You are on page 1of 6

th

Paper No 7 International Conference on Multiphase Flow


ICMF 2010, Tampa, FL USA, May 30-June 4, 2010

Lift and drag on agglomerates attached to walls

Robyn Larsen1, Dmitry Eskin2, Jos Derksen1


1
Chemical & Materials Engineering, University of Alberta, Edmonton, AB, Canada T6G 2V4, rlarsen@ualberta.ca,
jos@ualberta.ca
2
Schlumberger DBR Technology Center, Edmonton, AB, Canada T6N 1M9, deskin@slb.com

Keywords: agglomerates, simulation, lattice-Boltzmann, immersed boundaries, shear flow, asphaltenes

Abstract

With a view to modeling asphaltene deposition on pipeline walls we perform direct numerical flow simulations revealing drag
and lift forces on spheres and agglomerates in shear flow attached to walls. The direct simulations are based on the
lattice-Boltzmann method for solving the flow equations, combined with an immersed boundary method for representing the
no-slip conditions at the solid-liquid interfaces. For single spheres good agreement with analytical, low Reynolds number
results of lift and drag due to Leighton & Acrivos (1985) is observed. If scaled with the agglomerate size, non-dimensional
drag on agglomerates is roughly 30% lower than for a solid sphere, and drag varies over a range of 30% without clear trends in
terms of the number of primary spheres in the agglomerate or its voidage. At low Reynolds numbers, dimensionless lift is
much weaker than drag and is relatively more sensitive to the (random) relative placement if the primary spheres forming the
agglomerate.

Introduction of the computational domain on the hydrodynamic forces.

Asphaltene particle deposition in turbulent pipe flow is a Nomenclature


complex, multi-scale process involving various
transport-related phenomena: the formation of primary a,aa primary sphere radius, agglomerate radius
particles, their growth as a result of agglomeration (Kusters, dij deformation tensor
1991), the migration towards and subsequent deposition of F, F* drag force, non-dimensional drag force
agglomerates on the wall as well as the potential detachment lift force, non-dimensional lift force
of agglomerates from the wall as a result of hydrodynamic L, L*
forces. This work focuses on the latter aspect as it may be Lx , Ly , Lz flow domain dimensions
crucial in the prevention of agglomerate deposition. n number of primary spheres in agglomerate
We perform direct numerical simulations (DNS) of shear pressure, dimensionless pressure
p, p*
flow around agglomerates attached to walls. The
agglomerates are composed of equally sized spheres. For Re, Rea Reynolds number, agglomerate Reynolds
assessing the potential of the shear flow to detach the number
agglomerates, in the simulations we measure the drag and u0 moving wall velocity
lift force on the agglomerates. We assume that the ui , u fluid velocity component, fluid velocity
agglomerates have a near-spherical shape, i.e. high fractal magnitude
dimension. x,y,z Cartesian coordinates
The lattice-Boltzmann flow simulator (LBM) (Chen & Greek letters
Doolen, 1998; Eggels & Somers, 1995; Somers, 1993) γɺ , γɺ f overall shear rate, deformation rate in fluid
combined with an immersed boundary method (IBM)
ε voidage
(Derksen & Van den Akker, 1999) to represent the
spherically shaped solid-liquid interfaces was employed. ν kinematic viscosity
Prior to agglomerate simulations, the computational method ρ fluid density
was calibrated and verified by single sphere computations.
In the first place, a calibration of the hydrodynamic radius
for the primary spheres was done according to the approach Background on asphaltenes
of Ladd (1994). In the second place we used the
low-Reynolds number analytical solution of Leighton & Crude oil is a colloidal system consisting of saturates,
Acrivos (1985) for the lift and drag force acting on a single asphaltenes, resins and aromatics. Asphaltenes are colloidal
sphere attached to a flat in shear flow wall to validate our particles in the oil phase, a structure of which has not been
numerical approach. This exercise also provided completely understood yet. The asphaltene particles
quantitative information regarding the influence of the size precipitate from oil due to depressurization below the

