You are on page 1of 22

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/239408986

Smooth particle hydrodynamics: Status and future potential

Article  in  Progress in Computational Fluid Dynamics An International Journal · January 2007


DOI: 10.1504/PCFD.2007.013000

CITATIONS READS

186 2,929

5 authors, including:

Paul W. Cleary Mahesh Prakash


The Commonwealth Scientific and Industrial Research Organisation The Commonwealth Scientific and Industrial Research Organisation
400 PUBLICATIONS   13,608 CITATIONS    98 PUBLICATIONS   1,964 CITATIONS   

SEE PROFILE SEE PROFILE

Joseph Ha Nick Stokes


Department of Environment, Land, Water and Planning The Commonwealth Scientific and Industrial Research Organisation
48 PUBLICATIONS   1,107 CITATIONS    52 PUBLICATIONS   991 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Biophysical modelling of the environment View project

EnSym - Environmental Systems Modelling Platform View project

All content following this page was uploaded by Mahesh Prakash on 21 April 2015.

The user has requested enhancement of the downloaded file.


70 Progress in Computational Fluid Dynamics, Vol. 7, Nos. 2/3/4, 2007

Smooth particle hydrodynamics: status and future


potential

Paul W. Cleary*, Mahesh Prakash, Joseph Ha,


Nick Stokes and Craig Scott
CSIRO Mathematical and Information Sciences,
Private Bag 10, Clayton South, VIC 3169, Australia
E-mail: Paul.Cleary@csiro.au E-mail: Mahesh.Prakash@csiro.au
E-mail: Joseph.Ha@csiro.au E-mail: Nick.Stokes@csiro.au
E-mail: Craig.Scott@csiro.au
*Corresponding author

Abstract: SPH is a powerful mesh free method that is now able to solve very complex
multi-physics flow and deformation problems in a broad number of fields. This paper
concentrates on the use of SPH to simulate a broad range of complex industrial fluid
flow problems. These include free surface fluid flow for the generation of digital content,
geophysical flows such as volcanic lava flows and tsunamis, several types of die casting
(gravity, high pressure and ingot casting), resin transfer moulding and flow in porous media,
mixing of particulates in liquid, pyrometallurgy and slurry flow in semi-autogenous grinding
mills. The strengths and weaknesses of SPH will be explored and future opportunities for using
the method to make major modelling advances are discussed.

Keywords: smoothed particle hydrodynamics; free surface fluid flow; die casting; ingot casting;
tsunamis and flooding; dam break; particulate-fluid mixing; reacting gas-fluid-solid flows;
mineral processing fluid flow.

Reference to this paper should be made as follows: Cleary, P.W., Prakash, M., Ha, J., Stokes, N.
and Scott, C. (2007) ‘Smooth particle hydrodynamics: status and future potential’, Progress in
Computational Fluid Dynamics, Vol. 7, Nos. 2/3/4, pp.70–90.

Biographical notes: Paul Cleary received his PhD in Applied Mathematics from Monash
University in 1988. He joined CSIRO in 1991 working in the area of industrial CFD. This CFD
group, which he now leads, develops and applies novel computational methods such as DEM and
SPH to industrial, mineral processing and geophysical flow problems involving fluids and
particulate solids.

Mahesh Prakash is a Research Scientist at CSIRO Mathematical and Information Sciences,


Australia. His research interests include application of grid-free methods to industrial
flow problems, turbulence modelling using Smoothed Particle Hydrodynamics (SPH) and
non-Newtonian flow modelling using SPH.

Joseph Ha is a research scientist at the CSIRO Mathematical and Information Sciences.


He is interested in applying computational methods to solve industrial problems. His research
interests include wave propagation, combustion, turbulence, metal forming, two-phase flow and
smoothed particle hydrodynamics.

Nick Stokes is an emeritus chief research scientist at CSIRO Mathematical and information
sciences. He has ongoing interests in algorithm development for SPH and DEM.

Craig Scott is a Computational Scientist at CSIRO Mathematical and Information Sciences,


Australia. His research interests include the development of object oriented software frameworks,
tools and algorithms for high performance computing, with a focus on particle-based simulations.

1 Introduction approximations. The other is based on kernel based


interpolation such as smoothed particle hydrodynamics.
Mesh free methods have attracted much attention recently.
SPH is commonly described as a ‘particle’ method
Two distinct directions are followed by these methods.
for modelling coupled fluid flows, solid structure
One is based on field approximations such as radial basis
deformation and heat transfer. The particles represent
functions, element free Galerkin and moving least square

Copyright © 2007 Inderscience Enterprises Ltd.


