A Critical Review on Predicting Fouling in RO Desalination

Abstract: RO membrane fouling is one of the main challenges that membrane manufactures, the scientific community and industry professionals have to deal with. The consequences of this inevitable phenomenon have a negative effect on the performance of the desalination system. Predicting fouling in RO systems is key to evaluating the long-term operating conditions and costs. Much research has been done on fouling indices, methods, techniques and prediction models to estimate the influence of fouling on the performance of RO systems. This paper offers a critical review evaluating the state of industry knowledge in the development of fouling indices and models in membrane systems for desalination in terms of use and applicability. Despite major efforts in this field, there are gaps in terms of effective methods and models for the estimation of fouling in full-scale RO desalination plants. In existing models applied to full-scale RO desalination plants, neither the spacer geometry of membranes nor the efficiency and frequency of chemical cleanings which play an important role in the performance of this process are considered.


Introduction
Despite improvements and advances in our knowledge of water desalination, one of the main challenges of membrane technology -particularly in reverse osmosis (RO) technology -has been how to deal with membrane fouling [1][2][3].
Membrane fouling results from the accumulation of undesirable materials on, in or near the membrane and involves one or more of the following types [3,4]: (a) particulate and colloidal matter deposition on membrane surface [5]; (b) organic fouling [6]; (c) scaling and inorganic fouling [7]; and (d) biofouling due to adhesion and bacterial growth on the surface of the membrane generating a layer of gel [8].
The consequences of this inevitable phenomenon has a negative effect on the performance of the desalination system (decline in water production over time for constant pressure operations or an increase in required feed) that requires costly pretreatment, higher operating pressures, and frequent chemical cleanings, which can damage membranes, degrade permeate quality, and hasten membrane replacement.This additionally increases water cost and energy consumption [9,10].Therefore, one of the most important challenges is to understand the factors involved in membrane fouling and the subsequent reduction of permeate flux that is inevitably associated with membrane processes.
A great deal of research has been carried out to this field in the last 30 years and although desalination technology is being extensively studied, much remains to be done and researched in the field of membrane fouling.Research that has been undertaken focuses on 6 key areas: (1) characterization of foulant agents by autopsy studies of membrane elements; (2) understanding of fouling mechanisms; (3) indices for predicting fouling; (4) modeling for full-scale systems, (5) optimization of pre-treatment and chemical cleaning and (6) optimizing the membrane material and enhanced module design.The first four areas attempt to address directly how fouling occurs and how to predict it, while the others focus more on the mitigation and prevention of fouling, as for example through the use of antifouling membranes [11][12][13][14][15].
Focusing on attempts to address directly and predict model membrane fouling, several fouling prediction tools and techniques have been developed to describe membrane fouling [16][17][18][19][20].The traditional and most widely applied fouling indices in RO systems are the Silt Density Index (SDI) and the Modified Fouling Index (MFI).However, these indices have limitation in predicting RO fouling rate [21][22][23].
Some recent research has focused on modifying these methods in order to evaluate fouling potential [17,22,24,25], while another research focus is the proposal of prediction models based on experience in full-scale RO desalination plants [26][27][28][29].
This paper provides a critical review evaluating the state of industry knowledge in the development of fouling indices and models in membrane systems for desalination in terms of use and applicability.

Membrane fouling indices
Silt Density Index (SDI) and Modified Fouling Index (MFI) are common parameters or indices to determine the fouling potential (mainly colloidal) of feedwaters in RO systems.Microfiltration (MF) membranes with a pore size of 0.45 mm, which is larger by several orders of magnitude than the pore size of the RO membranes, are used to calculate these indices.Although these indices were developed to evaluate RO membrane fouling, they can also serve as reference in the evaluation of fouling in porous membranes like MF and ultrafiltration (UF).
These indices are based on conventional and dead-end filtrations, while commercial applications are performed in cross-flow filtration.This implies that the flow conditions in the module are not taken into account , though this is a crucial parameter in the optimization of the process.However, the experimental determination of these data is very simple and frequently used.

