Stepwise membrane fouling model for shear-enhanced filtration of alfalfa juice: experimental and modeling studies

Abstract

Alfalfa juice is predominantly created from the production of fodder pellets for cattle. Due to its high nutritive value and abundant sources, an inappropriate treatment may cause important environmental problems. Membranes are used to separate and concentrate leaf protein from alfalfa juice. However, membrane fouling reduces process efficiency and hinders wide application. This study investigated the fouling of MF and UF membranes by foulants of alfalfa juice. Permeate flux gradually increased with TMP in a stepwise pattern. A stepwise multisite Darcy's law model (SMDM) was proposed to simulate the stepwise multisite fouling process. Besides, the resistance coefficient and compressibility for different steps and sites were calculated to explain the complex fouling process. The effects of feed composition, membrane and hydraulic conditions on the stepwise fouling process were investigated. These various factors have a strong effect on the fouling process of industrial membranes. A series of long term tests were utilized to study flux decline and membrane fouling at various steps and sites of the fouling process. The results of this investigation can help us to understand the fouling process of alfalfa juice and facilitate membrane fouling control.

---

1. Introduction

Alfalfa is a common perennial vegetable and has been widely cultivated worldwide.¹ Owing to its high crude protein content (about 2600 kg ha⁻¹), high yield and high feed value, alfalfa is used as a raw material for the production of fodder pellets for cattle.² During the production process, the green crop is chopped, pressed and dried, producing much alfalfa juice. It is usually considered as waste and used as a fertilizer to spread on fields, causing serious environmental pollution problems.³ In fact, leaf proteins in alfalfa juice containing 50% lipophilic proteins and 50% hydrophilic proteins, have a balanced aminogram and a high digestibility. They possess significant functional properties, such as foaming, emulsifying and jellifying agents.^4–6 Due to its high nutritive value, abundant sources and absence of animal cholesterol, alfalfa extracted juice can be concentrated and used to produce high quality nutrition food for animals and human consumption with effective separation and concentration methods, while eliminating water pollution.⁷

As a promising method to separate dissolved and suspended matter according to their molecular scales, membrane technologies have been widely applied in various chemical and biochemical processes of food industry and wastewater treatment. They can treat meat and fish products, vegetal extracts and juices, dairy products and effluents,⁸ because this process involves no chemical agents and phase change is more economic and environmentally friendly.^9,10 A few previous studies^3,11–14 used UF to separate and concentrate crude protein from waste leaf extraction juice. Koschuh et al.¹² found that UF had a higher concentration capacity of crude protein, in comparison with heat coagulation and centrifugation. By combining UF and discontinuous diafiltration, Fernandez et al.^11,13 claimed that the obtained product had a much higher dry protein content and salt reduction was significantly reduced, which is more suitable for use in balanced animal feed formulations. In our previous study,³ dead-end filtration and dynamic filtration modules were used to separate and concentrate leaf protein in alfalfa juice. Serious flux decline caused by concentration polarization and subsequent membrane fouling occurred during concentration process, increasing costs and restricting its sustainable operation and industrial application.¹⁵ Membrane fouling is due to deposition of suspended matter on membrane surface or inside membrane pores, which adds an extra resistance to filtration.¹⁶ Flux decays at constant TMP or requires an increasing TMP to maintain a constant flux fouling resistance lowers membrane performance and can be removed by implementation of various flux decline control strategies, but operation cost increases.¹⁵ However, the selection of a suitable flux decline control strategy requires an analysis of membrane fouling in depth.

The main fouling mechanisms for MF and UF of proteins are cake formation, pore blocking, gel formation and adsorption.¹⁷ Pore blocking and adsorption occurred on membrane surface or into membrane pores when proteins deposit and block membrane pores.¹⁸ Cake formation and gel formation are produced by the accumulation of proteins on membrane surface.¹⁹ In initial stage of filtration, proteins are adsorbed to membrane, thus pore blocking and adsorption are main fouling mechanisms. But after a period of filtration, a cake layer or gel forms and covers the pores of the membrane.¹⁶ Therefore, for a long filtration process, such as protein concentration in industry, pore blocking and adsorption can be negligible, while cake formation and gel formation are considered to be the most severe fouling mechanisms. In industrial filtration, a high TMP is usually selected to improve flux behavior and production efficiency. The increase of TMP compresses fouling layer, increases irreversible fouling, reduces flux behavior and affects membrane sustainable use. The cake layer can be highly compressed and affect filtration performance. A number of filtration models^20–23 have been employed to describe fouling processes on membrane surface. Factors affecting adsorption process were also evaluated. However, limited studies have focused on interactions of TMP with compressibility of cake layer. Lee et al.²⁰ studied a mathematical flux decline model for cross-flow UF of colloidal suspensions. The compressibility factor is related to porosity of cake layer and TMP. Carrère et al.²¹ investigated a fouling model for clarification of lactic acid fermentation broths by cross-flow MF. The compressibility factor and filtration time were exponentially related. Kim et al.²² used UF to filtrate a synthetic waste water containing natural organic matter and described fouling mechanisms and cake compressibility coefficients of various organic wastewaters. Although these findings are encouraging, the compressibility of a protein cake layer on membrane surface may not be completely understood and a compressibility index alone can't reflect the complexity of fouling process.

