Erosion prediction in slurry pipeline tee-junctions

G.J. Brown
*
Alcoa World Alumina Australia, Kwinana Renery, Kwinana, Western Australia 6167, Australia
Received 1 December 1999; received in revised form 1 November 2000; accepted 24 April 2001
Abstract
A signicant erosion problem is found to occur on steel blanks located inside tee-junctions in a slurry
pipeline system. A 3-D computational uid dynamics (CFD) model is developed to predict the motion of
caustic liquor and bauxite particles through a tee-junction using an EulerianEulerian continuum approach
in conjunction with the ke turbulence model. Initial simulations assuming a uniform inlet ow are unable
to demonstrate the cause of the observed erosion. However, subsequent modelling with a swirling inlet ow,
based on a more thorough assessment of the upstream vessels and piping, results in the prediction of an
accumulation of particles on the steel blank at the centre of a slow-moving vortex, the location of which is
in excellent agreement with the observed wear on the plant. Furthermore, the predicted wear location is
found to be insensitive to the assumed level of inlet swirl and the numerical scheme employed. The multi-
phase CFD model is also used to assess several potential solutions to the erosion problem, resulting in the
replacement of the tee-junctions with a pivoting elbow design which is found to exhibit signicantly reduced
erosion rates on the plant. These results demonstrate the eectiveness with which CFD techniques can be
used in the solution of industrial erosion problems. They also highlight the potential sensitivity of modelling
results to inlet boundary condition assumptions and emphasise the need to adequately account for up-
stream inuences when applying CFD techniques to the simulation of industrial ows. 2002 Elsevier
Science Inc. All rights reserved.
Keywords: Erosion; Computational uid dynamics; Slurry; Tee-junction; EulerianEulerian multi-phase model
1. Introduction
At Alcoas Pinjarra alumina renery in Western Australia steel blanks located inside
tee-junctions in a slurry pipeline system are used to switch the ow between two possible paths
Applied Mathematical Modelling 26 (2002) 155170
www.elsevier.com/locate/apm
*
Fax: +61-8-9410-3166.
E-mail address: gary.brown@alcoa.com.au (G.J. Brown).
0307-904X/02/$ - see front matter 2002 Elsevier Science Inc. All rights reserved.
PII: S0307- 904X( 01) 00053- 1
(Fig. 1). The slurry being transported in this section of the plant consists of bauxite particles in a
hot caustic soda solution. The bauxite particles have a high silica content and are hence highly
abrasive.
A signicant erosion problem was found to occur on the steel blanks in the tee-junctions when
the blanks were located in the standard operating position (Fig. 1). It was found that a con-
centrated hole was being worn through these blanks inside 13 weeks of operation, allowing slurry
into the by-pass piping (Fig. 2). The slurry would then stagnate and cool inside the by-pass piping,
resulting in the creation of a hard scale that rendered the by-pass piping unusable.
The presence of such signicant and highly localised erosion on the steel blanks was unexpected
because the blanked end of the tee-junction was expected to be a region of near-stagnant ow.
This lack of knowledge regarding the cause of the erosion made it dicult for plant engineers to
develop a solution, other than to recommend the use of harder materials for the blanks. Con-
sequently, a computational uid dynamics (CFD) study was initiated to determine the cause of
the erosion and to assist in the design of modications to the piping system.
This paper describes the use of the commercial package CFX-4.2 [1] to predict the motion of
caustic liquor and bauxite particles through a single tee-junction using an EulerianEulerian
continuum approach in conjunction with the ke turbulence model. In particular, the inuence of
the assumed inlet conditions to the tee-junction is discussed.
Nomenclature
C inter-phase momentum transfer coecient
C
D
drag coecient
d
p
particle diameter
g gravitational acceleration
k
f
uid turbulent kinetic energy
L
s
characteristic length
p uid pressure
r
f
uid phase volume fraction
r
p
particulate phase volume fraction
Re
p
particle Reynolds number
S
t
Stokes number
u
f
uid velocity
u
p
particle velocity
V
s
characteristic velocity
b particle mass loading
e
f
uid turbulence dissipation rate
l
eff
eective viscosity for uid phase
l
f
uid dynamic viscosity
l
lp
arbitrary laminar viscosity for particle phase
q
f
uid density
q
p
particle density
156 G.J. Brown / Appl. Math. Modelling 26 (2002) 155170
Single phase ow predictions are also conducted using a dierential Reynolds stress turbulence
model, in conjunction with a second-order accurate dierencing scheme (Van Leer), to conrm
that the model results are qualitatively insensitive with respect to the turbulence model and dif-
ferencing scheme selected.