1
th
Paper No 7 International Conference on Multiphase Flow
ICMF 2010, Tampa, FL USA, May 30-June 4, 2010

so-called asphaltene onset precipitation pressure. In practice is to construct a simplified kinetic model that incorporate
the pressure gradually drops along a transport pipeline; the essential physics of mesoscopic processes in the fluid.
asphaltene particles precipitate and grow forming The kinetic equation provides many of the advantages of
agglomerates, which partially deposit on the pipe walls. molecular dynamics, including clear physical pictures, easy
Asphaltene deposition is generally considered a negative implementation of boundary conditions, and fully parallel
phenomenon that may cause significant technical problems algorithms (Chen & Doolen, 1998).
and economical losses (Hammami & Ratulowski, 2007). In The kinetic model comprises fictitious fluid particles
extreme cases, the asphaltene deposition may lead to residing on a uniform, cubic lattice. The fictitious particles
complete disruption of the pipeline. It is therefore important move in a specific set of directions, and collide to mimic the
to model the asphaltene deposition process to anticipate and behavior of a viscous fluid. The kinetic nature of the LBM
prevent (negative) critical process conditions. Attempts have introduces three important features as identified by Chen &
been made to provide methods and guidelines for predicting Doolen (1998). These are: (1) the convection operation of
and evaluating asphaltene flocculation and deposition in oil the LBM in phase (or velocity) space is linear; (2) the
reservoirs (Alkafeef et al 2005). incompressible Navier-Stokes equations can be obtained in
In this work, the focus is primarily on the hydrodynamic the nearly incompressible limit; (3) the LBM utilizes a
aspects of asphaltene agglomeration and deposition. In minimal set of velocities in phase space.
addition to surface effects, hydrodynamics is considered From an initial state the time-stepping is achieved in two
crucial for understanding agglomerate deposition. Direct sequential phases (a) streaming, where each fluid particle
numerical simulations of the simple shear flow around moves to the nearest node in the direction of its velocity;
spheres and agglomerates assembled of equally sized and (b) collision, in which particles arriving at a node
primary spheres attached to flat walls have been performed. interact and exchange momentum according to scattering
Lift and drag on spheres and agglomerates have been rules (Chen & Doolen, 1998).
determined. As discussed above, the LBM employs a uniform, cubic
The dimensionless coordinates of the parameter space we grid. In order to impose no-slip conditions at the surfaces of
will be partly exploring in this paper are (1) a Reynolds solid, spherical particles we use an immersed boundary
number, (2) the size of the asphaltene agglomerates method (IBM). The spheres are defined by a collection of
expressed in the number of primary spheres it is made of, points on their surface with mutual spacing a little less than
and (3) the void fraction of the agglomerate. The way we the lattice-spacing. The no-slip condition is achieved by
define Reynolds number and void fraction will be discussed calculating forces acting on the liquid such that the linearly
below. In addition to physical parameters there are interpolated velocity at the points defining the sphere’s
numerical parameters (such as spatial resolution of the surface gets zero (Ten Cate et al., 2002).
simulations and computational domain size) the effects of In our simulations, the spherical particles forming the
which will be explored as well. agglomerates have fixed positions. The forces determined in
the LBM are forces on the fluid required to maintain no-slip.
The opposite of the sum of these forces can thus be
Mesoscopic modeling of multiphase flow interpreted as the force exerted by fluid on the particles to
keep the latter stationary.
Mesoscopic modeling can be defined in terms of the length
scales it represents. The Greek word mesos means middle.
The meso scale is considered to be in between the
macroscopic scale of the world we live in (for our purpose
this typically is the diameter of the oil pipeline), and the
microscopic scale of individual molecules. For multiphase
flow processes, meso scales relate to the scales of bubbles,
particles, and droplets and their mutual interactions
(Derksen & Sundaresan, 2007).
Mesoscopic modeling is usually based on some sort of
kinetic representation of the physics involved. Instead of
considering the continuum differential equations,
mesoscopic modeling considers a representation based on Figure 1: Agglomerate consisting of equally sized primary
(sometimes real, sometimes fictitious) particles that interact spheres of radius a attached to a wall in a simple shear flow
according to simple rules (“particle-based modeling”). The generated by a moving upper wall. Lift force (L) and drag
lattice-Boltzmann method (LBM) is an example of a force (F).
mesoscopic method. It is (supplemented with an immersed
boundary method, IBM) the primary tool used in the A calibration procedure (as introduced by Ladd, 1994) is
research presented here. The simple update rules and needed to determine the effective radius of the spheres
locality of operations makes the LBM highly (commonly referred to as the hydrodynamic radius). The
computationally efficient and inherently suitable to be run hydrodynamic radius is recognized with the symbol a and is
on (massive) parallel computer platforms, allowing for given in lattice units. In our study a radius a=6 has been
simulations with high spatial and temporal resolution (Succi, used. Grid refinement performed in previous studies
2001). (Derksen, 2008; Derksen & Sundaresan, 2007) clearly
The LBM is a relatively novel way to do Computational indicates that for low to modest sphere-size-based Reynolds
Fluid Dynamics (CFD). The fundamental idea of the LBM numbers (up to order 10) an a=6 resolution is sufficient to