Actually, membranes fouling affinity scales and binding energies vary in different filtration stages, and the relationship between foulants and membranes may be different.²⁴ Besides, the mechanism for distribution and transportation of solutes from bulk solution to membrane is complicated.¹⁵ The exact process of protein fouling on membrane is still uncertain and the influencing factors need to be clarified. To fill the gap in the study reported here, we therefore propose to investigate the fouling process of leaf protein on membrane. To our knowledge, this is the first investigation to explore the fouling behavior of alfalfa juice filtration, and to discuss the environmental issues related to leaf protein concentration from alfalfa juice by membrane technology. A stepwise membrane fouling model is firstly proposed to simulate the fouling process. The effects of three main factors (feed water composition, membrane types and hydraulic conditions on membrane surface) affecting flux decline on fouling process are also estimated. Then, a series of long term tests were utilized to study the flux decline and membrane fouling at various fouling step process. The results should reveal the fouling process of alfalfa filtration and facilitate assessment and control of membrane fouling in membrane engineering.

2. Materials and methods

2.1 Test fluid

Alfalfa juice provided by Luzéal, Pauvres, France, was pre-filtered by a mesh of 0.4 mm pore size and mixed, then stored at the temperature of −20 °C until further use. In order to prevent serious membrane fouling, before experiment the juice was centrifuged at 4000 rpm for 10 min using a Sigma 3-16P device for separating the insoluble materials. The main characteristics of alfalfa juice are shown in Table 1.

Table 1 Main characteristics of alfalfa juice

Index	Alfalfa juice
Crude protein (g L^−1)	21
Chlorophyll a (mg L^−1)	12.38
Chlorophyll b (mg L^−1)	20.82
Dry matter (g L^−1)	86
Ash (g L^−1)	21
Turbidity (NTU)	600
Conductivity (ms cm^−1)	9.29
pH	5.8
Soluble matter (°Brix)	8.1
Density, ρ (g mL^−1)	1.20

---

2.2 Filtration modules

The rotating disk module (RDM), which consists of a disk mounted on a shaft and rotating near a fixed circular membrane, has been designed and built in our laboratory (Fig. 1). A flat membrane, with an effective area of 176 cm² (outer radius R = 7.72 cm, inner radius r = 1.88 cm), was fixed on the cover of the cylindrical housing in front of the disk. The disk equipped with 6 mm-high vanes can rotate at up to 2500 rpm, inducing high shear rates on the membrane surface. The shear rate could be adjusted by modifying the shear rate of the disk.

Fig. 1 Schematic of the rotating disk module (RDM) and of the disk with vanes.

MF and UF membranes fabricated by MICRODYN-NADIR GmbH were tested in the present study. According to manufacturer's information, their properties are summarized in Table 2.

Table 2 Properties of membranes tested

Membrane	Manufacturer	Surface material	Water permeability (L m^−2 h^−1 bar^−1)
P020F (UF 20 K)	Nadir	PH-PES	43
MV020T (MF 0.2 μm)	Nadir	PVDF	710

---

2.3 Experimental procedure

A new membrane was used for each series of experiments to ensure the same initial membrane conditions for the entire study. The membranes were soaked in deionized water for at least 24 h before use, and pre-pressured with deionized water for 1 hour under a pressure of 2 bar. After stabilization, the pure water flux of membranes was measured to calculate water permeability (L_p). Before experiments started, the feed was heated to 35 °C, and was fully recycled in the system at zero TMP, and this process lasted about 10 min for each test. Then experiments were performed with TMP stepping tests with full recycling. TMP was increased in steps from 0 to 3.0 bar for MF or 6.0 bar for UF at a fixed rotating speed with full recycling operation. Each TMP step lasted 10 min and the corresponding flux was measured every 2 min and TMP intervals were 0.25 bar. Tests were made for both MF and UF at rotating speeds of 2500, 2000, 1500, 1000 and 500 rpm, with volume reduction ratios (VRRs) of 1, 3 and 6.

