Chemical Engineering Journal 361 (2019) 690–702

Contents lists available at ScienceDirect

Chemical Engineering Journal

journal homepage: www.elsevier.com/locate/cej

3-D CFD-PBM coupled modeling and experimental investigation of struvite T

precipitation in a batch stirred reactor
⁎
Seyyed Ebrahim Mousavi, Mahbuboor Rahman Choudhury, Md. Saifur Rahaman
Department of Building, Civil, and Environmental Engineering, Concordia University, 1455 de Maisonneuve Blvd, West Montreal, Quebec H3G 1M8, Canada

H I GH L IG H T S

• AIncludes
3D CFD-PBM model developed to predict struvite formation in batch stirred reactor.
• Crystal size
hydrodynamic, thermodynamic, kinetic variables of struvite precipitation.
• Model will aid
distribution obtained from model matches experimental results.
• efficient reactor design for production of desired struvite crystals.

A R T I C LE I N FO A B S T R A C T

Keywords: A general model has been developed to elucidate the precipitation of struvite crystals in a batch stirred tank
Struvite precipitation reactor. The model, which evaluates reactor performance, also predicts crystal size distribution (CSD) over time
Computational fluid dynamics by considering the hydrodynamic, thermodynamic, and kinetic aspects of solution in the reactor. A
Population balance model Computational Fluid Dynamics (CFD) model was coupled with Population Balance Modeling (PBM) to model the
Stirred reactor
growth of crystals in the reactor. A thermodynamic equilibrium model for struvite precipitation was con-
Growth rate
solidated with the reactor model. While the equilibrium model provided information on supersaturation de-
velopment, the coupled CFD-PBM model captured the crystal growth kinetics and the influence of the reactor
hydrodynamics on the overall process. Size distribution is crucial as it determines distinct grades of final struvite
crystals, which are to be used as commercial fertilizer. In the simulation, the CFD flow field was solved through a
Eulerian multiphase approach and RNG-k-ɛ turbulence model. The population balance equation was solved using
a discretized form of the continuous partial differential equation, which transformed the continuous partial
differential equation into finite ordinary differential equations as per size classes, which were then solved si-
multaneously. The growth rate, as a function of the supersaturation index (SI), was employed in the model
through User Defined Function. The mean, standard deviation, and skewness of the model predicted CSD after
50 min were 20.81 μm, 9.61 μm, and 2.97, respectively and for the experimental CSD were 19.66 μm, 7.13 μm,
and 2.46, respectively. The predicted peak-size percent fraction revealed a deviation from experimental results
of 1.42%, 0.05%, 2.43%, 14.6%, 11.2%, 11.7%, 13.6%, and 14.2% at 0, 3, 10, 20, 30, 40, 50, and 60 min,
respectively.

1. Introduction growth in food demands. Consequently, augmenting soils with phos-

phorous-based fertilizers plays an indispensable role in permitting more
Discharge of significant amounts of phosphate from nutrient-rich food growth. Commercial fertilizers are produced primarily from
wastewater is problematic considering the potential for contamination phosphorous rock reserves, which are finite and non-renewable in
of groundwater and eutrophication of recipient water bodies. With an nature [3]. With increasing demands for phosphorous-based fertilizers
expected 40% rise in the world’s population by 2050 [1] food con- all around the world, availability of phosphorous is expected to become
sumption will increase rapidly in the near future, with consumption scarcer and more expensive. As a result, phosphorus recovery is re-
rates growing by 3.1% annually [2]. Without a doubt, naturally oc- ceiving increased amounts of attention as an alternative and cost-
curring phosphorous—found in the soil—cannot support this rapid competitive option for the production of phosphorous-based fertilizers.

⁎
Corresponding author.
E-mail address: [email protected] (Md. S. Rahaman).

https://doi.org/10.1016/j.cej.2018.12.089
Received 24 July 2018; Received in revised form 1 November 2018; Accepted 17 December 2018
Available online 18 December 2018
1385-8947/ © 2018 Elsevier B.V. All rights reserved.
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

Nomenclature Pi impeller power (W)

Ri production rate of ith species
A Debye-Huckel constant →
r position vector
a(V,V′) aggregation kernel (m3/s)
→
Ri interphase momentum exchange
Bagr birth rate due to aggregation Si source rate
Bbrk birth rate due to breakage →
U velocity vector
Ci ionic concentration 
vi phase velocity
CD drag coefficient V volume
C1ε k-ε turbulence model parameter Yi mass fraction of each species
C2ε k-ε turbulence model parameter Zi valency of corresponding elements
Cμ k-ε turbulence model parameter Si source rate
{Ci} ion activity of ion Ci →
U velocity vector
D diameter (m)
Dagr death rate due to aggregation Greek letter
Dbrk death rate due to breakage
Dm molecular diffusion α Phase volume fraction
Dt turbulent diffusivity β (V |V ') probability density function
ds diameter of solid particles γi activity of solution
d32 Sauter mean diameter (m) σk k-ε turbulence model parameter
→
Fi centrifugal forces σε k-ε turbulence model parameter
f drag function ε rate of turbulent energy dissipation (m2/s3)
Gk,m generation of turbulent kinetic energy θ Azimuthal angle (degree)
→ g gravitational acceleration (9.81 m/s2) μ viscosity (kg/m·s)
=
G(t) growth rate τeffi Reynolds stress tensor (N/m2)
g(V′) breakage frequency (s−1) ν number of daughter particles
I bulk fluid ionic strength ρ phase density
Ī¯ unit tensor τj particulate relaxation time
kg kinetic parameter
Ksp struvite solubility constant Subscripts
Kij interphase momentum exchange coefficient
k turbulent kinetic energy (m2/s2) agr aggregation
M torque (Nm) brk breakage
N kinetic parameter l liquid phase
Np impeller power number lam laminar
→
N angular velocity (rad/s) m mixture
n(V,t) number density depicting numbers of particle of size V at s solid phase
time t t turbulent
P pressure

Struvite produced from wastewater, which is known to provide an al- commercial fertilizer, it may be useful that it is produced in a size-
ternative source of both phosphorus and nitrogen, using a precipitation controlled manner to provide distinct grades for specific applications.
technique is a proven solution for phosphorus recovery and makes an Having size-controlled struvite production would help in post-produc-
interesting alternative for the fertilizer industry. tion grading, handling, and commercialization of the fertilizers.
In the chemical precipitation method, which is the most widely used Struvite precipitation modeling studies fall into three major cate-
approach due to its operational simplicity, struvite precipitates from gories: hydrodynamic-focused models, thermodynamic-focused models,
wastewater by addition of adequate amounts of Mg and by pH adjust- and kinetic-focused models. Most of the work completed in this field
ment in the reactor to create alkaline conditions suitable for struvite has covered one of these three mentioned aspects of the process. In
precipitation. Small scale plants employ stirred tank reactors, as they hydrodynamic-focused models, researchers investigated the hydro-
are easy to run and control [4–14], while large-scale plants implement dynamic behavior of the crystallizers in general. They focused on re-
fluidized bed reactors, as they can provide higher reactive surface areas actor configurations and studied the different parameters affecting the
and better solution turbulence [15–19]. The present study employed a hydrodynamic behavior of reactor such as impeller speed, residence
batch stirred reactor for a number of reasons, such as simplicity in time, feed concentration and location, temperature [23–33]. Some
setting up the experimental apparatus, shorter time required for each studies modeled the thermodynamic aspect of the process using in-
round of experimentation, and fewer reactant requirements. house codes or commercial thermodynamic solvers and investigated the
Struvite precipitation depends on a variety of complex aspects such thermodynamic parameters affecting struvite precipitation such as
as fluid hydrodynamics, solution thermodynamics, mass transfer, and struvite solubility, species concentration, pH effect, and saturation state
precipitation kinetics including the nucleation, growth, aggregation, of the solution [34–38]. Kinetic-focused models attempted to determine
and breakage of crystals [17,20–22]. Design and operation of an in- the rate of struvite precipitation, which is the rate at which struvite
dustrial scale crystallizer requires a reliable knowledge of these aspects, constituent ions (magnesium, ammonium, and phosphate) get con-
which have mostly been obtained from laboratory scale experiments. sumed to produce struvite. Such modeling can predict the variation of
An efficient model that can predict different aspects of the process and crystal size with time and/or optimized residence time and reactor
combine these mechanisms in one unified theory is highly desirous, volume. The predicted size can be regarded as the final struvite crystal
however to the authors’ knowledge this has not been presented in the size for a given time in a reactor volume. Some researchers suggested
literature as of yet. Since struvite is most commonly used as a first-order kinetics and second-order kinetics for struvite precipitation

