Chemical Engineering Science 153 (2016) 45–57

Contents lists available at ScienceDirect

Chemical Engineering Science

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

Modelling agglomeration and deposition of gas hydrates in industrial

pipelines with combined CFD-PBM technique
Boris V. Balakin a,b,n, Simon Lo c, Pawel Kosinski b, Alex C. Hoffmann b
a
Bergen University College, Mechanical and Marine Engineering Department, Bergen, Norway
b
University of Bergen, Physics and Technology Department, Bergen, Norway
c
CD-adapco, Multiphase Model Development, United Kingdom

H I G H L I G H T S

The CFD-PBM model of the oil–water–hydrate slurry is developed.

The model is validated with experimental data.

The model is compared with CSMHyK from Colorado School of Mines.

The flow patterns are described and analysed.

art ic l e i nf o a b s t r a c t

Article history: Hydrates of light hydrocarbons are frequently formed during the subsea petroleum production. These
Received 2 December 2015 crystalline ice-like solids may accumulate at concentrations sensitive from the flow assurance point of
Received in revised form view, increasing the overall pumping costs and imposing sufficient risk of the pipe blockage. Modern
7 May 2016
trend in the assessment of hydrate-related risks is the development of numerical models of multiphase
Accepted 6 July 2016
flows laden by hydrates. The present paper describes a computational fluid dynamic (CFD) model capable
Available online 8 July 2016
to simulate turbulent slurry of oil, water and gas hydrates. The population balance technique (PBM)
Keywords: coupled with CFD enables to predict such details of the process as the formation of hydrate phase, ag-
Agglomeration glomeration of formed solids and granular interactions within the hydrate phase. The simulation results,
Gas hydrate
validated with experimental data in terms of the slurry rheology, highlight flow patterns for a pipe
CSMHyK
system typical in oil industry. The model is in addition compared to the hydrate kinetics model from
Population balance modelling
Computational fluid dynamics Colorado School of Mines (CSMHyK).
Multiphase flow & 2016 Elsevier Ltd. All rights reserved.
Flow assurance

1. Introduction Considering the rheology of a hydrate-particle-laden multi-

Gas hydrates are formed upon the pressure-driven formation of
crystalline water around cages occupied by “guest” gas molecules,
for example methane, at temperatures below 10 °C. In nature such
conditions are found at the bottom of deep water reservoirs (Sloan
and Koh, 2008) and in permafrost (Makogon, 1997). Formation of
artificial gas hydrates in pipelines often accompanies petroleum
production. Having formed in petroleum lines, hydrates are
transported by the flow as solid particles dissipating energy during
collisions with the pipe wall and each other, which significantly
increases pumping costs and may also form plugs. Therefore their
presence is considered to be disadvantageous.
n
Corresponding author at: Bergen University College, Mechanical and Marine
Engineering Department, Bergen, Norway.
E-mail address: boris.balakin@hib.no (B.V. Balakin).
phase flow, it is possible to determine the apparent viscosity of the
formed suspension, which is higher than the viscosity in case no
hydrates were present. If the local flow conditions favour it, gas
hydrates may deposit on the pipe walls forming solid obstructions
and even plugging the pipes (Sloan and Koh, 2008; Balakin, 2010).
This occurs mainly due to gravity or centrifugal forces in elbows,
but deposition is also promoted by agglomeration of hydrate
crystals in cases where they are cohesive (Dieker et al., 2008). This
problem is increasing as production lines become longer and ex-
ploration moves to colder environments, indicating the need for a
reliable predictive tool, enabling accurate assessment of hydrate
plugging risks. Modern tools of this kind involve computational
fluid dynamics (CFD).
Early CFD-modelling of hydrates in pipes, from the 1980s, were
based on the “moving mesh” principle, where the formation of a
hydrate obstructions was modelled by morphing the boundaries of
the computational domain to follow the surface of the obstruction.

http://dx.doi.org/10.1016/j.ces.2016.07.010
0009-2509/& 2016 Elsevier Ltd. All rights reserved.
46 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57

Nomenclature Greek letters

B breakage rate (1/m3 s) α collision efficiency

^
B breakage rate constant (s y − 1/m3) γ shear rate (1/s)
cp specific heat (J/kg K) δ Kronecker delta
CD drag coefficient ϵ turbulent energy dissipation rate (m2/s3)
d diameter (μm) ϕ volume fraction
D rate of strain tensor with components Dl, m (1/s) κ von Karman constant
f fractal dimension λ thermal conductivity (W/m K)
f(r) particle size distribution function (1/m3) μ dynamic viscosity (Pa s)
Fa adhesive force (N) μa apparent viscosity (Pa s)
Fint solid pressure force per unit volume (N/m3) ρ density (kg/m3)
g acceleration due to the gravity (m/s2) s surface tension (N/m)
h specific enthalpy (J/kg) st turbulent Prandtl number
k turbulent kinetic energy (m2/s2) τ stress tensor with components τ l, m (Pa)
kCH4 growth rate constant (kg/m2 s K) ξ latent heat (J/kg)
M interphase momentum transfer term (N/m3)
∞
Mi = ∫ r if (r ) dr ith moment of particle size distribution Subscripts, superscripts
0
(mi/m3)
ṁ gas consumption rate (kg/m3 s) 0 primary particle
n stoichiometric coefficient a agglomerate
N rate of hydrate shell formation (1/m3 s) CH4 methane
O inter-phase energy transfer term (W/m3) e effective
p pressure (Pa) eq equilibrium
Pr Prandtl number hyd hydrate
q heat transfer coefficient (W/m2 K) H2 O water
Q energy source term (W/m3) i, j phase index
r particle radius (r) k component of water–hydrate phase
Re Reynolds number l, m coordinate index
t time (s) max maximum
T temperature (K) o oil
Δt time step (s) p particle
u local phase velocity (m/s) r relative
u mean flow velocity (m/s) T transposed
W molar mass (kg/mol) t turbulent
We Weber number w –h water–hydrate
xl Cartesian coordinates (m)
y breakage rate parameter
Cμ, C1, C2, σ k, σϵ turbulence model coefficients

