Numerical simulations of impact of membrane module design variables on aeration patterns in membrane bioreactors

Author’s Accepted Manuscript Numerical simulations of impact of membrane module design variables on aeration patterns in membrane bioreactors Xuefei Liu, Yuan Wang, T. David Waite, Greg Leslie www.elsevier.com/locate/memsci

PII: DOI: Reference:

S0376-7388(16)30915-2 http://dx.doi.org/10.1016/j.memsci.2016.07.011 MEMSCI14596

To appear in: Journal of Membrane Science Received date: 23 March 2016 Revised date: 6 July 2016 Accepted date: 7 July 2016 Cite this article as: Xuefei Liu, Yuan Wang, T. David Waite and Greg Leslie, Numerical simulations of impact of membrane module design variables on aeration patterns in membrane bioreactors, Journal of Membrane Science, http://dx.doi.org/10.1016/j.memsci.2016.07.011 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Numerical simulations of impact of membrane module design variables on aeration patterns in membrane bioreactors Xuefei Liu1, Yuan Wang1,2, T. David Waite2, Greg Leslie1* 1

UNESCO Centre for Membrane Science & Technology, School of Chemical Engineering,

University of New South Wales, Sydney 2052, Australia 2

Water Research Centre, School of Civil & Environmental Engineering, University of New

South Wales, Sydney 2052, Australia *

Corresponding

author.

Tel.:

+61-2-9585-6092;

fax.:

+61-2-9385-5966;

[email protected]

Abstract Computational Fluid Dynamics (CFD) models incorporating sludge rheology model and porous media sub-models were used to simulate the impact of eight design variables, including fibre orientation, filtration tank geometry and aeration design, on the hydrodynamics and aeration patterns in a pilot scale membrane bioreactor (MBR) fitted with commercially available membrane modules. Comparison of simulated with experimental grid averaged flow velocities measured using Particle Image Velocimetry (PIV) in an aerated bench scale MBR differed by approximately 6%. The assessment of different turbulence models revealed that it was difficult to accurately simulate local variations in turbulence induced by air bubbles due to limitations of Reynolds averaging Navier Stokes (RANS) models. Simulation results revealed that MBR’s fitted with hollow fibres in a vertical orientation in the filtration zone experienced 25% more membrane surface shear than 1

horizontally oriented fibres at the same aeration intensity. The inclusion of baffles around the membrane modules promoted turbulence and increased shear in the upper section of the membrane module by approximately 30%. This is important for control of fouling along sections of the hollow fibre membrane that experience the highest transmembrane pressure and are more susceptible to fouling. Rotating the nozzle aperture to face the bottom of the tank increased the homogeneity of shear stress on the lower half of the module and increased shear on the upper half of the membrane surface. This effect was amplified by locating the air diffusers 100 mm below the bottom of the module and aligning the aeration pipe in parallel with the direction of the fibre bundle. This research demonstrates the capability of using Computational Fluid Dynamics to optimise the design of the filtration zone comprising the membrane module, aeration system and MBR tank dimensions. Keywords: Membrane bioreactor, Computational Fluid Dynamics, Aeration, Membrane module design, Particle Image Velocimetry

Glossary of terms Symbols A

Total membrane surface area (m2)

C

Log-layer constant (in Equations 1 and 5, natural logarithms used)

Cij and Dij

prescribed matrices in the porous media model

d1

Distance between aerators and the adjacent membrane in x- direction (m)

d2

Distance between aerators and the bottom of membrane module (m)

2

Kloss

Friction loss coefficient (m-1)

Kperm Permeability coefficient (m-2) k

von Karman constant

L

Length (m)

n

Distance between the first and second grid points off the wall (m)

p

Pressure drop across membrane module (Pa)

Ut

Flow velocity tangent to the wall at a distance of y from the wall (m/s)

u

Eulerian flow velocity (m/s)

u+

Flow velocity inside boundary layer (m/s)

u𝜏

Friction velocity (m/s)

V

Free volume of the membrane module (m3)

w1

Distance of membrane module to wall (m)

w2

Width of membrane module (m)

y+

Dimensionless distance from wall

∆y

Thickness of boundary layer (m)

Greek Symbols α

Orientation of nozzle direction to y+ axis



Orientation of aeration tubes to x axis 3

µ

Dynamic viscosity (Pa.s)

θ

orientation of fibre to y axis

𝜌

Density of fluid (kg/m3)

𝜏

Wall shear stress (Pa)

1

Introduction

Membrane bioreactors (MBR) use microporous membranes in lieu of secondary clarifiers and media filters for solid-liquid separation in wastewater treatment applications [1]. The membranes are located in a Filtration Zone, which includes tankage containing membrane modules and coarse bubble aeration system [2]. Hydraulic capacity of the MBR process is controlled via aeration induced turbulence which prevents accumulation of mixed liquor suspended solids on the membrane surface [3-5]. Commercial MBRs are available in different configurations based on membrane orientation, aerator aperture size, aerator position and free volume between membrane modules and tank walls. The overarching design goal is to create a spatially uniform velocity gradient in the filtration zone to limit localised fouling and promote even distribution of filtrate flux over all the available membrane area [4, 6]. However, in the absence of performance data collected under controlled conditions, it is difficult to assess which design achieves the highest surface shear while simultaneously optimising the footprint of the filtration zone and power input for the aeration system. This can be attributed to the complex conditions in the filtration zone and the interdependence of the effect of each design variable on bubble induced shear. The effect of bubble characteristics, membrane module orientation and the geometry of filtration tank has been investigated by empirical measurement of flux decline [7, 8], liquid