691
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

[8,39–47]. One drawback of these proposed kinetic rates is that the including all of the aspects mentioned above (i.e., hydrodynamic,
nucleation and growth effect cannot be differentiated from the kinetic thermodynamic, and kinetic) in one model. This study developed a 3-
model and has an unknown combinational effect on growth and nu- dimensional (3-D) computational fluid dynamics (CFD) model that in-
cleation. In other studies, researchers investigated the precipitation tegrates the kinetic rate resulting from the population balance equation
kinetics based on an assumed growth rate or in a few cases, nucleation with flow hydrodynamic and solution thermodynamic equations and
and/or aggregation rate as a function of supersaturation state of the predicts the variation of struvite particle size distribution over time for
solution and reported order-dependency of the equation [20,22,48–53]. the first time. Development of such a holistic model may better enable
Few researchers have presented a holistic model that covers more design of efficient reactors and selection of operational parameters in
than one aspect of the process and predicts crystal size distribution. reactors for the production of struvite crystals within a particular size
Since crystal size is an important parameter in controlling the quality of range for use as commercial fertilizer.
produced struvite, the proposed model should accurately predict it. Ali
and Schneider [20] estimated the crystal growth rate considering the 2. Experimental methodology
thermodynamic aspect and power relation between growth rate, su-
persaturation index, and supersaturation ratio. By using their estimated 2.1. Solution thermodynamics
coefficients, they predicted average particle size within a 10% margin
of error, which is an acceptable level of prediction. Their work has Precipitation of Magnesium Ammonium Phosphate (MAP) hexahy-
limitations including assumption of a vigorous mixing in the system, drate (MgNH 4 PO4 . 6H2 O), commonly known as struvite, occurs via the
which was deemed enough to overlook the hydrodynamic effect on the following simplified chemical reaction:
outcome and a lack of size distribution prediction. Hanhoun et al. [22]
Mg2 + + NH+4 + PO34− + 6H2 O → MgNH 4 PO4 . 6H2 O (1)
proposed kinetic rates for growth and nucleation using a population
balance-based model coupled with a thermodynamic model. They Critically, supersaturation of the solution is the key to struvite
predicted crystal size distribution (CSD) using a reconstruction model, precipitation. This parameter depends on solution pH and total ele-
which involved Laplace transformation of the number density function mental composition of reactive species. As multiple ions are involved in
describing the CSD in a one-dimensional batch crystallization model supersaturation in struvite case, defining solute concentration can be
[22,54,55]. The results, obtained by comparison between the crystal complex. To overcome this issue and to have a parameter to describe
size produced during experiments (analyzed by laser granulometry and the state of supersaturation, solute concentration is defined in terms of
morphometry) and the values obtained by the numerical method of the ion activity product (IAP ), as detailed in the following equation:
reconstruction of size distribution, show an under-prediction of the CSD
with a significant percentage of error. The main limitations of this study IAP = {Mg2 +}{NH+4 }{PO34−} (2)
are the ignoring of hydrodynamics in the process, overlooking of ag- where {Ci} represents the ion activity of ion Ci .
gregation and breakage, and observation of a high degree of numerical By definingIAP , the degree of supersaturation as described with the
deviations in the results. Galbraith and Schneider [56] developed a supersaturation index (SI) is as follows:
process model incorporating solution thermodynamics, nucleation,
SI = log(IAP/K sp) (3)
growth and aggregation mechanisms, and a population balance model
(PBM) to simulate struvite precipitation. A series of experiments carried where K sp is the solubility product which shows the equilibrium state.
out by Galbraith et al. [52], yielded an ensemble data set and this data Therefore, when the IAP is less than K sp , the solution is unsaturated
set was used to regress the model’s kinetic parameters for nucleation, and when IAP is more than K sp , the system is supersaturated and nu-
crystal growth, and aggregation. A key limitation of this study was the cleation or/and growth can occur. Over time, the system will ultimately
uncertainty of the distribution of particles originating from the limita- return to an equilibrium state from a supersaturated state. The equili-
tion of the experimental method. However, the limitation of the data brium solubility product used in this work is K sp = 13.26 [34] fixed at
obtained for parameter estimation was not essentially a limitation of 25 °C and atmospheric pressure.
the proposed model framework [52]. Rahaman et al. [17] presented a Estimation of IAP for struvite has been made from the total initial
comprehensive mathematical model for a fluidized bed crystallizer in- molar concentrations of magnesium, nitrogen, phosphorus, and calcu-
cluding aspects of thermodynamics, precipitation kinetics, and reactor lated equilibrium pH from charge balance. For this calculation, these
hydrodynamics in one model. Their results predicted removal effi- species must be considered [17]: MgPO−4 , MgHPO4 , MgH2 PO+4 , Mg2 +,
ciencies of constituent ions in an acceptable level of agreement with the MgOH+, H3 PO4 , H2 PO−4 , HPO24−, PO34−, NH+4 , H+, OH−, and NH3 (aq). The
experimental data. The model predicted the removal efficiencies of corresponding equilibrium equations as well as thermodynamic con-
phosphate, ammonium, and magnesium with a maximum deviation of stant values for these ions and complexes are provided in Table 1.
9%, 3%, and 10%, respectively, when compared to the experimental Since analytical determination of magnesium, ammonia, and phos-
observations [17]. They also predicted the average size of produced phate provide the total concentration of these species in solution, it is
struvite crystals in the reactor. The study compared the average size of necessary to calculate ion activity by utilizing the appropriate activity
produced struvite crystals, obtained from model prediction, to that
obtained from experiments and observed a 20–30% deviation in the Table 1
model predicted values. Lack of size distribution prediction was a sig- Equilibrium constants for complexes.
nificant drawback for this work. Galbraith et al. [52] estimated the
Equilibrium pK (25 °C) Reference
kinetic rate coefficient values for nucleation, growth, and aggregation
rate using a data set generated from 14 batch experiments. Their model [Mg2 + ][PO34− ]/[MgPO− 4.92 [57]
4]
covers solution thermodynamics and particle development tracking [Mg2 + ][H2 PO− + 0.45 [58]
4 ]/[MgH2 PO4 ]
through the application of a PBM. They predicted the time varying CSD [Mg2 + ][ HPO34− ]/[MgHPO4 ] 2.9 [59]
by employing estimated kinetics rates. Prediction of size distribution [Mg2 + ][OH− ]/[MgOH+ ] 2.56 [58]
within a 20–30% error range, as well as predicting the most common [H+ ][H2 PO− 2.15 [60]
4 ]/[H3 PO4 ]
size at every time interval are important achievements of their work. [H+ ][HPO24− ]/[H2 PO− 7.198 [60]
4]
Avoidance of the complexity of hydrodynamic effects by reducing their [H+ ][PO34− ]/[HPO24− ] 12.375 [60]
impact is the main limitation of their study. [NH3 ][H+]/[NH+ 9.3 [61]
4]
A review of struvite precipitation literature shows the lack of a [H+ ][OH−]/[H2 O ] 13.997 [60]
general model that presents particle size distribution of struvite by

692
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

coefficient and dissociation constants. The activity coefficient can be The experiments were carried out in a 1-L glass reactor with a round
determined using a variety of empirical equations. Here, the Davies bottom (radius 142.5 mm), baffled with four removable, equidistant
equation has been used to calculate the activity coefficient, since it is stainless steel baffles. Use of baffles improves the fluid mixing, avoids
capable of performing activity calculations for higher ionic strengths the creation of a dead zone occurring in the reactor, and also helps
[62]: create a quick dynamic pH response. The reactor was well mixed with a
Rushton impeller, rotating at a constant speed of 100 rpm in all ex-
I1/2 ⎤
− log γi = AZ i 2 ⎡ − 0.3I periments. The well-mixed reactor (BIO FLO 110, New Brunswick
⎢
⎣ 1 + I1/2 ⎥
⎦ (4) Scientific Co., Inc., Edison, NJ, USA) used in this study had inbuilt
where γi is the activity of solution, Z i is the valency of corresponding mechanisms for measuring the pH in the system and for keeping the
elements, and A is the Debye-Huckel constant, which has a value of temperature constant in the system. The BIO FLO 110 reactor’s power
0.509 at 25 °C [62]. requirements are 115 Volts, 50–60 Hertz, 20 VA. In this study, a syn-
I is the bulk fluid ionic strength which can be calculated by the thetic solution containing the constituent elements of struvite was used.
following defined relation, based on ionic concentration, Ci and re- The reagents used in the preparation of the synthetic solution were
spective charge, Z i for all ions: Sigma ACS reagent grade NH4H2PO4 and MgCl2·6H2O, and Milli-Q
deionized (DI) water. Also, 0.5 M NaOH solution was used to adjust the
1
I=
2
∑ Ci Z2i (5)
pH. The temperature was held at 25 °C in all experiments.

