Abstract

Identifying the fouling degree of a membrane bioreactor (MBR) provides guidance on the determination of suitable membrane cleaning methods. There is still a lack of knowledge on the effects of powdered activated carbon (PAC) refresh ratio on the MBR fouling mechanism. Major fouling mechanisms of an MBR with constant flow rate at different PAC replenishment ratios were investigated by individual and combined mechanistic fouling models. The root mean square errors were employed to assess the prediction accuracy of the used fouling models. The combined models showed better prediction. The cake–complete model provided far better fits of the transmembrane pressure data, and provided good fits of other individual model predictions regardless of the PAC refreshment ratio. Fourier transform infrared spectroscopy, scanning electron microscopy and energy dispersive X-ray spectroscopy confirmed that the cake layer clogging was the main fouling mechanism followed by complete blockage and standard plugging. The cake–complete model may be used to predict the fouling mechanisms in PAC/MBR systems.

INTRODUCTION

Membrane bioreactor (MBR) technology is considered a well-established, mature technology with many full-scale plants around the world treating municipal and industrial wastewater (Krzeminski et al. 2017). However, membrane fouling is responsible for higher operation costs of MBRs due to the requirement for frequent cleaning of the foulants and the replacement of the membrane. Therefore, to operate the MBR more efficiently and economically, the development of predictive/simulative models of the membrane filtration process has become an important area in recent years (Liu & Kim 2008; Madaeni et al. 2010). Models play an important role in explaining the phenomenon of membrane fouling (Liu & Kim 2008). Furthermore, the model in practical application assists in the determination of the optimum operation conditions to minimize fouling and determination of the frequency of membrane cleaning (Mah et al. 2012).

Specific fouling mechanisms were divided into four blockage types based on the difference of membrane fouling progress (Hermia 1982), including complete blocking, intermediate blocking, cake layer clogging and standard blocking mechanism. They were widely used to make a description and prediction of the types of fouling mechanisms (Bolton et al. 2006; Chen et al. 2011). Considering the complicated situation of the MBR (biology, physics, chemistry, hydrodynamics, etc.), the fouling phenomenon is too complex to be described using a single fouling mechanism. Hu et al. (2018) reported that the single blocking model cannot accurately explain the membrane fouling of MBR system.

In recent years, a few studies have made the effort to develop new mixed blockage models (a model containing several fouling mechanisms) to more accurately explain and predict the complex mechanism in the MBR system (Bolton et al. 2006; Kim et al. 2013). Bolton et al. (2006) developed two-stage fouling models and deduced five types of mixed blockage models, finding that the combined models provided better prediction for the biological fluids filtration data of the filtration process. The combined model predictions exhibited higher credibility than those of individual models regardless of sludge retention time (Hu et al. 2017). Kim et al. (2013) introduced a new composite model and fitted the experimental data of the MBR system to find that the complex model was the best explanation for the complex membrane fouling in the actual MBR system.

Powdered activated carbon (PAC) was used to reduce membrane fouling in the MBR. Addition of PAC reduced the compressibility of the sludge floc and extended the membrane filtration cycle, and then reduced sludge deposition on the surface of the membrane (Khan et al. 2012). Thus, PAC could improve sludge characteristics and effluent quality and reduce membrane fouling. However, aged PAC deteriorated filtration performance rather than reduced fouling (Remy et al. 2010). Therefore, it is necessary to steadily renew the aged PAC. Our earlier work found out that an MBR with optimum PAC replenishment exhibited low fouling propensity and prolonged filtration. Whether PAC replenishment affects the membrane fouling mechanism needs further research.

The objective of this work was to evaluate the viability of the individual and combined fouling models used for the fit of experimental transmembrane pressure (TMP) data at constant flow. TMP profiles were obtained in bench-scale MBRs with different PAC replenishment ratio under continuous long-term operation. The prediction accuracy of the fouling models have been compared and discussed. Scanning electron microscopy and energy dispersive X-ray spectroscopy (SEM-EDS) was employed to investigate the distribution and composition of contaminants on the membrane. It is expected that identification of the dominant fouling mechanism will provide guidance for determining appropriate membrane cleaning methods in PAC hybrid MBR systems.