Bondarev et al. (1982) modelled crystallization of hydrates from
the flow of wet natural gas to the cold walls of an industrial line
considering the one-dimensional Stefan problem of the process.
The simulations showed peculiar periodic variations of the thick-
ness of the hydrate layer formed, which were explained by the
competing influence of the Joule–Thomson effect in the gas phase
and diminishing heat transfer in the hydrate phase. The authors
indirectly confirmed the simulation results by observations of in-
dustrial systems similar to the one they had simulated. The model
was not, however, able to predict such details of the flow as the
radial profiles of the gas velocity, temperature and turbulence
intensity. Sean et al. (2007) report a three-dimensional numerical
model of the dissociation of a gas hydrate obstruction under the
influence of a laminar flow around it. The reduction of the ob-
struction size was accounted for by moving the boundaries of the
computational mesh. Although this model was validated against
experiments, the overall process geometry and the problem
complexity was rather far removed from industrial conditions. It
should be noted, in general, that the Stefan problem for the
growing hydrate boundary describes a mechanism, which is not
entirely consistent with the plug formation scenario relevant for
the petroleum industry. In the petroleum industry gas hydrate
shells are formed at the surface of droplets dispersed in the oil
phase (Zerpa, 2013). Techniques based on the Stefan problem are,
however, relevant for studies of hydrate plug dissolution.
A set of one-dimensional CFD-models for gas hydrate flow as-
surance was developed in Colorado School of Mines (CSM) (Zerpa
et al., 2012; Zerpa, 2013). The models were able to predict the
kinetics of hydrate phase formation based on “pressure–volume–
temperature” (PVT) data and flow patterns computed by the dy-
namic multiphase flow simulator OLGA (Kinnari et al., 2008). The
technique was incorporated into OLGA as an additional routine
named “The Colorado School of Mines Hydrate Kinetics” (CSMHyK)
model, which is presently considered as one of the most realistic
and complete dynamic hydrate prediction tools, validated with
experimental data on the industrial scale. CSMHyK accounted for
many details of the process-related phenomena, adopting a
rheological approach to describe hydrate slurries reported by
Sinquin et al. (2004), which was further updated in CSM by a set of
in-house empirical relations for the adhesive force between hy-
drate particles and the hydrate primary particle size. These ex-
pressions were utilized for computation of the average size of
hydrate agglomerates and subsequently the apparent viscosity of
the slurry, which was finally returned to OLGA. Thus the entire
model couples the particulate and continuous phases making it
possible to predict the sudden viscosity increase due to hydrate
 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57
formation, potentially resulting in pipeline blockage.
Nevertheless a detailed analysis of the physical assumptions
upon which the CSMHyK model is based—such an analysis is
presented in the methodology section of the present paper—
makes it possible to identify a less-than-adequate account of the
kinetics of the development of the hydrate particle size distribu-
tion, which may result in an inappropriate description of the
system temporal evolution. Finally, the CSMHyK model does not
incorporate expressions for the granular stress induced in the
hydrate phase, which may be causing inappropriate sedimentation
patterns in case the model is up-scaled to three dimensions,
namely packing of the hydrate phase beyond the maximum
packing fraction.
A three-dimensional CFD-model of hydrate deposition from the
bulk of the gas pipeline was developed by Jassim et al. (2008),
where the particles were considered as separate Lagrangian ob-
jects, while the gas flow was treated in a Eulerian reference frame.
The model was integrated into the commercial CFD package FLU-
ENT. Jassim et al. (2008) demonstrated an influence of the hydrate
particle size on the particulate phase velocity profile and the
overall pipeline penetration efficiency (Crowe, 2006), which in-
dicated the importance of the temporal evolution of the particle
size, due to e.g. agglomeration. Even though the model was vali-
dated against experimental data, it was “one-way”-coupled so that
the gas phase was not influenced by the particles.
A set of three-dimensional CFD models for the deposition of
hydrates from turbulent flow was developed in our previous pa-
pers (Balakin et al., 2010a, 2011), as subroutines built into the
commercial CFD package STAR-CD (CD-adapco, 2006). These
models, considering both liquid and solid phases as Eulerian
continua, were based on a standard expression for the apparent
viscosity of suspensions, which was implemented by appropriately
adjusting the “solid phase viscosity” (which is required by the CFD
package due to the nature of the multiphase model used) so that
the viscosity of the suspension followed a standard expression.
The collisional stresses induced in the hydrate phase were con-
sidered, and an expression for the average hydrate agglomerate
size, proposed by Muhle (1996), was incorporated. Although the
models were two-way coupled and validated with the in-house
experimental data (Balakin et al., 2010c), they accounted neither
for the kinetics of the formation of hydrate phase nor for the ki-
netics of hydrate particle agglomeration in an explicit way.
The parallel development reported by Lo (2011), showed a
detailed CFD-model of the hydrate formation in a four-phase pipe
flow of oil–water–gas–hydrate based on the commercial CFD code
STAR-CD (CD-adapco, 2006). The model was built in a three-di-
mensional reference frame, but projected in two dimensions by
symmetry boundary conditions to limit computational costs. The
phases were entirely coupled in terms of the interfacial transfer of
mass, momentum and energy. The model was, however, missing
the rheological and granulometric aspects of the process. An in-
teresting model for hydrate phase formation and evolution in a
multiphase pipe jet has been reported recently by Labois et al.
(2014). This model, incorporated in the commercial package
TransAT, is two-way coupled in terms of the continuity, mo-
mentum and energy. It is also three-dimensional and considers a
rheological expression for the slurry. In addition a limited-slip law
for the adhesive hydrates was applied at the walls of the compu-
tational domain, simulating deposition due to adhesion. The
model was qualitatively validated against benchmarks, but a
quantitative validation was not reported. Agglomeration of hy-
drate particles was not considered.
Even though significant efforts have been made to directly
model agglomeration of cohesive particles by means of CFD using
the distinct element principle (DEM) (Cundall and Strack, 1979),
DEM is, at present, not yet applicable at industrial scales due to the
47
high computational costs. Another paper from CSM (Yang et al.,
2003) should, however, be mentioned as one of the rare efforts at
DEM-modelling of flow laden by gas hydrates. A simpler and also
faster alternative is the population balance approach (PBM)
(Hounslow et al., 1988; Ramkrishna, 2000), which deals with the
agglomerate size distribution (or moments of the size distribution)
instead of a separate numerical treatment of single particles as in
DEM. There is a variety of PBM-models considering the develop-
ment of the hydrate particle size distribution in water-filled stirred
reactors. The models by Englezos et al. (1987) and Malegaonkar
et al. (1997) consider the nucleation and growth of gas hydrates in
such a system, estimating the hydrate–water interfacial area by
means of PBM, allowing determination of the hydrate-growth rate
constant by fitting the simulation results to experimental data. The
model, however, does not account for agglomeration of hydrates
and the growth rate constant was found to be overestimated
(Skovborg, 1993), and the influence of the formation of hydrates
on the entire flow was not considered, i.e. the model was “one-
way” coupled. An improved “one-way” coupled PBM-model of gas
hydrate particle size development in a batch was developed by
Herri et al. (1999) who, in addition to nucleation and growth of
hydrate particles, considered agitation-induced agglomeration and
hydrodynamic fragmentation. The model (Herri et al., 1999), based
on the method of moments, was validated with experimental data,
but again did not account for the influence of the formation of the
solid phase on the mixture viscosity. Although the model in-
corporated an expression that related the growth rate constant to
the averaged flow parameters assuming diffusion of methane in
the water phase, which made it more physically realistic than
previous PBM-models of hydrates, the approach by Herri et al.
(1999) did not account for spatial non-uniformities of the flow
field and hydrate volume fraction profiles in the batch.
A more detailed population balance model was developed by
Balakin et al. (2009), where a turbulent flow-loop laden with hy-
drates was modelled using the method of moments, similar to the
model described by Herri et al. (1999). The model assumed the
system to consist of two sections with different values of shear
rate: (i) a pump section where hydrodynamic fragmentation was
dominant and (ii) a pipe section where both agglomeration and
fragmentation took place in parallel with nucleation and particle
growth. The model was validated against flow loop data from lit-
erature, but, as in the above-mentioned PBM-models, it did not
account for the slurry viscosity change due to the presence of the
solids phase. The work by Fidel-Dufour et al. (2005) considered
evolution of a hydrate–oil suspension in a bench-scale flow loop
where PBM-modelling was developed in order to account for the
change of the apparent viscosity. The resulting modification of the
shear rate was, however, not coupled with the PBM. The model
was validated against in-house experimental data. The deposition
of hydrate particles was not modelled.
The PBM-models mentioned so far (Englezos et al., 1987;
Malegaonkar et al., 1997; Herri et al., 1999; Balakin et al., 2009;
Fidel-Dufour et al., 2005) were based on in-house codes, which did
not account for the three-dimensional nature of the process and
were therefore less accurate as a tool for industrial flow assurance.
The modern trend to model complex reactive multiphase flows
at industrial scale is to merge CFD-models, which provide detailed
insight into the flow pattern, quantifying the local parameters that
govern the interfacial transfer of the mass, momentum and energy
as input to PBM models. Most commonly the PBM-models in-
corporated into CFD are focused on the prediction of the mean
agglomerate size and the total number of solid particles within a
computational cell, which automatically returns the interfacial
area. Even though there is a significant number of combined CFD-
PBM models developed so far (Marchisio et al., 2006; Lo and
Zhang, 2009; Balakin et al., 2014), they are not directly focused on
48 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57
hydrate-related phenomena. The model by Balakin et al. (2010b)
considers a combined CFD-PBM model for hydrate agglomeration
in a multiphase flow loop. The model is based on a single-moment
principle computing the total number of agglomerates within a
computational cell by means of a simplified PBM technique, while
the flow pattern is simultaneously predicted with CFD. The mean
size of the hydrate agglomerates in a computational cell was ob-
tained in the model assuming a unimodal particle size distribution
within the cell, which was considered to be valid for relatively fine
computational grids. Although the model was validated with ex-
perimental data in terms of the predicted particle size, the visc-
osity of the slurry was not modified as a consequence of the
presence of the hydrate phase. Moreover, as in most of the PBM-
models involved in the hydrate studies, the agglomeration and
breakage rate constants were found by fitting the model results to
experimental data and this limits the applicability of the CFD-PBM
approach.
The present paper describes a practical CFD-PBM model for
agglomeration and deposition of hydrate particles in turbulent oil-
dominated flows. At first we are focused on the validation of
CSMHyK (Zerpa et al., 2012) against several independent experi-
mental benchmarks. The model is then improved and adopted for
incorporation into an in-house developed PBM model, where the
agglomeration and breakage rate constants are determined theo-
retically rather than empirically, making the technique potentially
relevant for a wide range of industrial scenarios. The PBM-model is
further coupled with a CFD-model similar to one reported in Lo
(2011), which is built in the commercial package STAR-CCM þ. The
resulting three-dimensional CFD-PBM model is finally equipped
with a rheological expression for the apparent viscosity of a hy-
drate-laden suspension, “two-way” coupled in terms of the inter-
phase phenomena and accounts for the intra-phase granular
interactions.
2. Methodology
2.1. CSMHyK rheological model
The rheological model incorporated in CSMHyK (Zerpa et al.,
2012) is based on an apparent viscosity concept, which assumes
that the viscosity of the hydrate particle–liquid slurry follows the
rheological expression by Snabre and Mills (1993):
1 − ϕe
μr = , particle while the buoyancy force acting on smaller particles is
⎡⎣ 1 − (ϕe /ϕmax ) ⎤⎦2
where μr is the relative viscosity of suspension, that is, the ratio of
the apparent viscosity of the suspension, μa, to the liquid-phase
viscosity, μo: μr = μa /μo , ϕmax ( = 0.57) is the maximum packing
fraction of hydrate particles. The value of 0.57 has been used in
CSMHyK (Zerpa et al., 2012) and is reasonable in practise in spite
of a likely wide particle size distribution in view of the fact that the
hydrate agglomerates will have a relatively low fractal dimension
and therefore a low sphericity (Hoffmann and Finkers, 1995).
Further ϕe is the effective volume fraction of the hydrate ag-
glomerates, where the agglomerates consist of primary particles.
The theory (Snabre and Mills, 1993) assumes that the liquid is
immobilized inside aggregates. This is realized in the model as:
⎛ d ⎞3 − f
ϕe = ϕhyd ⎜ a ⎟ ,
⎝ d0 ⎠
where ϕhyd is the volume fraction of the primary hydrate particles,
d0 is the diameter of primary hydrate particle, da is the diameter of
the agglomerates and f is the fractal dimension of the agglomer-
ates. Following granulometric studies performed for gas hydrates
(Sinquin et al., 2004) the fractal dimension is in the interval 2.2–
2.8, while Zerpa (2013) used the constant value of f¼ 2.5.
According to the “core-shell” model of hydrate formation by
Zerpa et al. (2012), hydrate phase is formed at the surface of water
droplets for water-in-oil emulsions, building a shell that further
grows inward into the droplet to form hydrate particles. Thus the
primary hydrate particle size is given by a mean size of water
drops dispersed in the oil carrier phase. As determined in ex-
periments by Boxall et al. (2012) for turbulent pipe flow, this size is
related to the diameter of the pipe d, and the system Weber
number, We = ρo u 2d/σ , where ρo is the oil phase density, u is the
mean pipe flow velocity and s is the water–oil surface tension by
the following two expressions:
3
d0 = 0.063dWe− 5 (3)
5
for the case where We < 0.0674 (drop size dominated by ed-
Re 4
dies of the inertial subrange of turbulence). For the dispersions
with the drops size correlated with eddies from the viscous sub-
range:
1 3
d0 = 0.016dRe 2 We− 5 (4)
The maximum agglomerate diameter, da, is given in Muhle
(1996) by using a relation for the maximum agglomerate size:
1
⎡ F d 2 − f ⎤4 − f
a 0
da,max = ⎢ ⎥ ,
⎢⎣ μo γ ⎥⎦ (5)
where Fa is the adhesive force between hydrate particles and γ is
the shear rate. The standard strategy (see e.g. Genovese, 2012) is to
use the apparent viscosity, μa (see the description under Eq. (1))
instead of μo in this relation. It is important to note that this
strategy may be questionable: the suspension viscosity describes
the bulk properties, while the real particle–fluid and particle–
particle interactions occur at smaller scales. As a result the visc-
osity used in the model (μa) may be overpredicted on the inter-
particle scale and a question arises whether the carrier phase
viscosity should not be used instead. There exists an analogue to
this argument in terms of the density. An analysis was presented
by Ruzicka (2006) for the buoyancy of dispersions where it was
shown that the buoyant force acting on a large particle immersed
into the liquid–solid dispersion of smaller, primary particles, starts
to become proportional to the density of the suspension when the
large particle is approximately 10 times larger than the primary
(1) proportional to the density of the continuous phase. A similar ef-
fect may be valid for the viscosity, such that the continuous phase
viscosity should be used on the single-particle scale. We return to
this issue later in the paper.
Assuming that the average size of the hydrate agglomerate, da,
is proportional to the maximum size da,max , it is possible to obtain
an implicit relation for daeq (Zerpa et al., 2012):
⎡ 3− f ⎤
2
4− f
( daeq/d0 ) −
(
Fa ⎢⎣ 1 − ϕhyd /ϕmax )( d eq
a /d0 ) ⎥⎦
= 0,
d02 μo γ ⎡⎣ 1 − ϕhyd ( daeq/d0 )
3− f ⎤
⎦ (6)
which constitutes an expression for the equilibrium agglomerate
size, that is, it predicts the steady-state particle size after infinite
time of the agglomeration- and break-up process where agglom-
eration, due to the adhesive force, is balanced by break-up due to
(2)
the disruptive shear force. Eq. (6) reduces to a quartic equation if
f¼ 2.5 and can thus be solved analytically as shown in the
Appendix.
An expression for the adhesive force, Fa, was obtained by Die-
ker et al. (2008) from micromechanical studies of the adhesion
 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57 49