Solving the above equations must be done with a numerical ap-

proach due to their complexity and non-linearity. In this work, a 2.3. Experimental procedure
commercial spreadsheet software (Microsoft Excel 2003) was used to
solve all of the equilibrium equations and constants in Table 1 along Different initial values of SI were selected in the experiments to
with the Eqs. (1)–(5), based on iterative calculations for a given pH or keep the initial state of the batch reactor in the metastable zone to limit
SI. A Microsoft Visual Basic code was used in the Excel platform to nucleation. First, the constituent ions were prepared by dissolving
perform the iterative calculation. This approach was previously used in 0.005 M of Mg, N, and P in 1 L DI water. Through adjusting the pH by
Rahaman [63]. addition of 0.5 M NaOH, different initial pH and SI values were ob-
Considering struvite as a salt, the solubility at a given temperature tained, and a pH of 8 with corresponding SI value of 0.93 were chosen
depends on two major factors: species concentration and solution pH. for the main experiment. For each experiment, the seed was prepared as
The thermodynamic model results for minimum solubility pH in dif- described in Section 2.5.
ferent initial concentrations verified by Ohlinger’ solubility data [64] Samples of the crystals produced during the experiment had to be
are presented in Fig. 1. Together with the maximum threshold pH for taken at different times to analyze the CSD. Having proper agitation in
the beginning of spontaneous nucleation of struvite derived from a the reactor assures that each sample CSD is representative of the re-
series of experiments conducted by Ali and Schneider [65], Fig. 1 is actor. A pipette with an inlet diameter of 3 mm, which is very large
showing three zones for precipitation of struvite: (1) Unsaturated zone: considering the size range of the crystals measured in this study
crystals dissolve when added to the solution, and therefore, precipita- (10–120 μm), was used for sampling. A consistent and accurate sample
tion is impossible, (2) Metastable zone: the solution is saturated, and volume (50 ml) was collected each time. For sampling during an ex-
crystal growth can occur by seed addition, but the amount of Gibbs free periment, the pipette tip was inserted to about mid-depth in the reactor
energy is not enough for appreciable nucleation. However, nucleation and careful balance was maintained during the sampling to not touch
depends on time and provided sufficient time, some nucleation might the impeller blades or agitate the sample with the pipette tip. The
occur in this zone. Therefore, the growth phenomenon is predominant present study did not perform any assessment on the variability of
in this zone and (3) Labile zone or unstable zone: the solute con- samples originating from the sampling methodology used. A particle
centration exceeds the equilibrium value considerably, and sponta- size analyzer (PSA) was used to analyze each sample immediately after
neous nucleation can occur. Crystal growth can be rapid in this region. removal from the reactor. Non-return of the samples to the reactor did
The middle zone (zone between the red line and dashed line, in lead to a decrease in reactor volume, especially towards the end of
Fig. 1) shows the operating zone for precipitation. The pH range be-
tween the minimum and maximum limit shows the metastable zone.
Crystallization initiated by seed addition to the metastable zone is
heterogeneous [65].
In this study, the initial molar concentration of species and solution
pH were chosen in such a way that the initial state of solution stayed in
the metastable zone. Precipitation occurring in this zone means that
crystal growth is the dominant mechanism [48]. Also choosing the
lowest possible pH in this zone will hinder crystal aggregation, as it has
been reported that with increasing supersaturation, crystals are more
adhesive and, accordingly, there is a higher chance of crystal ag-
gregation [66]. This approach assumes that mass transfer between the
liquid phase and solid phase, which appears as a change in crystal size,
occurs due to the growth of initial seeds while other phenomena, such
as nucleation, have minimal effect.

2.2. Experimental setup

The experimental work aims to determine the crystal size distribu-

tion at different time intervals. Batch experiments were performed, as
this is easier in regards to parameter control. In a set of batch experi- Fig. 1. Determination of different saturation state zones for Struvite. Minimum
ments, where distinct experiments are carried out, it is easier to control struvite solubility ( ) is collected from Ohlinger et al. [64] and threshold pH for
parameters such as temperature and pH during each experiment run. spontaneous nucleation ( ) is collected from Ali and Schneider [65].

693
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

experiment. Still, no samples were returned to the reactor as the entire 3.1.1. Continuity equation
experiment is a dynamic process, which involves continuous growth of Most of the processes taking place in a stirred tank reactor are under
crystals. Returning the sample to the reactor would have introduced a turbulent regime. Therefore, the Reynolds averaged mass and mo-
some crystals that stayed outside the reactor in a different growth mentum equations must be considered. The continuity equation for
conditions and thus may have interfered with the representative CSD each phase is as follows:
evaluation. In order to avoid such interference and measure the CSD in a) For liquid phase:
this dynamic process at different time intervals, the samples were not →
∂
returned in the reactor after collection for particle size analysis. (ρ αl ) + ∇ . (ρl α1 U1) = 0
∂t l (6)

2.4. Measurements b) For solid phase:

∂ →
A laser scattering particle size distribution analyzer form HORIBA (ρ α s) + ∇ . (ρs α s Us ) = 0
∂t s (7)
(Model partica LA-950V2, Japan) was used to measure the crystal size
distributions. The laser scattering PSA initially measures volume-based where αl is the volume fraction of liquid phase, αs is the volume fraction
→
fractions in solution and then calculates number-based distribution of of solid phase, ρ is the density, and U is the velocity vector.
particles using the manufacturer’s software. Use of a laser scattering The liquid phase l and solid phase s are assumed to share the whole
particle size analyzer may lead to large errors, especially on the small space in proportion to their volume such that sum of their volume
end of size range, due to lack of resolution in analysis. The detection fractions is equal to one:
range of this machine ranges from 10 nm to 3 mm, which covers the αl + α s = 1 (8)
expected particle size range in this work. Also, the lower range of
measurement may not be valid for experimental conditions in case
nuclei formation takes place. pH measurement was performed by a 3.1.2. Momentum equation
Mettler Toledo pH probe (Model 405-DPAS-SC-K8S/225, USA). The momentum conservation equation for each phase is written as:
a) For liquid phase:
2.5. Seed preparation ∂   =   →
(ρ αl Ul ) + ∇ . (ρl αl Ul Ul ) = αl ∇p + ∇ . τeff1 + Rl + Fl + ρl αl g
∂t l (9)
As the initial solution is in the metastable zone, the experiments
were seeded to limit primary nucleation. To uniformly disperse seeds, b) For solid phase:
wet seeding was chosen. To prepare the seed, 0.005 mol of NH4H2PO4 ∂ → → → = → → →
(ρ αs Us ) + ∇ . (ρs αs Us Us ) = αs ∇p + ∇ . τeffs + Rs + Fs + ρs α s g
and MgCl2·6H2O were dissolved in 1 L DI water and the whole system ∂t s (10)
was stirred with a magnetic stirrer. Then, 0.5 M NaOH was added to →
raise the pH until the first precipitation of crystals was detected vi- where p is the pressure shared by the two phases and Ri represents the
sually. The NaOH was incrementally added and a final volume of about interphase momentum exchange terms. The interphase momentum
3–4 ml was added in different trials to visually observe precipitation. transfer is due to interfacial forces acting and interactions between
Then the solution was filtered and the crystals were dried overnight in a water and solid particles.
→
glass desiccator at room temperature (22 ± 2 °C) in order to measure The term Fi represents the centrifugal force and Coriolis accelera-
the distribution of seed crystal size, which shows the initial size dis- tion force term applied rotation and is written as:
tribution of particles, 4 random samples were collected from the dried → → → → →
F1 = −2α i ρi N × U1 − α i ρi N × (N × →
r) (11)
seed stock and then dispersed in 4 tubes containing DI water with a pH
=
of 8.5. The pH of the DI water was kept alkaline (pH 8.5) to avoid any The Reynolds stress tensor is based on the mean velocity gra-
τeffi
dissolution of struvite crystal in solution. The seed had a solid con- dients using the Boussinesq hypothesis:
centration of 3 g/L and was sonicated for 5 min to make sure that no
= → →T 2 → =
substantial aggregation between the seed particles took place and was τeffi = αi (μlam, i + μt , i )(∇U1 + ∇Ui ) − αi (ρi ki + (μlam,i + μt ,i ) ∇ . U1 ) I
reflected in the CSD measurement. The Branson Digital Sonifier 450 3
(Branson Ultrasonic Corp., Danbury, CT, USA) was used to sonicate the (12)
solution. The power output of the sonifier is up to 400 W at 100%
amplitude. In this study, the solution was sonicated at 70% amplitude
for 5 min. Although the present study did not perform any parametric
studies to assess influence of sonication time on aggregation between
seed particles, the impact of the 5 min of sonication was assumed to be
sufficient to prevent significant aggregation between the seed particles.
CSD measurement was performed for all four of the random samples
and the average was chosen as the initial size distribution for seed.
Fig. 2 shows the CSD results for four samples and the average chosen as
the initial size distribution for seed.

3. CFD modeling

3.1. Mathematical modeling and governing equations

The mathematical model of liquid-solid flow was developed via the

3D Eulerian multiphase approach. In the Eulerian approach, the two
phases are considered mathematically as interpenetrating continua. As
the two phases occupy separate volumes in the same 3-D space, the
concept of volume fraction is valid, and these two sets of equations are Fig. 2. Number-based initial crystal size distribution (CSD) of seeds, (a) 4
closed by constitutive relations or by the application of kinetic theory. random samples and (b) average initial CSD used in the CFD model.