Smooth particle hydrodynamics: status and future potential 71

‘blobs’ of discretised fluid or solids that move around in at CSIRO Mathematical and Information Sciences for
response to the fluid or solid stresses produced by the the past 15 years. Industrial applications are typically
interaction with other particles. Importantly, SPH does not incompressible, partially enclosed but with critically
use any fixed grids or meshes to track the fluid or important free surface behaviour including wave
calculate the fluid velocities. Formally, SPH is a Lagrangian propagation, fragmentation and splashing. They are also
continuum method for solving systems of partial differential often coupled to one or more other physical processes
equations. The fluid (or solid) is discretised and the (such as multi-phase, heat transfer, solidification, curing,
properties of each of these fluid/solid elements are attributed porous media and MHD). SPH is ideally suited for solving
to their centres, which are then interpreted as particles. SPH many of these problems and often has intrinsic advantages
uses an interpolation kernel to smooth the values of any over traditional grid based methods. Validation is more
information held by the particles giving smooth continuous difficult for complex industrial fluid flow applications and
interpolated fields (e.g., smooth density or pressure fields our validation efforts have been focused on ingot casting,
from the discrete values of density and pressure at the high pressure and low pressure die casting. In this paper, we
particles). These fields are used in the solution of the briefly review the SPH method (as used by us) and then
appropriate governing equations. present recent results from several challenging industrial
SPH was originally developed in the 1970s to solve problems, highlighting the advantages of SPH. Finally, we
compressible astrophysical problems (see Gingold and will comment on future applications and areas where we see
Monaghan, 1977). The method has been developed SPH having strong potential for significantly extending the
extensively over the past three decades to solve a broad boundaries of what is currently possible with conventional
array of such astrophysical problems (for details see a recent CFD simulation.
review article by Monaghan (2005)) and has become one of
the standard techniques for modelling astrophysical fluid
flow (e.g., Evrard, 1988; Hernquist and Katz, 1989; 2 SPH model description
Couchman et al., 1995). The method has also been extended
A brief summary of the SPH method is presented here.
to incompressible enclosed flows (Monaghan, 1994; Cleary
For more comprehensive details refer to Monaghan (1992).
and Monaghan, 1993; Morris et al., 1997; Cummins and
The interpolated value of a function A at any position r can
Rudman, 1999). Examples of application include heat
be expressed using SPH smoothing as:
conduction (Cleary and Monaghan, 1999), natural
convection in a cavity and Rayleigh-Benard convective Ab
instability (Cleary, 1998). Other applications involving free A(r ) = ∑ mb W (r − rb , h). (1)
b ρb
surface flows have also been solved using the SPH method.
These applications include a dam and the generation where mb and rb are the mass and density of particle b and
of a wave in two dimensions in Monaghan (1994), high the sum is over all particles b within a radius 2h of r. Here
pressure die casting, gravity die-casting and ingot casting W(r, h) is a C2 spline based interpolation or smoothing
(see Cleary, 1997; Cleary et al., 2000, 2002, 2003; Cleary kernel with radius 2h that approximates the shape of a
and Ha, 2000, 2001, 2003; Ha et al., 1999; Grandfield et al., Gaussian function. The gradient of the function A is given
2003; Prakash et al., 2005a). by differentiating the interpolation Equation (1) to give:
The flexibility and Lagrangian nature of SPH has
Ab
been exploited successfully to model a wide range of ∇A(r ) = ∑ mb ∇W (r − rb , h). (2)
problems. These include viscoelastic (Ellero et al., 2002) b ρb
and elastoplastic flow (Chen, et al., 2001; Gray et al., 2001),
Using these interpolation formulae and suitable finite
magnetohydrodynamics (Meglicki, 1994; Price and
difference approximations for second order derivatives, one
Monaghan, 2004a; Price and Monaghan, 2004b), explosion
is able to convert parabolic partial differential equations
(Liu et al., 2003a, 2003b), powder flow (Sugino and
into ordinary differential equations for the motion of the
Yuu, 2002), green water (Gomez-Gesteira et al., 2005),
particles and the rates of change of their properties.
solidification (Monaghan et al. 2005), turbulence
The SPH continuity equation, taken from Monaghan
(Monaghan, 2002) and solid fragmentation (Libersky et al.,
(1992, 1994), is:
1997; Parshikov et al., 2000; Rabczuk and Eibl, 2003).
Controlled but simplified small scale experiments are dρ a
often performed in laboratories and this data can be used = ∑ mb ( v a − v b ) ⋅ ∇Wab (3)
dt b
for careful and rigorous validation. Examples of such
validation for SPH include gravity currents on a ramp by where ρa is the density of particle a with velocity va and mb
Monaghan et al. (1999), solitary waves breaking on a beach is the mass of particle b. We denote the position vector
by Monaghan and Kos (1999) and the Scott Russell’s wave from particle b to particle a by rab = ra – rb and let
generator by Monaghan and Kos (2000). See Monaghan Wab = W(rab, h) be the interpolation kernel with smoothing
(2005) for references to many other validation examples. length h evaluated for the distance |rab|. This form of the
The SPH method and its application to solving continuity equation is Galilean invariant (since the positions
industrial problems has been a key focus of the CFD group and velocities appear only as differences), has good
72 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

numerical conservation properties and is not affected by free dH a 4mb ka kb r ⋅∇ W