4

flow velocities using Particle Image Velocimetry (PIV) [9-11], and characterisation of membrane surface shear using electrochemical/electrodiffusion methods [12-16]. For example, a 15% reduction in flux decline was obtained by rotating fibre orientation relative to the tank wall through 90o from horizontal to vertical [17, 18]. Direct measurement of shear forces on Teflon tubes representing bundles of hollow fibres using electrodiffusion methods found that the average shear increased from 0.3 Pa at a gas flowrate of 2mL/min to 0.8 Pa at a gas flowrate of 35 mL/min [12]. Similarly, increasing the fibre packing density produced “dead zones” where fibres received minimal shear forces [19], however, the occurrence of these dead zones could be reduced by changing the location of the aerator port from the centre of the module to the corner of the module [9]. Similarly, the installation of baffles at the periphery of the filtration zone was found to increase the crossflow velocity in the vicinity of the fibres [20]. The shear profile in a MBR is transient and non-uniformly distributed in the filtration zone. Different statistical parameters such as time-averaged shear, standard deviation and amplitude have been used to quantify shear for different MBR configurations [21]. While these empirical studies can be used to compare different design features (such as the effect of baffles or membrane orientation) the data is not representative of the hydrodynamic conditions in a commercial MBR module and would have limited utility in the design of a full scale filtration zone [5]. A more systematic approach is required to identify the optimum design that can achieve superior shear profile whilst minimising power input for aeration [20]. Numerical modelling enables a priori evaluation of different design variables by changing the geometry of the computing domain and boundary conditions in complex turbulent multiphase flow [6]. While numerical simulations of surface loading rates, settling velocities and velocity gradients have been used to design full scale conventional solid-liquid separation processes in wastewater treatment [22], these techniques have not been applied widely on

5

MBRs. Khalii-Garakani et al. [23] modelled the use of baffles in the filtration zone to constrain air flow around flat sheet modules and found the change in baffle angle from 90o to 85o could increase the shear stress on the membrane surface. The majority of the numerical studies have focussed on the movement of gas bubbles (or slugs) along tubular or flat sheet membrane modules and have been used to model temporal changes in spatial variation of shear as a function of bubble size, channel dimension and geometry for Newtonian fluids [2428] in order to relate shear rate induced by gas bubbles to the permeate flux [29], and to evaluate the impact of bubble frequency, size and shape on liquid velocities and shear stress on the membrane surface [30-35]. Notwithstanding this body of work, there remain inconsistencies on the ability to reduce fouling by optimising the aeration pattern (gas flowrate) [33]. Consequently, it is difficult to provide definitive recommendations with regard to the most effective bubbling regime that could be used to reduce membrane fouling. In this paper, various design variables including characteristics of fibre bundle (fibre diameter and packing density), module geometry (orientation and shape of module) and configuration of aeration systems are systematically evaluated using Computational Fluid Dynamics. The objective of the studies described here was to evaluate which combination of features of the hollow fibre membrane module, filtration tank, and aeration system that can achieve higher and more homogeneous shear in the filtration zone at the same aeration energy input. A Computational Fluid Dynamics (CFD) approach using rheological and porous media submodels described elsewhere [36] was used to simulate pressure drop across the hollow fibre membrane bundle for a range of conditions that are relevant for application of MBR’s in municipal wastewater treatment. Particular emphasis is given to identifying the capabilities of current state-of-the-art turbulence models in capturing the high degree of turbulence induced by bubbly flow by comparing the simulated data using three different types of turbulence models with experimentally measured data using Particle Image Velocimetry (PIV). 6

2

Theory

2.1 shear stress and liquid flow velocity Shear stress on the membrane surface can be calculated from the tangential flow velocity outside the boundary layer using the wall-function approach [37, 38], where the flow velocity inside the boundary layer (u+) is given by:

u 

 

Ut 1  In y   C u k

(1)

where uτ is the friction velocity, Ut is the flow velocity outside the boundary layer, k is the von Karman constant and y+ is the dimensionless distance from the wall, which is defined as:

y 

yu 

(2)

where ∆y is the thickness of boundary layer. The standard definition of y+ generally used in CFD can be written as:

y 

 u  n 

(3)

where n is the distance between the first and second grid points off the wall [38]. In ANSYS CFX®, a “Scalable Wall Function” is used for all -based turbulence models while an “Automatic Near-Wall Treatment” is used for all -based models, which result in slightly different treatments in y, to achieve an optimum model accuracy and robustness [38]: For the scalable wall function, y  n / 4 ; and for automatic wall treatment, y  n

7

The friction velocity is defined as:

 u     

1/2

(4)

By substituting Equation 2 and Equation 4 into Equation 1, we obtain

Ut 

 yu  1 u In   u C k   

(5)

The wall shear stress, , is a function of Ut, , and  and is directly linked to the flow velocity outside the boundary layer, Ut. In this work Ut is used to represent the shear stress on the membrane surface for sludge with the same viscosity and density. 2.2 Porous media model The resistance caused by the hollow fibre membrane bundle to the fluid flow can be modelled using a porous media approach [39]. A momentum source term, Si, is added to the governing momentum equations with the pressure drop considered to be proportional to the fluid velocity; i.e. 3  3  1 Si     Dij vJ   Cij vmag v j  2  j 1  j 1

D

1

(6)

(7)

K perm

C  Kloss

(8)

8

where Kperm is permeability, Kloss is friction loss coefficient and D and C are prescribed matrices. The first term on the RHS of this equation is the viscous loss term and the second is the inertial loss term. The pressure drop resulting from a hollow fibre membrane module is determined by the permeability of the whole module. As such, in a manner similar to the modelling of tube banks, the viscous term can be eliminated, yielding the simplified form of the porous media equation: 3 p j 1   Kloss vmag v j L 2 j 1

(9)

where the inertial loss coefficient, KlossJ, is a function of the Reynolds number (i.e. a function of the density, velocity and viscosity of the fluid) and affected by various characteristics of the hollow fibre bundles including spacing (i.e. packing density) and diameter, which can be determined experimentally for different hollow fibres and module packing densities. 3