694
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

where μlam, i is the laminar viscosity of ith phase and μt , i is the turbulent k2
μt,m = ρm Cμ
viscosity of ith phase. ε (21)
where Cμ is constant. Other fundamental turbulent flows model constant
3.1.3. Interface momentum exchange values have been determined from previous experiments. The most
The two phases interact through momentum transfer term [Eqs. (9) widely accepted standard values for these constants are as follows [67]:
and (10)]. This momentum exchange occurs through interfacial forces
acting on liquid and solid. These forces- considered as sources or sinks C1ε = 1.44
in the momentum equation-must be formulated separately. C2ε = 1.92
In this work a simple form of interaction has been used to solve the Cμ = 0.09
interphase force: σk = 1.0
n n
σε = 1.3
  
∑ Rji = ∑ Kji (vi − vj )
j=1 i=1 (13)
3.1.5. Species transport equation
where →vi and →
vj are the phase velocities and Kji = Kij is the interphase The conservation equation for the ith species is
momentum exchange coefficient, which can be written in the following
∂ ⎛ Dt ⎞
equation: (ρ Y)
i + ∇ . (ρ Vk Y)
i = ∇ . ρDm + ⎜ ∇Yi + Ri + Si
⎟

∂t ⎝ Sc t⎠ (22)
α i α j ρj f
Kji = where Yi is the ith species mass fraction, Sct is turbulent Schmidt
τj (14) number, Dt is the turbulent diffusivity, Dm is molecular diffusion, Ri is
where τj is particulate relaxation time defined as: the rate of species i production by chemical reaction, and Si is the rate
of addition by any extra sources. The default Sct number is 0.7 [68].
ρj d2j
τj =
18μi (15) 3.1.6. Population balance equations
The population balance equation (PBE) provides a simple form of
where dj is the diameter of solid particles.
population continuity for any phenomena. It can be derived as a bal-
And f is the drag function, which contains the drag coefficient CD .
ance for particles in a physical space. The general form of PBE is defined
There are different drag models used throughout the literature and the
as:
difference between them relies on the difference between the ways they
are calculating this f function. Between different drag models, Syamlal- ∂
[n(V, t)] + ∇ . [→
u n(V, t)] = Bagr − Dagr + Bbrk − Dbrk
O’Brien offers more accurate results in this type of study as compared to ∂t (23)
the other methods [67].
where Dagr and Bagr are death rate and birth rate of particles, respec-
The Syamlal-O’Brien model:
tively, of volume V due to aggregation, while Dbrk and Bbrk represent the
CD Res αl death rate and birth rate of particles of volume V due to breakage,
f= 2
24vr,s (16) respectively. Aggregation rate and breakage rate are defined as follows:

Drag coefficient derived by Dalla Valla: 1 V

Bagr =
2
∫0 a(V − V ', V ')n(V − V ')n(V ')dV '
(24)
2
⎛ 4.8 ⎞ ∞
CD = ⎜0.63 + Dagr = ∫0 a(V, V ')n(V)n(V ')dV '
⎝ Res / vr,s ⎟⎠ (17) (25)

In this model, based on measurements of the terminal velocities of Bbrk = ∫Ω v

vg(V ') β (V|V ')n(V ')dV '
(26)
particles in settling or fluidized beds, the relative Reynolds number can
be derived, which has a correlation with volume fraction: Dbrk = g(V)n(V) (27)
 
ρl d s |vs − vl | As Dagr and Bagr describes the aggregation rate and Dbrk and Bbrk
Res = describes the breakage rate. To link population balance modeling of the
μl (18)
secondary phase with fluid dynamics, Sauter mean diameter (d32) will
where l is for the liquid phase, s is for the solid phase, and ds is the be calculated by the ANSYS Fluent and used to represent the particle
diameter of the solid phase particles. diameter of the solid (secondary phase) in each calculation step. Sauter
mean diameter can be obtained from:
3.1.4. Turbulence modeling – RNG k – ɛ model
∑ ni d3i
In this work, the renormalized group k-ɛ model was used. For the d32 =
continuous phase—liquid phase in this study—the RNG k-ɛ model was ∑ ni d 2i (28)
used and the governing equations are written as below: Which illustrates the diameter of a sphere in a collection of equi-
∂ μt,m sized spherical objects that has the same “volume/surface area” ratio as
(ρ k) + ∇ . (ρm→
vmk) = ∇ . ⎛ ∇k⎞ + Gk,m − ρm ε
⎜ ⎟
the collection of polysized particles in the system. In this work, for
∂t m ⎝ σk ⎠ (19)
solving the PBE the discretized method, also known as the method of
∂ μt,m classes or sectional method, which was developed by Ramkrishna [69],
(ρ ε ) + ∇ . (ρm→
vmε ) = ∇ . ⎛ ∇ε ⎞ + C1ε Gk,m − C2ε ρm ε
⎜ ⎟
Hounslow et al. [70] and Lister et al. [71], is implemented. In this
∂t m ⎝ σε ⎠ (20)
method, the continuous particle population is discretized into a finite
where Gk, m is generation of turbulent kinetic energy due to the mean number of size intervals or bins. Although this approach has the ad-
velocity gradients. The effective turbulent transport varies with the vantage of computing particle size distribution directly, it requires a
effective Reynolds number (or eddy scale) allowing the model to better priori knowledge of the bin sizes and may need a large number of classes
handle low-Reynolds number and near-wall flows. In the high-Reynolds to cover the whole range of particle sizes. The size intervals or bins
number, as used this work, turbulence viscosity of the ith phase cal- were fixed on the basis of experimental size ranges obtained from the
culated form PSA. The PSA provided us CSD on the basis of different size ranges. We

695
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

selected 25 size ranges, which encompass all the experimental CSD 3.3. Method of solution
observations in this study. 25 size classes—the exact same sizes as re-
ported by PSA—have been applied to the model through a probability Modeling the impeller rotation is complex as there is relative motion
density function (PDF) file to cover the entire potential range of particle between the stationary baffles and rotating impeller blades. The two
sizes. most common approaches for this type of modeling are the Multiple
Reference Frame (MRF) and Sliding Mesh (SM) models. In these
models, the solution domain is divided into two main regions: (1)
3.1.7. Growth rate
Rotor: inner region containing the rotating impeller and (2) Stator:
Crystal growth rate can be described as a function of the solution
outer region containing the stationary baffles and vessel walls. For the
supersaturation index SI (defined earlier) as follows:
MRF model, steady-state calculations are performed with a rotating
G(t) = dl/ dt = k gSIn (29) frame of reference in the impeller region and a stationary frame of
reference in the outer region. For the SM model, the impeller region
where k g and n are the kinetic coefficients. The difference in solubility slides relative to the outer region in discrete time steps and time-de-
of the various faces of a single crystal, or of different size particles of the pendent calculations are performed using implicit or explicit inter-
same material, are not large enough to influence crystal growth unless polation of data at consecutive time-points. Although the SM method is
the crystals are less than 1 μm [72]. Therefore, we assumed that the more accurate for the modeling of actual phenomenon such as impeller
struvite crystals suspended in the solution grow at a similar rate. In this rotation as it is time dependent, it is computationally demanding. In
work, as explained earlier, the initial solution was prepared in a way this work, the MRF method was used and the MRF zone was created
that the solution was in the metastable zone so that crystallization was around the Rushton blades and the shaft as shown in Fig. 4.
not governed by nucleation. Therefore, the kinetic model has been se- In this work, all computational work was performed in double
lected based on the assumption that nucleation and aggregation are precision mode using the commercially available CFD code package,
negligible and a growth rate was used. The values of the coefficients (kg ANSYS Fluent v.17.1. The pressure-based solver is used to numerically
and n) were taken from the average of the ranges as reported by Ali and solve the continuity and momentum equations for velocity and pressure
Schneider [20]. The exact values of kg and n used in this study were fields. A coupled SIMPLE scheme was chosen for the pressure-velocity
48 μm/h and 1.66, respectively, and the growth rate was added into the coupling method. This algorithm uses a relationship between velocity
simulation through User Defined Function (UDF). and pressure correction to enforce conservation of continuity and to
attain the required pressure field. All governing equations, described
earlier, were discretized by ANSYS Fluent software using the finite
3.2. Solution domain, mesh and boundary conditions
volume approach with a First-Order upwind discretization scheme.
Then, an Eulerian multiphase method implemented in the software was
The same reactor, as used in the experiment, was modeled in si-
applied. Liquid phase was considered the continuous phase (primary
mulation. The reactor was built of glass and had a diameter of 124 mm
phase) and the solid was considered the dispersed phase (secondary
and a height of 90 mm. The bottom radius of the reactor is 142.5 mm. A
phase). The time step size used in the model was initially 0.001 s and, as
standard six-blade Rushton turbine impeller agitated the reactor. The
clearance between the centrally located impeller and the bottom of the solution converged, the time step size was gradually increased to
0.01 s to save computational time. The total computational time was
reactor was held at 35 mm. Further geometric details are provided in
Fig. 3. about 8 weeks for simulation of 60 min (3600 s) of physical time. All
calculations were performed with 48 paralleled processors in the high-
No slip wall boundaries were set to all walls. For the reactor upper
surface, pressure outlet was assumed as fixed to the local atmosphere. performance computing facilities of CalculQuebec and
ComputeCanada.
As it is important in such modeling to employ an adequate number of
computational cells, the geometry divided into more than 2 × 105 3-D
tetrahedral computation discrete control volumes or cells by the com- 4. Results and discussion
mercial grid generation tool, ANSYS Workbench v.17.1. To assure that
the results were independent of the mesh size, four different tetrahedral 4.1. Grid independence results
grid configurations were chosen to check mesh independency. Also, in
order to capture the complex flow behavior around both the impeller Four different grid configurations were chosen to understand the
tip and at the baffle edges, a refinement factor was used in these re- influence of mesh configuration on the flow prediction. The detailed
gions. The difference between these four grids resulted from the grid profile for each grid case and the grid study results are presented in
size and the refinement factor used, as seen in Table 2. Table 2.

Fig. 3. Schematic views of the reactor geometry and impeller used for this study.