=∑ Tab ab2 a 2 ab2
surfaces or density discontinuities. The use of this form of dt b ρ a ρ b k a + kb (rab + η )
the continuity equation is very important for predicting free (7)
2mbξ µ a µb r ⋅∇ W
surface flows. −∑ (ν ab + rab ) 2 ab2 a 2 ab2
As two particles approach each other, their relative b ρ a ρ b ( µ a + µb ) (rab + η )
velocity is negative (as is the gradient of the kernel) so that where the first summation is the heat conduction term and
there is a positive contribution to dρa/dt. If this rate of the second summation is the viscous heating. The enthalpy
change is positive then the density of particle a rises, per unit mass is defined by:
leading to a positive pressure that pushes the particles
T
apart again. If two particles move apart, then their densities
H = ∫ c p (θ )dθ + L[1 − f s (T )] (8)
decrease creating a negative pressure that pulls the particles
0
back towards each other. This interplay of velocity and
density/pressure ensures that the particle remains ‘on where cp is the temperature dependent specific heat, L is
average’ equally spaced and that the density is close to the latent heat and fs(T) is the volume fraction of material
uniform, so that the fluid is close to incompressible. that is solid at temperature T, kb is the conductivity, and
The momentum equation when converted to SPH form Tab = Ta – Tb.
becomes the acceleration for each particle a: Equation (7) has an explicit conductivity which can be
temperature dependent and ensures that heat flux is
dv a  P P  automatically continuous across material interfaces, such as
= g − ∑ mb  b2 + a2  between a die and a liquid metal. This allows multiple
dt b  ρb ρ a 
(4) materials with substantially different conductivities and
ξ 4µ a µb v ab rab  specific heats to be accurately simulated.
−  ∇ aWab
ρ a ρb ( µ a + µb ) rab2 + η 2  To simulate confined or partially confined fluid flow,
such as is typically found in industrial and geophysical fluid
where Pa and µa are pressure and viscosity of particle a and flows, the modelling of physical boundaries is important.
vab = va – vb. Here ξ is a factor associated with the viscous Boundaries in SPH can be modelled in a range of ways, but
term; Cleary (1996), η is a small parameter used to smooth here we use approximately evenly spaced SPH particles at
out the singularity at rab = 0 and g is gravity. which a Lennard-Jones type force (a very steep polynomial
Since the SPH method used here is quasi-compressible variation of force with displacement) is applied in the
one needs to use an equation of state, giving the normal direction. In the tangential direction, the particles are
relationship between particle density and fluid pressure. included in the summation for the shear stress using
A suitable one is: Equation (4) (with the pressure gradient term set to zero) to
give non-slip boundary conditions on the walls. Details are
 ρ γ  contained in Cleary and Monaghan (1993). Boundary
P = P0   − 1 (5) particles are created by generating triangular meshes
 ρ 0  
from either Computer Aided Design (CAD) models
of the geometry for industrial applications or from
where P0 is the magnitude of the pressure and ρ0 is the
topographic surfaces in the form of Digital Terrain
reference density. For water or molten liquids we use γ = 7.
Models (DTM) for geophysical and coastal engineering
This pressure is then used in the SPH momentum
applications. This approach is flexible and powerful and
Equation (4) to give the particle motion. The pressure scale
with some care, allows good representation of nearly any
factor P0 is given by:
boundary surfaces.
γ P0
= 100 V 2 = cs2 (6)
ρ0 3 Free surface flows
where V is the characteristic or maximum fluid velocity. In this section, we will consider one interesting application
This ensures that the density variation is less than 1% and of SPH. It is recognised that the animation and rendering of
the flow can be regarded as incompressible. fluid is a challenging problem in computer graphics.
The SPH heat equation is based on the internal energy To obtain photo-realism in the animation of fluid,
one developed in Cleary and Monaghan (1999), but Computational Fluid Dynamics (CFD) is increasingly used
modified to use an enthalpy formulation for solidifying to provide the required degree of realism. Realistic
metals (see Cleary et al., 1998) and including contributions animation of fluid using SPH consists of three parts;
from viscous heating. numerical simulation of the physical flow by SPH,
Smooth particle hydrodynamics: status and future potential 73

generation of high quality surface meshes with specific The simulation was carried out using a particle size of
required properties from the SPH particle locations; and 15 mm. At the start of the simulation, there were 331,860
rendering of the images (typically using a ray tracing boundary particles. This number increases by 154,363 for
method, which in this case was POVRAY). Three examples every second of simulation time. By 8.48 s of simulation
of our recent work in this area are presented below; water time, the total number of particles was 1,639,950 making it
flooding a stairway, water flooding a corridor and the a large simulation. The computation was carried out on a
pouring water into a glass. 3199 MHz Intel Xeon CPU with 4126 MB of memory and
had a run time of 558 hours. There are 418 frames in the
3.1 Water flooding a stairway animation up to 8.48 s of simulation time. The simulation
time between consecutive frames is 0.02 s. The total CPU
The dimension of the stairway is 1.8 m wide, 3.4 m deep time used for surface mesh generation was 328 hours on a
and 4.4 m high. The top of the stairs is connected to a 1300 MHz Itanium 2 CPU. The rendering of the animation
corridor which is 5 m long, 0.5 m wide and 1.6 m high. frames with four views was 356 hours on an Itanium 2.
The fluid viscosity used was 0.01 Pa s. Water is injected
from one end of the corridor at 4 m/s to simulate the source
3.2 Water flooding a T junction corridor
of flooding. The water initially floods the corridor. When
the water reaches the entrance to the stairway, some of it The second example involves the flooding of a T junction
flows on down the corridor and some of it surges down into corridor. The viewer is located at the base of the T looking
the stairwell. When the water hits the end of the corridor, up towards the two side branches at the top of the T.
it is deflected upwards and back in the direction of the The two side branches are 2.5 m high, 1.0 m wide and 2.0 m
source in the form or a breaking wave. The two opposing long whereas the central corridor is 2.5 m high 3.0 m wide
streams collide near entrance to the stairway and create a and 6.0 m long. The fluid viscosity was 0.01 Pa s. A wall of
crest. Initially the crest is very large and fast moving but water is generated in each of the side corridor moving at
decreases in size and becomes almost stationary towards the high speed towards the middle of the T. The simulation is
end of the simulation. Figure 1 shows six views of the scene carried out using 50 mm particles. The number of particles
at two different stages of the flow into the stairwell. increases over time, with 800,000 particles present after 5 s
of simulation time. This took about 12 days on a single
Figure 1 Four views of water flooding a stairway at two processor 667 MHz alpha workstation.
different stages of flooding Figure 2 shows the flooding process. Each of the surging
walls of water break forming two waves, one near the
ground and one high in the air. These are symmetric and
collide in the middle of the far end of the T, as seen in the
first frame of Figure 2. The colliding waves propel water
down the central corridor towards the viewer. The collision
of the upper waves generates a hollow swirling tube of
water that jets along the corridor, quickly collapsing into the
main flood surging along the floor.