2.4 Analytical methods

Turbidities of permeate were measured with a Ratio Turbidimeter (Hach, USA). Conductivity was measured with a Multi-Range Conductivity Meter (HI 9033, Hanna, Italy) and pH was measured with a pH Meter (MP 125, Mettler Toledo, Switzerland). Dry matter was determined by measuring the weight loss after drying samples at 105 ± 2 °C for 5 h in an oven. Soluble matter measurements (°Brix) were done, at room temperature, by means of a digital refractometer PR-32α (ATAGO Co., Ltd, Japan). The crude protein concentration in solution was determined using the Kjeldahl method. To convert organic nitrogen, except ammonium, the factor 6.25 was used.⁸ Chlorophyll A and B were measured by a spectrophotometric method (Biochrom Ltd., Cambridge Science Park, Cambridge, Angleterre).

2.5 Calculated parameters

The permeate flux (J) is calculated by:where A is the effective membrane area (m²), V is the total volume of permeate (m³), and t is the filtration time (h).

VRR is defined as:

where V₀ and V_R are initial feed volume and retentate volume, respectively.

The mean TMP is obtained by integrating the local pressure p_c (Pa) over the membrane area as follows:

where ρ is the density of the fluid (g L⁻¹), κ is the velocity factor (0.89) for this RDM system and R is the housing inner diameter (m).

2.6 Data analysis

Replicate analyses gave errors of ±5 to 10%. TMP-flux profile data were fitted to the developed model. The data were analyzed by nonlinear regression using Origin 8. Statistical analyses were undertaken through one-way ANOVA and least significant difference analysis for comparisons of treatment means with p < 0.05.

3. Results and discussion

3.1 Flux behavior

The flux behavior of MF (MV020) and UF (UP020) under different rotating speeds and VRRs are shown in Fig. 2. The selected rotating speeds creating the shear rates represent general hydrodynamic conditions on membrane surface. Due to higher driving force at higher TMP, flux increases step by step when TMP rises, while these stepwise fouling behaviors are apparent and increasing patterns are highly different. As presented in Fig. 2, these increasing patterns can be divided into three categories: (1) fouling pattern 1: for MF of 2500 and 2000 rpm at VRR = 1 and UF of 2500, 2000 and 1500 rpm at all VRRs, flux kept increasing linearly and only a small amount of deviation occurred at the end; (2) fouling pattern 2: for MF of 1500 rpm at VRR = 1 and 2500, 2000 and 1500 rpm at VRR = 3 and 6 as well UF at 1000 and 500 rpm, flux kept increasing linearly at initial stage, then the growth rate decreased and finally, the flux tended to be stable and reaches a plateau; (3) fouling pattern 3: for MF at 1000 and 500 rpm at all VRRs, after a very short rising period, the flux reached a plateau.

Fig. 2 Permeate flux as a function of TMP at various rotating speeds and VRRs for MF ((a), (c) and (e)) and UF ((b), (d) and (f)).

Visual inspection results significantly indicate different nonlinearities in all operational curves of permeate flux versus TMP. This depends on different compressibilities of fouling layer and subsequent filtration resistances. In traditional fouling theory,²⁵ three zones related with membrane fouling have been proposed to describe the operational curves of permeate flux versus TMP. In zone 1, for low TMP, only concentration polarization exists with adsorption fouling, and as a subcritical zone, the flux is lower than the critical one, thus no obvious fouling occurs and flux increases linearly with TMP. At zone 2, cake formation and consolidation as well pore blocking and protein adsorption at membrane occur and flux increases slower. For zone 3, the cake layer is compressed by TMP and irreversible fouling improves, while the flux reaches a plateau (limiting flux).⁹ This fouling theory and the related mathematical models can only explain the fouling pattern 2, but they are not adequate to represent the stepwise-shaped curves for fouling patterns 1 and 3. For fouling pattern 1, zone 1 is dominates and no zone 3 occurs, whereas, with respect to fouling pattern 3, zones 2 and 3 are definitely dominated and zone 1 is more ephemeral. Besides, traditional fouling theory just forces on the simple change trend between flux and TMP, but the stepwise fouling process is ignored. In fact, the main core issues are fouling improvement and compressibility of fouling layer, but there is no index to reflect the change of fouling layer in traditional fouling theory. Therefore, in Section 3.2, a stepwise fouling model is firstly proposed to simulate stepwise fouling process and explain various fouling patterns with different feed water compositions, membrane types and hydraulic conditions on membrane surface, while fouling formation and cake compressibility during different stages can be explained.

3.2 Modeling of the stepwise fouling process

In previous studies,^9,25–28 some fluxes and fouling mathematical models were identified for predicting critical flux and limiting flux. The feed water composition, temperature and cross-flow velocity have an influence on behaviors of stepwise fouling. However, to our knowledge, no flux and fouling models consider a stepwise phenomenon in fouling process. As shown in Fig. 2, it is interesting to note that the curves of flux versus TMP can be divided into several nonlinear sections. Each section can represent a type of Darcy's law model (DM).