MATERIAL AND METHODS

MBR setup

Four identical laboratory-scale MBRs were used and fed with synthetic wastewater (Zhang et al. 2019). The synthetic wastewater consisted of glucose 344 mg/L, urea 33 mg/L, (NH4)2SO4 120 mg/L, KH2PO4 15 mg/L, K2HPO4 20 mg/L, MgSO4 · 7H2O 24.4 mg/L, CaCl2 · 2H2O 12 mg/L, FeCl3 · 7H2O 1.5 mg/L, ZnSO4 · 7H2O 0.15 mg/L, CuSO4 · 5H2O 0.05 mg/L, and Pb(NO3)2 0.26 mg/L, and NaHCO3 was added to adjust the pH value. The membrane flux in the MBR was kept at 15 L/(m2 h) on a 2 min-off/8 min-on cycle. The mean pore radius of the polyvinylidene fluoride (PVDF) hollow-fiber membrane (MOF-1B-W, MOTECH, China) was 0.2 μm and its effective area was 0.3 m2. The oxygen concentration was controlled at 2–4 mg/L in the tank. The hydraulic residence time was 6.1 h for all the MBRs and the operating temperature was room temperature. A resolution pressure sensor was used to monitor the TMP. The replacement ratios of PAC were 0%, 1.25%, 1.67% and 2.50% for the MBRs, which corresponded to a PAC residence time of ∞, 80, 60 and 40 d, respectively. Sludge retention time (SRT) was the same as the retention time of PAC. The corresponding MBRs were named as MBR-A, MBR-B, MBR-C and MBR-D, respectively. The properties of the mixed liquor and the performance of MBRs are presented in Table 1. The chemical oxygen demand (COD) and NH4+-N concentrations in effluent were lower than 20 mg/L and 0.5 mg/L, respectively. Their corresponding removal rate was stable at 95 ± 2% and 96 ± 3%. There were no significance differences in COD and NH4+-N removal efficiency between the MBRs.

Table 1

Mixed liquor properties and MBR performance

MBR MLSS (g/L) Polysaccharide (mg/gVSS) Protein (mg/ gVSS) COD removal rate (%) NH4+-N removal rate (%) 
8.2 ± 1.8 0.7 ± 0.3 3.4 ± 1.7 94.3 ± 1.0 94.2 ± 0.8 
6.7 ± 1.0 1.2 ± 0.6 4.0 ± 2.6 95.4 ± 0.8 98.8 ± 1.1 
7.1 ± 0.9 1.1 ± 0.6 2.9 ± 1.2 95.4 ± 0.5 98.9 ± 1.5 
6.7 ± 0.8 1.0 ± 0.6 3.4 ± 1.6 95.0 ± 0.4 97.1 ± 0.6 
MBR MLSS (g/L) Polysaccharide (mg/gVSS) Protein (mg/ gVSS) COD removal rate (%) NH4+-N removal rate (%) 
8.2 ± 1.8 0.7 ± 0.3 3.4 ± 1.7 94.3 ± 1.0 94.2 ± 0.8 
6.7 ± 1.0 1.2 ± 0.6 4.0 ± 2.6 95.4 ± 0.8 98.8 ± 1.1 
7.1 ± 0.9 1.1 ± 0.6 2.9 ± 1.2 95.4 ± 0.5 98.9 ± 1.5 
6.7 ± 0.8 1.0 ± 0.6 3.4 ± 1.6 95.0 ± 0.4 97.1 ± 0.6 

MLSS, mixed liquor suspended solids; VSS, volatile suspended solids; COD, chemical oxygen demand.

Fouling models

Four single blockage models deduced from Darcy's law were considered: standard clogging, complete clogging, intermediate blockage, and cake layer clogging (Bolton et al. 2006). In order to explain and predict the complex membrane fouling phenomenon in MBR systems mixed models consisting of two single blocking models were used. The models are shown in Table 2.