2
th
Paper No 7 International Conference on Multiphase Flow
ICMF 2010, Tampa, FL USA, May 30-June 4, 2010

capture the essential physics of sphere-fluid interactions. center of the domain. They collide inelastically and
. eventually form a spherically shaped agglomerate. For n≥16
typically such agglomerates have a voidage of ε≈0.62,
Problem definition where the voidage is defined as void space volume fraction
in the smallest possible sphere (with radius aa ) fully
We perform direct numerical simulations of the simple shear 3
flow around agglomerates consisting of equally sized,  a 
enclosing the agglomerate: ε = 1 − n   . Less dense
spherical primary particles attached to walls (see Figure 1  aa 
for a typical flow system) and determine the lift and drag
agglomerates are made by contracting spheres with a
force on the agglomerates, as well as the distribution of
slightly larger radius ( a + δ ) and subsequently changing the
hydrodynamic forces on the primary spheres throughout the
radius back to a.
agglomerate. Shear is generated by placing a wall parallel to
In the agglomerate simulations the Reynolds number based
the wall the agglomerate sits on at distance Lz and moving
γɺ a 2
that wall with constant velocity u0 so that in steady state on the agglomerate radius ( Re a = a ) was kept to a
u0 ν
γɺ = . The agglomerates considered have a nearly constant (low) value of Re a = 0.09. Also dimensions of the
Lz
spherical shape. Given the (small) size of asphaltene flow domain Lx ⋅ Ly ⋅ Lz
agglomerates (typically, not larger than a few tens of (streamwise·spanwise·wall-normal) relative to aa (i.e. the
micrometers) and the slow flow at such close distances to
Lx Ly Lz
the pipeline wall, we only consider low Reynolds number aspect ratios , , ) were kept constant based on
flows. The (non-zero) value of the Reynolds number does aa aa aa
matter given the inertial nature of the lift force and its results of single-sphere simulations (see the next section).
proportionality to Re for spheres attached to walls (Leighton The two variables in the agglomerate simulations were the
& Acrivos, 1985) and to Re away from walls (Saffman, number of primary spheres in the agglomerate (in the range
1965 & 1968; McLaughlin, 1993). 2 to 64), and the voidage of the agglomerate (ε in the range
0.62 to 0.82).

Figure 3: Randomly distributed primary particles are


Figure 2: A sphere attached to a wall in a simple shear flow. contracted to the center of a cubic domain to generate an
agglomerate.
Direct numerical simulations were first preformed on single
spherical particles attached to walls (see Figure 2 for the
flow configuration). For this situation at low Reynolds Results & discussion
γɺa 2
numbers (defined as Re = ) there is an analytical Forces on a single spherical particle on a flat wall
ν To validate our numerical approach we started with
solution of Leighton & Acrivos (1985). The lift and drag simulations of single, spherical particles attached to a flat
L wall in a simple shear flow at a low Reynolds number
force stemming from this solution are = 9.22 Re
νργɺ a 2 (Re=0.012). The results of these simulations can be
compared to the analytical results due to Leighton &
F
and = 32.1 respectively. The analytical solution Acrivos (1985).
νργɺa 2 Illustrations of the flow around a single sphere in terms of
considers an unbounded flow domain. We use the analytical velocity vectors and the pressure field are given in Figures 4
solution to validate our numerical procedure, and to and 5 respectively. Dimensionless pressure p* is defined
establish aspect ratios (computational domain size in
p
streamwise, spanwise, and wall-normal direction over as p* ≡ .
ρ ( γɺ Lz )
2
particle radius) for which the numerical solution approaches
independency of domain size. Once appropriate aspect ratios As expected, the numerical results in terms of the lift and
(i.e. aspect ratios beyond which the hydrodynamic forces are the drag significantly depend on the size of the
largely insensitive of the domain size) were found, we were computational domain, see Figure 6. In this figure we
able to move from single spherical particles to agglomerates introduce dimensionless lift and drag according to
attached to walls. L F
The agglomerates are made by randomly releasing a pre-set L* ≡ , F* ≡ so that L* = 9.22 and
number (n) of equally sized spheres in a three-dimensional νργɺ a 2 Re νργɺ a 2
domain (as shown in Figure 3), and attracting them to the F * = 32.1 represents the Leighton & Acrivos solution.