The flux influence of additional hydraulic resistance can be described by DM as follows:

where μ is the dynamic viscosity of feed (pa s) and R_m and R_f are the hydraulic resistances (m⁻¹) of membrane and all fouling layer including cake layer and concentration polarization, respectively.

The fouling resistance can be described as the product of the area-specific amount of fouling,¹⁶ W (kg m⁻²), and the specific resistance of fouling, α (m kg⁻¹) as described in eqn (5).

R_f = αW (5)

The fouling layer is compacted with the increase of TMP, further reducing permeate flux. In some previous studies,^20,23,29 the fouling layer of protein is considered to be highly compressible, i.e. the specific resistance, α₀ (m kg⁻¹) is pressure independent and can be described as follows:

α = α₀(TMP)^β (6)

where α₀ is a constant determined by the particle size²⁹ and β is the compressibility index of fouling layer (varying from 0 to 1).

Eqn (4) can be rewritten as:

γ = α₀W (8)

where γ (m⁻¹) represents resistance coefficient of fouling layer and can be fitted by experimental data.

This is a reasonable assumption based on the mono-fouling process. The DM, which assumes uniform fouling affinity scales and binding energies between foulants and membrane for fouling sites, is applied to homogeneous membrane surfaces.^30,31 In its general form, the multisite DM takes into account several kinds of active sites on membrane surface of fouling, and is useful for a fouling process with more generally homogeneous membrane surfaces.^32,33 Foulants in feed comprises various organic and inorganic matters and the membrane surface is associated with various pits, edges, and other discontinuities.³² Therefore, it is more reasonable to assume that not all surface sites on membrane are identical. Besides, a variety of site types might exist, with varying affinity scales and binding energies for fouling process.

During fouling process, it is speculated that all kinds of unoccupied sites are in excess and the foulant molecules deposit and bind preferentially to sites with greater affinity. As the concentration of foulants in feed increases, fouling behavior to the higher binding energy sites on membrane surface becomes saturated and excess foulant molecules are compelled to deposit on other sites with lower affinity. Thus a stepwise fouling process with the sum of multiple fouling steps can be obtained and perceived to explain a group of fouling sites with mono DM/multisite Darcy's law model (MDM)-type fouling processes. In order to better explain fouling process of alfalfa juice, a stepwise multisite Darcy's law model (SMDM) was proposed in this investigation.

The MDM can be written as follows:

where β_j and γ_j are resistance coefficient of fouling layer and compressibility index of fouling layer for site type j, j = 1, 2, …, n. When n = 1, the DM can be obtained. Eqn (9) is merely applicable to the single-step fouling process. When more steps are involved, a critical TMP (TMP_i) is introduced, with subscript i denoting the fouling step. When TMP in the solution is above TMP_i, eqn (9) can be modified to the following:where J_i is additional permeate flux during step i.

Eqn (10) can be utilized to describe the fouling process at each step. This equation can only be used when TMP > TMP_i. If TMP < TMP_i, then it means that step i has not been reached, thus TMP for step i could not be considered and should be equal to 0. Eqn (10) can be rewritten as follows:

If TMP > TMP_i, ;

If TMP < TMP_i, J_i = 0

For all of the fouling steps, the related mathematical expression of SMDM can be given as follows:

The limiting flux (J_lim) is the maximum plateau from all steps in entire fouling process (eqn (13)):

A schematic illustration of an SMDM is shown in Fig. 3. For example, if the data can be characterized as a three-step fouling process including a single DM-type fouling for the first step, a three-site MDM-type fouling for the second step, and a two-site MDM-type fouling for the third step, then corresponding SMDM can be described as follows:

where β₁ and γ₁ are DM parameters for single-site fouling in the first step, β₂, β₃, β₄, γ₂, γ₃ and γ₄ are MDM parameters for three-site fouling in the second step; β₅, β₆, γ₅ and γ₆ are MDM parameters for two-site fouling in the third step, respectively.

Fig. 3 A schematic illustration of SMDM for fouling process.