between cyclopentane hydrate particles: ⎛ ϕH O ϕhyd ⎞

ṁ CH4 = 6kCH4 ⎜ 2 − ⎟ ΔT ,
d0 ⎝ d0 da ⎠ (14)
Fa = [0.0017 (7.7 − ΔT ) + 0.0007],
2 (7)
where d0 is given by Eqs. (3) and (4), da is the diameter of hydrate
where ΔT is the system subcooling relative to the hydrate equili- agglomerate which is calculated using the PBM-model described
brium curve. Eq. (7) is not, however, based on real industrial in the following subsection and kCH4 is a growth-rate constant. Eq.
petrochemical systems due to complexity of performing micro- (14) is derived assuming that the methane does not diffuse into
mechanical studies under high-pressure conditions. the droplets covered by hydrate shell, i.e. the consumption of
methane from the oil phase is significantly hindered by the shell.
Further internal growth of the hydrate in the water drops isolated
2.2. Eulerian–Eulerian CFD-model
by shells is enabled by methane dissolved in water.
The momentum equations are given as (Balakin et al., 2010a):
The flow of the hydrate–water–oil suspension is modelled as a
stream of two Eulerian media co-existing in a common simulation →
∂(ϕi ρi ui ) →→ →
+ ∇·(ϕi ρi ui ui ) = − ϕi ∇p + ϕi ρi g + ∇·ϕm ( τ̲ i + τ̲ it )
domain (Balakin et al., 2010a). According to the proposed defini- ∂t
tion of flow, the phases are therefore described by two separate → →
+ Mi + δi, w − h Fint, w − h, (15)
sets of Navier–Stokes equations as discussed in e.g. Balakin et al.
(2011) and Lo (2011). Following the scenario of the hydrate for- →
where p is pressure, g is acceleration due to gravity and τ and τ t
mation in oil-dominated flows proposed in Zerpa et al. (2012), the are the molecular and turbulent stress tensors, δ i, w − h is the Kro-
suspension is presented as a composition of two phases: the necker operator The absolute value of the inter-phase momentum
continuous oil phase with the methane dissolved is indexed as i¼ o transfer terms is equal and oppositely directed for the phases, such
→ →
and the dispersed phase, consisting of the water droplets, which that Mo = − Mw – h . In the present model they are given as:
are covered by hydrate shells, indexed as i = w − h. The continuity
→ 3 ⎛ ϕ ϕhyd ⎞ → → ⎛ → →⎞
equations are then written as follows: Mi = πρo ⎜ CD, H2 O H2O + CD, hyd ⎟ | ui − uj | ⎜ ui − uj ⎟ ,
4 ⎝ d0 da ⎠ ⎝ ⎠ (16)
∂ϕi ρi ⎛ →⎞
+ ∇·⎜ ϕi ρi ui ⎟ = 0,
∂t ⎝ ⎠ (8) where j is the index denoting the “other” phase of the system. The
drag coefficient is given according to Schiller–Naumann, see e.g.
→
where ui is the velocity of phase i and t is the time. The sum of the Crowe et al. (1998) accounting for the dual structure of the water–
volume fractions is equal to unity, ϕo + ϕw – h = 1. hydrate composite:
The formation of the methane hydrate phase in the droplets is
described by the chemical reaction involving the dissolved me- 24 ⎛ ⎞
CD, k = ⎜ 1 + 0.15 Rep0.687
,k ⎟
thane and the water: Rep, k ⎝ ⎠ (17)
nCH4 CH4 + n H2 O H2 O → nhyd HYD, (9) for Rep, k < 1000 and CD, k = 0.44 for Rep, k > 1000. The particle
Reynolds number for the component k (water droplet or hydrate
where nCH4 ( = 1), n H2 O ( = 5.75), and nhyd ( = 1) are the stoichio-
particle or agglomerate) is:
metric coefficients. Then the continuity equations for the hydrate
→ →
and water fractions are given as: ρo dk | uw – h − uo|
Rep, k = ,
∂ϕw – h ρw – h fhyd Whyd nhyd
μo (18)
→
∂t
( )
+ ∇· ϕw – h ρw – h ui fhyd = ṁ CH4
WCH4 nCH4
,
(10) where d H2 O = d0 , dhyd = da and μo ( = 1.3 mPa s is the dynamic
viscosity of the oil phase. The “solid pressure” that accounts for
interparticle interactions in the hydrate phase is applied to the
∂ϕw – h ρw – h fH2 O → ⎛ Whyd nhyd ⎞ momentum equation of the dispersed phase (Balakin et al., 2011):
∂t
( )
+ ∇· ϕw – h ρw – h ui fH2 O = ṁ CH4 ⎜⎜ 1 −
⎝
⎟⎟,
WCH4 nCH4 ⎠ (11) → ⎡ ⎤
Fint, w − h = − exp ⎢⎣ −600 (ϕmax − ϕw – h ) ⎦⎥ ∇ϕw – h.
(19)
where fhyd + fH2 O = 1 are the mass fractions of hydrate and water
in the water–hydrate composite, Whyd ( = 119.65 kg/mol) is the The viscous stress tensor τ for each phase is calculated as:
molar mass of methane hydrate, WCH4 ( = 16.04 kg/mol) is the 2 ∂u l
molar mass of methane and the terms involving ṁ CH4 are source τil, m = 2μa Dil, m − μa δl, m il ,
3 ∂x (20)
terms due to the formation of hydrate.
⎛ ∂u l ∂u m ⎞
The density of the hydrate–water composite is given as: where Dil, m = 21 ⎜ ∂xmi + il ⎟ is the rate of strain tensor and the su-
⎝ ∂x ⎠
ρw – h = ρ H2 O ϕ H2 O + ρhyd ϕhyd , (12) perscripts denote the contravariant vector and tensor components.
The apparent viscosity of the suspension, μa, is given by Eqs.
where ϕ H2 O and ϕhyd are volume fractions of water and hydrate,
(1) and (2).
and ρ H2 O and ρhyd ( = 807.77 kg/m3) are their densities. The turbulent stress τ̲ it is calculated by k –ϵ turbulence sub-
Based on a volume-balance within a computational cell
models applied to each phase separately (Balakin et al., 2011):
ϕw – h = ϕ H2 O + ϕhyd = const it is possible to derive an expression for
the volume fraction of hydrate: ⎛ 2 ∂u l ⎞ 2
τil, m; t = μit ⎜ 2Dil, m − δl, m il ⎟ − ρi ki δl, m.
⎝ 3 ∂x ⎠ 3 (21)
ρ H2 O ϕw2 – h fhyd
ϕhyd =
(
ρhyd + fhyd ϕw – h ρ H2 O − ρhyd ) (13) The turbulent viscosity is computed as:

where ṁ CH4 is the rate of methane consumption into the hydrate ki2
μit = Cμ ρi ,
phase and is computed following Boxall et al. (2012): ϵi (22)
50 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57

where Cμ = 0.09.
The transport equations for the turbulent kinetic energy, ki, and
the rate of turbulent energy dissipation, ϵi are:

∂(ϕi ρi ki ) → ⎛ ϕ (μ + μ t )∇ki ⎞
+ ∇·(ϕi ρi ui ki ) = ∇·⎜⎜ i a i
⎟⎟ + ϕi (G − ρi ϵi )
∂t ⎝ σk ⎠ (23)

∂(ϕi ρi ϵ i ) → ⎛ ϕ (μ + μ t )∇ϵ i ⎞ ϵ
+ ∇·(ϕi ρi u i ϵ i ) = ∇·⎜⎜ i a i ⎟⎟ + ϕi i (C1G − C2 ρi ϵ i ),
∂t ⎝ σϵ ⎠ ki (24)
→ → →
where G = μa (∇ ui + (∇ ui )T ) : ∇ ui ,
C1 = 1.44 , C2 = 1.92, σ k = 1.0,
σϵ = κ 2/[(C1 − C2 ) Cμ ] with κ = 0.4187.

The energy equations are given separately for each phase in a

fashion a similar to Lo (2011):

∂ϕi ρi hi ⎡ ⎛ μt ⎞⎤
→
∂t
( )
+ ∇· ϕi ρi ui hi − ∇·⎢ ϕi ⎜⎜ λ i ∇Ti + i ∇hi ⎟⎟ ⎥ = Oi + Q hyd
⎣⎢ ⎝ σt ⎠ ⎥⎦ (25)

where h is the specific enthalpy, λ is the thermal conductivity,

st( ¼0.9) is the turbulent Prandtl number. The first term in the Fig. 1. Computational grid and boundary conditions. Gravity is directed down-
right-hand-side of Eq. (25) describes interphase energy transfer: wards. The view is 3-D with perspective, with the inlet closest to the viewer, such
that the inlet and the outlet are equal in size. The domain shown is a flat, rec-
⎛ ϕH O ϕhyd ⎞ tangular section with the narrow top and bottom walls no-slip, solid walls and the
Oi = 6 ⎜ 2 qo, H2 O + qo, hyd ⎟ ( Ti − Tj ), broad front and back faces are planes of symmetry.
⎝ d0 da ⎠ (26)

where qo, k is the inter-phase heat transfer coefficient given by moreover, periodic with respect to the outlet such that the phase
Ranz–Marshall correlation: velocity, volume fraction, component mass fractions, k, ϵ and the
scalar M0 from Eq. (29) computed at the model outlet were all
λo ⎛ ⎞
qo, k = ⎜ 2 + 0.6 Rep0.5 0.3
, k Pro ⎟, transposed to the inlet by in-house modifications to the program.
dk ⎝ ⎠ (27) An infinite pipe was, in this way, simulated with this inlet
where Pro = cp, o μo /λo is the oil phase Prandtl number. boundary condition.
The second term in the right-hand-side of Eq. (25) represents The initial conditions, equivalent to the experimental condi-
the amount of heat (per unit volume) due to formation of hydrates tions without hydrates (fhyd = 0) of Fidel-Dufour et al. (2005) for
since the reaction (Eq. (9)) is exothermic (Sloan and Koh, 2008): the water–dodecane flow saturated with methane, are presented
in Table 1.
Whyd nhyd
Q hyd = mCḢ 4 ξhyd,
WCH4 nCH4 (28)
2.3. PBM-model
where ξhyd is the latent heat of hydrate formation.
The numerical solution of Eqs. (8)–(25) (in line with Eq. (29) The PBM model accounts for the evolution of hydrate particle
discussed in the following subsection) was performed with the use size distribution via the 0th moment of the number particle size
of implicit SIMPLE technique (CD-adapco, 2006) incorporated into distribution, M0, which is the total number of particles per unit
the commercial CFD-package STAR-CCMþ with following relaxa- volume of suspension (Hounslow et al., 1988):
tion coefficients: 0.1 for pressure, 0.3 for velocity, 0.5 for phase
dM0 16
volume fraction, 0.8 for k –ϵ submodel, 0.9 for the component mass =B− αγM3 M0 + N ,
dt 3 (29)
fraction, phase energy and Eq. (29). The equations were discretized
temporally with the second-order Euler technique marching by where B is the rate of hydrodynamic fragmentation of hydrate
10 ms. The spatial discretization is done with the upwind scheme agglomerates. The second term on the right-hand-side is the rate
on the computational grid shown in Fig. 1. The mesh consisted of of agglomerate production dependent on the third moment of
2800 one-mm rectangular cells with 25% near-wall refinement. particle size distribution given by M3 = 3ϕhyd /(4π ), which reflects
The model is scaled to 2D in the following way: a rectangular the particle volume fraction, and α is the collision “efficiency”, that
computational domain, 1 m long and 10.2 cm high is defined, and is, the probability that a given collision leads to agglomeration. N is
symmetry boundary conditions are specified at two opposite sides the rate of conversion of water droplets to primary hydrate
of the domain, as for flow between two parallel plates. This mimics
a flow system reported in Fidel-Dufour et al. (2005) with a rea- Table 1
sonable computational effort. The boundary conditions at the top Initial conditions.
and bottom walls are no-slip with constant temperature
p, MPa ui , m/s ϕw–h Ti, °C
To = Tw – h ( = 3 °C), and those at the in- and outflow boundaries are
7.00 0.50 0.13 4.00
periodical in the sense explained further. A pressure boundary
condition was used at the model outlet with a prescribed tem- ki, J/kg ϵi , m2/s3 ρw , kg m−3 ρo , kg m−3
perature To = Tw – h ( = 3 °C), while the rest of the model scalars were 2.55·10−3 4.94·10−2 1000.00 768.00

assigned zero gradients there. The inlet boundary condition was μ w , mPa s λw , W/m K λo, W/m K λhyd , W/m K
set at a constant temperature of 4 °C, which is slightly higher than 1.00 0.62 0.14 0.34
that at the walls and the outlet in order to produce physically
cp, w, kJ/kg K cp, o, kJ/kg K cp, hyd, kJ/kg K ξhyd, J/g
realistic temperature gradients, which were supposed to drive the 4.20 2.20 2.10 446.00
formation of hydrates. The inlet boundary condition was,
 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57 51

particles and γ = ϵo ρo /μo is the shear rate. 20

18 Cameirao et al., 2011 CSMHyK
The rate of fragmentation is given by:
16
^
B = B γ yM3, (30) 14
12
where y ( = 2.89) is a parameter used in the models of Balakin 10

d0
^ 8
et al. (2010b), Marchisio et al. (2006) and Soos et al. (2007). B is
6
the breakage rate constant, which is unknown (see below). The
4
rate of primary particle/droplet production can be estimated as-
2
suming that the water droplet converts to hydrate rapidly and the 1 2 3 4 5 6
volume of liquid water core in the primary hydrate particle is in- run number
significant compared to the volume of hydrate within the particle:
Fig. 2. Mean water droplet size predicted by CSMHyK and compared with ex-
mCḢ 4 Whyd nhyd 6 perimental data for runs 1–6 from Cameirao et al. (2011). γ ¼ 764–1376 s  1, ϕhyd
N= . ¼3.9–15.9%, μo ¼ 1.8–3.3 mPa s, ΔT = 5.0–5.8 °C , u = 0.96–1.72 m/s and
ρhyd WCH4 nCH4 πd03 (31) ρo = 824–902 kg/m3.