Methods

3.1 CFD simulation 3.1.1 Physical MBR system A three-dimensional model of the filtration zone of a pilot scale MBR (Table 1) operating on primary settled sewage at the Bondi Sewage Treatment Plant (Sydney Water, Australia) was developed using commercial CFD code (ANSYS CFX 14.5). Details of the operating conditions, influent and effluent water quality and fouling behaviour of the pilot plant can be found elsewhere [40]. The model was used to simulate the effect of membrane, tank and aerator design variations on membrane surface shear stress (Figure 1). These included orientation of the membranes in 9

the tank (), bubble size, angle of the aerator nozzle to the horizontal (), alignment of the tubular aerator to the membrane (), horizontal space between tubular aerator and adjacent membrane (d1) and vertical distance between aerator tube and base of module (d2). In addition, the presence or absence of a dedicated baffle could be simulated by varying the free space around the membrane module based on the ratio of distance of the membrane to the tank wall (w1) and the width of membrane module (w2). With regard to the effects of bubble size, since the bubble size is not only affected by air flowrate, but also by other parameters such as nozzle configuration, simulations were performed using constant bubble number at various air flowrates, as well as using constant air intensities that produce different numbers of bubbles. Conditions for the reference simulation were based on the current configuration of the membrane filtration zone (250 L capacity), which consisted of four vertical, curtain shaped, PVDF hollow fibre membrane modules (10.4 m2, Origin Water, Beijing, China) encased in a stainless steel baffle (L x W x H = 0.49 m x 0.276 m x 0.73 m) with three tubular coarse bubble aerators, positioned at the base of the baffle, perpendicular to the curtain, delivering process air at an intensity of 4.70 Nm3/hr (Table 1). A subsequent 18 simulations, performed by varying a single parameter, were used to assess the change in the shear profile relative to the reference conditions (Table 2). 3.1.2 Model Development Discretisation of the computational domain involved use of a hexahedron mesh with five inflation layers for the membrane module as well as the near membrane zone to increase the local resolution of the boundary layer for shear stress calculations, and a tetrahedron mesh for the remaining domain, resulting in a total of 660,000 elements. A mesh independent test was performed by refining the grid size of the membrane module from 5 mm to 3 mm. A

10

maximum error of 5% in the flow quantity (gas hold-up and near membrane flow velocity) was observed between the finer grid and coarser grid. Details of meshing methods and grid independence check can be found in Supplementary information (Table s1). The CFD model used in this work (Figure 2) was composed of a core numerical hydrodynamics model for the liquid and gaseous phases, empirical sludge rheology models to account for the behaviour of the mixed liquor [36] and porous media models to represent the resistance of membrane modules to the fluid flow [36, 39]. The core hydrodynamics model implemented the Eulerian model to simulate the liquid and gas components of the multiphase flow with the water and sludge mixture treated as a single liquid phase [36]. Three different types of turbulence models, namely, the RNG k- model, the Shear-stress transport (SST) turbulence model, and a model recently developed by McClure et al [41], which included additional terms in the turbulence models were employed to simulate the turbulence flow while a zero equation model was used for the dispersed gas phase. A scalable wall function and automatic wall treatment with default model constants as described in the ANSYS CFX modelling guide [38] were used in the RNG k- model and the SST model, respectively. Details of multiphase models for the sludge mixture and gaseous phase can be found elsewhere [36, 42]. The membrane modules were modelled as four individual liquid outlets with a negative source term for mass removal (i.e. a constant permeate flowrate). The hydraulic resistance caused by the hollow fibre membrane bundle was simulated by incorporating the inertial term of the porous media model while the viscous term was neglected (Equation 9). Empirical correlations obtained from our previous work [36], which were calibrated for hollow fibres with diameters ranging from 1.3 to 2.4 mm and packing densities from 200 to 560 m2/m3, and for non-Newtonian fluids with viscosities from 0.8 to 2.1 mPa.s at 100 s-1 at 11

velocities from 0 to 0.35 m/s, were used to model the macroscopic characteristics of the hollow fibre membrane module in the fluid domain: For flow direction parallel to the hollow fibre bundles pressure drop is defined as, p 1  1.98Re0.51 DH 1.53  u 2 L 2

(11)

and, for flow direction perpendicular to the hollow fibre bundles pressure drop is defined as,, A  p  13.5   V u  L  

0.905

1  u 2 2

(12)

where DH is the hydraulic diameter of the hollow fibre membrane bundle, A is the total membrane surface area, V is the free volume of the module (volume not occupied by hollow fibres), and  is the viscosity of the liquid phase. Information with regards to the experimental calibration of the sludge rheology models can be found in the Supplementary information. 3.2 Experimental validation of CFD model using Particle Image Velocimetry (PIV) CFD simulated liquid flow velocity was compared against liquid velocity measured in a 31.25 L bench scale reactor with the same aspect ratio as the pilot plant using Particle Image Velocimetry (PIV) (Figure 3). The bench scale reactor was fitted with three equally spaced curtain shaped hollow fibre membrane modules (L x W x H = 0.11 m x 0.01 m x 0.30 m) (OriginWater, Beijing, China) positioned in the centre of the Perspex tank. A circular ceramic disk (7cm diameter) was used to generate 3 mm diameter bubbles at a flow rate of 0.06 Nm3/hr (Table 1). This aeration rate resulted in an average gas holdup of 0.12% for the entire tank, which ensured the motion of the florescent particles could be captured by the high speed camera (X-Stream VISIONTM XS-4 high speed camera with Nikon Nikkor 35mm lens, 12

shutter speed 1/90 s, f-number f/1.4D). A light sheet produced from a laser emitting at 532 nm wavelength was used to illuminate a 360 cm2 measurement plane with origin coordinates located at x = 5 mm, y = 100 mm and z =-15 mm (Figure 3). Red fluorescent polyethylene microspheres (1.045-1.055 g/cc in density, Cospheric LLC, Santa Barbara, CA, USA) of sizes ranging from 10 to 45 m were used as tracer particles with the maximum intensity of the emission spectrum of these particles at 605 nm. A Hoya 25A red filter was installed in front of the high speed camera to shield the strong reflected green laser light from bubbles and membrane fibres so that only the red fluorescent reflection from the fluorescent particles were collected by the camera. The camera was positioned at 90o to the measurement plane with a sensor gain of -6dB to minimise the conflicts between light intensity and particle visibility. Two hundred images of particle trajectories were collected for each measurement grid and processed with VIDPIV 4.6 (ILA, Juelich, Germany) software to obtain an averaged local flow velocity in the measurement grid. 4