The developed SMDM was employed to explain the fouling process of alfalfa juice for MF and UF. The estimated parameters based on experimental data by nonlinear regression analysis are shown in ESI Tables S1 and S2.† The results imply that the membrane surface comprises several distinct site types with various affinities for foulants in alfalfa juice. The SMDM method is suitable for both MF and UF. Moreover, in the first step for two membrane types, the DM can best fit the operational curves of permeate flux versus TMP, indicating only that one site type would interact with foulants of alfalfa juice in this step. Such fouling sites for MF and UF have a higher affinity for foulants. In the second step, the two-site DM fits satisfactorily experimental data for both MF and UF. For the third step, both one-site DM and two-site DM give a high fitting degree. It indicates the existence of multiple site types on these filtration experiments. With lower rotating speed or higher VRR, the number of site types and fouling steps increases. At low rotating speed, shear rate on membrane surface decreases and more serious concentration polarization occurs, as well more foulants deposit on membrane and have more interaction with various membrane sites and the subsequently formed fouling layers are multiple and cumulative. Besides, a high concentration of foulants increases with VRR, which also causes more sites and multiple steps in the fouling process.

The β and γ for MF and UF vary with each step. This variation can be attributed to the differences of feed water composition, hydrodynamic condition and membrane properties.¹⁵ It should also be noted that each type of membrane site can include a wide population of fouling sites. However, these sites have similar affinities.³² For most steps and sites of MF, the β values were lower than these of UF, but γ values were higher than these of UF, implying that the filtration resistance of MF was lower than that of UF, and the fouling layer compressibility was higher than that of UF, which could be explained as follows. Due to smaller membrane pore, UF had higher rejection of foulants and had more serious concentration polarization than MF, subsequent leading to thicker fouling layer and greater filtration resistance on membrane. In filtration process, as a driving force, flux could compress fouling layer. Because of much higher flux, the fouling layer of MF was highly compressed, thus MF had higher fouling layer compressibility than UF. Besides, the β and γ in the first step are much lower than these in the second and third steps, due to higher concentration polarization and membrane fouling at higher TMP,³⁴ especially above critical flux zone or in limiting zone.²⁵ Therefore, different site types at the second and third steps could contribute a significant amount to the membrane fouling of alfalfa juice, even if the fouling affinity of site types in the first step may be greater.

3.3 Fouling performance of alfalfa juice filtration process

Feed water composition, membrane types and hydraulic conditions on membrane surface are three main influencing factors affecting concentration polarization and membrane fouling.³⁴ In this study, VRR reflects the foulants concentration in alfalfa juice. Leaf proteins and lipids have a higher concentration with higher VRR, as well as feed has higher foulant concentration. MF and UF are different membrane types. Rotating speed is positively correlated with shear rate, which determines hydraulic conditions on membrane surface.³⁵ As shown in Fig. 2, VRR, membrane type and rotating speed have strong effects on fouling characteristics. However, the effect of feed water composition, membrane types and hydraulic conditions on the stepwise fouling process is still unknown. To better study the stepwise fouling mechanism and influencing factors involved in the filtration process, the related fouling characteristics of SMDM were further analyzed.

3.3.1 Effect of feed water composition. In industrial applications, the feed water usually includes various components (e.g., colloids, particles, soluble organics, and various electrolytes). These components characteristics and their interactions among multiple-components in feed water may change the morphology and hydrophobicity of a membrane, affect van der Waals forces between solutes, hinder solute back-diffusion, and accelerate the growth of foulants.¹⁵ Some parameters including pH, alkalinity and ionic strength have been characterized as significant factors influencing fouling performances.

In the concentrating process of alfalfa juice, different VRRs have different feed composition and own various fouling processes and characteristics. The effect of VRR on stepwise fouling process of alfalfa juice was investigated through batch experiments in VRR 1, 3 and 6 for MF and UF. Fig. 4 and 5 depict the resistance coefficient and compressibility of different VRRs, respectively. The data show that fouling behavior of alfalfa juice on MF and UF vary with VRRs. At the same rotating speed, the resistance coefficient values increased with VRRs, because at higher VRR, more feed leaf proteins and other large particles were concentrated, then higher concentration of foulants created thicker fouling layer and greater resistance coefficient, in accordance with their flux behavior. Besides, the resistance coefficient herein is in the order: step 3 > step 2 > step 1, implying that at higher TMP operation, more foulants deposited on membrane surface and a thicker layer formed, so a more complex fouling process and a higher resistance coefficient occurred. As for the compressibility, their values decreased with increasing VRR. In alfalfa juice concentration process, as an important driving force, the permeate flux pushed foulants toward membrane surface and compressed the cake layer, therefore a higher flux improved the compressibility.³⁶ At higher VRRs, a thicker cake layer formed and reduced the flux, thus compressibility was eliminated.

Fig. 4 Resistance coefficient in fouling process on (a) VRR = 1, (b) VRR = 3 and (c) VRR = 6 at different rotating speeds.

Fig. 5 Compressibility in fouling process on (a) VRR = 1, (b) VRR = 3 and (c) VRR = 6 at different rotating speeds.