Figure 2 Progressive flooding of a corridor


74 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

There is substantial splashing near the junction leading to a (or mini-bore) and flows back. The flow on the base is
choppy and very agitated free surface of water flowing highly three dimensional with fluid flowing back
towards the viewer (third frame). The water then reaches the predominantly around the sides of the base of the glass.
solid wall at the base of the T and can be seen rising up to When the return flow reaches the original jet of fluid,
form a breaking wave that then surges back along the the resulting collision generates a surge of waves across the
corridor. At this point the viewer becomes submerged by the free surface leading to the formation and ejection of a large
rising flood waters. droplet from the water in the middle of the base of the glass.
In the final frame, the water level can be seen rising,
3.3 Pouring fluid into a glass but the flow is still not quiescent with significant internal
recirculation generated by the fast flow of the thin sheet of
The third example is a quasi-benchmark type of problem in water down the side of the glass under the free surface and
the digital content industry. It is the filling of a glass with a across the base of the glass, then welling up at the far side
liquid such as water or milk. The glass is 60 mm high and and generating almost standing waves on the free surface.
56 m in diameter and is filled by an obliquely oriented jet As the glass fills, this effect steadily weakens.
(of diameter 10 mm) generated above and to the side of the
glass and initially moving at 0.6 m/s. The fluid viscosity
3.4 Geophysical flows
used was 0.01 Pa s. The simulation used 0.5 mm particles.
The number of particles reached 300,000 after 2.25 s of Many problems in environmental sciences involve the
simulation time. large-scale movement of solids and fluids. They are often
Figure 3 shows some snapshots from the filling of the irregular in their timing, difficult to observe and measure,
glass. In the first frame the jet can be seen accelerating and involve multiple types of physical process on a broad range
thinning as it moves on a parabolic trajectory and collides of spatial and temporal scales and can be catastrophic in
with the side of the glass about half way up. The water their consequences. Computational modelling can play an
spreads out into a fan like sheet around the curved surface important role both in helping to understand the nature of
of the glass. The divergent sheet is unstable and breaks up the fundamental processes involved, and in predicting the
with an identifiable ligament structure consisting of rope detailed outcomes of various types of event in specific
like bands of fluid with gaps in between. These appear to be locations. Environmental fluid and particulate flows can be
a wave like instability with the ligaments of fluid being at characterised as large scale and transient, with complex free
equal radial distances from the contact point with the glass. surface behaviour upon geometrically complex topography.
This means that the ligament structure has a circular arc They typically also involve multiple types of physics on
shape, being lower directly below the incidence point of the multiple scales with significant convective transport.
jet and higher on the sides. The fragmentation pattern of the Particle based modelling methods have specific advantages
water is very transient. over traditional grid or mesh based continuum methods for
these types of problems. These include highly accurate and
Figure 3 Pouring fluid into a glass non-diffusive prediction of complex free surface behaviour
including wave motion, fragmentation and splashing;
accurate and automatic convection of material; and the
straight-forward inclusion of multi-scale multi-physics.

3.5 Flow after a dam break


One geophysical flow that can be simulated using SPH is a
dam-break. Typically such simulations are carried out with
two dimensional shallow water equations using a finite
difference method (e.g., Frazao et al., 1999; Zoppou and
Roberts, 2000). With these approximations, it becomes
impossible to resolve the detailed three dimensional features
of the flow, which can be important in many dam-break
situations, particularly when the topography is complex.
This is a traditional test problem for SPH and generally is
used also in two dimensions (but now in a vertical rather
than horizontal plane) to examine the flow of an initial
square of water along a horizontal flat plane (Monaghan,
1994) With SPH it is now easy to resolve the full three
dimensional flow features of a dam-break using real
In the second frame, the ligament structure in the spreading topography from DTM resulting in more accurate
fan of fluid is completely established. The fluid flows over predictions of propagation time of the flood wave and the
the bottom of the glass as thin sheet. This reaches the far maximum height to which the water reaches the banks
wall, is deflected up and falls back as a breaking wave (Prakash et al., 2001).
Smooth particle hydrodynamics: status and future potential 75