The collision efficiency is given by Potanin (1991) for cluster–

water–hydrate slurry flow an Archimedes flow loop at water-cuts
cluster agglomeration when da/d0 > 3:
in the range 10–50%. The suspension viscosities were determined
1.7 ⎡ d /d −0.15 − 0.2⎤3/2 , from experimental measurements of the pipe pressure drop. The
α=
⎡⎣ ln ( da/d0 ) ⎤⎦1/4 ⎣
( a 0) ⎦
chord-length distribution of hydrate particles, obtained empiri-
(32)
cally by the focused beam reflectance method (FBRM), makes it
1/3
where da = 2 ( M3/M0 ) . possible to calculate the mean particle size (Cameirao et al., 2011).
Eq. (32) can, for simplicity, be replaced by a power-law fit: The volume fraction of hydrate was estimated via the gas
−0.421
α = 1.3276 ( da/d0 ) (R2 = 0.9995). The collision efficiency for consumption.
da/d0 < 3 is given by a linear interpolation between the points FBRM measurements of the mean droplet size are shown in
α ( da/d0 = 1) = 1.0 and α ( da/d0 = 3) computed from Eq. (32) re- Fig. 2. They are compared with the model studied in our paper,
sulting in an expression α = 1.0668 − 0.0668 ( da/d0 ). namely Eq. (3). The model and the experiment differ by 41%. Fig. 3
^ shows FBRM measurements of the mean agglomerate size for
The unknown B is found assuming the existence of an equili- experimental runs 1–6 from Cameirao et al. (2011) and compared
brium condition when all possible hydrate former is consumed, so with CSMHyK predictions, that is, Eq. (6). The underestimation of
that the production of new primary particles has stopped and the
experimental particle sizes by the model (discrepancy of 52%)
agglomerates have become so large that they no longer grow. Thus
could be attributed to the factors discussed above, namely the use
the total number of particles does not change (dM0/dt )eq = 0:
of μa instead of μo in Eq. (6), see Zerpa et al. (2012), alternatively it
^ 16 eq eq eq can be due to their adoption of the effective volume fraction
B γ yM3eq = α γM3 M0 ,
3 (33) concept. It is important to note that Eq. (6) was derived based on
an expression for the maximum size of agglomerate and this can
where M3eq
is taken from experiment in the present study, while
the water cut can be taken during engineering estimates. The term also lead to differences between the model and the experiments.
The viscosity measured experimentally by Cameirao et al.
“water cut” refers to the volume fraction or volume percentage of
water in the 3-phase mixture M3eq = 4ϕw – h /3π |t = 0 . (2011) is given together with the model-predicted viscosities, i.e.
M0eq and αeq are calculated assuming the equilibrium size of Eqs. (1) and (2), in Fig. 4. The discrepancies do not exceed 38%
agglomerate daeq is given by Eq. (6): M0eq = 8M3eq/(daeq )3 and which, together with the discrepancies in the determination of the
−0.421 mean particle size, can be thought of as tolerable for industrial
α eq = 1.3276 ( daeq/d0 ) . Finally the expression for the breakage
flow-assurance studies. According to the proposed scenario of gas
rate constant becomes:
hydrate formation of Zerpa et al. (2012), which is also adopted
eq
^ 16 eq ( 1 − y) 8M3 here, the rheological expression given in Eq. (1) should, however,
B= α γ .
3 daeq3 (34) include the volume fraction of water drops covered by hydrates
rather than the volume fraction of the hydrate phase itself. Since
Eq. (29) was solved numerically starting from a value of M3 model predictions based on this assumption agree reasonably well
extrapolated from experimental data to the first time-step of the with experiment, this indicates that the solidified drops were
8M3
simulation using M0 = .
d03
t=0 t =Δt 30 Cameirao et al., 2011
CSMHyK
25
3. Results and discussion
20
3.1. Validation of CSMHyK rheology
d/d0

15

The objective of this section is the validation of the CSMHyK 10

rheological model (discussed in Section 2.1) by comparison with
experimental data from literature. Since the present section does 5
not focus on the kinetics of hydrate formation, the rheology of the 0
flow is based on the equilibrium volume fraction of the hydrate 1 2 3 4 5 6
phase determined experimentally, while the agglomerate dia- run number
meter and the suspension viscosity is computed by the CSMHyK Fig. 3. Mean particle size predicted by CSMHyK compared with experimental data
model. for runs 1–6 from Cameirao et al. (2011). γ = 764–1376 s−1, ϕhyd = 3.9–15.9% ,
The first dataset considered is from Cameirao et al. (2011) for μo = 1.8–3.3 mPa s, ΔT = 5.0–5.8 °C , u = 0.96–1.72 m/s and ρo = 824–902 kg/m3.
52 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57
10
9 7
8 CSMHyK corrected
viscosity [mPa·s]
7
6
5
4 3
3 Cameirao et al., 2011
2 CSMHyK
1 0
1 2 3 4 5 6
run number
Fig. 4. Viscosity of suspension predicted by CSMHyK and compared with experimental
data for runs 1–6 from Cameirao et al. (2011). γ¼ 764–1376 s  1, ϕhyd ¼ 3.9–15.9%, μo
¼ 1.8–3.3 mPa s, ΔT = 5.0–5.8 °C , u = 0.96–1.72 m/s and ρo = 824–902 kg/m3.
likely fully converted into hydrate so their volume fraction was
comparable to that of water drops.
The next validation is performed by comparison with experi-
mental data from Camargo (2001), where the agglomeration of
hydrate particles was studied in a batch and a flow loop simulta-
neously. Fig. 5 shows how the experimental results compare to
model predictions as a function of shear rate, controlled by varying
the system velocity. The calculated viscosity is significantly un-
derestimated for the low shear rates, which is explained by an
overestimation of the agglomerate size, which, in turn, leads to
ϕe > ϕmax , taking the suspension viscosity model to the non-phy-
sical side of a singular point. This is due to the fact that the pri-
mary droplet size predicted by CSMHyK for the two lowest ex-
perimental shear rates (viscous regime) are in the range 20–
40 μm, significantly exceeding experimental sizes reported by
Camargo (2001). This, in turn, leads to an unrealistic increase in Fa.
It is therefore suggested here to use a drop size of 4.9 μm instead
of the model-predicted values. This size comes from averaging the
sizes predicted by the model for the inertial subrange (last 3 ex-
perimental points) and leads to correct qualitative behaviour of
the model, the discrepancy then reduces to 18%.
The last validation is performed using the experimental dataset
from the flow-loop experiments by Fidel-Dufour et al. (2005) re-
porting rheological studies of the transient crystallization of hy-
drates, performed in a pumpless pipe system filled with a gas–
water–dodecane suspension. The apparent viscosity of gas hydrate
suspension was determined via the pressure measurements dur-
ing hydrate formation at different water cuts. The experimental
apparent viscosity at the moment of maximum gas conversion is
given in Fig. 6 as a function of the volume fraction of hydrate,
demonstrating an underestimation as compared to the experi-
ments for high hydrate amounts in the system. This could be ad-
dressed to the fact, confirmed in the original experimental work
(Fidel-Dufour et al., 2005), that agglomeration of hydrates was
Fig. 5. Relative viscosity of suspension as a function of shear rate predicted by CSMHyK,
using the original model and after correcting the primary particle size to 4.9 μm. Model
results are compared with experimental data from Camargo (2001). γ¼ 63–400 s  1, ϕ
hyd ¼ 23.8%, μo = 60 mPa s, ΔT = 2.5 °C , u = 0.4–2.5 m/s and ρo = 760 kg/m .
3 kCH4 = 4·10−8 kg (m2 s K)−1.
8
Fidel-Dufuor et al., 2005
CSMHyK
6
5
4
2
1
3 4 5 6 7 8 9 10
Fig. 6. Relative viscosity of suspension as a function of hydrate volume fraction
predicted by CSMHyK, using the original model and after correcting adhesive force
with Eq. (35). Model results are compared with experimental data from Fidel-
Dufour et al. (2005). γ = 392 s−1, ϕ = 3.5–9.6%, μo = 1.3 mPa s, ΔT = 6.0 °C ,
σ = 76 mN/m , u = 0.5 m/s and ρo ¼768–790 kg/m3.
driven by capillary bridges between water-shell covered particles.
The applicability of Eq. (7) is thus questionable.
CSMHyK was then corrected in the present study by modifying
the expression for the adhesive force into one accounting for ca-
pillary effects, following Rabinovich et al. (1998):
π
Fa = σd0 cos (θ ),
6 (35)
where θ is the wetting angle. Eq. (35) assumes that volume of the
bridge is given by the volume of a water droplet, while the mini-
mum inter-particle separation is equal to d0, since the model takes
into account the “roughness” of agglomerates, which are formed by
the primary particles. The implementation of this expression into
Eq. (6) leads to better results than those of the standard CSMHyK.
This is shown in Fig. 6 where the discrepancy is significantly re-
duced for high water cuts/hydrate volume fractions.
The transient experimental viscosity is shown in Fig. 7 for 13%
water cut. The experimental log on the volume fraction of hydrate
ϕ was used as input into CSMHyK. It can be seen from the figure
that the initial stage of the process is associated with a significant
increase in the apparent viscosity, which is explained by the
combined action of agglomeration and the hydrate phase forma-
tion. At longer times the viscosity goes to a quasi-steady value due
to the hydrate-former having been consumed. A significant un-
derestimation (38%) of the experimental results is demonstrated
by CSMHyK, both in terms of the process kinetics and the absolute
value of the final relative viscosity.
The temporal behaviour of the agglomerate size predicted by
CSMHyK is shown in Fig. 8. Since CSMHyK predicts the equilibrium
agglomerate size at any given time, it returns a high value initially
and a reduction in the size as the volume fraction of hydrate
Fig. 7. Relative viscosity of suspension as a function of time predicted by CSMHyK
(using Eq. (7) for adhesive force) and the present population balance model (ad-
hesive force given by Eqs. (7) and (35)) for 13% water cut. Model results are com-
pared with experimental data from Fidel-Dufour et al. (2005). γ = 392 s−1, ϕhyd ¼ 0–
6.8%, μo = 1.3 mPa s, ΔT = 6.0°C , σ = 76 mN/m , u = 0.5 m/s and ρo = 768 kg/m3 ,
 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57 53