Experimental validation of CFD model

Experimental validation of the CFD model was undertaken by comparing Particle Image Velocimetry (PIV) measurements from a bench scale reactor with results of simulations for a model reactor of identical geometry (membrane module, tank, and aerator) and operating parameters. Local liquid velocities were extracted from the model simulation at a 100 mm line in the x-direction starting from x = 5 mm (i.e. Line 1 in Figure 3), and at a 300 mm line in the y-direction starting from y = 100 mm (i.e. Line 2 in Figure 3). The simulated data was compared with 57 flow velocities measured by PIV at an interval of 2.19 mm on Line 1, and 145 flow velocities collected at an interval of 2.06 mm on Line 2, respectively. An average error of 12.8% and 7.6% was found between the simulated and PIV measured flow velocities in the x- and y-direction, respectively (Figure 4). When the porous media model was disabled, the averaged error increased to 26.7% and 18.0% for the x- and y-direction, respectively 13

(Figure s2(a) in the Supplementary information), indicating the importance of using the porous media model to account for the resistance caused by the hollow fibre membrane bundle to the fluid flow. The possibility of improving the accuracy of the model by increasing grid resolution and by using the Shear-stress transport (SST) turbulence model was also evaluated. A maximum of 7% difference in liquid velocity was observed when the grid size was refined from 10 mm to 5 mm (Figure s2(b) in the Supplementary information), indicating the numerical solution was grid-independent. The SST model could not capture the local small-scale turbulent fluctuations observed by the PIV measurements with a 21% of further decrease in accuracy compared to the RNG k-model (Figure s2(c) in the Supplementary information). The RNG k-model, which considers the effect of swirl on turbulence and hence improves the accuracy for modelling swirling flows, has been widely used to model the liquid phase in MBRs, including flow structures in the whole tank [43], movement of Taylor bubble through a tubular membrane [29], and the impact of fibre arrangement on hydrodynamics of the system [35]. The SST model uses the standard k-model in the bulk flow and incorporates the transport of turbulent shear stress while using the k- model in the boundary layer. Prieske et al. [31] examined the relationship between circulation velocity and aeration flowrate in a pilot-scale MBR using the SST model. However, the PIV validation studies indicated that both the RNG k-model and the SST model could not accurately describe small-scale random vorticity apparent in the PIV data. This was attributed to the inherent limitation of the Reynolds averaging Navier Stokes (RANS) models. Such models approximate the effects of turbulence on a larger scale flow field, while the real turbulence flow induced by aeration involves turbulence length scales much smaller than the smallest finite volume mesh [38]. Additional turbulence induced by the dispersed phase in the continuous phase is not covered in the current commercial simulation package due to the lack 14

of accurate mathematical models able to account for the interphase transfer of k and /ω. As such, this aspect not included in the CFD model developed in this work. Recently, McClure et al [41] have attempted to include additional terms in the turbulence models to achieve more accurate results in describing the bubble-induced turbulence in bubble column bioreactors. However, it was found that the modified models resulted in more significant errors and therefore are not suitable to the current systems (Figure s2(c) in the Supplementary information). One of the reasons why these turbulence models including the modified models resulted in deviations from PIV measured data was because the effects of bubble aggregation and break-up on the local turbulence were not addressed in this work. In addition, turbulence caused by the lateral fibre movement, which can be simulated using Fluid-Structure Interaction [44], was not included in the current model. A comparison of the simulated data with the mean flow velocities measured by PIV averaged at 20 mm intervals showed that the average error reduced to 7.7% and 4.5 % for the x- and ydirection, respectively. Therefore, investigations on the accuracy of the CFD model used in the current study suggested that although the model (two-phase Eulerian approach, and RNG k- model coupled with sludge rheology models and porous media model) may underestimate the randomness of turbulence, it is capable of quantifying the grid averaged flow variables in submerged hollow fibre membrane filtration vessels. 5

CFD modelling results and discussion

5.1 Membrane orientation Hollow fibres in commercial modules can either be horizontally aligned (θ = 900) or vertically aligned (θ = 00) relative to the side walls of the tank. The CFD simulated area weighted membrane surface shear stress was 0.76 (± 0.28) Pa for the reference condition. This value was consistent with the values obtained from studies employing comparable 15

aeration intensities [14] and CFD simulated shear stress in the range of 0.3 – 0.9 Pa reported by Ratkovich et al. [45]. The area-weighted average liquid velocity tangent to the membrane surface (i.e. Ut, abbreviated as liquid velocity in later sections) was found to decrease by 25 % (0.129 m/s vs. 0.097m/s) by rotating θ from 00 to 900 under the same aeration intensity (Figure 5), indicating a lower aeration efficiency for a horizontally aligned membrane module. These findings were consistent with the conclusion drawn by Chang et al. [7] based on empirical bench scale observations. The implication of this finding is that the resistance to fluid flow and the pressure drop experienced in a module with fibres in normal orientation to liquid flow was much larger than for fibres in a parallel orientation to the flow at the same packing density [36, 39]. Consequently, horizontally aligned membrane modules increase flow resistance which results in a reduction in liquid velocity through the filtration zone and reduces circulation of liquid flow around the membrane modules. 5.2 Use of Baffles in the Filtration Zone 5.2.1 Evaluation of the effects of baffles in the filtration zone of the pilot scale MBR Baffles can be placed in the filtration zone between module racks or cassettes in order to concentrate the flow of bubbles in the volume occupied by the membranes. CFD simulations indicate that insertion of a baffle into the filtration zone increase the area-weighted liquid velocity from 0.113 to 0.129 m/s (14% increase) (Figure 6). A modest increase of 14% is not a compelling reason for the inclusion of a baffle in the filtration zone, however, liquid velocity in the upper quarter of the membrane module increased by 32% following the inclusion of a baffle (Figure 6). Increasing liquid velocities in the upper section of a vertical membrane module justifies the use of baffles as this section is closer to the suction pump and experiences a higher transmembrane pressure which can make this region more susceptible to fouling [17]. Moreover, inspection of the liquid velocity vector profiles shows that the liquid