3
th
Paper No 7 International Conference on Multiphase Flow
ICMF 2010, Tampa, FL USA, May 30-June 4, 2010

To get to a situation for which the lift and the drag are domain sizes) we will be dealing with in the agglomerate
virtually insensitive of domain size we first fixed the simulations.
domain size in streamwise direction (Lx=64), and
wall-normal direction (Lz=31) and systematically increased
the spanwise size (Ly). For Ly>64 drag and lift get
insensitive of Ly (see Figure 5, top panels). Subsequently we
fixed Ly to 71 and checked the sensitivity with respect to Lx.
At Lx around 80 lift and drag tend to constant values. Then
the wall-normal distance was increased and at Lz=100 lift
and drag converged to virtually constant values of 7.8 and
34.6 respectively. These numbers are within 15% of the
analytical solution.

Figure 6: Normalized lift force L* (left) and drag force


F * as a function of the size of the flow domain as defined
in Figure 2. The top panels have Lx=64, Lz=31, while Ly
varies. The middle panels have Ly=71, Lz=31, while Lx
varies. The bottom panels have Lx=81, Ly=71, Lz varies.
Re=0.012; a=6.

Figure 4: Velocity vector field of the flow around a single Agglomerates attached to a flat wall
sphere attached to a wall in simple shear flow. Agglomerates are characterized by the number of primary
particles (n) they consist of, and by their porosity. In order
to determine the porosity the radius of the agglomerate aa
(the radius of the smallest sphere that fully encloses the
agglomerate) needs to be determined. Given an agglomerate
with n primary particles and voidage ε, the flow conditions
are fully defined by the Reynolds number based on aa:
γɺa 2 n
Re = a (with aa = a 3 ); Re=0.09. The
ν 1− ε
agglomerate simulations have constant domain size relative
 L Ly L 
to aa:  x , , z  =(10.8,10.2,13.7). The single-sphere
 aa aa aa 
results indicated that for these aspect ratios the drag and the
lift do not depend much on the size of the flow domain.
As an example we now analyze the flow at Re=0.09 around
an agglomerate that consists of n=64 primary spheres and
that has a porosity of ε=0.75, see Figures 7 to 10. In Figure
7 we show the velocity field in the xz-cross section. The
upper panel (showing the entire cross section) illustrates the
laminar nature of the flow. The large domain size in
z-direction (Lz) is required to avoid too significant impact of
the upper wall on the lift and the drag force. The lower
panel of Figure 7 zooms in on the region around the
Figure 5: Non-dimensional pressure p* around a single agglomerate and has a logarithmic velocity magnitude scale
sphere attached to a wall at Re=0.012. Top panel: xz-cross so that it becomes clear that the velocities inside the
section through sphere center; bottom panel: xy cross agglomerate are extremely small. The latter is reiterated in
section through sphere center. the velocity vector plot in the xy-plane in Figure 8.

Reasons for the deviation from the analytical solution relate


to the (still) finite Reynolds number and the relatively
modest spatial resolution: the sphere radius a corresponds to
6 lattice spacings. This resolution was chosen because of the
large numbers of spheres per agglomerate (and related large

4
th
Paper No 7 International Conference on Multiphase Flow
ICMF 2010, Tampa, FL USA, May 30-June 4, 2010

structure is a secondary effect. Due to the limited flow


through the agglomerates (Figures 7, 8) this is not
surprising.
At low Reynolds numbers, lift is a much weaker force than
the drag force, and lift will only marginally impact
detachment of agglomerates attached to walls. Compared to
solid spheres, agglomerates feel a lift force that is roughly
50% of the lift force a solid sphere with radius aa would
experience.

Figure 8: Velocity vectors in the xy-plane at z=4a above the


bottom plate. Agglomerate with n=64, ε=0.75.

Figure 7: Velocity magnitude contours in the xz cross


section around an agglomerate with n=64 primary spheres,
and ε=0.75. Top panel: the entire xz cross section & linear
color scale; bottom panel: lower portion of cross section &
logarithmic color scale.

The pressure distribution around (i.e. outside) the


agglomerate (Figure 9) is very similar to the distribution
around a solid sphere (Figure 5). Inside the agglomerate
small irregularities in the pressure field can be observed that
are the result of the specific relative placement of the
primary spheres in the agglomerate. Finally, liquid
deformation (and thus viscous stress) inside the agglomerate
is very weak, see Figure 10. In this figure γɺ f denotes the
generalized deformation rate in the fluid γɺ f = 2dij dij ,

1  ∂u j ∂ui 
with d ij =  +  the deformation rate tensor.
2  ∂xi ∂x j

These flow visualizations indicate that the flow around this
agglomerate can be largely considered to be the flow around
a solid sphere (with a rough surface) of radius aa.
The results of all our simulations in terms of dimensionless
lift L* and drag F * are given in Figure 11. Dimensionless
drag is in the range 17 to 26, i.e. some 60% of the drag on a
solid sphere attached to a flat wall in shear flow. There is Figure 9: Non-dimensional pressure p* around an
not a clear trend with respect to the number of primary agglomerate with n=64, ε=0.75. Top panel: lower part of the
spheres in the agglomerate or its voidage. The essential xz-plane through the center of the agglomerate; bottom
parameter is the size of the agglomerate while its internal panel: entire xy-plane at z=4a.