In membrane filtration of alfalfa juice, the overall influence of feed composition on the fouling behavior can be considered as a combination of multiple influences. The various components in feed, such as particles, colloids, soluble organics, and various electrolytes, may interact with each other and have various foulant components, pH and ionic strength, thus affecting fouling behavior.¹⁵ This study uses various VRRs to reflect the variation of feed composition and suggests that VRR can have a measurable influence on fouling behavior. The feed composition may differ spatially and temporally in alfalfa juice disposal sites, due to differences in site-operation patterns, alfalfa juice productions and natural conditions. It is important to consider such local characteristics in the related in fouling assessment.

3.3.2 Effect of membrane type. Membrane characteristics including pore size, surface electric charge, porosity, hydrophobicity, hydrophilicity and roughness have a pronounced influence on fouling behavior.¹⁵ As seen in Fig. 4 and 5, a MF membrane has a lower resistance coefficient than a UF one, however, its compressibility is higher. In comparison with UF, MF had a bigger pore size and lower rejection for solutes, thus less foulants accumulated on membrane surface and thinner cake layer formed. Owing to the much higher flux, MF compressed the cake layer more than UF. In addition, UF had had a larger number of steps and fouling sites than MF, especially for VRR = 1, therefore, UF had a more complex fouling process. This is because UF had higher concentration of foulants on membrane surface and its fouling formation mechanism had more multiple steps and more complex effects. These results suggested that the membrane type is a significant factor affecting the fouling process of alfalfa juice and a selecting the right membrane can reduce fouling behavior and optimize filtration process.

3.3.3 Effect of hydraulic condition. Hydraulic conditions, dominated by shear, affects concentration polarization and membrane fouling, because it can influence the hydrodynamic and mass transfer rate on membrane surface. Typically, high shear rate can reduce the boundary layer thickness and concentration polarization, and then retard membrane fouling.¹⁵ In this study, when rotating speed rose from 500 rpm to 2500 rpm, the corresponding shear rate increased from 0.17 × 10⁵ s⁻¹ to 3.28 × 10⁵ s⁻¹.

As shown in Fig. 4 and 5, the resistance coefficient and compressibility decreased with rotating speed. The increase of rotating speed could eliminate concentration polarization and deposition of foulants on membrane, thus fouling resistance on membrane reduced. At the same time, high flux produced by rotating speed provided a higher driving force for compressing fouling layer and decreasing compressibility. In addition, the number of steps and fouling sites increases with the decrease of shear rate. At lower shear rate, concentration polarization was higher and the interaction between foulants and membrane was more complex, thus the fouling layer resistance was bigger and fouling process needed more steps and fouling sites. Therefore, a high shear rate cannot only control concentration polarization and membrane fouling resistance, but also decrease the steps and sites in fouling process. This is of special interest for applications to fouling control of membrane protein concentration.

3.4 Long-term tests

In order to study the membrane fouling and flux decline in long term filtration for different fouling steps, long term tests (5 h) under various were performed. Operating TMP for steps 1 and 2 was selected according to critical TMP in ESI Tables S1 and S2.† For step 3, TMP was increased to the maximum value (MF: 3 bar and UF 6 bar). Fig. 6 and Table 3 show the results of 5 h tests at selected conditions. For both rotating speeds and membranes, the permeate flux at step 1 stabilized very quickly and remained almost constant during the 5 h test. While for the operating at step 2 and step 3, flux decayed clearly at first 60 min, and then kept reducing with a slight rate. This indicates that step 1 achieved a fouling balanced state more easily and faster than other steps. Higher rotating speed improved flux behavior, but could not accelerate fouling balanced state. Because rotating speed reduced concentration polarization, but it did not change the barrier force (F^f–m_barrier and orF^f–f_barrier) during fouling layer formation. Percentages of flux decline (FD) and hydraulic permeability loss (PL) for the tests of Fig. 6 are shown in Table 3. For all operating strategies, the FD values of 2000 rpm were smaller than that of 1000 rpm, but for PL, 2000 rpm was higher than 1000 rpm, which could be explained as follows. In this case, FD was mainly governed by concentration polarization and membrane fouling (irreversible fouling and reversible fouling),³⁷ so a higher shear rate on membrane at higher rotating speed decreased FD by reducing polarized layer and cake fouling. As for PL, it was controlled by pore adsorption and irreversible fouling,³⁸ and as at 2000 rpm, permeate flux was much higher and more solutes passed through membrane pores and led to more pore blocking fouling, thus decreasing PL. Therefore, PL values did not behave as FD. As mentioned above, MF had more serious fouling than UF, thus its FD and PL were higher. Furthermore, for step 1, FD and PL were highly low, implying that long term runs did not obviously deteriorate filtration efficiency. But for step 2 and 3, more adsorption and irreversible fouling occurred, thus FD and PL greatly increased and significantly affected filtration process. Generally, in long term operations, with the increase of step, membrane fouling enhanced stepwise, decreasing FD and PL gradually.