Table 2

Equations of the nine fouling models

No. Model Equation 
Standard  
Cake  
Complete  
Intermediate  
Cake–complete  
Cake–intermediate  
Complete–standard  
Intermediate–standard  
Cake–standard  
No. Model Equation 
Standard  
Cake  
Complete  
Intermediate  
Cake–complete  
Cake–intermediate  
Complete–standard  
Intermediate–standard  
Cake–standard  

P is pressure at time t (kg/ms2), P0 is pressure at time t = 0 (kg/ms2), J0 is the initial flux, Ks is the standard blocking constant, Kc is the cake layer clogging constant, Kb is the complete blocking constant and Ki is the intermediate blocking constant.

The fitting degree of model predicted values and the actual values was compared by the root mean square error (RMSE). RMSE can reflect the error between the model and the actual value. The unit of RMSE can be unified as Pa or kPa, so the difference between membrane fouling types can be determined by comparing the size of RMSE values among different MBRs. In this work, the RMSE unit was unified as kPa. RMSE was calculated by the equation: 
formula
(1)
where and are the measured and the corresponding predicted data, respectively.

Analytical methods

The fouled membrane was obtained from the membrane module when TMP was 40 kPa. The membrane samples were dried in desiccators at room temperature.

Fourier transform infrared spectroscopy (FTIR) (VERTEX 70, Bruker, Germany) was used to study structural information of the foulants extracted from the membrane module. Membrane characterizations for the verification of the combined model included SEM and EDS (SU8010, Hitachi, Japan). The dry membrane sample was immersed in liquid nitrogen and fractured carefully. Then the sample was fixed on the metal carrier and a layer of gold was sputtered in a vacuum. The morphology of the membrane was observed by scanning electron microscope. During SEM observation, EDS analysis was used to obtain elemental composition for further understanding of the contaminants. In this study, EDS analysis was performed by SEM and its associated EDS analyzer. The average value of each element is obtained by the average of three measurements per sample.

RESULTS AND DISCUSSION

Blocking model analysis

The experimental data TMP versus time was used for the prediction of the membrane fouling. Moreover, macroscopic TMP data reflected the state of the reactor, including mechanical cleaning and adsorption due to aeration and PAC refreshment in the MBR. Experiment data of TMP versus time for the four MBRs was simulated by nine models listed in Table 1. The data were fit with the classical blocking laws and combined models as shown in Figures 1 and 2, respectively. RMSE values were also obtained and are presented in Table 3.

Table 3

RMSE obtained by the fit of the fouling models in the MBRs

Model MBR-A MBR-B MBR-C MBR-D 
Intermediate 2.54 1.58 0.92 2.82 
Cake 4.51 4.66 5.31 4.31 
Standard 8.2 6.33 5.05 8.11 
Complete 19.61 23.48 35.3 32.53 
Cake–complete 0.82 0.44 0.55 0.51 
Cake–intermediate 1.32 1.21 0.9 0.95 
Intermediate–standard 1.39 1.16 0.89 1.43 
Cake–standard 1.75 1.33 1.19 1.39 
Complete–standard 8.2 6.33 5.86 8.11 
Model MBR-A MBR-B MBR-C MBR-D 
Intermediate 2.54 1.58 0.92 2.82 
Cake 4.51 4.66 5.31 4.31 
Standard 8.2 6.33 5.05 8.11 
Complete 19.61 23.48 35.3 32.53 
Cake–complete 0.82 0.44 0.55 0.51 
Cake–intermediate 1.32 1.21 0.9 0.95 
Intermediate–standard 1.39 1.16 0.89 1.43 
Cake–standard 1.75 1.33 1.19 1.39 
Complete–standard 8.2 6.33 5.86 8.11 
Figure 1

Pressure vs. time data fit with the four classical individual models.

Figure 1

Pressure vs. time data fit with the four classical individual models.

Figure 2

Pressure vs. time data fit with the combined model.

Figure 2

Pressure vs. time data fit with the combined model.