696
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

Table 2
Different applied grids and number of cells.
Grid Case No. 1 2 3 4

5 5 5
Number of cells 1.9 × 10 2.2 × 10 3.5 × 10 4.5 × 105
Grid size (mm) 5.5 5 5 4
Refinement factor around baffles 1 1 1 1

around impeller 1 1 2 2
Liquid phase velocity at point (1) (m/s) 0.0153 0.0266 0.0282 0.0288
Liquid phase velocity at point (2) (m/s) 0.233 0.256 0.259 0.260
Liquid phase velocity at point (3) (m/s) 0.0680 0.0764 0.066 0.0571
Average Liquid phase velocity (m/s) 0.0619 0.0623 0.0624 0.0625

where ρ is the density (kg/m3), μ is viscosity (kg/m·s), DI is the tip-to-

tip impeller diameter and N is the impeller speed (s−1) .

Pi (W) = 2π N(s−1)M(Nm) (31)

where Pi is the power needed for rotating the impeller (W ), N is the

speed of rotation (s−1) and M is the torque (Nm ). The power numbers
were then calculated as dimensionless numbers to generalize power
factor for any case as follows:
Pi (W)
Np =
ρ (kgm−3)N(s−1)3D(m)5 (32)

where ρ is the fluid density (kgm−3)

and D is impeller diameter (m ). It
should be pointed out that for Reynolds number less than 10, Eq. (32) is
not valid.
Fig. 4. 3D Schematic view of the reactor domains (a) computational grids
To cover a complete range of fluid regimes, the simulation starts
employed, (b) MRF zones and selected points for mesh independency check.
with Reynolds number = 10 and a corresponding speed of 0.2 rpm. The
The selected points for mesh independency check are (1) a point near the top of
one stationary baffle in the stator zone, (2) a point on the middle-outside edge other Reynolds numbers were simulated, by changing the speed of
of a rotating blade of the impeller in the rotor zone, (3) a point on the bottom agitation. Reynolds numbers of 10, 100, 200, 1000, 5000, 10,000,
wall of the reactor below the impeller in the stator zone. 25,000, and 50,000 were simulated. For the turbulent regime, the RNG
k-ε model was chosen, as it is used for further steps, and for Reynolds
numbers lower than 200, laminar flow was chosen since Reynold
The liquid phase velocity at three different points (points 1, 2, and
number = 200 is the boundary of laminar flow in stirred tanks with
3) shown in Fig. 4 and the volume-averaged liquid phase velocity over
Rushton impellers [77]. The results for this comparison are shown in
the whole domain were chosen to study the grid sensitivity of the
Fig. 5.
model. As the differences between numerical results in grid case no. 2, 3
As is apparent, the model prediction for single phase flow is in
and 4 were not very high in most cases when compared to the same
agreement with the Rushton et al. [76] experimental data. In this study,
differences observed with grid case no. 1; grid case no. 2 was chosen for
the single phase and multiphase simulation results were obtained using
the simulation in this study as it runs faster.

4.2. Flow pattern prediction validation

For any CFD simulation, it is desirable to validate the CFD predic-

tion of the flow field before commencing any further steps. In this work,
this validation was performed by experimental results of other re-
searchers. Geometry-independent parameters such as impeller power
number, overall flow pattern, and velocity magnitude distribution were
chosen as parameters to validate the hydrodynamic behavior of the CFD
model.

4.2.1. Impeller power number validation

The impeller power number Np has been commonly used to check
the validity of CFD simulation of single phase flow in stirred tanks
[73–75]. In this work, simulation was conducted for both single phase
flow (water) and multiphase flow (mixture) as a validation technique
for multi-phase flow, and power results were compared with experi-
mental values reported by Rushton et al. [76]. Power numbers were
calculated from the model prediction torque for a wide range of Rey-
nolds numbers, from laminar to turbulent flow regimes, and based on
the following equations:
Fig. 5. Comparison of experimental and CFD model (single phase and multi-
ρND2I phase) predicted impeller power number with experimental data provided by
Re =
μ (30) Rushton et al. [76].

697
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

a round-bottomed reactor. The reactor had a rounded bottom with a volume fraction of the dispersed phase was 1.47 × 10−5 when the seed
large radius (142.5 mm) as compared to the height and diameter of the crystals were introduced to the reactor. At the end of 60 min, the solid
reactor. As a result, the height of the bottom-curved part (15.3 mm) was volume fraction of the dispersed phase increased by about three times
not very large as compared to the overall height of the reactor (90 mm). to 4.32 × 10−5.
It can be hypothesized, though not determined here, that the impact of
a large-radius curved bottom, which constituted a small part of the total 4.3. Crystal size distribution (CSD) variation by time
reactor height, offered minute deviations in power number prediction
for different Re number when compared to the results obtained with a CSD was experimentally measured at a number of time intervals:
flat bottom reactor by Rushton et al. [75]. In addition to this, errors in after 3, 10, 20, 30, 40, 50 and 60 min from starting the precipitation
extraction of numerical data from a graph in the original Rushton et al. (seeding) and the results were compared with the model CSD output in
[75] paper, using a web-based plot digitizer, may have led to some Fig. 8.
errors in the plotting of the solid line in Fig. 5. Since both these axes are As was demonstrated, the results predicted by the model closely
in logarithmic scale, a small deviation might result in more agreement follow the same pattern as the experimental values. In 0 min (Seed)
to the predicted data obtained in this study. Nevertheless, the predicted distribution figure, a small discrepancy between model prediction and
data showed similar trends to those observed by Rushton et al. [75]. experimental measurements was observed. This disagreement can be
Similar results obtained for multiphase flow show satisfactory agree- explained as follows: for the initial distribution of particles (or seed size
ment with the single-phase results, as well as experimental data. This distribution), the software is limited to “volume-based distribution”,
correspondence in the single phase and multiphase flow results was which was measured by the particle size analyzer. As mentioned earlier,
expected since the solid volume fraction in this work is in the range of use of the laser scattering particle size analyzer (which initially mea-
10−5. Although in the present study a two-phase system was modeled sures volume-based distribution and then calculates number-based
due to the presence of a solid fraction and a liquid fraction in the re- distribution using a software) may lead to large errors, especially at the
actor, it can be argued that a single-phase model could have also suf- lower end of the size range. It is worth pointing out that this numerical
ficed for similar result output under such solid volume fraction sce- error will add on and become part of the errors in the CSD results ob-
nario. The highest percentage of deviation (∼17%) occurred at tained at later time intervals.
Re = 10. A possible reason for this discrepancy is the effect of baffles While the magnitudes of the crystal size distribution fractions do not
used for development of mixing conditions which can disturb the la- match the experimental results in every case, the shape of the model-
minar flow regime in Reynolds number = 10. predicted distributions is very similar to the experimental results. Also,

4.2.2. Overall flow pattern

An assessment for the accuracy of the CFD model in predicting flow
field could be conducted by comparing the velocity vector configura-
tion with the laser Doppler velocimetry (LDV) results of previous stu-
dies. Costes and Couderc [78] investigated the hydrodynamic char-
acteristics of the flow induced by a standard Rushton impeller. They
reported mean velocities in two vertical planes; the median plane be-
tween two baffles (θ = 45°) and the plane of the baffles (θ = 0°). Fig. 6
(E1/CFD1 and E2/CFD2) shows the CFD model velocity prediction in
comparison with Costes and Couderc [78] results in the plane of baffles
and in a median plane between two baffles, respectively.
The circulation pattern of the flow is correctly reproduced. The
position of the recirculation centers is in accordance with the experi-
mental results, showing that as we expected, Rushton turbine is a radial
flow type of impeller, as the axial velocity components are almost
vertical at the top of the blades. Also, the intensity and direction of the
radial flow jets from impeller tip toward the reactor walls were also
correctly predicted.
It is important to note that the flow pattern and recirculation ar-
rangement are quite independent of the stirrer speed and of the size of
the reactor as reported by Costes and Couderc [78] and also the qua-
litative nature of the multiphase flow field around the impeller and
shape of the flow jet is similar to that of single phase flow, especially for
low solid concentration as investigated by Wadnerkar et al. [79].
Another parameter, which a CFD model should be able to predict, is
the velocity magnitude distribution in a stirred tank. Here, the contour
for the velocity is shown in Fig. 7, which presents the velocity profile as
predicted by the CFD model. As expected, velocity magnitude decreases
according to the increased radius and the highest magnitude velocity
occurs at the impeller tip. Contour plots of the solid volume fraction in
the reactor (at minute 60) on a vertical symmetry plane and three
different horizontal planes located at 20 mm, 50 mm, and 80 mm from
the bottom of the reactor are shown in Fig. S1 (Supporting
Information). It can be observed that, for most part, the inhomogeneity
in the solid phase volume fraction of the reactor fell within a narrow Fig. 6. Comparison of experimental results by Costes and Couderc [78] flow
range between 3.90 × 10−5 and 4.55 × 10−5. Larger solid phase vo- pattern with flow pattern prediction obtained by the CFD model at (E1 and
lume fractions were present in the zone below the rotating impeller CFD1) vertical plane containing a baffle (θ = 0°) and (E2 and CFD2) a vertical
(Fig. S1) due to the low velocity in that region. The initial overall solid plane between baffles (θ = 45°).