In addition, the multi-phase CFD model is used to assess several potential solutions to the
erosion problem.
Fig. 1. Schematic of slurry pipeline tee-junction.
Fig. 2. Hole in steel blank. Direction of inlet ow to tee-junction is indicated.
G.J. Brown / Appl. Math. Modelling 26 (2002) 155170 157
2. Particle transport model selection
The three most common particle transport models, and those most prevalent in commercial
CFD codes, are the EulerianEulerian, Lagrangian and driftux models. All three of these
models could potentially be applied to the simulation of a mineral processing slurry, depending on
the exact slurry characteristics and the ow geometry in question. It is therefore worth reviewing
the advantages and disadvantages of each model with respect to this application.
2.1. EulerianEulerian models
EulerianEulerian models are also generally referred to as multi-uid or continuum models
because the particles are treated as an additional continuous phase. An additional set of con-
servation equations is solved for the particulate phase and coupling between the two phases takes
place through inter-phase transfer terms in the two sets of conservation equations.
This approach is ideally suited to the modelling of slurries with moderate to high particle
concentrations, where two-way particleuid coupling and possibly particleparticle interactions
are important, because source terms can readily be formulated in terms of spatial gradients in
phase velocities and volume fractions [2]. These features have been exploited, for example, in the
modelling of thickener vessels used in the mineral processing industry [3].
The main limitation of the EulerianEulerian approach is in its application to systems with a
large particle size distribution. Additional continuity and momentum equations are required for
every additional particle size simulated, hence signicantly increasing the complexity of the model
formulation and the computational overhead. As a result, the EulerianEulerian approach is
usually applied in situations where the important features of the ow can be discerned through the
simulation of a single representative particle size.
2.2. Lagrangian models
In the Lagrangian approach Eulerian continuum equations are still solved for the uid phase
(the carrier liquid in the case of a slurry) but Newtonian equations of motion are solved to de-
termine the trajectories of individual particles (or groups of particles). Each particle can have
dierent properties (size, shape, density, initial conditions) thus allowing the simulation of slurries
involving large particle size distributions or dierent ore types.
In ows where the particle loading is high enough that signicant two-way particleuid
coupling would be expected, the particle trajectories can be evaluated in a coupled loop with the
continuous phase solution. However, achieving realistic coupling between the particle and uid
phases requires adequate resolution of the spatial distribution of the particles, which in turn re-
quires the calculation of a large number of trajectories. This makes use of the Lagrangian ap-
proach computationally expensive for ows in which two-way coupling must be considered. In
addition, in Lagrangian simulations the volume fraction occupied by the particle phase is gen-
erally ignored in the solution for the continuous phase. This assumption would obviously be
invalid at anything other than dilute particle loadings.
These features of the Lagrangian approach make it ideally suited to cases in which the de-
termination of particle classication is important and in which the solids loading is low, thus
158 G.J. Brown / Appl. Math. Modelling 26 (2002) 155170
allowing two-way coupling and the volume occupied by the particles to be neglected and the
particle trajectories to be calculated as a post-process. A recent example in the mineral processing
industry has been the prediction of spiral concentrator performance [4].
2.3. Driftux models
Driftux or algebraic slip models involve the solution of Eulerian continuum equations for a
single uid phase which exhibits a variable density based upon the local particle volume fraction
summed across all particle sizes (i.e. the uid density is the eective slurry density). The particles
are assumed to be continuously slipping with respect to this carrier uid at a constant velocity due
to gravitational and/or centrifugal forces and the distribution of particles is obtained through the
solution of a single scalar transport equation for the volume fraction of each particle size con-
sidered.