3.6 Tsunami inundation of the coastline The tsunami wave was generated by a wave-maker and has
initial height of 40 m and a speed of 54 km/h. This is a 3D
A tsunami is a series of waves generated by an impulsive
analogy of the 2D wave-maker simulated in Monaghan
disturbance that vertically displaces the water column.
(1994). Figure 4 shows the key stages of the tsunami with
Earthquakes, landslides, volcanic eruptions and explosions
initial inundation of the nearby coastline, later inundation of
can generate tsunamis. Tsunamis can savagely attack
the more remote coast, early receding of the floodwaters
coastlines, causing devastating property damage and loss of
close by and the significant flow of water up along the main
life, as evidenced in the Indian Ocean tsunami of December
protected valley. By 120 s, (Fig. 4(e)), the surge of water in
2004. The extent of damage depends on the strength of the
the main valley has continued inland along both forks and
wave generated by the tsunami. Therefore, the prediction of
can now be seen passing beyond the top of the viewing
these waves can prove to be extremely useful in minimising
frame. This represents a travel distance inland in excess of
the loss of life and property.
2 km along a valley that was almost completely protected
Simulations of a single tsunami wave were carried out
from the original incident wave by the large mountain spur
using SPH on a coastal section of a real topography
in between. This type of behaviour was observed in the
(Cleary and Prakash, 2004), as shown in Figure 4. Fluid
Indian Ocean 2004 tsunami in regions where the bathymetry
particles (the blue region in the foreground) were set up on a
structure leads to focusing of the waves into such regions.
portion of the topography to represent the ocean. The ocean
bed slopes away from the coast with the depth of water
increasing with distance from the coast. The interpolation 3.7 Volcanic lava flow
length for the particles was fixed at 13.0 m with the particle Although most volcanic lava flows do not result in loss of
spacing being 10 m (on average). The total number of fluid human life, they can potentially cause enormous damage to
particles used in the simulation was around 140,000. property. Lava flows can bury homes and agricultural land
under tens of meters of hardened black rock. People are
Figure 4 Tsunami inundation of a coastline
rarely able to use land buried by lava flows or sell it for
more than a small fraction of its previous worth. Typical
examples of lava flows are from the Kilauea and Mauna Loa
Volcanoes in Hawaii. The speed at which lava moves across
the ground depends on several factors, including
• type of lava erupting and its viscosity
• steepness of the ground over which it travels
• whether the lava flows as a broad sheet, through
a confined channel, or down a lava tube
• the rate of lava production
• its rate of cooling/solidification.
Numerical simulation is again a useful tool with which to
evaluate various scenarios. Earlier simulation efforts have
used grid based methods such as the finite volume method
as in Itoh et al. (2000) or the finite element method as in
Klingelhöfer et al. (1999), using either a 2D model or a
simplified 3D model. Gray and Monaghan (2004) used SPH
to investigate the cracking of magma chambers. Here, we
present a 3D simulation using SPH on real topography.
The thermal model includes heat conduction within
the lava, conductive heat losses to the ground and radiative
heat losses from the exposed free surface of the lava. The
solidification model uses a temperature dependent viscosity
and is fully coupled to the fluid flow. See Cleary and
Prakash (2004) for further details. The variable viscosity
model of Klingelhöfer et al. (1999) was used to give the
variation of viscosity with temperature, with properties
chosen to match Hawaiian basaltic lava. The eruption
temperature of the lava was fixed at 1500 K and the ground
temperature was fixed at 300 K. At 1500 K, the viscosity of
the lava is then about 20 Pa s. Even a modest decrease
in temperature to about 1200 K causes the viscosity to
76 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

increase to about 860 Pa s, which significantly affects its Soon after the start (Fig. 5(a)) the lava spurts out of the
ability to flow. crater at a high speed and forms a circular dome of fluid that
Figure 5 shows the progression of the lava flow from the rises up and begins to fall back to the ground due to gravity.
time when the molten lava is first ejected from a crater near At this stage there is very little cooling of the lava which
the coast. The lava colour shows its temperature with red remains around 1500 K. In Figure 5(b), the lava has
indicating high temperature and low viscosity (starting at reached ground level, spreads across the topography and
about 1500 K and 20 Pa s) while black indicates a lower collects in the nearby valleys of the mountainous terrain.
temperature of around 800 K with a high viscosity. The At this stage the lava is still very hot and free flowing and
ground is shaded by altitude so that the topography can be has lost little of its momentum. In Figure 5(c), the lava has
seen, with black being high altitude and dark grey being low substantially filled two of the nearby valleys and the leading
altitude. The surrounding ‘sea’ is coloured blue. edges of the lava are approaching the sea. The speed of the
lava front has started to decline due to the cooling and the
Figure 5 Front view of lava flow with variable viscosity model. consequent increase in viscosity. Lava has started to collect
Dark grey indicates high temperature and low in a third, more distant and deeper valley on the right.
viscosity; black indicates a lower temperature and high
Figure 5(d) shows the progression of the lava into the ‘sea’.
viscosity
The shading of the lava in the last third of its passage to the
sea is darker, illustrating the cooling of the material. Much
of the lava is close to black when it is deposited in the ‘sea’,
indicating that its temperature is approximately 800 K and it
is close to solidifying. As the lava solidifies and its velocity
decreases, fresh lava from behind catches up to the leading
front and flows over these earlier layers leading to a
thickening of the lava sheets.