16

in the region above the aerators is displaced by the rising bubbles travelling to the top of the tank within the membrane channel. The baffle constrains and directs the rising flow and divides the flow field into two regions: an ascending flow region inside the baffle and a descending flow region outside the baffle (Figure 6a and c). When the baffle was removed, the ascending flow region became more dispersed and moved towards the wall (dark blue zone in Figure 6b and d) resulting in a lower flow velocity in the ascending flow zone which reduced the effectiveness of the aeration system. These results suggest that the use of a baffle affords the opportunity to reduce the potential of cake layer formation in this region. 5.2.2 Evaluation of the size of membrane module relative to filtration tank Results from section 4.2.1 suggest baffles prevent the movement of air bubbles away from the membrane modules and promote high shear conditions at the membrane surface, particularly in the regions most pronne to fouling. The inclusion of baffle plates in full scale MBRs would incur additional capital cost. An alternative approach is to optimise the distance between the wall of the filtration tank and the membrane module in order to achieve a similar effect as that induced by the pressence of a baffle. CFD simulations were performed by varying the ratio of the distance of the membrane module to wall (w1) to the width of membrane module (w2). In these simulations, air diffusers were positioned parallel to the membrane elements (i.e. = 00, d1 = 0 mm). Increasing the ratio w1/w2 from 0.2 to 0.6 increased the area-weighted liquid velocity by 43% (from 0.035 m/s to 0.05 m/s) (Figure 7a,b,c). No further improvement in shear conditions was observed by increasing this ratio to 0.8. Varying the distance between the module and the tank wall alters the circulation of upward flow past the fibres, however, even at the optimum value of 0.6, the effect on average liquid velocity was not as great as that achieved with a baffle (Figure 7d). This suggests that other factors, such as achiveing an optimum tank foot

17

print, should be considered over effects of tank dimensions on surface shear, when designing tankage for the filtration zone of an MBR. 5.3 Aeration system 5.3.1 Bubble size Simulated area-weighted average liquid velocity along the membrane surface increased from 0.129 m/s to 0.150 m/s when the bubble size decreased from 5 mm to 3 mm under a constant air flowrate of 4.7 Nm3/h (equal to the aeration rate in a previous study of an MBR using full scale modules [40]). An increase of bubble size to 10 mm resulted in a further decline in liquid velocity to 0.098 m/s (Figure 8 upper). These findings are in qualatitve agreement with experimental work performed on a flat sheet MBR [46] and several bench scale hollow fibre MBRs in which bubble size was varied at a constant air flow [12, 21, 47]. However, these results are inconsistent with the effect of bubble size on wake turbulence, which increases for larger bubbles in a rising column of air [9]. To clarify this point, the current study assumed that the bubbles did not collide, coalesce or break-up and maintained a constant diameter (3, 5, and 10 mm) through the entire simulation. Therefore, at the same aeration intensity, reducing the bubble diameter resulted in an increase of gas hold-up and the total interfacial area between bubbles and liquid phase which increased the turbulence induced by the rising bubbles. Conversely, simulations were also performed at constant bubble number density, which showed that the area-weighted average liquid velocity increased from 0.129 m/s to 0.399 m/s as bubble size increased from 5 mm to 10 mm. This is consistent with observations that coarse bubble could produce larger shear stress on the membrane surface than fine bubbles[9]. The increase in liquid velocity (and shear stress) was found to occur at all sections along the length of the membrane (i.e. in the positive y direction; Figure 8 lower).

18

However, it is important to note that the production of coarse bubbles at the same number density is attended by higher power consumption by the air blowers. 5.3.2 Nozzle direction Air diffusers can be oriented either with the nozzle facing the bottom of the tank (i.e. the reference condition,  = 180o) or with the nozzle facing the membrane elements ( = 0o). When α = 00 the development of dead zones in the lower section of the membrane modules was more pronounced when liquid velocities were less than 0.05 m/s (with minimum shear of 0.03 Pa)

(Figure 9a). Changing the nozzle direction to 1800 was found to reduce the

occurrence of dead zones and increase membrane shear stress by 12% (Figure 9b). Directing the nozzles toward the bottom of the tank facilitates the distribution of air bubbles and creates vortices and therefore promotes turbulence before the bubbles enter the hollow fibre bundles. 5.3.3 Orientation and spacing between aeration tubes In the reference condition (i.e. the current design of the air diffusers and membrane module of the pilot plant), high liquid velocities were confined to the region above the diffusers but not widely spread in the horizontal direction (Figure 10a), This creates an unsparged, low velocity area in the lower half of the membrane cassettes which would be conducive to solids accumulation and clogging of the hollow fibre bundle (Figure 10a). Solids accumulation has been observed in experimental studies that used similar aerator layout, and was eliminated by adjusting the interval orientation of the aerators [33]. In these simulations, it was possible to eliminate the uneven distribution of liquid velocity by locating the air diffusers parallel to the membrane cassettes ( = 00) (Figure 10b and 10c). Compared to perpendicular aerators, the average flow velocity was slightly increased in the upper section of the membrane cassettes for the case of  = 00 and d1 = 0 mm (i.e. aerators parallel and in-line with membrane modules). Membrane modules configured in “top-out” permeation mode exhibited higher 19