This technique has obvious advantages in cases where the EulerianEulerian approach is
computationally too prohibitive due to the need to consider multiple particle sizes and yet the
Lagrangian technique is also unsuitable because the particle concentration is high enough
to signicantly inuence the uid ow. However, because of the need to calculate a xed particle
slip velocity the use of driftux models is generally limited to geometries in which either grav-
itational and/or centrifugal forces dominate, such as in gravity settling vessels [5] and hydro-
cyclones [6,7].
2.4. Particleuid coupling
Selection of the correct particle transport model for a particular application can be assisted by
rst calculating the particle mass loading, b, and the Stokes number, S
t
.
The particle mass loading is expressed as:
b
particulate mass per unit volume of flow
fluid mass per unit volume of flow

r
p
q
p
r
f
q
f
; 1
where r is a volume fraction, q is a density and the subscripts p and f refer to the particle and uid
phases, respectively. Signicant two-way particleuid coupling is generally expected for particle
mass loadings greater than 0.2 and values greater than 0.6 indicate that signicant particle
particle interactions are likely in at least some parts of the ow domain [8].
The degree to which the particle motion is tied to the uid motion can be determined through
evaluation of the Stokes number. This is dened as the ratio of the particle response time due to
viscous drag to a characteristic turbulent eddy time in the carrier uid. This can be expressed as:
S
t

q
p
d
2
p
V
s
18l
f
L
s
; 2
where d
p
is the particle diameter, l
f
is the dynamic viscosity of the carrier uid and V
s
and L
s
are
characteristic velocity and length scales in the ow [8].
For large values, S
t
> 2:0, the particulate ow is highly inertial and, in a conned geometry,
would be dominated by particlewall interactions, whereas for values less than 0.25 the eect of
particlewall interactions on the particle ow is essentially negligible because the particles are
G.J. Brown / Appl. Math. Modelling 26 (2002) 155170 159
more tightly coupled to the uid through viscous drag. At Stokes numbers below 0.05 the particles
and carrier uid are strongly coupled and the particles would be expected to approximately follow
the uid ow. For very small values, say S
t
<< 0:01, particle inertia is essentially negligible and
the particles would be expected to respond almost instantaneously to any changes in the uid ow
[8,9].
2.5. Model selection
For the tee-junction ow to be evaluated in the current study the application of a driftux
model was not considered because the geometry is such that the particle motion would not be
dominated by either gravity settling or centrifugal separation.
The particle mass loading at inlet was calculated to be 0.175 which indicated that it may have
been possible to neglect two-way particleuid coupling. However, some particle settling was
expected to occur in the blanked end of the tee which would increase the particle mass loading in
this area to a level where two-way coupling might inuence the uid ow. Furthermore, pre-
diction of erosion on the steel blank would obviously require adequate resolution of the particle
distribution in an area which is o the mean ow path. Therefore, the use of a Lagrangian model
would be computationally expensive due to the need to calculate a large number of particle
trajectories in a coupled solution.
The Stokes number based on the inlet velocity, pipe diameter and mean particle size was cal-
culated to be 0.026, which indicated that particlewall interactions would have a negligible eect
on the particle ow and hence that no special treatment of the particle boundary conditions was
required.
In consideration of the above analysis an EulerianEulerian model was selected for the tee-
junction study.
3. Erosion modelling
Following the concepts established by Finnie and Bitter [10,11], erosion of a solid surface due
to particle impacts can be considered to be due to two separate mechanisms, namely deformation
wear and cutting wear.
Deformation wear occurs when repeated particle impacts at high angles plastically deform the
surface layers of the material, eventually causing material loss through surface fragmentation.
Cutting wear occurs due to particle impacts at small angles, with a scratch or cut being formed on
the surface if the shear strength of the material is exceeded.
The total erosion rate at a particular point on a surface is found by summing the contributions
due to the deformation and cutting mechanisms and depends on the properties of the material,
with deformation wear being more signicant for hard, brittle materials and cutting wear being
more signicant for softer, ductile materials. For standard commercial grade steels, as used in the
tee-junction in this study, peak erosion rates have been measured to occur at impact angles of 25
30, indicating that cutting wear dominates [11].