Individual models

Figure 1 shows the predicted results of the experimental data fitted by the single models. It is clear that the intermediate blocking model provided better fits for the data of the four MBRs, which had the lowest RMSE as shown in Table 3. The smaller the RMSE value of the fouling model, the closer the fouling of membrane module in actual operation is to this type of fouling mechanism. The superiority of the models provided followed the order: intermediate > cake > standard > complete model. The difference between the measured and predicted data was large for the complete model; thus, there were relatively larger RMSEs for the complete model. This may be because the complete blocking mechanism dominated only in the early stage of membrane fouling (Duclos-orsello et al. 2006). The second largest value of RMSE was observed for the standard model, which happened in the final stage of fouling. These results indicated that the intermediate blocking played a dominant role during the operating time of the TMP data. Additionally, it is clear that the PAC replenishment ratio could not affect the individual fouling mechanism for the MBRs.

Combined models

The single model cannot accurately and effectively identify the type of membrane fouling mode under certain conditions (i.e. SRT) (Sabia et al. 2016). Figure 2 shows the predicted result using the combined models for the TMP data.

The RMSE values of the combined mechanistic models were lower than those obtained by the single models. The combined mechanical model well reflected the comprehensive effects of various fouling mechanisms in the MBR system, and had a better ability to predict membrane fouling. The trend of data predicted by the combined model appeared to be similar except for the complete–standard model. The cake model combined with other individual models provided better fits than did the other combined models. Moreover, the cake–complete model provided better fits, and was superior to the other combined models. This finding was consistent with the result reported by Bolton et al. (2006) who found that the combined cake–complete model provided the best fits of both the sterile filtration data and nearly the best fit of the virus filtration data, which suggested that the cake–complete would be useful for fitting data from a wide range of solutions and membrane types. The results also suggested that the cake–complete model provided better fits for the data of the four MBRs operated in this study regardless of PAC refreshment ratio. These results revealed again that PAC refreshment had no effect on the membrane fouling mechanism of the PAC-MBR system. The cake–complete model may be applicable to a system in which the model is consistent with the fouling mechanism observed by experiments.

The possibility of the cake–complete model to fit the prediction of each individual model is also demonstrated in Figure 3. The cake–complete predictions were the same as the complete blocking model or cake layer clogging model. It should be noted that the cake–complete model could make predictions that were almost the same as those of the standard or intermediate blocking models. This result indicated that the cake–complete model will be useful for fitting data similar to each individual model.

Figure 3

Cake–complete model (solid lines) fit to the predictions of the four individual models for TMP in the MBRs.

Figure 3

Cake–complete model (solid lines) fit to the predictions of the four individual models for TMP in the MBRs.

The contributions of the component models to the combined models were evaluated by using the fit parameters (Kc, Kb, Ki, Ks). The specific model constants of the cake–complete model fit for the predictions of the combined models are listed in Table 4.