local permeate flux and hence higher fouling potential at sections closer to the permeate header (i.e. the upper sections) [17]. Therefore, any increase in shear stress over this region would reduce cake layer build-up. Interestingly, a 6.7% decrease in average flow velocity in the upper half of the membranes was observed when the aerator was positioned parallel to the membranes ( = 00) while alternating the location in between two membrane elements (i.e. d1 = 23 mm). Therefore, by improving the distribution of near membrane flow velocity without decreasing the shear stress in the upper half of the membranes, aeration diffusors positioned parallel and in-line with membrane modules exhibited superior performance compared with other aerator orientations and spacing. 5.3.4 Distance between aeration pipe and membrane module Simulations were performed by changing the distance from the aeration pipe to the membrane curtains (d2) from 20 to 100 mm. A maximum distance of 100 mm was selected to ensure the aerators were located within the lower edge of the baffle. Results showed that the area-weighted liquid velocity increased 11% from 0.126 m/s to 0.140 m/s, and became more homogeneously distributed along the membrane (in the y-direction) with the increase of d2 from 20 to 100 mm. Increasing d2 to 100 mm facilitated the distribution of shear on the lower half of the membrane without compromising the shear on the upper half of the module (Figure 11). Increasing the distance between the diffusers and modules provided additional space for air bubbles to disperse horizontally before entering the membrane module resulting in an even distribution of the air volume fraction and shear stress profile along the fibre bundle.

20

6

Conclusions



A three-dimensional two-phase CFD model, which incorporated rheology models to represent the behaviour of mixed liquor, and porous media models to describe the macroscopic characteristics of hollow fibre membrane module to the fluid field, was used to assess the effects of different design variables, including module configuration, filtration tank geometry, and aeration systems on the efficiency of bubble induced shear at the membrane surface.



The accuracy of the modelling technique was assessed and validated using Particle Image Velocimetry. Results showed that: − The model (two-phase Eulerian model, and RNG k- model) was capable of simulating the average shear stress; the accuracy of simulation can be improved by incorporating both the sludge rheology model and porous media model; − The difficulty in accurately modelling small-scale turbulence, may be attributed to the inherent limitations of RANS models, effects of bubble deformation and break-up, and effect of fibre movement on the local turbulence, and cannot be improved by increasing the grid resolution or using different turbulence models. Notwithstanding this, current CFD model provides insight on the effect of different design parameters on the area-averaged membrane surface shear. Further improvements on the algorithms are required to improve modelling of aeration induced turbulence flow at small turbulence length scales.



CFD simulated results revealed that: − MBR systems configured with hollow fibres in a vertical orientation in the filtration zone present less resistance to the fluid flow and produce larger shear on the membrane surface;

21

− The inclusion of baffles around the membrane modules promotes turbulence and increases shear in the upper section of the membrane module by approximately 30%. This is important for control fouling along sections of the hollow fibre membrane that experience the highest transmembrane pressure with these sections being more susceptible to fouling; − At a fixed aeration rate, higher shear can be achieved by reducing aperture size to generate smaller bubbles, however, at constant bubble number density shear increases with increasing bubble size; − Rotating the nozzle aperture to face the bottom of the tank increases the homogeneity of shear stress on the lower half of the membranes and produces a comparatively larger shear on the upper half of the membrane surface. This effect is amplified by locating the air diffusers 100 mm below the bottom of the module and aligning the aeration pipe in parallel with the direction of the fibre bundle. − 

This research demonstrates the capability of using Computational Fluid Dynamics to optimise the design of the filtration zone comprising the membrane module, aeration system and MBR tank dimensions and features including baffles.

Acknowledgements Funding provided by the Australian Research Council, Water Research Australia, Sydney Water and Beijing Origin Water through ARC-Linkage project LP100100056 is gratefully acknowledged. The authors would like to thank Sydney Water for facilitating the access to the wastewater treatment plants and Beijing Origin Water for providing membranes. Special thanks to Dr Tracie Barber from UNSW School of Mechanical and Manufacturing Engineering for assistance on PIV measurements. 22