The other critical factor aecting wear is the particle impact velocity, with both cutting and
deformation wear being proportional to impact velocity raised to a power n determined through
160 G.J. Brown / Appl. Math. Modelling 26 (2002) 155170
physical tests. In general n is found to vary between 2.0 and 3.0 depending on both the surface and
particle materials.
These concepts were embodied by Finnie [10] in a simplied erosion model relating erosion rate
to particle mass ux, impact angle and impact velocity. This model or its variants have been used
successfully in many CFD erosion studies, for example in the examination of uidised beds using
an EulerianEulerian model [12].
An erosion model has not been implemented in the current study because the cause of the
erosion was found to be evident from an examination of the particle distribution and ow patterns
within the tee-junction, as presented in the results below. However, the author intends to extend
this study to include the implementation of an erosion model in the near future.
4. EulerianEulerian (multi-uid) model description
4.1. Governing equations
A two-phase EulerianEulerian model has been established for the liquid phase (caustic soda
solution) and a single representative bauxite particle size, in this case selected to be particles of 150
lm diameter. The full solids mass loading is assumed to exist at this particle size.
Turbulent dispersion of the particles is neglected and the particle mass loading in the majority
of the ow domain is considered to be low enough to allow the eect of particles on uid phase
turbulence and particleparticle interactions to also be neglected. As a result of this last as-
sumption the particle phase is declared to be laminar and is attributed a small dynamic viscosity.
This has the eect of making the diusion term in the particle phase momentum equation neg-
ligible.
As a result of these assumptions and for steady ow the required continuum equations for
conservation of mass and momentum are:
r r
f
q
f
u
f
0; 3
r r
p
q
p
u
p
_ _
0; 4
r r
f
q
f
u
f
u
f
r r
f
l
eff
ru
_ _
ru
T
__
f
r
f
rp C u
p
_
u
f
_
r
f
q
f
g; 5
r r
p
q
p
u
p
u
p
_ _
r r
p
l
lp
ru
_ _
ru
T
__
p
r
p
rp C u
f
_
u
p
_
r
p
q
p
g; 6
where it should be noted that l
lp
is an arbitrarily small laminar viscosity for the particle phase.
In addition, turbulence closure in the uid phase is achieved through solution of the standard
ke model, with the k and e equations taking the form:
r r
f
q
f
u
f
/ r r
f
l
eff
r
/
r/
_ _
r
f
S
/
; 7
where / represents either k or e, r
/
is the turbulent diusivity of / and S
/
is a source term.
The inter-phase momentum transfer coecient C is calculated through consideration of uid
particle drag. The particle drag coecient is given by:
G.J. Brown / Appl. Math. Modelling 26 (2002) 155170 161
C
D

24
Re
p
1
_
0:15Re
0:687
p
_
0 < Re
p
61000 8
with the particle Reynolds number dened as:
Re
p

r
f
q
f
d
p
ju
p
u
f
j
l
f
: 9
4.2. Boundary conditions
All variables are dened at the tee-junction inlet. The particles are also assumed to be uniformly
distributed and, due to the low Stokes number, the particle velocity distribution is assumed to be
identical to that for the uid phase. A zero gradient condition is applied at the outlet.
Standard no-slip wall functions are applied at all solid surfaces for both the uid and particle
phases. Use of no-slip wall functions for the particle phase is justied in this case on the basis of
the low Stokes number (0.026) which indicates that there is a strong coupling between the particles
and the uid and hence that particlewall rebound characteristics will have a negligible impact on
the particle ow.
4.3. Computational domain and numerical procedure
The tee-junction geometry is discretised using a block-structured non-orthogonal Cartesian
mesh. Due to the circular pipe sections a 5-block mesh structure is adopted to reduce the non-
orthogonality of the mesh elements. The computational mesh is shown in Fig. 3.
Fig. 3. Computational domain for tee-junction model.
162 G.J. Brown / Appl. Math. Modelling 26 (2002) 155170
The multi-phase conservation equations for mass, momentum and uid turbulence are solved
within the commercial code CFX-4 [1] using a nite volume technique based on the SIMPLEC
algorithm [13]. Inter-phase coupling is achieved using Spaldings inter-phase slip algorithm (IPSA)
[14]. Convection terms in the momentum equations are discretised using the rst-order hybrid
scheme.