agglomerated particle is given in Fig. 8 for two different models of

the adhesive force. It can be seen from the figure that mean par-
ticle size gradually increases using the PBM models, which is
physically realistic. This occurs in contrast to the CSMHyK pre-
diction, in which the particle size is always the equilibrium size,
and a transient development of the particle size is not accounted
for as it is in PBM models. In CSMHyK the particle size decreases as
the apparent viscosity increases with time due to the continuous
formation of hydrate.

3.3. CFD-PBM simulations for the pipe flow

Fig. 8. Mean particle size as a function of time predicted by CSMHyK (using Eq. (7))
and compared with the present population balance model (using Eqs. (7) and (35)) The CFD-submodel used in the present work was validated
for 13% water cut. γ¼ 392 s  1, ϕhyd ¼ 0–6.8%, μo = 1.3 mPa s, ΔT = 6.0 °C ,
σ = 76 mN/m , u = 0.5 m/s and ρo = 768 kg/m3 , kCH4 = 4·10−8 kg (m2 s K)−1.
against experimental data on turbulent pipe flows of water–hy-
drate slurry (Balakin et al., 2010a,c, 2011). The results of the
increases. This is not physically realistic, the agglomerate size combined CFD-PBM simulation for the recirculated system, which
should equal the primary particle size initially before agglomera- is described in the methodology section, are presented below.
tion has had time to take place, and then increase. The CFD/PBM The initial conditions for the simulation of the recirculation of
model is able to reflect that, and the particle size therefore starts at phases within the computational domain are uniform velocity and
the primary particle size. Using the expression of Dieker, (7) for volume fraction throughout the system. However, within 100 s
the cohesive force, the PBM model converges to the same size as after beginning the simulation the water phase segregates towards
the CSMHyK model for larger times. the bottom of the domain under the influence of gravity, as shown
in Fig. 10. The steady velocity profile for both phases did not
change significantly due to the settling of the water phase, because
3.2. PBM-simulations in a unit cell
both phases are assigned comparable viscosities. The temperature
profile was slightly asymmetric with an extremum towards the
At first the CFD-PBM model defined in the present paper was
bottom of the pipe. The concentration of water is very high in the
tested to validate the population balance sub-model. In order to do
bottom of the pipe, leading probably to a phase-inversion there,
this, a lumped model consistent of a unit computational cell where
such that the assumption of an oil-rich system breaks down in that
the constant shear rate γ taken from the experiments by Fidel-
region. CSMHyK does not have this problem, since it does not
Dufour et al. (2005) was prescribed. The gravity term was not in-
model the segregation. We note, however, that, according to the
cluded into the momentum equations for both phases so that the
study reported in Herri et al. (1999), the average size of hydrate
phases did not separate during simulations. The mass transfer
particles formed during the homogeneous nucleation and growth
coefficient kCH4 = 4·10−8 kg (m2 s K)−1 was selected by fitting the
in a water-continuous system at a shear stress comparable to ours
history of water conversion predicted by the model to experi-
is very similar to the primary size used in our model. The quan-
mental results as it is shown in Fig. 9. As it follows from the figure,
titative error in primary size due to this phase-inversion in part of
the simulation results slightly overestimate the experiment at the
the computational domain is therefore very small.
initial stages of hydrate production, which could be explained by
According to the experiments by Fidel-Dufour et al. (2005), the
not optimal selection of the system temperature, since, according
time taken for the hydrate phase formation was significantly
to experimental log, it was in the interval 2.5–4.5 °C. The average
longer than the time taken for the water to segregate. As is shown
discrepancy with experiment was, however, less than 14%.
in Fig. 11, where the contours of the hydrate phase in the axial
The results of the PBM-modelling are shown in terms of the
cross-section of the pipe are shown for 300, 600 and 900 s of
temporal history of relative viscosity in Fig. 7, discussed partly
process time, the production of hydrate phase is concentrated in
above, for two different expressions of the adhesive force, that is,
the same region as the layer of water that has formed. The max-
Eqs. (7) and (35). The best fit of experimental data is observed
imum concentration of the hydrate phase is initially located close
when the capillary force is used instead of that given in Eq. (7)
to the cold wall, while it moves upwards as the solidification
resulting in a reduction of discrepancy down to 9%. Eq. (35) was
process progresses.
therefore selected for CFD-PBM simulations of the process.
Fig. 12 illustrates the spatial distribution of relative agglomerate
The temporal evolution of the PBM-predicted mean diameter of
size (scaled with the primary droplet size), da/d0 , along a vertical
line cutting the pipe axis, given as a function of the distance from
the bottom of the pipe (y co-ordinate of Cartesian system) at 300,
600 and 900 s of process time. It can be seen in the figure that the

Fig. 9. Conversion of the water phase as a function of time predicted by

CFD-PBM model and compared with experiment for 13% water cut. γ ¼ 392 s  1,
ϕhyd ¼ 0–6.8%, μo = 1.3 mPa s, σ = 76 mN/m , u = 0.5 m/s and ρo = 768 kg/m3 , Fig. 10. Contours of water volume fraction in the vertical, axial cross-section of the
kCH4 = 4·10−8 kg (m2 s K)−1. pipe, with a settled layer at the steady-state case.
54 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57
t=300 s
t=600 s
t=900 s
Fig. 11. Contours of hydrate volume fraction in the vertical, axial cross-section of
the pipe at 300, 600 and 900 s of the process.
maximum of the particle size is at the bottom of the pipe because
of the local compaction of the dispersed phase there due to seg-
regation under the influence of gravity. The formation of ag-
glomerates was, however, hindered in the vicinity of the wall so
that the agglomerate size was slightly less there (especially during
the latter stages of agglomeration). This can be explained by the
most intensive fragmentation of particles taking place in the
neighbourhood of the wall.
The profile of the aggregate size according to the granulometric
sub-model of CSMHyK (Eq. (6)) is also shown in the figure for
comparison. The CFD-PBM model calls this parameter only for the
^
calculation of the breakage rate constant B (Eq. (34)). The max-
imum particle size for the CSMHyK model is located at the centre
of the pipeline where the shear rate, γ, is minimal while it is
smaller in the near-wall region where the turbulent energy dis-
sipation ϵo is at a maximum. Asymmetry in the particle size profile
is observed in addition since enhanced fragmentation is taking
place within the deposited layer which could happen due to the
collisional mechanistic fragmentation simulated via the hydro-
dynamic breakage and defined by the apparent viscosity. The size
of the agglomerates predicted by CSMHyK is significantly larger
than the one computed by the present CFD-PBM model since, as it
was discussed above, it models the equilibrium particle size rather
than the instantaneous value. Since CSMHyK does not account for
convective transfer of the particle size, as the CFD-PBM model
does (Eq. (29)), it is not able to predict gravity-induced collection
of agglomerates at the pipe bottom. The temporal variation of the
particle size is different for both models. The CSMHyK-predicted
diameter reduces with time due to the increase of the suspension
viscosity, which is caused by the production of hydrate particles.
The diameter profile tends to be flattened with time.
The particle size predicted by the CFD-PBM model increases
from 300 to 600 s of process time due to production of hydrate
phase. It is then slightly reduced at 900 s because of the increase in
Fig. 12. Mean relative agglomerate size da scaled with the primary droplet size d0
as a function of distance from the bottom of pipe. The CFD-PBM model results are
compared with CSMHyK prediction. Model prediction for 300, 600 and 900 s of
process time. Eq. (35) for the capillary attractive force is used both in CSMHyK and
the CFD-PBM model.
the breakage rate, which, as mentioned, is dependent on CSMHyK
prediction of the particle size. Physically this phenomenon could
be explained by the mechanical fragmentation of agglomerates
followed by the increase of the hydrate phase volume fraction. The
CSMHyK-predicted particle size reduces in time because of the
increase in the flow viscosity.
The process of hydrate particle agglomeration is further illu-
strated in Fig. 13, where the contours of the collision efficiency are
shown for a cross-section normal to the axis of pipe at 300, 600
and 900 s of process time. As seen in the figure, the lowest ag-
glomeration probability is observed at the bottom of the pipe, i.e.
in the zone where the largest agglomerates are accumulated so
that the inter-particle collisions are more dominated by inertia
there and the collisions leading to bounce-off are more frequent.
The maximum collision efficiency is located above the segregated
bed, where the agglomerate size is smallest. The zone with re-
duced collision efficiency expands towards the centre of the pipe
with time as agglomerate sedimentation proceeds.
The profile of the relative suspension viscosity, μr, along a
vertical line cutting the pipe axis is depicted in Fig. 14 for 300, 600
and 900 s of process time. The CFD-PBM model prediction is
compared with the relative viscosity computed by CSMHyK, where
Eq. (35) was used for the adhesive force (due to the better corre-
spondence with the experiments shown previously in this paper).
The CSMHyK-predicted viscosity values are not used by the CFD
submodel but are provided to compare both techniques.
It follows from the figure that the maximum of the viscosity
 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57 55