Fig. 6 Permeate flux versus time (5 h) under various operating strategies for (a) MF and (b) UF (VRR = 6 solution and 25 °C).

Table 3 Flux decline (FD) and permeability loss (PL) for long term tests^a,b

Membrane	Step	1000 rpm		2000 rpm
		FD (%)	PL (%)	FD (%)	PL (%)
a NE, no exist because this step does not exist in SMDM.b The FD values were calculated from the average flux during first hour and fifth hour.
MF	Step 1	6.3	4.5	5.3	5.2
	Step 2	9.2	18.6	8.8	19.9
	Step 3 (6 bar)	13.4	29.4	11.9	31.1
UF	Step 1	5.1	3.9	4.8	4.1
	Step 2	8.6	14.2	7.9	17.3
	Step 3 (6 bar)	12.7	22.8	NE	NE

---

3.5 Simulation results

With the fitting parameters of fouling process (compressibility and resistance coefficients) in ESI Tables S1 and S2,† SMDM (eqn (12)) can be used to simulate permeate flux. As shown in Fig. 2, the simulation of SMDM and experimental data are very similar. Therefore, SMDM can not only calculate the fouling process parameters, but also simulate permeate flux effectively.

4. Conclusions

The investigation related to food processing by membrane technology is an emerging area in food engineering research. Alfalfa was an important protein resource and MF and UF were used to separate and concentrate leaf protein from alfalfa juice. This study focused on analyzing fouling behavior during membrane filtration process, which could help membrane fouling control and improve filtration efficiency. Since the foulants deposited on membrane presented important stepwise patterns, their flux behavior and membrane fouling showed various stepwise trends. With the elevating of TMP, fouling increased step by step, while fouling trends varied with different operating conditions. As a significant method for fouling behavior analysis, a developed SMDM was proposed to identify fouling variation and help simulate this stepwise fouling process during filtrating alfalfa juice. Fouling process was divided into several steps and sites, as well their compressibility and resistance coefficients were calculated to explain the complex fouling process. Moreover, feed composition, membrane characteristic and hydraulic condition played an important role in fouling behavior. In this study, the effect of VRR, membrane type and shear rate on fouling process was analyzed using SMDM. In membrane engineering, the overall influence of various factors on fouling behavior could complicate the processes. Besides, a series of long term runs which operated at various fouling step processes were conducted to verify flux behavior including flux decline and permeability loss. The results have significant implications for fouling assessment and fouling control in membrane engineering plants. SMDM can also be applied to other membrane applications as an alternative to describe fouling mechanism. Future studies are necessary to define the characteristics of foulants and their interactions with membrane to obtain a more theoretical approach for analyzing fouling formation in filtration process.

Nomenclature

A	Membrane surface (m^2)
J	Permeate flux of solution (L m^−2 h^−1)
Lpa	Membrane water permeability after filtration or cleaning (L m^−2 h^−1 bar^−1)
Lpi	Initial membrane water permeability (L m^−2 h^−1 bar^−1)
N	Rotating speed (rpm)
pc	Measured peripheral pressure (Pa)
Qf	Permeate flow rate (m^3 h^−1)
R	Housing inner radius (m)
Rout	Outer radius (m)
Rin	Inner radius (m)
Rm	Hydraulic resistances of membrane (m^−1)
Rf	Hydraulic resistances of fouling layer (m^−1)
R^2	Coefficient of multiple determination
TMP	Transmembrane pressure (bar)
T	Filtration time (h)
V	Total volume of permeate (L)
VRR	Volume reduction ratio (L)
V0	Initial feed volume (L)
VR	Retentate volume (L)
α	Specific resistance of fouling (m kg^−1)
β	Compressibility index of fouling layer (varying from 0 to 1)
κ	Velocity factor
γ	Resistance coefficient of fouling layer (m^−1)
ρ	Density of the fluid (g L^−1)
ω	Angular velocity of rotating disk (rad)
μ	Dynamic viscosity of feed (pa s)