698
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

crystals. The magnitudes of percent fractions in most of the bins in

every time interval are in the reasonable range of experimental error.
The percent fractions for other sizes in the distribution, obtained from
the model, showed 0–20% deviation from experimental results. The
mean (μm), standard deviation (μm), and skewness of the model pre-
dicted and experimental CSDs after 0, 3, 10, 20, 30, 40, 50 and 60 min
are provided in Table S1, showing similar trends. As crystal size in-
creases with experimental time, the standard deviation and skewness
also increases in the experimental condition due to the uncertainties
involved in experimental sampling and measurements.
For most cases, the model over-predicts (up to 14%) the size of the
crystals in different bins and time intervals, which may be due to the
Fig. 7. Velocity contours on (a) a horizontal plane at the middle of impeller (b) presence of nucleation in the reactor. The nucleation mechanism results
a vertical symmetry plane. in creating finer crystals in comparison with the existing grown crystals.
It can be hypothesized here that overlooking nucleation mechanisms
and assuming that all the mass transfer resulting from only crystal
the position of the peak in each graph—which shows the typical size of
growth leads to over-prediction of the “growth/(growth + nucleation)”
crystals at each time interval—is mostly in good agreement with ex-
ratio in mass transfer; which results in the prediction of larger crystal
perimental data. The model predicted peak-size percent fraction, which
sizes. The present study did not measure the possible contributions of
shows the typical size of crystals at each time interval, revealed a de-
overlooking of nucleation mechanism and the effect of other un-
viation from experimental results of 1.42%, 0.05%, 2.43%, 14.6%,
certainties (i.e. limitations in model numerical calculations and devia-
11.2%, 11.7%, 13.6%, and 14.2% at 0, 3, 10, 20, 30, 40, 50, and
tions in experimental results) in the observed over-prediction of crystal
60 min, respectively. These deviances could be a combined contribution
size by the model.
of: (a) the uncertainties involved in the numerical techniques employed
The experimental data uncertainties (i.e. uncertainties involved in
for the coupled CFD-PBM model, and (b) the eccentricities in the ex-
sampling crystals from the reactor) and CSD measurement fluctuations
perimental results. The present study did not evaluate the individual
might have become more significant as time passed. This hypothesis,
contributions of the uncertainties from numerical techniques occurring
though not validated in this study, is made from the relatively higher
in the CFD and PBM models. Nonetheless, these CSD observations from
degrees of deviations observed in the CSD measurements at times
experimental results and model predictions can be regarded as a posi-
50 min and 60 min. One possible reason for these variations in experi-
tive result given the inherent limitations of the numerical techniques in
mental data, especially at time 50 min and 60 min could be due to the
the models and the uncertainties incurred when sampling reactor

Fig. 8. Experimental averaged ( ) and CFD model predicted (–––) number fraction based particle size distribution at time = 0 (Seed), 3, 10, 20, 30, 40, 50, and
60 min.

699
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

4.4. Supersaturation state prediction

The results of SI variation over time are compared with the values
predicted by CFD model in Fig. 10. To find the SI values at every ex-
perimental time interval, pH measurement was carried out with ex-
treme care in this study. The total concentration of Mg, N, and P are
constant in the reactor. With formation of struvite, elemental con-
centrations will decline in the liquid phase and increase in the solid
phase. Any change in pH will affect the value of SI. Therefore, the value
of pH was measured every 5 min from t = 0 (seeding moment) to
t = 60 min in all experiments. Then the thermodynamic equilibrium
model calculated the SI value from liquid phase elemental concentra-
tions.
The decrease in pH shows that precipitation does occur, since with
precipitation, constituent ions will move from liquid phase to solid
phase and change the speciation of the solution. Based on Eq. (1), as
struvite precipitates, it releases hydrogen ions in solution, causing a
drop in pH. The model predicted pH (via charge balance) and the
measured pH values are compared in Fig. 10(a) and a maximum de-
viation of about 0.2 were observed between them. The developed
Fig. 9. Experimental and CFD predicted particle average size – number fraction model successfully predicts the SI values with acceptable accuracy. The
weighted. maximum amount of error was 24%, belonging to the supersaturation
index at 60 min.
use of the laser scattering technique to measuring crystal size dis-
tribution in the dilute solution. As time passes, and precipitation ma- 5. Conclusions
tures, the solution becomes more cloudless (i.e. more transparent) as
compared to initial time solution samples. As a result, the laser scat- The integrated 3-D CFD-PBM, developed in this study, provides a
tering particle size analysis technique offers less certainty. general approach to obtain CSD of struvite by considering hydro-
Fig. 9 shows the experimental and model predicted average crystal dynamic, thermodynamic, and kinetic aspects of solution chemistry in a
sizes. Since struvite crystals are the product of this process, knowing the batch stirred reactor. Size distribution is essential as it controls a dis-
average size of the product at different time intervals is an important tinct quality of final struvite crystals, which is linked to its commercial
parameter to control the process and produce products within desirable application as fertilizer. Hence the development of such general model
size range. It can be observed that the averages of the last two ex- will enable the design of efficient reactors for the production of spe-
perimental measurements (at 50 min and 60 min) were lower than that cially sized struvite crystals and maximize the economic value of pro-
of the previous observation (at 40 min). This may be due to the fact that duced fertilizers. The model estimated the time-varying crystal size
there is more chance of breakage of produced crystals during the later distribution of particles within an acceptable level of agreement with
stages of the experiment. As the particle size increases, it can break due experimental measurements. Results showed that ignoring nucleation,
to collision with baffles, impellers, and each other. So, the average and considering just the growth phenomena, merged with experimental
experimental measurements were lower in the last two readings. uncertainties and limitations of the numerical techniques employed for
It can be observed that the scatter of the experimental data mostly the coupled model, mostly caused over-prediction (up to 14%) of the
encompasses the model prediction at different times. Although average size of crystals in different time intervals and different size ranges. As a
crystal size is over-predicted in most cases as discussed before, it fol- result, the model though followed the same trend, but over-predicted
lows the same pattern as real data and the highest error percentage, of the average crystal size in the reactor in different time intervals,
about 10%, occurs at time 20 min. showing a maximum deviation of 10% at the time of 20 min. The
general pattern and the position of peaks in CSD curves were found to
be in good agreement with the experimental measurements. The model

Fig. 10. Experimental and CFD Model predicted (a) solution pH-time variation, and (b) Supersaturation index-time variation.