4 Die casting
The following section presents some examples of our
application of the SPH method to the simulation of
industrial flow problems in die casting which has
been a significant area of focus for our SPH development.
Three specific die casting processes are presented – gravity
die casting, high pressure die casting and ingot casting.
These applications illustrate the particular advantages of
SPH over grid based methods. These include complex free
surface and material interfacial behaviour are modelled
easily and naturally; there are no constraints imposed
on the geometry of system; and the physical equations
(such as solidification) can be applied to individual
particles. In particular, high pressure die casting involves
momentum-dominated flows with highly complex free
surface behaviour which are extremely difficult to simulate
with conventional numerical methods.

4.1 Gravity Die Casting (GDC)


Gravity Die Casting (GDC) processes are capable of making
complicated high integrity components, such as wheels,
cylinder heads, engine blocks and brake callipers, at lower
cost than most other casting methods. Cycle times for GDC
are shorter than for the sand casting process leading to
larger quantities of castings produced per unit time. Surface
finish and internal quality (particularly pertaining to
porosity) are also better using GDC. Improvements
to both product quality and process productivity can be
brought about through improved die design. These include
developing more effective control of the die filling and
thermal performance.
Smooth particle hydrodynamics: status and future potential 77

Ha et al. (1999) compared isothermal SPH simulations Figure 6 Flow pattern showing overflow of metal from the top
with water analogue modelling of gravity die casting for a half of the spout for second stage optimised wheel
design at 30 tonnes/hr. Top frame shows fluid coloured
complex die in two different orientations. The SPH
by velocity and the bottom frame at a later time is
simulations were found to be able to very well capture the coloured by oxide content
free surface wave behaviour, jet oscillation and the fine
details of the flow. Ha et al. (2000) then compared SPH and
MAGMAsoft simulation results of GDC with experimental
results from X-ray imaging. The X-ray imaging technique
was successfully developed to visualise the flow of molten
aluminum in a predominantly 2D steel die. The SPH
simulation results were again seen to capture much more
closely the fine detail of the free surface motion, including
plume shape and the correct relative heights of the surface
levels in the die cavity. Both these validation tests indicated
that the SPH predictions were of high accuracy and this
gives us confidence when applying the method to the other
die casting processes.

4.2 Ingot casting


This process is used to turn the stream of primary
aluminium from a smelter into transportable ingots
with the best quality (as judged by their appearance) at the
lowest possible cost. Typically, this involved the flow of
aluminium from a launder into a rotating wheel which has
external nozzles that dispense measured volumes of liquid
aluminium into ingot moulds that move at constant speed in
a line below the wheel caster. 4.3 High Pressure Die Casting (HPDC)
Significant performance improvements have been made
using a new wheel filling system that resulted from a High Pressure Die Casting (HPDC) is an important method
combination of SPH modelling and pilot scale testing. for manufacturing high volume, low cost metal components.
This work is reported in more detail in Prakash et al. Molten metal is injected at both high speed (30–100 m/s)
(2005a). The SPH modelling was used as the primary design and under high pressure through complex gate and runner
tool, passing through several design iterations to understand systems and into the die. In order to obtain homogeneous
the flow fundamentals and to progressively improve cast components with minimal porosity and void formation
performance. The best concept from the simulation design one needs to have a good die design with accurate
stage was then refined through a detailed program of pilot positioning of air vents which enable the release of most of
scale testing and a final round of SPH simulation. The new the entrapped air. The position of the air vents, in, turn
ingot casting wheel design is now able to operate at 50% depends on the gating system used and the geometry of the
higher throughput with an expected reduction in the oxide die. Numerical simulation offers a powerful and cost
content by about 53% compared to that of existing designs. effective way to study the effectiveness of different die
Figure 6 shows two snapshots of the flow of aluminium designs and filling processes.
through the new casting wheel. The line of ingot moulds are A key advantage that SPH has over other competing
moving to the right and the wheel is rotating clockwise. methods for modelling some industrial heat and mass flows
Aluminium flows from the launder located behind the wheel is that complex free surface and material interface
(which is made transparent so that the launder area is behaviour, including fragmentation, can be modelled easily
visible) into the wheel, through the innovative nozzles and and naturally with low resolutions and produce accurate
into the dies. This design minimises oxide formation by predictions with no mass lost or numerical diffusion of the
minimising splashing, minimising free surface exposure to interfaces. The applications of SPH to 2D HPDC were
the air and imparting as little momentum to the aluminium reported by Cleary et al. (2000) and Ha and Cleary (2000).
as it enters the ingot mould as possible. These SPH results compared favourably with water
78 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

analogue experiments of Schmid and Klein (1995). Figure 8 Filling of an engine rocker cover by HPDC
The applications of SPH to 3D HPDC were reported by
Cleary and Ha (2000, 2001, 2003), Cleary et al. (2002), and
Ha et al. (2004). Ha et al. (2003) compared the 3D SPH and
MAGMAsoft simulation results of an automotive piston
head with water analogue experimental results. The
Magmasoft predictions were at most only qualitatively
indicative, whereas the SPH predictions showed good
agreeemnt with the experiments, particularly in predicting
the flow of liquid into different parts of the die, the relative
strengths of jets and the time at which back filling flow
trapped air in pockets within the die. The SPH results were
better able to capture the fine detail of the fluid motion and
splashing, particularly the relative rates of flow around
sharp bends and through thin sections.

4.3.1 Rocker cover


Figure 7 shows the casting configuration for an engine
rocker cover. It is about 750 mm × 250 mm in area with
3 mm section thickness. A tangential runner system feeds
fluid through a long 2 mm high gate at a speed of 12 m/s.
The particle size used for this simulation was 1.25 mm.