Reference [1] S. Judd, The MBR book: Principles and Applications of Membrane Bioreactors in Water and Wastewater Treatment, 1st ed., Elsevier, Amsterdam; Boston; London, 2006. [2] K. Yamamoto, M. Hiasa, T. Mohmood, T. Matsuo, Direct solid-liquid separation using hollow fiber membrane in an activated sludge aeration tank, Water Sci. Technol., 21 (1989) 43-54. [3] Z.F. Cui, S. Chang, A.G. Fane, The use of gas bubbling to enhance membrane processes, Journal of Membrane Science, 221 (2003) 1-35. [4] E. Braak, M. Alliet, S. Schetrite, C. Albasi, Aeration and hydrodynamics in submerged membrane bioreactors, Journal of Membrane Science, 379 (2011) 1-8. [5] L. Böhm, A. Drews, H. Prieske, P.R. Bérubé, M. Kraume, The importance of fluid dynamics for MBR fouling mitigation, Bioresour. Technol., 122 (2012) 50-61. [6] Y. Wang, T.D. Waite, G.L. Leslie, 15 - Computational fluid dynamics (CFD) analysis of membrane reactors: modelling of membrane bioreactors for municipal wastewater treatment, in: A. Basile (Ed.) Handbook of Membrane Reactors, Woodhead Publishing, 2013, pp. 532568. [7] S. Chang, A.G. Fane, S. Vigneswaran, Modelling and optimising submerged hollow fibre membrane modules, AIChE Journal, 48 (2002) 2203-2212. [8] S.H. Yoon, H.S. Kim, I.T. Yeom, Optimization model of submerged hollow fiber membrane modules, Journal of Membrane Science, 234 (2004) 147-156. [9] A.P.S. Yeo, A.W.K. Law, A.G. Fane, Factors affecting the performance of a submerged hollow fiber bundle, Journal of Membrane Science, 280 (2006) 969-982. [10] A.P.S. Yeo, A.W.K. Law, A.G. Fane, The relationship between performance of submerged hollow fibers and bubble-induced phenomena examined by particle image velocimetry, Journal of Membrane Science, 304 (2007) 125-137. [11] N. Liu, O. Zhang, G.-L. Chin, E.-H. Ong, J. Lou, C.-W. Kang, W. Liu, E. Jordan, Experimental investigation of hydrodynamic behavior in a real membrane bio-reactor unit, Journal of membrane Science, 353 (2010) 122-134. [12] C.V.V. Chan, P.R. Bérubé, E.R. Hall, Shear profiles inside gas sparged submerged hollow fiber membrane modules, Journal of Membrane Science, 297 (2007) 104-120. [13] K. Zhang, Z. Cui, R.W. Field, Effect of bubble size and frequency on mass transfer in flat sheet MBR, Journal of Membrane Science, 332 (2009) 30-37. [14] B.G. Fulton, J. Redwood, M. Tourais, P.R. Bérubé, Distribution of surface shear forces and bubble characteristics in full-scale gas sparged submerged hollow fiber membrane modules, Desalination, 281 (2011) 128-141. [15] L. Böhm, A. Drews, M. Kraume, Bubble induced shear stress in flat sheet membrane systems—Serial examination of single bubble experiments with the electrodiffusion method, Journal of Membrane Science, 437 (2013) 131-140. [16] L. Böhm, M. Kraume, Fluid dynamics of bubble swarms rising in Newtonian and nonNewtonian liquids in flat sheet membrane systems, Journal of Membrane Science, 475 (2015) 533-544. [17] S. Chang, Fane, A. G., The effect of fibre diameter on filtration and flux distribution relevance to submerged hollow fibre modules, Journal of Membrane Science, 184 (2001) 221-231. [18] S. Chang, Optimisation of Submerged Hollow Fibre Membrane Modules, in, University of New South Wales, Sydney, 2001. 23

[19] P.R. Bérubé, E. Lei, The effect of hydrodynamic conditions and system configurations on the permeate flux in a submerged hollow fiber membrane system, Journal of Membrane Science, 271 (2006) 29-37. [20] E. Nguyen Cong Duc, L. Fournier, C. Levecq, B. Lesjean, P. Grelier, A. Tazi-Pain, Local hydrodynamic investigation of the aeration in a submerged hollow fibre membranes cassette, Journal of Membrane Science, 321 (2008) 264-271. [21] C.C.V. Chan, P.R. Bérubé, E.R. Hall, Relationship between types of surface shear stress profiles and membrane fouling, Water Research, 45 (2011) 6403-6416. [22] Metcalf, Eddy, Wastewater Engineering, Treatment and Reuse, 4th ed., McGraw - Hill, New York, 2003. [23] A. Khalili-Garakani, M.R. Mehrnia, N. Mostoufi, M.H. Sarrafzadeh, Analyze and control fouling in an airlift membrane bioreactor: CFD simulation and experimental studies, Process Biochemistry, 46 (2011) 1138-1145. [24] C. Cabassud, S. Laborie, J.M. Laine, How slug flow can improve ultrafiltration flux in organic hollow fibres, Journal of Membrane Science, 128 (1997) 93-101. [25] R. Ghosh, Z.F. Cui, Mass transfer in gas sparged ultrafiltration: upward slug flow in tubular membranes, Journal of Membrane Science, 162 (1999) 91-102. [26] Z.F. Cui, K.I.T. Wright, Gas-Liquid 2-Phase Cross-Flow Ultrafiltration of Bsa and Dextran Solutions, Journal of Membrane Science, 90 (1994) 183-189. [27] M. Mercier, C. Fonade, C. LafforgueDelorme, How slug flow can enhance the ultrafiltration flux in mineral tubular membranes., Journal of Membrane Science, 128 (1997) 103-113. [28] P. Wei, K. Zhang, W. Gao, L. Kong, R. Field, CFD modeling of hydrodynamic characteristics of slug bubble flow in a flat sheet membrane bioreactor, Journal of Membrane Science, 445 (2013) 15-24. [29] T. Taha, W.L. Cheong, R.W. Field, Z.F. Cui, Gas-sparged ultrafiltration using horizontal and inclined tubular membranes--A CFD study, Journal of Membrane Science, 279 (2006) 487-494. [30] N. Ndinisa, A. Fane, D. Wiley, D. Fletcher, Fouling Control in a Submerged Flat Sheet Membrane System: Part II - Phase Flow Characterization and CFD Simulations, Separation Science and Technology, 41 (2006) 1411-1445. [31] H. Prieske, A. Drews, M. Kraume, Prediction of the circulation velocity in a membrane bioreactor, Desalination, 231 (2008) 219-226. [32] H. Prieske, L. Böhm, A. Drews, M. Kraume, Optimised hydrodynamics for membrane bioreactors with immersed flat sheet membrane modules, Desaliantion and Water Treatment, 8 (2010) 270-276. [33] A. Drews, H. Prieske, E.-L. Meyer, G. Senger, M. Kraume, Advantageous and detrimental effects of air sparging in membrane filtration: Bubble movement, exerted shear and particle classification, Desaliantion, 250 (2010) 1083-1086. [34] L. Martinelli, C. Guigui, A. Line, Characterisation of hydrodynamics induced by air injection related to membrane fouling behaviour, Desalination, 250 (2010) 587-591. [35] S. Buetehorn, D. Volmering, K. Vossenkaul, T. Wintgens, M. Wessling, T. Melin, CFD simulation of single- and multi-phase flows through submerged membrane units with irregular fiber arrangement, Journal of Membrane Science, 384 (2011) 184-197. [36] X. Liu, Y. Wang, T.D. Waite, G. Leslie, Numerical simulation of bubble induced shear in membrane bioreactors: Effects of mixed liquor rheology and membrane configuration, Water Research, 75 (2015) 131-145. [37] B.E. Launder, D.B. Spalding, Lectures in mathematical models of turbulence, Academic Press, London, New York, 1972. [38] Ansys, CFX- Solver Theory Guide, Release 14.5. ANSYS, Inc., in, 2012. 24