5. Tee-junction EulerianEulerian model results
Initial simulations of the tee-junction were undertaken using an assumption of a uniform inlet
ow and particle distribution. The resulting ow patterns are shown in Fig. 4 and the particle
distribution in Fig. 5.
It can be seen that a slow moving recirculation is created in the blanked end of the tee as the
main ow moves from the inlet leg to the discharge leg. Particles entrained in this recirculation
from the main ow move down the back wall of the tee and across the face of the blank, before
being deposited at the bottom of the front wall of the blanked area the vertical component of
uid velocity next to the wall being insucient to suspend the particles. By comparison with Fig. 2
it can be seen that this result is inconsistent with the highly localised and asymmetric erosion
observed on the plant.
In light of this result a more thorough assessment of the piping and vessel layout upstream of
the tee-junction was undertaken. It was subsequently concluded that there was signicant po-
tential for the inlet ow to be swirling, because the slurry is drawn from the cylindrical upstream
vessel through a single centrally located underow pipe.
Fig. 4. Fluid phase streamlines. Multi-phase tee-junction model with uniform inlet ow.
G.J. Brown / Appl. Math. Modelling 26 (2002) 155170 163
Further simulations were conducted with a swirl velocity component added to both the uid
and particle phases at the inlet to the computational model. Solid body rotation was assumed,
with the peak tangential velocity occurring at the pipe wall. The particle distribution at inlet was
still assumed to be uniform. In the absence of any data regarding the potential level of swirl in the
ow an initial simulation was conducted with the peak tangential velocity at inlet set equal to the
inlet axial velocity. The results achieved are illustrated in Figs. 6 and 7.
In Fig. 6 streamlines for the uid phase have been initiated from several positions at the tee-
junction inlet and also from several positions on the surface of the steel blank. It can be seen that
the rotation of the inlet ow generates a vortex in the blanked region of the tee, with the vortex
centred about a point remarkably close to the site of the observed erosion on the plant (Fig. 2). In
addition, a plot of particle concentration reveals an accumulation of particles on the surface of the
steel blank at the same point (Fig. 7).
Further analysis of particle phase streamlines revealed that particles entering the blanked end of
the tee are eventually entrained into the centre of the vortex before being drawn up the vortex
core. This suggests that the highly localised erosion seen on the steel blank in the plant is due to
repeated particle impacts at highly acute angles as particles are entrained into the centre of the
vortex. Given the small bed of solids at the centre of the vortex there may also be the potential for
erosion due to friction between particles sliding in the bed and the surface of the blank.
Given the lack of data regarding the level of swirl in the inlet ow on the plant, simulations
were also conducted with the peak tangential velocity at inlet set to 1.5x and 0.5x the inlet axial
velocity. In both cases a vortex was found to form in the blanked end of the tee, centred about the
same point as in the initial simulation, and with an accumulation of solids on the steel blank at the
Fig. 5. Particle concentration (v=v). Multi-phase tee-junction model with uniform inlet ow.
164 G.J. Brown / Appl. Math. Modelling 26 (2002) 155170
core of the vortex. As expected, there are quantitative variations in the results, with rotational
velocities in the vortex increasing and the solids concentration in the core decreasing as the inlet
Fig. 6. Fluid phase streamlines. Multi-phase tee-junction model with swirling inlet ow.
Fig. 7. Particle concentration (v=v). Multi-phase tee-junction model with swirling inlet ow.
G.J. Brown / Appl. Math. Modelling 26 (2002) 155170 165
swirl increases. However, these results show importantly that the predicted ow is qualitatively
insensitive to the assumed level of inlet swirl.
6. Tee-junction single phase dierential stress results
Fluid phase streamlines were found to be almost identical in the ke multi-phase runs, to those
in single phase ke runs. This indicates that the relatively low particle mass loading in the majority
of the ow (b < 0:2), and even the higher mass loadings in the blanked end of the tee, do not result
in any signicant alteration to the uid ow eld due to two-way coupling. It was therefore
concluded that it would be sucient to use a single phase model to check the sensitivity of the
results to the numerical scheme selected.