t=300 s 3.5
t=300 s
3

relative viscosity
PBM
2.5 CSMHyK (capillary force)

1.5

t=600 s 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 9 9.5
y (mm)

11
10
t=600 s
9

relative viscosity
8 PBM
7 CSMHyK (capillary force)
6
5
4
3
t=900 s 2

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 9 9.5
y (mm)
16
14
t=900 s

relative viscosity
12 CSMHyK (capillary force)
10 PBM

8
6

Fig. 13. Contours of the collision efficiency α in the axial cross-section of the pipe at 4
300, 600 and 900 s of process time. 2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 9 9.5
predicted by both models is again close to the bottom of the pipe. y (mm)
This is obviously due to the fact that the amount of hydrates lo- Fig. 14. Relative viscosity as a function of distance from the bottom of pipe. The
cated there is much larger than in bulk of the flow. Both models CFD-PBM model results are compared with CSMHyK prediction at 300, 600 and
predict an increase in the suspension viscosity with time due to 900 s of the process. Eq. (35) for the capillary attractive force is used both in
continuous production of the hydrate phase. At the early stages CSMHyK and the CFD-PBM model.

the suspension viscosity predictions of CSMHyK are higher than

the PBM-predictions. This is explained by larger values of particle
size returned by CSMHyK, as discussed above. The situation
changes at the later stages, when the CFD-PBM predicted particle
size leads to domination over the CSMHyK estimate of viscosity.
The layer of deposited particles is evidently responsible for dis-
sipation of the momentum resulting in the local velocity reduction
of both phases. This is illustrated in Fig. 15, where the contour plot
of the oil phase velocity magnitude is shown. A reduction of the oil
speed is observed in the settled layer, while the maximum is
progressively shifted towards the top wall. The profile becomes
highly asymmetric.
4. Conclusions
The present paper describes a combined CFD-PBM numerical
model, which is capable of predicting the formation, agglomera-
tion and deposition of gas hydrates in petroleum lines and related
equipment, whatever the geometry of the apparatus. The model is
based on the combination of Eulerian–Eulerian CFD technique
with provisions to treat the flow of three phases (oil–water–hy-
drate) with the population balance method (PBM), which accounts
for agglomeration of cohesive hydrate particles. The PBM-part of
the model is based on the CSMHyK, a validation of which against
several independent experimental cases was performed in the
paper. CSMHyK-predicted sizes of the primary water droplets and
the equilibrium size of hydrate agglomerates are used as an input
into the agglomeration and breakage kernels of the PBM-model.
The CFD-submodel is coupled with the PBM-part via a rheological
expression, which accounts for agglomeration using the concept of
effective volume fraction, that is, assuming immobilization of the
carrier fluid in agglomerates. The resulting model is capable of
reproduction for the kinetics of the hydrate phase agglomeration
in more realistic way than CSMHyK. It has to be noted, however,
that the present study did not deal with the CSMHyK sub-model
developed for the OLGA software but with equations documented
in the research literature.
The CFD-PBM model was validated with the integral flow
parameters obtained during flow loop experiments on hydrate
phase formation in straight pipes. A detailed CFD-study of these
experiments resulted in the spatial and temporal description of
the process, which was found to be influenced by the deposition of
hydrate phase. The bed of gas hydrates formed in the horizontal
pipes under the mutual influence of agglomeration and gravity is
characterized with the increase of the flow viscosity which, in
turn, is responsible for the plugging of the flow there. It is evident
that progressive sedimentation accompanied with the continuous
formation of the hydrate phase leads to blockage of the pipe at a
certain point in time.
We note that in this paper we have concentrated on aspects of
the CSMHyK model where there is scope for improvement. In all
other respects our calculations follow the principles of the
CSMHyK model, and this work should therefore be seen as a
contribution to the CSMHyK model.
It is worth mentioning that the sensitivity of the present model
is to be studied in a separate article for an industrial system whilst
the present paper was focused on the general concept of hydrate-
laden flow modelling. Generally, it follows from the simulation
56 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57

Appendix
t=300 s
Eq. (6) could be simplified to:
2
A ( 1 − BZ )
Z3 − = 0,
1 − CZ (36)

where Z = x 0.5, A = Fa/d02 μo γ ,

B = ϕhyd /ϕmax and C = ϕhyd . It is further
transformed to the canonical form:
aZ 4 + bZ 3 + cZ 2 + dZ + e = 0 (37)
t=600 s
with the coefficients a = − C , b¼ 1, c = − AB2, d = 2AB , e = − A
with discriminants:
Δ0 = c2 − 3bd + 12ae (38)

and
Δ1 = 2c3 − 9bcd + 27b2e + 27ad2 − 72ace . (39)

The following coefficients are further found:

t=900 s
3
Δ1 + Δ12 − 4Δ03
Q= ,
2 (40)

8ac − 3b2
p= ,
8a2 (41)

Fig. 15. Contours of the oil phase velocity magnitude in the axial cross-section of 1 2 1 ⎛ Δ ⎞
S= − p+ ⎜Q + 0⎟
the pipe at 300, 600 and 900 s of the process. 2 3 3a ⎝ Q⎠ (42)

