 exhibited the maximum surface negative charge compared to other membranes. It can be concluded that the negative charge density of CNC-Ser is improved by − COOH and − NH 2 functionalities on the top layer rather than CNC, causing a high negative charge on the surface of M5. In point of fact, the results of surface zeta potential are in consistent with the results of WCA presented in Table 2 and also are a good justification for obtaining the highest antifouling ability and flux recovery for M5 (Fig. 6) (Hoseinpour et al. 2018).
Flux recovery ratio (FRR) is a decent criterion to evaluate the performance of the prepared membranes besides PWF. Therefore, FRR was calculated for different prepared membranes and is presented in Fig. 8 (with milk powder solution as foulant). From the figure, the lowest FRR was reported as 69.54% for the bare membrane, whereas the membranes embedded with CNC and CNC-Ser showed an improvement in FRR. The highest values of FRR were achieved for M3 and M5. This result could be explained by surface zeta potential so that based on Fig. 7, M3 and M5 represented more negative surface charge, inhibiting cake layer formation. As a fact, the viscosity of the casting solution is relatively low at low loading additives which led to a fast migration of the additives over phase inversion, resulting more hydrophilicity, more surface free energy (Table 2), and more negative charge on the membrane surface. Moreover, FRR was reported as 93.41% and 83.34% for M5 and M3, respectively. The better performance of M5 could be explained by hydrogen bonding between − COOH and − NH 2 groups of the modified CNC and H 2 O molecules which caused an increased membrane capability against cake layer formation on the membrane surface.
A reduced trend in FRR could be observed with increasing additive loadings, particularly for the unmodified CNC as a result of unpleasant dispersion and agglomeration in the membrane matrix. This reduction also could be related to surface roughness, so that the higher surface roughness caused more foulant cake layer formation on the membrane surface. From Table 3 and Fig. 5, the surface of M3 and M5 was smoother rather than the other membranes which is consistent with FRR results. Fouling took place for the other membranes with higher surface roughness intensively, resulting lower FRR.
Moreover, in order to present more analysis about fouling phenomenon, reversible and irreversible fouling (milk powder solution as foulant) were calculated for each membrane and are presented in Fig. 8. From the figure, higher reversible fouling which is temporary and the lower irreversible fouling known as permanent fouling were obtained for the membranes embedded with CNC-Ser compared to the membranes with CNC. This outcome verified the better performance of CNC-Ser for improving the fouling characteristics of the bare membrane. As can be seen from the figure, the bare membrane has the most irreversible fouling which caused a weak performance for applying it during several frequent filtrations. On the other hand, as observed in the figure over optimum loading of additives, the reversible fouling of the bare membrane increased from 30.22 to 43.67% and 53.22% for M3 and M5, respectively, while the irreversible fouling decreased notably from 30.45 to 16.65% and 6.58% for M3 and M5, respectively (Gholami et al. 2020a(Gholami et al. , 2020bVatanpour & Sanadgol 2020).

Oil water removal
In this section, the performance of the prepared membranes in terms of oil rejection was investigated. In this mean, three different feed concentrations (100, 300, 500 ppm oil) were examined for all membranes in which the results are presented in Table 4. From the data of Table 4, the least FRR and flux values were reported for the bare membrane which resulted from its high roughness, low antifouling ability, low porosity, and its hydrophobic nature. The membrane embedded with CNC-Ser (modified additive) showed better performance in terms of flux and FRR compared to the membranes embedded with CNC in which these results are in a good agreement with the data presented in Fig. 8. Further improvement in the FRR and flux of the membranes with CNC-Ser can be related to the presence of − COOH and − NH 2 groups in serine (M5) which led to more surface charge and hydrophilicity and also less surface roughness.
Besides, the performance of the prepared membranes for oil removal was assessed by COD removal as presented in Table 4. From COD removal results, the least efficiency for COD removal was reported for the bare membrane (M1) in all three levels of oil concentration. Similar to FRR and Flux data, the rejection performance of the membranes embedded with CNC-Ser was higher rather than the membranes modified with CNC in all three levels of oil concentration. The maximum COD removal was reported as 99% for M5 even for the least oil concentration (100 pm), indicating that M5 could be suggested to apply over the long-term filtration of oily wastewaters to provide a nearly stable high rejection and PWF (Gholami et al. 2020a(Gholami et al. , 2020b).

Long-term performance
In this section, the long-term performance of the bare membrane, M3 and M5, for oil removal from Kermanshah polymer production plant wastewater in a dead-end setup and a cross-flow setup was investigated.

Dead-end setup
Flux data for long-term filtration of distillated water and Kermanshah polymer production plant wastewater alternately for M1 (bare membrane), M3 (optimal modified membrane with CNC), and M5 (optimal modified membrane with CNC CNC-Ser) is presented in Fig. 9a for 1100 min. It is obvious from the figure that the performance of the bare membrane is not satisfied and it could not be a proper case for longterm performance as blocking took place for M1 after nearly three cycles. The main reason for this soon blocking of M1 is the hydrophobic nature of the bare PES membrane which adsorbs oily components that led to the formation of cake layer. Another reason is related to high surface roughness, causing trapping foulant agents. For modified membranes, seven cycles of distilled water were examined, and almost repeatable data were achieved. A gentle decreasing trend in water flux was observed for the modified membranes particularly M5. The better performance of M5 is related to the presence of serine, offering more negative charge, antifouling ability, and permeation flux.

Cross-flow setup
Polymer production plant wastewater was passed through a cross-flow setup for 600 min to evaluate the performance of M1, M3, and M5, and the flux data are presented in Fig. 9b. Also, to evaluate flux recovery of the membranes after 600-min filtration of polymer production plant wastewater, distillated water was passed through setup. It should be mentioned that oil was rejected over 97% for all three membranes, while M1 presented the least water flux around 10 kg/m 2 .h similar to dead-end setup, proving the inability of the bare ultrafiltration membrane to get stable oil rejection. Water flux for M3 and M5 was nearly stable around 40 and 85 kg/m 2 .h, respectively. As M5 showed the highest water flux with acceptable stability and the highest flux recovery after 600 min, it could be suggested as a qualified ultrafiltration membrane to get a high stable rejection and water flux for treating Kermanshah polymer production plant wastewater.   Moreover, in order to have a good comparison with the literature, some reported data is presented in Table 5. From Table 5, the rejection efficiency of the present research is similar to other studies, whereas the amounts of PWF and FRR are relatively high rather than the others. It should be noticed that the maximum PWF was 150 kg/m 2 .h reported from a study with TEG as an additive. Unlike the high amount of PWF, FRR value was relatively low about 45.3% for this research. In overall, cellulose nanocrystals-serine showed a good performance for oil rejection with high PWF and FRR.

Conclusion
In this study, water flux and antifouling characteristics of PES ultrafiltration membrane were improved with blending CNC and modified CNC with serine amino acid (CNC-Ser). The presence CNC and CNC-Ser progressed the hydrophilicity, porosity, negative surface potential, water flux, and flux recovery ratio of the bare PES ultrafiltration membrane. However, CNC-Ser with − COOH and − NH 2 functional groups demonstrated more significant effect on the membrane properties and oil removal performance. The optimally modified membrane (0.1wt% of CNC-Ser) showed the least WCA (41.37°), the highest PWF (91.56 kg/m 2 .h in deadend setup and 105 kg/m 2 .h in cross-flow setup) and FRR (99.2%), and great long-term filtration performance in both dead-end (1100 min) and cross-flow setup (600 min).

Fig. 9
Long-term filtration for unmodified and optimal modified membranes in dead-end (a) and cross-flow (b) setups