A single phase simulation was conducted for the swirling inlet ow case using a dierential
Reynolds stress model (DSM) in conjunction with the second-order accurate Van Leer dier-
encing scheme, on the basis that this has been shown to give signicantly improved accuracy for
swirling ows on non-orthogonal Cartesian grids when compared to use of the standard ke
turbulence model and hybrid dierencing [15].
The predicted ow patterns are shown in Fig. 8. Comparison with the ke results in Fig. 6
shows that there is some variation in the structure and location of the vortex, but that the overall
nature of the ow does not change signicantly. On the basis of this result, highly localised
erosion would still be expected at a site in close agreement with the multi-phase ke results and the
observed erosion on the plant. This demonstrates that the results are relatively insensitive to the
exact numerical scheme selected.
Fig. 8. Fluid phase streamlines. Single phase tee-junction model with swirling inlet ow (DSM+Van Leer).
166 G.J. Brown / Appl. Math. Modelling 26 (2002) 155170
7. Industrial solution
Having identied the cause of the highly localised erosion on the steel blank, the Eulerian
Eulerian CFD model was used to assess several potential solutions to the erosion problem.
The rst option investigated was to move the steel blank to the next available ange position in
the piping, a further 300 mm below the bottom of the tee-junction. The simulation results showed
that this would result in a far less structured and more slowly moving rotation of the ow in the
blanked end of the tee-junction. They also showed that the particle concentration on the steel
blank would be higher, but that the particles would be more uniformly distributed over the
surface of the blank. This suggested that highly localised erosion on the steel blank was unlikely to
occur, but the lower velocities and higher solids concentration indicated that there was the po-
tential for a scaling problem to occur (the slurry in this area of the plant will grow scale in areas
where it is allowed to stagnate and cool). Because of the simplicity of this option a trial was
conducted on the plant and the CFD predictions were again found to be accurate. No erosion of
the steel blank occurred after a signicant period in operation, however the scale growth on the
blank was such that it eectively became cemented to the piping ange, making it extremely
dicult to remove in the event that the by-pass piping needed to be utilised.
Fig. 9. Schematic of pivoting elbow design.
G.J. Brown / Appl. Math. Modelling 26 (2002) 155170 167
By altering the length of the pipe spool beneath the tee-junction it may have been possible to
position the blank at a point where erosion was minimised without introducing a scaling problem,
however this may have required considerable trial and error on the plant and hence this option
was not investigated further.
The second option investigated was to replace the tee-junction with a combination of 30 and
60 elbows as shown in Fig 9. In this concept, two xed 30 elbow sections would be tted to the
normal vertical piping leg and the by-pass line and a 60 elbow section could be turned through
180 (using an overhead jib crane) to connect the upstream piping to either ow path. A simu-
lation of this option was conducted for the normal ow conguration. The resulting ow patterns
are shown in Fig. 10 and the particle distribution in Fig. 11.
It can be seen that the elbow design allows the ow to continue to rotate as it passes between
the horizontal and vertical piping legs without the formation of any unusual localised ow pat-
terns. The particles tend to form a band or rope as they move through the bend, resulting in
higher particle concentrations on some areas of the piping walls but the particles do not accu-
mulate at any single point in the bend and they essentially move tangentially to the walls, hence
minimising the erosion risk.
On the basis of the CFD results the elbow design was trialed on the plant and was found to be
highly successful, with no erosion problems being evident after an extended period of operation.
As a result, the by-pass tee-junctions in the equivalent location on other production units have
been systematically changed to the pivoting elbow design since the initial trial.
Fig. 10. Fluid phase streamlines. Multi-phase elbow model with swirling inlet ow.
168 G.J. Brown / Appl. Math. Modelling 26 (2002) 155170
8. Conclusions
A 3-D CFD model based upon the commercial tool CFX-4 has been used to investigate the
cause of highly localised and asymmetric erosion found to occur on steel blanks within slurry
pipeline tee-junctions. The motion of caustic liquor and bauxite particles through a tee-junction
has been predicted using an EulerianEulerian multi-phase approach in conjunction with a ke
turbulence model.