5
th
Paper No 7 International Conference on Multiphase Flow
ICMF 2010, Tampa, FL USA, May 30-June 4, 2010

References

Chen, S., Doolen, G.D. Lattice Boltzmann method for fluid


flows. Annual Rev. Fluid Mech., Vol. 30, 329 (1998)

Derksen, J. & Van den Akker H.E.A. Large-eddy


simulations on the flow driven by a Rushton turbine. AIChE
J., Vol. 45, 209 (1999)
Figure 10: Fluid deformation rate γɺ f around and inside
the agglomerate with n=64, ε=0.75. Derksen, J.J. & Sundaresan, S. Direct numerical simulations
of dense suspensions: wave instabilities in liquid-fluidized
beds. J. Fluid Mech., Vol. 587, 303 (2007)

Derksen, J.J. Flow induced forces in sphere doublets. J.


Fluid Mech., Vol. 608, 337 (2008)

Eggels, J.G.M. & Somers, J.A. Numerical simulation of free


convective flow using the lattice- Boltzmann scheme. Int. J.
Heat Fluid Flow, Vol. 16, 357 (1995)

Hammami, A. & Ratulowski, J. Precipitation and deposition


of asphaltenes in production systems: A flow assurance
overview. In Asphaltenes, Heavy Oils, and Petroleomics;
Mullins, O. C., Sheu, E.Y., Hammami, A. & Marshall, A. G.,
Eds., Springer, New York (2007)

Kusters, K.A. The Influence of Turbulence on Aggregation


of Small Particles in Agitated Vessels. PhD Thesis,
Eindhoven University of Technology, Netherlands (1991)

Ladd, A.J.C. Numerical simulations of particle suspensions


via a discretized Boltzmann equation. Part 2. Numerical
results. J. Fluid Mech., Vol. 271, 311 (1994)
Figure 11: Dimensionless lift (top) and drag on
agglomerates as a function of the number of primary Leighton, D. & Acrivos, A. The lift on a small sphere
particles; voidages are as indicated. touching a plane in the presence of simple shear flow.
ZAMP, Vol. 36, 174 (1985)

Conclusions McLaughlin, J.B. The lift on a small sphere in wall-bounded


linear shear flows. J. Fluid Mech., Vol. 246, 249 (1993)
Motivated by the potential role of hydrodynamic forces on
the detachment of agglomerates off solid walls we Alkafeef, S.F., Al-Medhadi F., Al-Shammari, A.D. A
performed simulations of the lift and the drag force on Symplified Method to Predict and Prevent Asphaltene
agglomerates in shear flow over a flat surface. The Deposition in Oilwell Tubings: Field Case, SPE Production
simulation procedure was first calibrated by modeling a and Facilities, May, 126-132 (2005).
single, solid sphere and comparing the simulation results
with the analytical solution of Leighton & Acrivos (1985). Saffman, P.G. The lift on a small sphere in a slow shear flow.
These calibration simulations revealed the significant role of J. Fluid Mech., Vol. 22, 385 (1965) & Correction. J. Fluid
the size of the flow domain on the lift and the drag. For Mech., Vol. 31, 624 (1968)
sufficiently large domains our simulations showed fairly
good accuracy (errors in the 15% range). Somers, J. A. Direct simulation of fluid flow with cellular
Agglomerates, even the ones with loose structure (voidage automata and the lattice-Boltzmann equation. App. Sci. Res.,
of 75%) showed limited flow through them and therefore Vol. 51, 127 (1993)
the drag and the lift force could be well characterized by the
agglomerate size. Succi, S. The lattice Boltzmann equation for fluid dynamics
and beyond, Clarendon Press, Oxford (2001)

Acknowledgements Ten Cate, A., Nieuwstad, C.H., Derksen, J.J. & Van den
Akker, H.E.A. PIV experiments and lattice-Boltzmann
Support of the Schlumberger DBR Technology Center is simulations on a single sphere settling under gravity. Phys.
gratefully acknowledged. Fluids, Vol. 14, 4012 (2002)

You might also like