results that the probability of hydrate plug formation could be and

reduced increasing the agitation, which would in parallel reduce b3 − 4abc + 8a2d
the capability of hydrates, once they are formed, to deposit and q= .
8a3 (43)
agglomerate. The de-hydration of industrial line would reduce the
amount of hydrates formed on the water droplets while the The first group of the roots is given by:
thermal regime management is questionable and is a subject for a b 1 q
separate study. The reason is that e.g. heating of the pipe would Z1,2 = − −S± − 4S 2 − 2p + .
4a 2 S (44)
reduce the amount of hydrate phase, but increase the adhesive
force in between the hydrates. These roots are, however, complex with values of A, B and C
There are several assumptions, which have to be considered in that postulate the problem and therefore considered as physically
the end. Most of them come from the adoption of CSMHyK into the meaningless.
PBM-model: it is presently unclear whether it is best to use the The second group of the roots is:
apparent viscosity of the suspension or the liquid viscosity when b 1 q
Z3,4 = − +S± − 4S 2 − 2p − ,
the equilibrium size of agglomerates is found from the balance of 4a 2 S (45)
the adhesive and the shear forces. Moreover, a better rheological
where both of them are real. Z3 results, however, in non-physically
concept may be formulated taking into account the possibility of
large da/d0 , while Z4 is the exact solution of Eq. (6).
partial permeability of agglomerates by the carrier phase. The
process of nucleation and further growth of the hydrate shell on
the surface of water droplet is presently unclear, the model as-
References
sumes fast uniform nucleation and almost complete conversion of
droplets. The diffusion of the hydrate-former into the water phase
Balakin, B.V., 2010. Experimental and Theoretical Study of the Flow, Aggregation
is assumed to be independent of the local flow parameters around and Deposition of Gas Hydrate Particles (Ph.D. thesis). University of Bergen.
the droplet, while a more realistic turbulence model of large-eddy Balakin, B.V., Hoffmann, A.C., Kosinski, P., 2009. Population balance model for nu-
cleation, growth, aggregation and breakage of gas hydrate particles in turbulent
or detached-eddy type could be used instead of k –ϵ . flow. AICHE J. 56, 2052–2062.
Balakin, B.V., Hoffmann, A.C., Kosinski, P., 2011. Experimental study and computa-
tional fluid dynamics modeling of deposition of hydrate particles in a pipeline
with turbulent water flow. Chem. Eng. Sci. 66, 755–765.
Acknowledgements
Balakin, B.V., Hoffmann, A.C., Kosinski, P., 2014. Coupling STAR-CD with a popula-
tion-balance technique based on the classes method. Powder Technol. 257,
We gratefully acknowledge funding received from the Norwe- 47–54.
Balakin, B.V., Hoffmann, A.C., Kosinski, P., Hoiland, S., 2010a. Turbulent flow of
gian–Estonian Research Co-operation Programme EMP230 for our hydrates in a pipeline of complex configuration. Chem. Eng. Sci. 65, 5007–5017.
work. The internal research allowance from the Bergen University Balakin, B.V., Hoffmann, A.C., Kosinski, P., Istomin, V.A., Chuvilin, E.M., 2010b.
College is appreciated. The authors wish to thank Professor Ama- Combined CFD/population balance model for gas hydrate particle size predic-
tion in turbulent flow. In: Proceedings of the International Conference on Nu-
deu K. Sum from Colorado School of Mines for useful discussions merical Analysis and Applied Mathematics (ICNAAM-2010), vol. 1, pp. 151–154.
on the problematics of the present work. Balakin, B.V., Pedersen, H., Kilinc, Z., Hoffmann, A.C., Kosinski, P., Hoiland, S., 2010c.
 B.V. Balakin et al. / Chemical Engineering Science 153 (2016) 45–57
Turbulent flow of Freon R11 hydrate slurry. J. Pet. Sci. Eng. 70, 177–182.
Bondarev, E.A., Gabysheva, L.N., Kanibolotskii, M.A., 1982. Simulation of the for-
mation of gas hydrate during gas flow in the tube. Izv. Akad. Nauk SSSR, Me-
khanika Zhidkosti i Gaza (Transl.) 5, 105–112.
Boxall, J.A., Koh, C.A., Sloan, E.D., Sum, A.K., Wu, D.T., 2012. Droplet size scaling of
water-in-oil emulsions under turbulent flow. Langmuir 28, 104–110.
Camargo, R., 2001. Rheological Properties of Hydrate Suspensions in Crude Oils (Ph.
D. thesis). University of Paris (in French).
Cameirao, A., Fezoua, A., Ouabbas, Y., Herri, J.M., Darbouret, M., Sinquin, A., Glenat,
P., 2011. Agglomeration of gas hydrate in a water-in-oil emulsion: experimental
and modelling studies. HLA (Open access).
CD-adapco, 2006. Methodology, STAR-CD Version 4.02. 200 Shepherds Bush Rd.
London.
Crowe, C., Sommerfeld, M., Tsuji, Y., 1998. Multiphase Flows with Droplets and
Particles. CRC Press, Boca Raton.
Crowe, C.T., 2006. Multiphase Flow Handbook. Taylor and Francis Group, FL.
Cundall, P.A., Strack, O.D.L., 1979. A discrete numerical model for granular assem-
blies. Geotechnique 29, 47–65.
Dieker, L.E., Taylor, C.J., Koh, C.A., Sloan, E.D., 2008. Micromechanical adhesion force
measurements between cyclopentane hydrate particles. In: Proceedings of the
6th International Conference of Gas Hydrates.
Englezos, S., Kalogerakis, N., Dholabhai, P.D., Bishnoi, P.R., 1987. Kinetics of forma-
tion of methane and ethane gas hydrates. Chem. Eng. Sci. 42, 2647–2658.
Fidel-Dufour, A., Gruy, F., Herri, J.M., 2005. Rheological characterisation and mod-
elling of hydrate slurries during crystallisation under laminar flowing. In:
Proceedings of the 5th International Conference of Gas Hydrates.
Genovese, D.B., 2012. Shear rheology of hard-sphere, dispersed, and aggregated
suspensions, and filler-matrix composites. Adv. Colloid Interface Sci. 171–172,
1–16.
Herri, J.M., Pic, J.S., Gruy, F., Cournil, M., 1999. Methane hydrate crystallization
mechanism from in-situ particle sizing. AICHE J. 45, 590–602.
Hoffmann, A.C., Finkers, H.J., 1995. A relation for the void fraction of randomly
packed particle beds. Powder Technol. 82, 197–203.
Hounslow, M.J., Ryall, R.L., Marshall, V., 1988. A discretized population balance for
nucleation, growth, and aggregation. AICHE J. 34, 1821–1832.
Jassim, E., Abdi, M., Muzychka, Y., 2008. A CFD-based model to locate flow re-
striction induced hydrate deposition in pipelines. In: Offshore Technology
Conference.
Kinnari, K., Labes-Carrier, C., Lunde, K., Hemmingsen, P.V., Davies, S.R., Boxall, J.A.,
Koh, C.A., Sloan, E.D., 2008. Hydrate plug formation prediction tool – an in-
creasing need for flow assurance in the oil industry. In: Proceedings of the 6th
International Conference on Gas Hydrates (ICGH 2008).
Labois, M., Pagan, N., Lakehal, D., Narayanan, C., 2014. Computational modelling of
subsea hydrate formation and associated risks and impact on flow assurance.
57
In: Proceeding of the 10th International Conference on CFD in the Minerals and
Process Industries (CFD-2014).
Lo, S., 2011. CFD modelling of hydrate formation in oil-dominated flows. In: Pro-
ceeding of the Offshore Technology Conference (OTC21509).
Lo, S., Zhang, D., 2009. Modelling of break-up and coalescence in bubbly two-phase
flows. J. Comput. Multiphase Flows 1, 23–28.
Makogon, Y.F., 1997. Hydrates of Hydrocarbons. PennWell Publishing Company,
Oklahoma, U.S.A.
Malegaonkar, M.B., Dholabhai, P., Bishnoi, P.R., 1997. Kinetics of carbon dioxide and
methane hydrate formation. Can. J. Chem. Eng. 75, 1090–1096.
Marchisio, D.L., Soos, M., Sefcik, J., Morbidelli, M., 2006. Role of turbulent shear rate
distribution in aggregation and breakage processes. AICHE J. 52, 158–173.
Muhle, K., 1996. Flock stability in laminar and turbulent flow. In: Coagulation and
Flocculation: Theory and Applications (Chapter 8).
Potanin, A.D., 1991. On the mechanism of aggregation in the shear flow of sus-
pensions. J. Colloid Interface Sci. 145, 140–157.
Rabinovich, Y.I., Esayanur, M.S., Moudgil, B.M., 1998. Capillary forces between two
spheres with a fixed volume liquid bridge: theory and experiment. Chem. Eng.
Sci. 53, 10992–10997.
Ramkrishna, D., 2000. Population Balances, Theory and Applications to Particulate
Systems in Engineering. Academic Press, San Diego, CA.
Ruzicka, M.C., 2006. On buoyancy in dispersion. Chem. Eng. Sci. 61, 2437–2446.
Sean, W.-Y., Sato, T., Yamasaki, A., Kiyono, F., 2007. CFD and experimental study of
methane hydrate dissociation. Part I. Dissociation under water flow. AICHE J.
53, 262–274.
Sinquin, A., Palermo, T., Peysson, Y., 2004. Rheological and flow properties of gas
hydrate suspensions. Oil Gas Sci. Technol. 59, 41–57.
Skovborg, P., 1993. Gas Hydrate Kinetics (Doctor thesis). Institut for Kemiteknik,
Danmarks Tekniske Hojskole, Lyngby, Denmark.
Sloan, E.D., Koh, C.A., 2008. Clathrate Hydrates of Natural Gases, 3rd ed. CRC Press,
FL, U.S.A.
Snabre, P., Mills, P., 1993. I. Rheology of weakly flocculated suspensions of rigid
particles. J. Phys. France 47, 1811–1834.
Soos, M., Wang, L., Fox, R.O., Sefcik, J., Morbidelli, M., 2007. Population balance
modelling of aggregation and breakage in turbulent Taylor–Couette flow. J.
Colloid Interface Sci. 307, 433–446.
Yang, S.-O., Graham, M., Sloan, A.D., 2003. Simulation of hydrate agglomeration by
discrete element method. In: Proceedings of the 15th Symposium on Ther-
mophysical Properties.
Zerpa, L.E., 2013. A Practical Model to Predict Gas Hydrate Formation, Dissociation
and Transportability in Oil and Gas Flowlines (Ph.D. thesis). Colorado School of
Mines.
Zerpa, L.E., Sloan, E.D., Sum, A.K., Koh, C.A., 2012. Overview of CSMHyK: a transient
hydrate formation model. J. Pet. Sci. Eng. 98–99, 122–129.