700
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

predicted percent fraction for the peak-size, which shows the typical potassium recovery from source separated urine through struvite precipitation,
size of crystals at each time interval, revealed a range of 0–14% de- Water Res. 41 (2007) 458–466, https://doi.org/10.1016/j.watres.2006.10.014.
[16] Q. Zhao, T. Zhang, C. Frear, K. Bowers, J. Harrison, S. Chen, Phosphorous recovery
viation from experimental observations. Compared to previous at- technology in conjunction with dairy anaerobic digestion, CSANR Research Report
tempts, which mostly showed average crystal size, the coupled CFD- 2010-001, 2010. Available from: http://csanr.wsu.edu/publications/
PBM model in this study made CSD prediction by considering three researchreports/cffreport.html.
[17] M.S. Rahaman, D.S. Mavinic, A. Meikleham, N. Ellis, Modeling phosphorus removal
different aspects (i.e., hydrodynamic, thermodynamic, and kinetic) of and recovery from anaerobic digester supernatant through struvite crystallization in
solution chemistry in a stirred reactor. Although not included in this a fluidized bed reactor, Water Res. 51 (2014) 1–10, https://doi.org/10.1016/j.
study, it would be interesting to investigate and compare the CSD watres.2013.11.048.
[18] A. Guadie, S. Xia, W. Jiang, L. Zhou, Z. Zhang, S.W. Hermanowicz, X. Xu, S. Shen,
predictive capabilities of the CFD model against other modeling ap- Enhanced struvite recovery from wastewater using a novel cone-inserted fluidized
proaches (e.g., a well-mixed system model). Such a well-mixed system bed reactor, J. Environ. Sci. (Beijing, China) 26 (2014) 765–774, https://doi.org/
will generalize the hydrodynamic features of the reactor and will re- 10.1016/S1001-0742(13)60469-6.
[19] C.C. Su, R.R.M. Abarca, M.D.G. de Luna, M.C. Lu, Phosphate recovery from flui-
duce computational requirements when compared to CFD models,
dized-bed wastewater by struvite crystallization technology, J. Taiwan Inst. Chem.
which was used in the present study. The proximity of model predic- Eng. 45 (2014) 2395–2402, https://doi.org/10.1016/j.jtice.2014.04.002.
tions to observed experimental results indicates the suitability of this [20] M.I. Ali, P.A. Schneider, An approach of estimating struvite growth kinetic in-
model in the design of a stirred tank reactor to produce struvite crystals corporating thermodynamic and solution chemistry, kinetic and process descrip-
tion, Chem. Eng. Sci. 63 (2008) 3514–3525, https://doi.org/10.1016/j.ces.2008.
in the desired size range. 04.023.
[21] L. Pastor, D. Mangin, R. Barat, A. Seco, A pilot-scale study of struvite precipitation
Acknowledgements in a stirred tank reactor: conditions influencing the process, Bioresour. Technol. 99
(2008) 6285–6291, https://doi.org/10.1016/j.biortech.2007.12.003.
[22] M. Hanhoun, L. Montastruc, C. Azzaro-Pantel, B. Biscans, M. Frèche, L. Pibouleau,
The authors would like to thank the Natural Sciences and Simultaneous determination of nucleation and crystal growth kinetics of struvite
Engineering Research Council (NSERC) of Canada for providing the using a thermodynamic modeling approach, Chem. Eng. J. 215–216 (2013)
903–912, https://doi.org/10.1016/j.cej.2012.10.038.
financial support required to conduct this study. The authors would also [23] Z. Wang, Z. Mao, C. Yang, X. Shen, Computational fluid dynamics approach to the
like to acknowledge CalculQuebec and ComputeCanada for providing effect of mixing and draft tube on the precipitation of barium sulfate in a continuous
high-speed computing facility for the present study. stirred tank1, Chin. J. Chem. Eng. 14 (2006) 713–722, https://doi.org/10.1016/
S1004-9541(07)60001-9.
[24] D. Logashenko, T. Fischer, S. Motz, E.D. Gilles, G. Wittum, Simulation of crystal
Appendix A. Supplementary data growth and attrition in a stirred tank, Comput. Visual. Sci. 9 (2006) 175–183,
https://doi.org/10.1007/s00791-006-0031-z.
[25] Z. Sha, P. Oinas, M. Louhi-Kultanen, G. Yang, S. Palosaari, Application of CFD si-
Supplementary data to this article can be found online at https://
mulation to suspension crystallization—factors affecting size-dependent classifica-
doi.org/10.1016/j.cej.2018.12.089. tion, Powder Technol. 121 (2001) 20–25, https://doi.org/10.1016/S0032-
5910(01)00369-2.
References [26] Z. Zhu, H. Wei, Flow field of stirred tank used in the crystallization process of
ammonium sulphate, ScienceAsia 34 (2008) 97–101, https://doi.org/10.2306/
scienceasia1513-1874.2008.34.097.
[1] UN, Concise Report on the World Population Situation in 2014, United Nations, [27] W. Wantha, A.E. Flood, Numerical simulation and analysis of flow in a DTB crys-
New York, 2014. tallizer, Chem. Eng. Commun. 195 (2008) 1345–1370, https://doi.org/10.1080/
[2] P. Heffer, M. Prud, Fertilizer Outlook 2014-2018, 82nd IFA Annual Conference, 00986440801963527.
International Fertilizer Industry Association (IFA), Sydney, Australia, 2014, pp. 1–7. [28] R. Plewik, P. Synowiec, J. Wojcik, A. Kus, Suspension flow in crystallizers with and
[3] S. Kataki, H.M. West, M.L. Clarke, D.C. Baruah, Phosphorus recovery as struvite: without hydraulic classification, Chem. Eng. Res. Des. 88 (2010) 1194–1199.
recent concerns for use of seed, alternative Mg source, nitrogen conservation and [29] M.S. Rahaman, D.S. Mavinic, Recovering nutrients from wastewater treatment
fertilizer potential, Resour. Conserv. Recycl. 107 (2016) 142–156. plants through struvite crystallization: CFD modelling of the hydrodynamics of UBC
[4] T. Zhang, L. Ding, H. Ren, Pretreatment of ammonium removal from landfill lea- MAP fluidized-bed crystallizer, Water Sci. Technol. 59 (2009) 1887–1892, https://
chate by chemical precipitation, J. Hazard. Mater. 166 (2009) 911–915, https:// doi.org/10.2166/wst.2009.214.
doi.org/10.1016/j.jhazmat.2008.11.101. [30] M. Al-Rashed, J. Wójcik, R. Plewik, P. Synowiec, A. Kuś, Multiphase CFD modeling:
[5] O. Tünay, I. Kabdasli, D. Orhon, S. Kolçak, Ammonia removal by magnesium am- fluid dynamics aspects in scale-up of a fluidized-bed crystallizer, Chem. Eng.
monium phosphate precipitation in industrial wastewaters, Water Sci. Technol. 36 Process. 63 (2013) 7–15, https://doi.org/10.1016/j.cep.2012.11.001.
(1997) 225–228, https://doi.org/10.1016/S0273-1223(97)00391-0. [31] C.V. Rane, A.A. Ganguli, E. Kalekudithi, R.N. Patil, J.B. Joshi, D. Ramkrishna, CFD
[6] I. Kabdasli, M. Gürel, O. Tünay, Characterization and treatment of textile printing simulation and comparison of industrial crystallizers, Can. J. Chem. Eng. 92 (2014)
wastewaters, Environ. Technol. 21 (2000) 1147–1155, https://doi.org/10.1080/ 2138–2156, https://doi.org/10.1002/cjce.22078.
09593330.2000.9619001. [32] Z. Jaworski, A.W. Nienow, CFD modelling of continuous precipitation of barium
[7] J.M. Chimenos, A.I. Fernández, G. Villalba, M. Segarra, A. Urruticoechea, B. Artaza, sulphate in a stirred tank, Chem. Eng. J. 91 (2003) 167–174, https://doi.org/10.
F. Espiell, Removal of ammonium and phosphates from wastewater resulting from 1016/S1385-8947(02)00150-X.
the process of cochineal extraction using MgO-containing by-product, Water Res. 37 [33] M.S. Rahaman, M.R. Choudhury, A.S. Ramamurthy, D.S. Mavinic, N. Ellis,
(2003) 1601–1607, https://doi.org/10.1016/S0043-1354(02)00526-2. F. Taghipour, CFD modeling of liquid-solid fluidized beds of polydisperse struvite
[8] M. Türker, I. Çelen, Removal of ammonia as struvite from anaerobic digester ef- crystals, Int. J. Multiphase Flow 99 (2018) 48–61, https://doi.org/10.1016/j.
fluents and recycling of magnesium and phosphate, Bioresour. Technol. 98 (2007) ijmultiphaseflow.2017.09.011.
1529–1534, https://doi.org/10.1016/j.biortech.2006.06.026. [34] K.N. Ohlinger, T.M. Young, E.D. Schroeder, Predicting struvite formation in di-
[9] M. Ronteltap, M. Maurer, W. Gujer, Struvite precipitation thermodynamics in gestion, Water Res. 32 (1998) 3607–3614, https://doi.org/10.1016/S0043-
source-separated urine, Water Res. 41 (2007) 977–984, https://doi.org/10.1016/j. 1354(98)00123-7.
watres.2006.11.046. [35] I. Çelen, J.R. Buchanan, R.T. Burns, R. Bruce Robinson, D. Raj Raman, Using a
[10] R. Yu, J. Geng, H. Ren, Y. Wang, K. Xu, Combination of struvite pyrolysate recycling chemical equilibrium model to predict amendments required to precipitate phos-
with mixed-base technology for removing ammonium from fertilizer wastewater, phorus as struvite in liquid swine manure, Water Res. 41 (2007) 1689–1696,
Bioresour. Technol. 124 (2012) 292–298, https://doi.org/10.1016/j.biortech.2012. https://doi.org/10.1016/j.watres.2007.01.018.
08.015. [36] M.S. Rahaman, D.S. Mavinic, N. Ellis, Phosphorus recovery from anaerobic digester
[11] N.M. Khai, Chemical precipitation of ammonia and phosphate from Nam son supernatant by struvite crystallization: model-based evaluation of a fluidized bed
landfill leachate, Hanoi, Iran, J. Energy Environ. 3 (2012) 32–36, https://doi.org/ reactor, Water Sci. Technol. 58 (2008) 1321–1327, https://doi.org/10.2166/wst.
10.5829/idosi.ijee.2012.03.05.06. 2008.721.
[12] H. Xu, P. He, W. Gu, G. Wang, L. Shao, Recovery of phosphorus as struvite from [37] M. Hanhoun, L. Montastruc, C. Azzaro-Pantel, B. Biscans, M. Frèche, L. Pibouleau,
sewage sludge ash, J. Environ. Sci. (Beijing, China) 24 (2012) 1533–1538, https:// Temperature impact assessment on struvite solubility product: a thermodynamic
doi.org/10.1016/S1001-0742(11)60969-8. modeling approach, Chem. Eng. J. 167 (2011) 50–58, https://doi.org/10.1016/j.
[13] E.L. Foletto, W.R.B. dos Santos, M.A. Mazutti, S.L. Jahn, A. Gündel, Production of cej.2010.12.001.
struvite from beverage waste as phosphorus source, Mater. Res. (Sao Carlos, Braz.) [38] N.J. Barnes, A.R. Bowers, A probabilistic approach to modeling struvite precipita-
16 (2013) 242–245, https://doi.org/10.1590/S1516-14392012005000152. tion with uncertain equilibrium parameters, Chem. Eng. Sci. 161 (2017) 178–186,
[14] S.G. Barbosa, L. Peixoto, B. Meulman, M.M. Alves, M.A. Pereira, A design of ex- https://doi.org/10.1016/j.ces.2016.12.026.
periments to assess phosphorous removal and crystal properties in struvite pre- [39] N.C. Bouropoulos, P.G. Koutsoukos, Spontaneous precipitation of struvite from
cipitation of source separated urine using different Mg sources, Chem. Eng. J. 298 aqueous solutions, J. Cryst. Growth 213 (2000) 381–388, https://doi.org/10.1016/
(2016) 146–153, https://doi.org/10.1016/j.cej.2016.03.148. S0022-0248(00)00351-1.
[15] J.A. Wilsenach, C.A.H. Schuurbiers, M.C.M. van Loosdrecht, Phosphate and [40] N.O. Nelson, R.L. Mikkelsen, D.L. Hesterberg, Struvite precipitation in anaerobic