[39] Y. Wang, M. Brannock, S. Cox, G. Leslie, CFD simulations of membrane filtration zone in a submerged hollow fibre membrane bioreactor using a porous media approach, Journal of Membrane Science, 363 (2010) 57-66. [40] Y. Wang, K.H. Tng, H. WU, G. Leslie, T.D. Waite, Removal of phosphorus from wastewaters using ferrous salts – A pilot scale membrane bioreactor study, Water Research, 57 (2014) 140-150. [41] D. McClure, J. Kavanagh, D. Fletcher, G. Barton, Development of a CFD Model of Bubble Column Bioreactors: Part Two – Comparison of Experimental Data and CFD Predictions, Chemical Engineering and Technology, 37 (2014) 130-140. [42] M. Brannock, Y. Wang, G. Leslie, Mixing Characterisation of Full-Scale Membrane Bioreactors: CFD Modelling with Experimental Validation, Water Research, 44 (2010) 31813191. [43] J. Saalbach, M. Hunze, Flow structures in MBR-tanks, Water Science and Technology, 57 (2008) 699-705. [44] X. Liu, Y. Wang, T.D. Waite, G. Leslie, Fluid Structure Interaction analysis of lateral fibre movement in submerged membrane reactors, Journal of Membrane Science, 504 (2016) 240-250. [45] N. Ratkovich, M. Hunze, I. Nopens, Hydrodynamic study of a hollow fiber membrane system using experimentally and numerically derived surface shear stresses, Multiphase Science and Technology, 24 (2012) 47-66. [46] A. Sofia, W.J. Ng, S.L. Ong, Engineering design approaches for minimum fouling in submerged MBR, Desaliantion, 160 (2004) 67-74. [47] A.G. Fane, A.P.S. Yeo, A.W.K. Law, K. Parameshwaran, F. Wicaksana, V. Chen, Low pressure membrane processes - doing more with less energy, Desaliantion, (2005).

Figure 1 Geometry of model domain and design variables Figure 2 Schematic diagram of CFD model development and components Figure 3 Schematic representation of flow measurements using Particle Image Velocimetry and the geometry of the CFD model developed for the PIV studies Figure 4 Comparison of CFD modelled and PIV measured liquid velocities at (a) different x locations (Line 1 in Figure 3); and (b) different y locations (Line 2 in Figure 3) Figure 5 Liquid velocity along membrane surface for different fibre orientation (a) θ = 0 o; (b) θ = 900

25

Figure 6 Liquid velocity contour profile in the membrane tank with vectors showing the ascending and descending flow region with a baffle (a) view from z+ axis and (c) view from x+ axis, and without baffle (b) view from z+ axis and (d) view from x+ axis Figure 7 Comparison of liquid velocity obtained at different w1/w2 ratios: (a) w1/w2 = 0.2; (b) w1/w2 = 0.6; and (c) w1/w2 = 0.8; (d) average liquid velocity near membrane surface Figure 8 Comparison of liquid velocity contour with different bubble sizes (a) 3 mm; (b) 5 mm; (c) 10 mm; upper – same aeration intensity; lower – same bubble number Figure 9 Combination of liquid velocity contour and vector profile for different nozzle direction: (a)

= 1800; (b)

= 180o; contour map indicates liquid velocity from low (blue)

to high (red), vectors indicate flow direction Figure 10 Contour of the liquid velocity along membrane surface for different orientations and layout of aeration tubes: (a)

= 90o; (b)

= 00, d1 = 0 mm; (c)

= 00, d1 = 23 mm; and

(d) average liquid velocity at different height (+y direction) of the membrane module Figure 11 Liquid velocity along the membrane surface at different air diffuser to module distance (d2), (a) d2 = 20 mm; (b) d2 = 50 mm; and (c) d2 = 100 mm Table 1 Configurational parameters of membrane filtration tank in pilot scale MBR and the bench scale membrane tank for CFD model validation MBRs Reactor dimension Membrane nominal pore size Total membrane area Packing density Aeration for membrane scouring Recirculation flow rate

Membrane filtration tank in pilot scale MBR 0.5m x 0.5m x 1.0m 0.1 – 0.3 μm

Bench scale membrane tank for PIV measurements 0.25 m x 0.25 m x 0.5 m 0.1 – 0.3 μm

10.4 m2 350 m2/m3 4.70 Nm3/hr

0.37 m2 350 m2/m3 0.06 Nm3/hr

45 mL/s

-------

26

Table 2 Design variables used in reference simulation and modified scenarios in CFD simulations Design variables 1. θ (degree) 2. Baffle 3. w1/w2 4. Bubble size (mm) 5. α (degree) 6.  (degree) 7. d1 (mm) 8. d2 (mm)

Standard scenario 0 present 0.3 5 180 90 N/A 50

Modified scenarios 90 absent 0.2, 0.4 – 0.8 3, 10 0 0 0, 23 20, 80, 100

Highlights 

Effects of 8 design variables on membrane surface shear were modelled using CFD



Variables included fibre orientation, tank geometry, and aeration system



Vertically orientated fibres with aerator in line with module created higher shear



Baffles constrain the recirculation of liquid flow and create higher shear



Comparison of simulated and PIV data identified opportunities for model development

27

28

29

30

31

32

33

34

35

36

37

38

39