Silt density index (SDI)
The SDI is used to predict the colloidal fouling potential of feedwaters in RO systems and the efficiency of pre-treatments.SDI measurement is performed using standard (ASTM D4189 [30]).The feedwater is filtrated in dead-end mode by an MF membrane with a diameter of 47 mm and pore size of 0.45 mm at a constant pressure of 207 kPa (30 psi).The two time intervals measured at the beginning of filtration are the initial(t i ) and final(t f ) time to collect 500 mL of permeate, respectively.The third time interval (t) can be 5, 10 or 15 min, which is the period between (t i ) and (t f ).SDI is calculated by the following Equation (1): Generally, membrane manufactures suggest a value below 3 for the SDI, but, 4 or 5 are also acceptable values.Most pre-treament studies are based on SDI 15 < 3. Standard ASTM D4189 [30] specifies that the membranes must have a mean pore size of 0.45 ± 0.2 mm, and the values of the SDI obtained with membranes of different suppliers, which present differences in their morphology SDI has its limitations, and a lack of reliability has been demonstrated in several studies [31][32][33].SDI is a estatic measurement of resistance assuming lineal permeate flux decline.This allows good results to be obtained when the water has a high quality -as the initial and final fluxes would be similar.However, the use of SDI may be not appropiate when the water has a high fouling potential, since SDI has no linear relation with the colloidal content.In this case, derivation of this index is very empirical and is not based on any mechanisms of fouling [31,34].For these reaons, SDI should not be used as input in the mathematical model to predict fouling rates [35].To overcome the limitations of SDI, J.C. Schippers and J. Verdouw [31] proposed a different parameter: membrane fouling index (MFI).

Modified fouling index (MFI)
MFI (also called as MFI 0.45 ) is a parameter based on the filtration mechanism of layer deposition or cake formation and takes into account the mechanism of reduction of flow that takes place in membrane systems.Therefore, it represents better the operating conditions of the membranes than the SDI and can be used to measure water with a high and low fouling potential.
MFI [36] is determined using similar equipments and procedures as the SDI, except that the volume of permeate water is measured in 30 s intervals over 15 min of filtration.In this period, the data of permeate volume and t are collected.A better understanding of the experimental data that is obtained is achieved by using Equation (2) as porpposed by J.C. Schippers and J. Verdouw [31].Equation ( 2) is based on resistances-in-series model and considers that fouling resistance is due to cake formation on membrane surfaces.Equation (2) shows a lineal relation between t/V (s/L) and V (L).The slope of this equation is the value of the MFI (Equation ( 3)).
where ∆p (Pa) is the transmembrane pressure, µ (Pa s) is the water viscosity, Rm (m −1 ) is the hydraulic resistance of the membranes , α (m/kg) is the specific resistance of the cake, A (m 2 ) is the membrane surface, V (L) is the volume and C b (kg/m 3 ) is the concentration of particles in feedwater.
MFI is determined in the second region of the curve t/V vs V (Figure 1).It can be divided into three stages: blocking filtration, cake filtration (lineal) and cake filtration with clogging and/or cake compressure.In case of a high concentration of colloids, the graph t/V vs. V is less lineal and the MFI is calculated from the first lineal region observed [37].
Membrane manufacturers suggest using MFI <1 s/L 2 and a maximum value of 4 s/L 2 for control the membrane fouling.Most studies have been based on a target value less than 1 s/L 2 .In practice, the calculation of MFI is complex, so in most cases the SDI is calculated.Some recent research has focused on modifying these methods in order to study the applicability of multiple MFIs to evaluated fouling potential of feed water in a full-scale RO plant [22].