Figure 7 Geometry of a rocker over

4.3.2 Cross member


Figure 9 shows the die and Figure 10 shows the filling of a
structural cross bar support. This part consists of several
sections with strengthening ribs and various cut outs. There
are nine gates distributed along the leading edge of the
Figure 8 shows the filling of this rocker cover. The long component and these are fed by a tangential runner system
gate has an area that is similar to the runner cross sectional with shock absorbers at the ends. The gate height is 2.6 mm
area, so the fluid experiences little accelaration in the gate and the particle size used for this simulation was 1.7 mm.
region. This results in little immediate splashing and The fluid sprays out from the gate along clear preferential
fragmentation of the fluid upon entry into the die. However, pathways, leaving regions with long enduring voids on the
the complex stepped contours of the die lead to long lived sides of the central body where insufficient fluid can be
void regions. This is essentially the result of the diverging supplied by this runner system. By 65 ms, the leading fluid
flow from the gate being unable to maintain a uniform front, reaches the fair side of the die and much of the structure on
so fluid is not able to flow into the more protected regions the near side of the die is substantially filled. The final area
of the die produced by the steps in the shape. Note that there to fill is in the middle of the far side of the die. This is a
is also clear evidence of preferential race tracking of fluid structurally critical location where porosity is highly
around the perimeter of the rocker cover. By 250 ms, undesirable. The simulation shows that, despite the presence
the die is substantially filled with critical voids remaining of a large central bridge which is intended to channel fluid
in two locations, one just to the side of the gate as seen into this region much earlier, the region is difficult to fill
in the fourth frame and one on the far side of the casting and could be subject to cold shuts, joining problems where
where recirculating fluid has created a long lived roughly the fronts of fluid meet from either side and from trapped
elliptical void. porosity.
Smooth particle hydrodynamics: status and future potential 79

Figure 9 Geometry of cross member The fluid in the blue slow moving region has a low void
fraction. The tower structure is substantially filled as is the
base plate on the opposite side. The last remaining area to
be filled is the central void region closest to the runner
system.

Figure 11 Filling of differential cover during HPDC

Figure 10 Filling of structural cross member by HPDC

4.3.4 Coupled solidification and flow

4.3.3 Differential cover The previous simulations were isothermal and the large cast
volume ensured that there was little solidification and
Here, the filling of an automotive differential cover is therefore viscosity variation throughout the filling. In the
presented. The differential cover is about 250 × 250 mm in next example, which is small and this, the solidification is
area and has a section thickness of about 6.5 mm. Liquid crucially important. The filling simulation was therefore
aluminium is fed into the die cavity through the curved performed using a fully coupled fluid and heat flow model.
gates that are 1.5 mm high. The particle size used in this Conduction was predicted in both the solidifying liquid
SPH simulation is 0.75 mm. When the cavity is completely metal and into and through the die. Temperature dependent
filled, the total number of particles is about 900,000. material properties, particularly viscosity and the release of
Figure 11 shows the filling pattern. Aluminium enters latent heat were included.
the die cavity through the four gates attached to a The item to be cast is a simple coaster which was
conventional tangential runner system with shock absorbers. 9 × 9 cm and 2 mm in thickness. A tangential runner with a
After entry into the die cavity, two of the streams quickly diamond shaped insert in the middle was used. The gate was
merge leaving three main streams moving directly away 0.9 mm thick and a particle size of 0.6 mm was used.
from the gate following the contours of the part and with Figure 12 shows the fluid temperature of the metal flow
long lasting voids in between. At 20 ms, some leading during filling. All the fluid is shown on the right side of the
fragments of liquid begin to collect in the circular structure die, so the colouring shows the surface properties. On the
on the right of the die (when viewed from above). left, only the material in the bottom half of the die is shown.
Significant amounts of fragmented fluid are now seen to The temperature shown on the top left surface therefore
collect in the raised ‘tower’ on the left side of the die. belongs to fluid in the mid-plane of the die. Surrounding
By 50 ms there has been significant back filling and the rear the fluid shown here is the solid die, but it is not shown.
third of the die is completely filled. The transition region For the coupled thermal flow, heat conduction from the
between the fast flowing streams moving away from the liquid metal into the die is critical since it determines the
gate and the back filling fluid occurs where the fluid colour solidification and therefore viscosity of the fluid. This has a
(representing speed) abruptly changes from yellow to blue. major impact on the predicted fill pattern.
80 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

Figure 12 Die cast filling of a coaster with fluid coloured metal on the sides of the die leads to a back filling flow
by temperature using a coupled flow and towards the gate. At 60 ms there are four distinct unfilled
solidification model
areas; a large section directly in front of the gate, one near
each of the side walls towards the end of the die and one
along the centreline of the die directly adjacent to the end
wall. At 80 ms, the back filling flow has closed the large
void region adjacent to the gate, but the increasing viscosity
of the solidifying metal makes it difficult for the fluid
to flow into the last two remaining large scale voids.
Eventually they do fill in this case, but this casting was
close to being a cold shut.