Simulations with an assumed uniform inlet ow were unable to identify the cause of the
erosion. However, simulations with a swirling inlet ow, based on a more thorough assessment
of the upstream vessels, showed an accumulation of particles on the steel blank at the centre of
a slow-moving vortex, the location of which is in excellent agreement with the observed wear
on the plant. This result was found to be qualitatively insensitive to the assumed level of inlet
swirl.
Comparison of the multi-phase results with single phase predictions has shown that the par-
ticles do not signicantly alter the uid ow in this case due to the relatively low solids mass
loading in the majority of the ow domain. On this basis, a single phase simulation has also been
conducted using a DSM in conjunction with a second-order accurate convective dierencing
scheme. The results obtained demonstrate that the predicted erosion site is also insensitive to the
numerical scheme employed.
Fig. 11. Particle concentration (v=v). Multi-phase elbow model with swirling inlet ow.
G.J. Brown / Appl. Math. Modelling 26 (2002) 155170 169
The multi-phase CFD model has also been used to assess several potential solutions to the
erosion problem, with the model results showing that the highly localised erosion could be
eliminated through replacement of the tee-junction with a pivoting elbow design. This design has
subsequently been implemented with signicant success on the plant.
These results demonstrate how CFD techniques can be used to signicant benet in the pre-
diction of industrial erosion problems and in the development of solutions to these problems. The
results also demonstrate how deceptively complex ows can be established inside simple geo-
metries due to non-uniformities in the inlet ow, thus highlighting the need to adequately account
for upstream inuences when applying CFD techniques to the simulation of industrial ows.
References
[1] AEA Technology, CFX-4.2 Flow solver user guide, AEA Technology plc, 1997.
[2] M.J. Andrews, P.J. ORouke, The multiphase particle-in-cell (MP-PIC) method for dense particulate ows, Int. J.
Multiphase Flow 22 (1996) 379402.
[3] R.B. Kahane, M.P. Schwarz, R.M. Johnston, Residue thickener modelling at Worsley Alumina, in: Proceedings of
International Conference on CFD in Mineral & Metal Processing and Power Generation, CSIRO, Melbourne,
Australia, July 34, 1997.
[4] B.W. Mathews, C.A.J. Fletcher, A.C. Partridge, Computational simulation of uid and dilute particulate ows on
spiral concentrators, Appl. Math. Modelling 22 (1998) 965979.
[5] Y. Li, M. Rudman, G. Brown, Particle transport in a bottom-feed separation vessel, Appl. Math. Modelling 22
(1998) 10231036.
[6] K.A. Pericleous, Mathematical simulation of hydrocyclones, Appl. Math. Modelling 11 (1987) 242255.
[7] H.D. Zughbi et al., Numerical and experimental investigations of wear in heavy medium cyclones, Miner. Eng.
4 (1991) 245262.
[8] C.A.J. Fletcher, Industrial CFD short course, Course Lecture Notes, University of New South Wales, June 27July
1, 1994.
[9] J.Y. Tu, C.A.J. Fletcher, Eulerian modelling of dilute particle-laden gas ows past tubes, Comput. Fluid Dyn. J.
5 (1996) 125.
[10] I. Finnie, Erosion of surfaces by solid particles, Wear 3 (1960) 87103.
[11] J.G.A. Bitter, A study of erosion phenomena: parts 1 and 2, Wear 6 (1963) 521 & 169190.
[12] D. Achim et al., Mathematical modelling of erosion in a uidised bed, in: Proceedings of 12th Australian Fluid
Mech. Conferecne, Sydney, Australia, December 1015, 1995.
[13] J.P. Van Doormal, G.D. Raithby, Enhancements of the SIMPLE method for predicting incompressible uid ows,
Numer. Heat Transfer 7 (1984) 147163.
[14] D.B. Spalding, Numerical computation of multi-phase uid ow and heat transfer, in: C. Taylor, K. Morgan
(Eds.), Recent Advances in Numerical Methods, vol. 1, Pineridge Press, Swansea, 1980, pp. 139167.
[15] N.A. Shore et al., Numerical aspects of swirl ow computation, Computational Techniques and Applications:
CTAC95, Melbourne, Australia, July 37, 1995.
170 G.J. Brown / Appl. Math. Modelling 26 (2002) 155170