Table 4

Constants of cake–complete model fit for the predictions of the combined models

Model MBR-A MBR-B MBR-C MBR-D 
Cake–complete Kb = 0.01515 Kb = 0.01456 Kb = 0.01080 Kb = 0.00964 
Kc = 0.16352 Kc = 0.11171 Kc = 0.05806 Kc = 0.12025 
Cake–intermediate Ki = 0.00445 Ki = 0.00753 Ki = 0.04515 Ki = 0.00081 
Kc = 0.01120 Kc = 0.00604 Kc = 7.94 × 10−7 Kc = 0.06289 
Cake–standard Ks = 0.03233 Ks = 0.02894 Ks = 0.01992 Ks = 0.02169 
Kc = 0.28202 Kc = 0.19884 Kc = 0.10835 Kc = 0.19567 
Complete–standard Kb = 1.18 × 10−11 Kb = 8.54 × 10−12 Kb = 8.68 × 10−18 Kb = 0.02296 
Ks = 0.03460 Ks = 0.03062 Ks = 0.02419 Ks = 0.09095 
Intermediate–standard Ki = 0.00361 Ki = 0.00361 Ki = 0.00361 Ki = 0.00361 
Ks = 0.12225 Ks = 0.11374 Ks = 0.0512 Ks = 0.20035 
Model MBR-A MBR-B MBR-C MBR-D 
Cake–complete Kb = 0.01515 Kb = 0.01456 Kb = 0.01080 Kb = 0.00964 
Kc = 0.16352 Kc = 0.11171 Kc = 0.05806 Kc = 0.12025 
Cake–intermediate Ki = 0.00445 Ki = 0.00753 Ki = 0.04515 Ki = 0.00081 
Kc = 0.01120 Kc = 0.00604 Kc = 7.94 × 10−7 Kc = 0.06289 
Cake–standard Ks = 0.03233 Ks = 0.02894 Ks = 0.01992 Ks = 0.02169 
Kc = 0.28202 Kc = 0.19884 Kc = 0.10835 Kc = 0.19567 
Complete–standard Kb = 1.18 × 10−11 Kb = 8.54 × 10−12 Kb = 8.68 × 10−18 Kb = 0.02296 
Ks = 0.03460 Ks = 0.03062 Ks = 0.02419 Ks = 0.09095 
Intermediate–standard Ki = 0.00361 Ki = 0.00361 Ki = 0.00361 Ki = 0.00361 
Ks = 0.12225 Ks = 0.11374 Ks = 0.0512 Ks = 0.20035 

The terms KcJ0, Kb/J0, Ki and Ks have units of m−1. Their contributions to the combined models can be compared according to their order of magnitude; i.e., when their contributions to the combined models are similar, the terms will have a similar magnitude (Bolton et al. 2006). For example, the contributions of the cake to the combined cake–complete, cake–intermediate, and cake–standard models were evaluated by KcJ0/(Kb/J0), KcJ0/Ki and KcJ0/Ks, respectively. The contributions of the complete blocking to the complete–standard model and of intermediate blocking to the intermediate–standard model were evaluated by Kb/(J0/Ks) and Ki/Ks, respectively. The cake model was the major component of the cake model combined with other single models as shown in Table 5. The standard model was the main contribution to the intermediate–standard model. It was also the main contribution to the complete–standard model, and the values of Kb/J0Ks ratio indicated that the complete blocking model had been the negligible component. Thus, the cake and standard blockings were the major contribution components of the specific combined models. These results suggested that cake and standard blockings were the fouling mechanisms.

Table 5

Contribution of the component models to the combined models

Model MBR-A MBR-B MBR-C MBR-D 
Cake–complete 16.9 12.0 8.4 19.5 
Cake–intermediate 3.2 1.0 2.2 × 10−5 9.7 
Cake–standard 10.9 8.6 6.8 11.3 
Complete–standard 2.7 × 10−10 2.2 × 10−10 2.9 × 10−16 9.7 × 10−12 
Intermediate–standard 0.7 0.6 0.9 0.3 
Model MBR-A MBR-B MBR-C MBR-D 
Cake–complete 16.9 12.0 8.4 19.5 
Cake–intermediate 3.2 1.0 2.2 × 10−5 9.7 
Cake–standard 10.9 8.6 6.8 11.3 
Complete–standard 2.7 × 10−10 2.2 × 10−10 2.9 × 10−16 9.7 × 10−12 
Intermediate–standard 0.7 0.6 0.9 0.3 

FTIR and SEM-EDX analysis

To better understand the fouling mode in the MBRs, the actual membrane situation was investigated.

FTIR analysis

FTIR analysis was performed to study the organic components on the fouled membrane surface and the result is presented in Figure 4.

Figure 4

FTIR spectra of the fouling layer on the fouled membrane surface.

Figure 4

FTIR spectra of the fouling layer on the fouled membrane surface.