701
S.E. Mousavi et al. Chemical Engineering Journal 361 (2019) 690–702

swine lagoon liquid: effect of pH and Mg: P ratio and determination of rate constant, [58] M.M. François, J.G.H. Morel, Principles and Application of Aquatic Chemistry,
Bioresour. Technol. 89 (2003) 229–236, https://doi.org/10.1016/S0960-8524(03) 1993.
00076-2. [59] A.W. Taylor, A.W. Frazier, E.L. Gurney, Solubility products of magnesium ammo-
[41] M. Quintana, E. Sánchez, M.F. Colmenarejo, J. Barrera, G. García, R. Borja, Kinetics nium and magnesium potassium phosphates, Trans. Faraday Soc. 59 (1963)
of phosphorus removal and struvite formation by the utilization of by-product of 1580–1584, https://doi.org/10.1039/tf9635901580.
magnesium oxide production, Chem. Eng. J. 111 (2005) 45–52, https://doi.org/10. [60] A.E. Martell, R.M. Smith, Critical Stability Constants, Plenum Press, New York,
1016/j.cej.2005.05.005. London, 1974.
[42] K.S. Le Corre, E. Valsami-Jones, P. Hobbs, S.A. Parsons, Kinetics of struvite pre- [61] V.L. Snoeyink, D. Jenkins, Water Chemistry, Wiley, USA, 1980.
cipitation: effect of the magnesium dose on induction times and precipitation rates, [62] J.W. Mullin, Crystallization, third ed., Butterworth-Heinemann, 1993.
Environ. Technol. 28 (2007) 1317–1324, https://doi.org/10.1080/ [63] M.S. Rahaman, Phosphorus Recovery from Wastewater through Struvite
09593332808618891. Crystallization in a Fluidized Bed Reactor: Kinetics, Hydrodynamics and
[43] M. Quintana, M.F. Colmenarejo, J. Barrera, E. Sánchez, G. García, L. Travieso, Performance, The University of British Columbia, 2009.
R. Borja, Removal of phosphorus through struvite precipitation using a by-product [64] K.N. Ohlinger, T.M. Young, E.D. Schroeder, Kinetics effects on preferential struvite
of magnesium oxide production (BMP): effect of the mode of BMP preparation, accumulation in wastewater, J. Environ. Eng. 125 (1999) 730–737.
Chem. Eng. J. 136 (2008) 204–209, https://doi.org/10.1016/j.cej.2007.04.002. [65] M.I. Ali, P.A. Schneider, A fed-batch design approach of struvite system in con-
[44] M.S. Rahaman, N. Ellis, D.S. Mavinic, Effects of various process parameters on trolled supersaturation, Chem. Eng. Sci. 61 (2006) 3951–3961, https://doi.org/10.
struvite precipitation kinetics and subsequent determination of rate constants, 1016/j.ces.2006.01.028.
Water Sci. Technol. 57 (2008) 647–654, https://doi.org/10.2166/wst.2008.022. [66] A.G. Jones, Crystallization Process Systems, first ed., Butterworth-Heinemann,
[45] E. Ariyanto, T.K. Sen, H.M. Ang, The influence of various physico-chemical process Boston, 2002.
parameters on kinetics and growth mechanism of struvite crystallisation, Adv. [67] C.V. Rane, E. Kalekudithi, J.B. Joshi, D. Ramkrishna, Effect of impeller design and
Powder Technol. 25 (2014) 682–694, https://doi.org/10.1016/j.apt.2013.10.014. power consumption on crystal size distribution, AIChE J. 60 (2014) 3596–3613,
[46] A. Capdevielle, E. Sýkorová, F. Béline, M.L. Daumer, Kinetics of struvite pre- https://doi.org/10.1002/aic.14541.
cipitation in synthetic biologically treated swine wastewaters, Environ. Technol. 35 [68] J.B. Joshi, V.V. Ranade, Computational fluid dynamics for designing process
(2014) 1250–1262, https://doi.org/10.1080/09593330.2013.865790. equipment: expectations, current status, and path forward, Ind. Eng. Chem. Res. 42
[47] N.-M. Chong, Q.-M. Thai, Optimization and kinetics of nutrient removal from (2003) 1115–1128, https://doi.org/10.1021/ie0206608.
wastewater by chemical precipitation of struvite, Desalin. Water Treat. 54 (2015) [69] D. Ramkrishna, Population Balances, first ed., Academic Press, 2000 doi:10.1016/
3422–3431, https://doi.org/10.1080/19443994.2014.908416. B978-012576970-9/50000-X.
[48] M.I.H. Bhuiyan, D.S. Mavinic, R.D. Beckie, Nucleation and growth kinetics of [70] M.J. Hounslow, R.L. Ryall, V.R. Marshall, A discretized population balance for
struvite in a fluidized bed reactor, J. Cryst. Growth 310 (2008) 1187–1194, https:// nucleation, growth, and aggregation, AIChE J. 34 (1988) 1821–1832, https://doi.
doi.org/10.1016/j.jcrysgro.2007.12.054. org/10.1002/aic.690341108.
[49] M. Harrison, M.R. Johns, E.T. White, C.M. Mehta, Growth rate kinetics for struvite [71] J.D. Lister, D.J. Smit, M.J. Hounslow, Adjustable discretized population balance for
crystallisation, Chem. Eng. Trans. 25 (2011) 309–314, https://doi.org/10.3303/ growth and aggregation, AIChE J. 41 (1995) 591–603, https://doi.org/10.1002/
CET1125052. aic.690410317.
[50] A. Triger, J.S. Pic, C. Cabassud, Determination of struvite crystallization mechan- [72] S. Srisanga, A.E. Flood, S.C. Galbraith, S. Rugmai, S. Soontaranon, J. Ulrich, Crystal
isms in urine using turbidity measurement, Water Res. 46 (2012) 6084–6094, growth rate dispersion versus size-dependent crystal growth: appropriate modeling
https://doi.org/10.1016/j.watres.2012.08.030. for crystallization processes, Cryst. Growth Des. 15 (2015) 2330–2336, https://doi.
[51] C.M. Mehta, D.J. Batstone, Nucleation and growth kinetics of struvite crystal- org/10.1021/acs.cgd.5b00126.
lization, Water Res. 47 (2013) 2890–2900, https://doi.org/10.1016/j.watres.2013. [73] A. Brucato, M. Ciofalo, F. Grisafi, G. Micale, Numerical prediction of flow fields in
03.007. baffled stirred vessels: a comparison of alternative modelling approaches, Chem.
[52] S.C. Galbraith, P.A. Schneider, A.E. Flood, Model-driven experimental evaluation of Eng. Sci. 53 (1998) 3653–3684, https://doi.org/10.1016/S0009-2509(98)00149-3.
struvite nucleation, growth and aggregation kinetics, Water Res. 56 (2014) [74] C. Bartels, M. Breuer, K. Wechsler, F. Durst, Computational fluid dynamics appli-
122–132, https://doi.org/10.1016/j.watres.2014.03.002. cations on parallel-vector computers: computations of stirred vessel flows, Comput.
[53] D. Crutchik, J.M. Garrido, Kinetics of the reversible reaction of struvite crystal- Fluids 31 (2002) 69–97, https://doi.org/10.1016/S0045-7930(01)00016-0.
lisation, Chemosphere 154 (2016) 567–572, https://doi.org/10.1016/j. [75] D.A. Deglon, C.J. Meyer, CFD modelling of stirred tanks: numerical considerations,
chemosphere.2016.03.134. Miner. Eng. 19 (2006) 1059–1068, https://doi.org/10.1016/j.mineng.2006.04.001.
[54] S. Qamar, Modeling and Simulation of Population Balances for Particulate [76] J.H. Rushton, E.W. Costich, H.J. Everett, Power characteristics of mixing impellers –
Processes, Faculty of Mathematics, Otto-von-Guericke University Magdeburg, 2008. Part II, Chem. Eng. Prog. 46 (1950) 467–476.
[55] S. Qamar, G. Warnecke, M.P. Elsner, A. Seidel-Morgenstern, A Laplace transfor- [77] J.M. Zalc, M.M. Alvarez, F.J. Muzzio, B.E. Arik, Extensive validation of computed
mation based technique for reconstructing crystal size distributions regarding size laminar flow in a stirred tank with three Rushton turbines, AIChE J. 47 (2001)
independent growth, Chem. Eng. Sci. 63 (2008) 2233–2240, https://doi.org/10. 2144–2154, https://doi.org/10.1002/aic.690471003.
1016/j.ces.2008.01.025. [78] J. Costes, J.P. Couderc, Study by laser Doppler anemometry of the turbulent flow
[56] S.C. Galbraith, P.A. Schneider, Modelling and simulation of inorganic precipitation induced by a Rushton turbine in a stirred tank: Influence of the size of the units—I.
with nucleation, crystal growth and aggregation: a new approach to an old method, Mean flow and turbulence, Chem. Eng. Sci. 43 (1988) 2751–2764, https://doi.org/
Chem. Eng. J. 240 (2014) 124–132, https://doi.org/10.1016/j.cej.2013.11.070. 10.1016/0009-2509(88)80018-6.
[57] C.W. Childs, C.W. Childs, A potentiometric study of equilibria in aqueous divalent [79] D. Wadnerkar, M.O. Tade, V.K. Pareek, R.P. Utikar, CFD simulation of solid–liquid
metal orthophosphate solutions, Inorg. Chem. 9 (1970) 2465–2469, https://doi. stirred tanks for low to dense solid loading systems, Particuology 29 (2016) 16–33,
org/10.1021/ic50093a017. https://doi.org/10.1016/j.partic.2016.01.012.

702