References

1. J. Volenec, S. Cunningham, D. Haagenson, W. Berg, B. Joern and D. Wiersma, Field Crop. Res., 2002, 75, 97–110.
2. L. Firdaous, P. Dhulster, J. Amiot, A. Gaudreau, D. Lecouturier, R. Kapel, F. Lutin, L.-P. Vézina and L. Bazinet, J. Membr. Sci., 2009, 329, 60–67.
3. W. Zhang, N. Grimi, M. Y. Jaffrin and L. Ding, J. Membr. Sci., 2015, 489, 183–193.
4. B. Lamsal, R. Koegel and S. Gunasekaran, LWT--Food Sci. Technol., 2007, 40, 1520–1526.
5. W. E. Barbeau and J. E. Kinsella, Food Rev. Int., 1988, 4, 93–127.
6. R. Douillard and O. De Mathan, in New and Developing Sources of Food Proteins, Springer, 1994, pp. 307–342.
7. Z. Xie, J. Huang, X. Xu and Z. Jin, Food Chem., 2008, 111, 370–376.
8. W. Koschuh, G. Povoden, V. H. Thang, S. Kromus, K. D. Kulbe, S. Novalin and C. Krotscheck, Desalination, 2004, 163, 253–259.
9. J. Luo, L. Ding, Y. Wan and M. Y. Jaffrin, J. Membr. Sci., 2012, 409, 276–284.
10. C. An and G. Huang, Environ. Sci. Technol., 2012, 46, 12742–12750.
11. S. Fernández, C. Menéndez, S. Mucciarelli and A. P. Padilla, Desalination, 2003, 159, 153–160.
12. W. Koschuh, G. Povoden, V. H. Thang, S. Kromus, K. D. Kulbe, S. Novalin and C. Krotscheck, Desalination, 2004, 163, 253–259.
13. S. S. Fernández, C. Menéndez, S. Mucciarelli and A. P. Padilla, J. Sci. Food Agric., 2007, 87, 1850–1857.
14. B. Bals and B. E. Dale, Biotechnol. Bioeng., 2011, 108, 530–537.
15. W. Zhang, J. Luo, L. Ding and M. Y. Jaffrin, Ind. Eng. Chem. Res., 2015, 54, 2843–2861.
16. E. Poorasgari, T. V. Bugge, M. L. Christensen and M. K. Jørgensen, J. Membr. Sci., 2015, 475, 65–70.
17. W. Zhang and L. Ding, Sep. Purif. Technol., 2015, 141, 160–169.
18. B. Cancino-Madariaga, R. Ruby, C. Astudillo Castro, J. Saavedra Torrico and M. Lutz Riquelme, Ind. Eng. Chem. Res., 2012, 51, 4017–4024.
19. S. Chellam and W. Xu, J. Colloid Interface Sci., 2006, 301, 248–257.
20. Y. Lee and M. M. Clark, J. Membr. Sci., 1998, 149, 181–202.
21. H. Carrère, F. Blaszkow and H. R. de Balmann, J. Membr. Sci., 2001, 186, 219–230.
22. H.-C. Kim and B. A. Dempsey, J. Membr. Sci., 2013, 428, 190–197.
23. R.-S. Juang, H.-L. Chen and Y.-S. Chen, J. Membr. Sci., 2008, 323, 193–200.
24. C. Y. Tang, T. Chong and A. G. Fane, Adv. Colloid Interface Sci., 2011, 164, 126–143.
25. C. L. Astudillo, Ind. Eng. Chem. Res., 2015, 54(1), 414–425.
26. G. L. Baruah and G. Belfort, Biotechnol. Prog., 2003, 19, 1524–1532.
27. G. Belfort, R. H. Davis and A. L. Zydney, J. Membr. Sci., 1994, 96, 1–58.
28. G. Samuelsson, I. Huisman, G. Trägårdh and M. Paulsson, J. Membr. Sci., 1997, 129, 277–281.
29. E. Piron, F. Rene and E. Latrille, J. Membr. Sci., 1995, 108, 57–70.
30. G. Foley, J. Membr. Sci., 2006, 274, 38–46.
31. Z. Li, T. Katsumi and T. Inui, J. Membr. Sci., 2011, 378, 485–494.
32. H.-C. Yang, J. Luo, Y. Lv, P. Shen and Z.-K. Xu, J. Membr. Sci., 2015, 483, 42–59.
33. N. Nady, M. C. Franssen, H. Zuilhof, M. S. M. Eldin, R. Boom and K. Schroën, Desalination, 2011, 275, 1–9.
34. C. Y. Tang, Y.-N. Kwon and J. O. Leckie, J. Membr. Sci., 2009, 326, 526–532.
35. W. Zhang, Z. Zhu, M. Y. Jaffrin and L. Ding, Ind. Eng. Chem. Res., 2014, 53, 7176–7185.
36. D. C. Sioutopoulos and A. J. Karabelas, J. Membr. Sci., 2015, 473, 292–301.
37. S. Meng, H. Winters and Y. Liu, Water Res., 2015, 83, 248–257.
38. S. Meng and Y. Liu, Water Res., 2013, 47, 6618–6627.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c6ra23212d

---