In Figure 4, C=O stretch (amide I) and N-H at the peak (1,627 cm−1) were attributed to the presence of proteins. The peaks around 1,200–900 cm−1 together with the peak at 3,427 cm−1 represented carbohydrates (Kimura et al. 2015). The fingerprint region between 900 and 600 cm−1 showed ring vibrations of nucleotides and aromatic amino acids, and the band at 1,386 cm−1 corresponded to C–O stretching of carboxylate groups attributed to the presence of uronic acids (Badireddy et al. 2010). The other substance included fats and lipids (peak near 2,927 cm−1). As can been seen from the relative intensity of the peak, the uronic acids in the cake layer of the fouled membrane for MBR-A were obviously more than those in the other three reactors, indicating that the renewal of PAC could slow down the accumulation of uronic acid pollutants in the cake layer. The mitigation effect was more obvious with the increase of PAC renewal rate. MBR-A had a stronger absorption peak at 1,141 cm−1 while the other three reactors had absorption peaks near 1,097 cm−1, which were ascribed to polysaccharides. The polysaccharide pollutants in the cake layer of MBR-A were relatively higher than those in the other reactors. In the aromatic region, the absorption peaks appeared in all four reactors near 833 cm−1 and 599 cm−1, suggesting that there were a lot of aromatic pollutants in the cake layer. Hence, the major organics in the fouling layers were identified as polysaccharides and proteins, indicating the existence of biopolymers in the cake layer. It can be concluded that the renewal of PAC could reduce the foulants in the cake layer. In addition, with the increase of PAC renewal ratio, the pollutants on the membrane surface decreased. This could be explained by the fact that the aging PAC and sludge extraction and the addition of new PAC aggravated the disturbance in the reactor, and the fluidized PAC led to scrubbing of the membrane surface during the filtration process. Moreover, the greater the renewal rate of PAC, the better the scouring effect.

SEM-EDS analysis

Figure 5 shows the SEM-EDS analysis of the fouled membrane surface. The surfaces of the fouled membranes were covered by granular pollutants and even cake layers with different thickness. The whole surface of the fouled membrane from MBR-control was overlapped by a large number of particles with uneven particle size. There were few contaminants on the surface of the membranes from MBR-B and some large particles existed, which seemed to occur as a thin layer on the surface. Much less contaminant presented on the fouled membrane surface of MBR-C, and the membrane pore could be seen. The surface of the fouled membrane from MBR-D had particle pollutant with larger size and a compact cake layer formed, resulting in a wrinkled membrane surface due to filtration resistance (Lim & Bai 2003). The particles with larger size blocked the membrane pore. It is inferred that optimal PAC replacement led to reduction of pollutants adsorption/deposition on the surface of the membrane. Cake layer clogging occurs when particles accumulate on the surface of a membrane in a permeable cake, increasing thickness, which adds resistance to the flow. Complete blocking assumes that particles seal off pore entrances and prevent flow. The results indicated that the fouling mechanisms included cake layer clogging and complete blocking.

Figure 5

Surface SEM photographs and SEM-EDS analyses of the membrane from the MBRs.

Figure 5

Surface SEM photographs and SEM-EDS analyses of the membrane from the MBRs.

Fluorine element as a base indicator of the membrane material, PVDF, was found to be decreased, confirming the formation of fouling covering the membrane surface. The N and O element contents increased from 0 and 4.7% of fresh membrane to 5.3–6.9% and 7.5–10.3% of the fouled membrane, respectively. The C element content also increased. Elements of C, N and O are the basic elements of microorganisms and metabolic organic matter, accounting for 59.3–62.5%, suggesting that there were large amounts of microorganisms and metabolites on the surface of the fouled membrane. In addition, an increase of C element indicated PAC accumulation as PAC became trapped on the membrane surface due to its higher hydrophobicity.

Excluding C, N and O, peaks were detected for K, Na, Mg, Ca, Al and Si, S, P, and Cl, whose contents were relatively lower. The inorganic elements may exist in the form of Na+, Al3+, K+, Ca2+ and SiO32−, PO43−, SO42−, and Cl. Thus, Ca-P, silicate and sulfate precipitates were the main inorganic component of the cake layer and particle pollutants (Yurtsever et al. 2017). Additionally, the inorganic elements could bridge the cells and biopolymer and help form a dense cake layer (Hu et al. 2017). Therefore, Al, Mg and Ca played a significant role in the cake layer formation. Additionally, the formed particle pollutants sealed off pore entrances and prevented flow, resulting in the complete blocking. These results suggested that the cake–complete model was the most useful, also shown by its ability to provide good fits of all data sets, and provide good fits of each of the other combined model predictions. It should be noted that the MBR-A and MBR-B suffered more seriously from fouling caused by inorganic matters, meaning that PAC replenishment reduced the formation of cake layer.