indices derived from MFI
J.C. Schippers y J. Verdouw [31] showed that the MFI depends on the membrane molecular weight cut-off.Few authors have developed procedures to calculate the MFI using membranes Fig. 1.Ratio of filtration time and filtrate volume (t/V) as a function of filtrate volume (V) [38].
with smaller pore size.Table 1 shows a summary of procedures for calculating the MFI, as well as the indices, parameters and methods used to measure the fouling potential.The advantages and disadvantages of each procedure have been commented on in works referenced in Table 1 and others [38,39].
S.F.E.Boerlage et al. [35] showed that the MF membrane (0.45 mm) used for the MFI was not suitable for fouling of small size colloids.This fouling can happen in RO membranes if the pre-treatment does not separate these particles.The same authors [35] developed the MFI − UF at constant pressure (MFI − UF const.pressure ).This procedure uses an UF membrane instead of an MF membrane to separate more particles, but it can take more than 20 h.
The aforementioned fouling indices have been measured at constant pressure, whereas most membrane systems works at constant flux.S.F.E.Boerlage et al. [40] further developed the MFI − UF const.pressure in order to adapt it to constant flux conditions.There resulted a noticeable difference in the duration of the test compared to the MFI − UF at constant pressure; the MFI − UF at constant flux (MFI − UF cont.flux ) could be obtained in 2 h.Recently, S. Khirani et al. [41] proposed the NF − MFI using an NF membrane to measure the MFI.As is shown in Table 1, the NF − MFI is measured at constant pressure.Khirani et al. [41] showed that fouling potential could be measured by the NF − MFI, even for small organic particles.Although this method is a step towards obtaining more realistic fouling indices, the mode of operation was still at constant pressure and dead-end flow.
Modified methods for measuring the MFI has a disadvantage that they require a long measuring time with more complex systems than SDI or MFI itself.The filtration mode is dead-end flow, so it is not close to real conditions in terms of hydrodynamic flux in RO process.Cross-flow hydrodynamic conditions influence the selective deposition of smaller particles or colloids, which are the most likely to be deposited on membranes, as illustrated in Figure 2.
Due to the balance between the convection flow and the backscattering of particles, the larger particles with higher backscattering speeds tend to move away from the surface of the membrane, whereas the smaller particles are preferably deposited as soiling agents.These cross-flow hydrodynamic conditions lead to a different composition and structure of the cake when compared

Dead-end
Cross-flow
to the final blind filtration [42].
Figure 2 shows the cross-flow filtration.All foulants in the feedwater are deposited or passed through the membrane, as in the case of the measurements of the SDI, MFI and MFI − UF, while in cross-flow filtration, foulants are fractionated by selective deposition.These hydrodynamic effects could lead to inaccuracies in the extent of SDI and MFI that is performed in dead-end flow.
To take into consideration the effect of small particles or colloids in the MFI, S.S. Adham y A.G. Fane [43] proposed the use of a selective MF membrane to be operated in cross-flow mode.They called this index the cross-flow sampler-MFI (CFS − MFI).After MF membrane filtration (colloid matter passes though this membrane), the MFI/SDI is measured as shown in Table 1.Althouh this method is a better approach, the cross-flow MF is separated from the measurement device in dead-end flow, so the CFS − MFI is determined in discontinued mode.M.A. Jaaved et al. [44] calculated the CFS − MFI in continuous operation mode with MFI (in dead-end flow) directly conected to the CFS.Recently, L.N. Sim et al. [45] applied the CFS to the MFI − UF cont. to simulate selective colloidal deposition in real RO systems.The proposed index is known as CFS − MFI UF and uses a UF membrane for the MFI.The particles that pass across the CFS and are deposited on the UF membrane will foul the RO membranes.J. Choi et al. [46] proposed procedures for measuring the MFI with different types of membranes (Table 1).The test was called the combination fouling index-MFI (CFI − MFI).It takes into account various foulant agents separated by different membranes.However, the proposed approach is not simple since several types of membranes are required.Although the different measurement systems of MFI improve the prediction of fouling in RO membranes, they are complex and require long times to be determined.

Fouling index I
In Equation (4), the term α • C b is a fouling potential measurement and is usually called fouling index I.If α and C b are known, I can be calculated using Equation 4 [53]: Following the theory of cake deposition or formation, when there is not compaction the value of cake resistance is R c .It can be rewriten as Equation (5) [31,53]:  -Fouling mechanisms: cake filtration -Test: ∆p -Fouling mechanisms: cake filtration -Test: t/(V/A) Characteristics: the test tries to take into consideration the organic matter in the feedwater.Disadvantages: the test is carried out under constant pressure and the deposition factor of particles in cross-flow is not considered.The total retention of organic matter is not achieved in this procedure.-Fouling mechanisms: cake filtration -Test:

CFS
The index I is related with MFI [31] with the parameters α and C b (Equation ( 6)): The MFI depends on the operating conditions of the filtration, ∆p and A according to Equation (6).A normalization in the same condition as SDI is required.Otherwise, I does not depend on operating conditions, so the parameter α does not vary as a result of effect of cake comprenssibility.It can be considered that I is already a normalized value of MFI, which depends on pressure and membrane surface (Equation ( 6)).However, values of MFI under different conditions of filtration with the same water sample are not the same as for I [41].Equation ( 5) is rewritten as follows: The fouling index can be interpreted as a fouling parameter referred to the increase in cake resistance (R c ) divided by specific permeate volume (V/A) (by cake formation as the only type of fouling mechanisms).
The value of R c of the deposited foulants on the membrane surface can be calculated knowing I and C b (Equation ( 6)).However, the specific resistance of the cake (or permeability of the cake) is affected by the pressure applied, and that effect can be represented (as a first approximation) by an empirical expression in the form of Equation ( 8) [54].
where α 0 is the cake specific resistance at reference pressure and ∆p is the pressure gradient working with the reference pressure.n is the comprensibility coefficient.The effects of pressure and compressibility on the characteristics of the cake and colloidal dispersion is a complicated topic that is still under investigation.Index I is defined by Equation ( 4) and its value is calculated by the experimental determination of MFI (Equation ( 6)).The parameter I is related to fouling potential of feedwater, which is defined by multiplication of two characteristics: its specific resistance α and concentration C b .

Fouling potential parameter (k fp )
L. Song et al. [50] defined a new standardization method for the determination of fouling potential in membrane processes.Initially, it was developed to evaluate the potential of colloidal fouling in UF membranes, but later was also applied in the characterization of fouling in large-scale RO processes [55][56][57].
Index k fp (Pa s/m 2 ) (called fouling potential) is defined by Equation ( 9): In Equation ( 9), J (m/s) is the specific permeate flux, and R * and R * t (Pa s/m) are the initial and final resistance of the membrane R * 0 .In this resistance, the resistive effect of the viscosity is included and is equivalent to multiplication of the resistance as it is usually considered, R (m −1 ), and the dynamic viscocity of the fluid mm (Pa s) (Equation ( 10 Peer-reviewed version available at Membranes 2017, 7, 62; doi:10.3390/membranes7040062 If the paramater k fp is assumed constant over time, it can be calculated using Equation (11): where v t is the total specific volume of parmeate over time t.