SEM-EDS and FTIR validated the results of the combined models analysis. The cake–complete model was proven to be capable of providing a detailed description of the fouling mechanism for the MBR tested in this work. The proposed method can provide guidance for determining the appropriate membrane cleaning methods by identifying the major fouling mechanism. The results of the study can be used as a reference for model-based analysis of the effect of different operating conditions on fouling mechanisms in MBRs. It is expected that the combined mechanistic fouling model facilitates the efficient and economic operation of the MBR process. However, the various models were applied to laboratory scale in a relatively short experimental time under well-defined operating conditions. Further research should be the model-based analysis of the fouling mechanism of a pilot plant MBR, fed by real municipal wastewater and operated over a long experimental period.

CONCLUSIONS

The applicability of the blocking models to the TMP data for the hybrid PAC-MBR at different PAC refreshment ratio was examined. The combined cake layer clogging and complete blocking model provided good fits for all data and each individual model's predictions. The most dominant fouling mechanism was cake layer clogging for the MBR with different PAC replenishment. FTIR and SEM-EDS analysis showed that inorganic precipitates, organic and metabolites accumulated on the membrane, which further confirmed the formation of filter cake. The proposed combined fouling model has the potential to improve the operation of the PAC-MBR process.

ACKNOWLEDGEMENT

This work was supported by the Fundamental Research Funds for the Central Universities of China (No. 2662018JC013, 2662017JC019).

REFERENCES

REFERENCES
Badireddy
A. R.
,
Chellam
S.
,
Gassman
P. L.
,
Engelhard
M. H.
,
Lea
A. S.
&
Rosso
K. M.
2010
Role of extracellular polymeric substances in bioflocculation of activated sludge microorganisms under glucose-controlled conditions
.
Water Res.
44
(
15
),
4505
4516
.
Duclos-Orsello
C.
,
Li
W. Y.
&
Ho
C. C.
2006
A three mechanism model to describe fouling of microfiltration membranes
.
J. Membr. Sci.
280
(
1–2
),
856
866
.
Hermia
J.
1982
Constant pressure blocking filtration laws application to power-law non-newtonian fluids
.
Trans. Ind. Chem. Eng.
60
(
3
),
183
187
.
Hu
B.
,
Zuo
X. T.
,
Xiong
J.
,
Bao
R. L.
&
He
J. J.
2017
Fouling modeling of the mixed liquor in MBR by the individual and combined models
.
Water Sci. Technol.
76
(
4
),
761
775
.
Hu
B.
,
Zuo
X. T.
,
Xiong
J.
,
Yang
H. M.
,
Cao
M. H.
&
Yu
S. L.
2018
Identification of fouling mechanisms in MBRs at constant flowrate: model applications and SEM-EDX characterizations
.
Water Sci. Technol.
77
(
1
),
229
238
.
Kimura
K.
,
Ogyu
R.
,
Miyoshi
T.
&
Watanabe
Y.
2015
Transition of major components in irreversible fouling of MBRs treating municipal wastewater
.
Sep. Purif. Technol.
142
,
326
331
.
Madaeni
S. S.
,
Hasankiadeh
N. T.
,
Kurdian
A. R.
&
Rahimpour
A.
2010
Modeling and optimization of membrane fabrication using artificial neural network and genetic algorithm
.
Sep. Purif. Technol.
76
(
1
),
33
43
.
Remy
M.
,
Potier
V.
,
Temmink
H.
&
Rulkens
W.
2010
Why low powdered activated carbon addition reduces membrane fouling in MBRs
.
Water Res.
44
,
861
867
.