Predictive performance models
These models are an alternative to fouling indices in the prediction of the fouling influence on RO systems.Some authors [26][27][28][29] have proposed equations to estimate the decline of the permeate flux (J w ) over time due to long-term variation of the water permeability coefficient (A).Generally, these correlations are applicable for the respective membrane type and for specific operating conditions.
One of the main drawbacks in the development of this type of model is the availability of long-term operating data for a wide range of operating conditions and different types of full-scale membranes.All models aim to describe the permeate flow decline over time or the variation of the normalized water permeability coefficient A n due to compaction, fouling etc.
A proposed model to predict the decline of J w due to membrane compaction was used by M. Wilf et al. [26] to estimate the J w decline in long-term (Equation ( 13)).Three years of experimental data from different sea water reverse osmosis (SWRO) desalination plants were used to identify the parameter of the model.They calculated the parameter for permeate flow decrements of 25 and 20%.
where m is a parameter with values between -0.035 and -0.041 [26] related to permeate flow decline of 20% and 25% respectively and t is the operating time in days.
Zhu et al. [27] also proposed a model (Equation ( 14)) to predict the coefficient A. This involves an exponential equation, but in this case a hollow fiber membrane was utilized (Dupont TM B-10) during one year of operating time.This correlation is not based on experiments but on model-based simulation: variable feed pressure (6.28-7.09MPa), constant feedwater concentration and temperature (35,000 mg/L and 27 • C respectively).Belkacem et al. [58] used the Zhu model in terms of membrane resistance increase.The membrane used was the BW30LE-440 Filmtec TM in a two stage desalination plant with re-circulation during one year of operation.
where τ is a correlative parameter, the value was 328 under the aforementioned operating conditions.
Abbas et al. [28] (Equation ( 15)) proposed a model to determine the variation of the normalized average water permeability coefficient A n = A/A 0 , where A 0 is the initial average water permeability coefficient.It was an exponential equation depending on three parameters and time, the utilized membrane was the BW30-400 Filmtec TM .Five years of operating data were used for the parameter identification.The feedwater temperature was between 28 and 30 where α = 0.68, β = 79 and γ = 201.1 for the aforementioned membrane and operating conditions.
A forth model was proposed by Ruiz-García et al. [29] (Equation ( 16)).They include the parameter k fp in the model and gave specific information about the behavior of the performance decline in long-term.They proposed a two stages pattern in the decline of A in RO systems.An initial stage I, where a more pronounced decline than stage II was shown.This is mainly due to membrane compaction, irreversible fouling (strongly adherent films) and k fp .The stage II is related to a gradual decrease mostly due to irreversible fouling, and frequency and efficiency of the chemical cleaning (CC).The model described the mentioned stages by the superposition of two exponential functions.The used about 3,300 operating days of a full-scale brackish water reverse osmosis (BWRO) desalination plant to fit the parameters of the model.They got three equations, one related to maximum values of the normalized water permeability coefficient (A n ) (Post chemical cleaning (Post-CC)), average and minimum values (Pre chemical cleaning (Pre-CC)).This allowed to obtain equations to estimate a range of values for the coefficient A n in time.
The first exponential function is three parameters dependent (δ 1 , τ 1 and k fp ), and it is related to the behavior in the stage I (Fig. 3), while the second is two parameters dependent (δ 2 , τ 2 and k fp ) and it is more related to the stage II (Fig. 3).The first function gets closer to zero as the stage I ends.The δ are related with the weight of each exponential, the lower is δ 1 and the higher is δ 2 the higher is A n when the desalination plant is stabilized.The τ concern to the decline in each stage (i.e.how fast is the irreversible effects (mainly fouling) affecting performance), the larger the value, the more constant is the function.Generally, the higher k fp results en a faster decline of A n in the stage I and II.They also carried out a comparison between the different models by using their experimental data.

Conclusions and perspective of future
The analysis of the different techniques, parameters, indexes and models that have been developed to date in the characterization and evaluation of RO membrane fouling potential, reveals the existence of gaps in effective methods for the characterization and evaluation of fouling.It seems that the efforts made to advance our knowledge have turned out to be ineffective in terms of the mitigation and control of membrane fouling due to gaps in effective methods for the characterization and evaluation of fouling.The task of developing reliable fouling prediction tools is extremely important for the desalination industry, since fouling is one of the main causes of performance decrease in full-scale RO desalination plants.There are different fouling rates that have been developed and are still used in the industry, but there remains much work to be done to improve these methods, indices and evaluation parameters.Among the weaknesses or deficiencies observed in the current methods of fouling assessment are the following: (a) Most conventional indexes, SDI and MFI are not appropriate.(b) (b) There are very few studies about indices or parameters applied directly to spiral wound membranes and feedwater with high salinity.Most of the studies are applied at laboratory scale with well controlled operating conditions, flat membrane systems and at low salinity.However, it is preferable for fouling potential to be determined with RO membranes and under operating conditions similar to those of full-scale desalination plants.(c) (c) Currently, the effect of cake-enhanced osmotic pressure (CEOP) has not been taken into account in measuring fouling potential.However, CEOP can contribute to a significant loss of performance, even more than the hydraulic resistance brought about by cake formation.
The aforementioned prediction models are based on long-term data of full-scale RO desalination plants under full-scale operating conditions.However, neither the spacer geometry of membranes nor the efficiency and frequency of chemical cleanings -which play an important role in the performance of this process -are considered in these models.

Fig. 3 .
Fig. 3. Schematic presentation of the two stages in A n decline.(I) initial more pronounced drop due to compaction and irreversible fouling;(II) gradual decline mainly caused by irreversible fouling [29].