5 Resin transfer moulding for composite


manufacture
There exists a high degree of commonality of issues
between injection moulding, RTM and other composite
manufacturing processes. Similar techniques can also be
used for plastic injection moulding, wax casting for creating
moulds for investment casting. Many of the issues are also
common to the just discussed die casting areas.
The key physics to be modelled include fluid flow,
curing of the polymer, heat transfer in the resin, the carbon
fibre preform and the mould, the exothermic nature of the
curing reaction and the effect of the carbon fibre porous
media on the flow. In RTM, the viscosity changes are
complex, since the viscosity decreases with increasing
temperature but increases with the state of cure. The cure
rate increases with temperature so a higher operating
temperature leads to lower initial viscosity (and therefore
easier flow through the preform) but higher viscosity later
(leading to more difficulty in fill). It is imperative that the
part be filled before the viscosity increases to such an extent
that the resin to flow into some sections is prevented,
leaving exposed fibre cloth.
From a modelling perspective, the key challenges are;
the discontinuity of the edge of the carbon fibre where the
domain changes from low permeability porous media to free
space (which causes significant hardship for pressure
solvers for grid based CFD methods); and the extremely
long timescales involved in the filling (hours). SPH has
been successfully used to solve this challenging problem.
Figure 13 shows an RTM experiment by Young and Lai
(1997) that has been used for validation of RTM prediction
capability. In this case, the carbon fibre preform in packed
Once the tangential runner fills and pressurises, fluid is into a thin flat mould. The domain is 80 mm high. A small
sprayed out into the die on 45 degree angled trajectories gap is left at the top (of the order of 1–4 mm) and oil is
towards the side walls. Liquid metal builds up along the side introduced under high pressure into the free space to the left
walls slowing as it cools and becomes more viscous. When of the porous media. The oil is able to preferentially flow
the metal makes contact with the colder top and bottom along the small gap leading to a race tracking effect. Oil
walls of the die, the cooling and solidification accelerate. diffuses back into the perform from the filled part of the gap
The central parts of the die cavity are filled with a fairly leading to a very characteristic free surface shape and
sparse but highly fragmented hot liquid metal. Along the defined rate of propagation. Three snapshots of this race
centreline of the die, splashing liquid metal from either side tracking flow, as predicted by SPH, are shown in Figure 13.
collide and stick to the top and bottom walls forming a The free surface shape and speed closely resembles that
fragmented line of prematurely solid metal. The build up of observed in the experiments.
Smooth particle hydrodynamics: status and future potential 81

Figure 13 Progress of oil through the porous media preform, This sucks solid pellets down towards the impeller.
showing the strong race tracking observed in the By 1.5 s, the pellets have started to cluster near the centre of
experiments at times t = 0.8, 1.4 and 2.1 s
the tank as a result of impeller starting to generate this
recirculation pattern. At 2.0 s, the pellets start being dragged
down into the fluid. The recirculating flow pattern becomes
fully established around 2.5 s and a significant proportion of
the pellets are being drawn down to the level of the
impeller. At 3.0 s, pellets are being flung outwards by the
impeller and are now being recirculated within the fluid.
This behaviour is in close agreement with the experimental
observations. This comparison is reported in more detail in
Prakash et al. (2005b).
This example clearly demonstrates the potential for
modelling multiphase systems for which the particulate
size is too large to validly use inter-penetrating fluid
approximations for the solid phase. This allows detailed
flow dynamics around each particulate to be well resolved.

Figure 14 Submergence and mixing of solid particulates in a


mixing tank for an impeller speed of 200 rpm using
a loading of 1.5 kg of 16 mm wooden pellets in a tank
of around 1 m diameter and 1 m height

6 Mixing of particulates in liquid


Mixing of particulate solids and liquids is a very important
application area. If the solids are very small (compared to
the domain size) and the particulate loadings are low
(so that the system is dilute or lightly loaded) then this type
of flow can typically be well modelled using good
conventional CFD. If, however, the particulate solids are
large or have a high loading then these methods are not
valid and the particulates need to be individually modelled.
Using conventional methods for such problems is
exceedingly expensive due to the need to re-mesh each
timestep to follow the fluid around the moving solids. SPH
with its Lagrangian nature and without the mesh restriction
is easily able to model such difficult systems.
To demonstrate this, we model a moderate loading of
reasonable size buoyant cylindrical pellets in a mixing tank
full of water and driven by a central impeller. The solid
pellets are represented by collections of SPH particles that
are constrained to move together. The forces on these
particles are determined in the same way as for fluid
particles but with an overriding constraint of rigid-body
motion. The simulations are validated by comparing the
results with physical modelling experiments that shows a
relationship between the submergence criterion of the solids
and parameters such as size of solids, solids loading and
impeller speed. The experiments were conducted using
water as the fluid and cylindrical blocks of wood with a
specific gravity of 0.5 with near identical conditions being
used for the simulations. For further details refer to
Prakash et al. (2005b).
In order to explore the behaviour of pellets at high
impeller speed, a simulation was run for a speed of
200 rpm and the flow predictions are shown in Figure 14.
The impeller generates a strong central downdraft of
water leading to a bulk recirculation flow within the tank.
82 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

7 Pyrometallurgy: gas, fluid and solids in Figure 15. When there is no difference in densities of the
two materials, the pellets are neutrally buoyant and
The use of Computational Fluid Dynamics (CFD) to model
remain in their original positions throughout the simulation
pyrometallurgical processes has traditionally been by grid
(as they should). This is independent of the inclusion of heat
based finite element or finite volume methods. These
transfer. There is very little movement of the pellets due to
simulation tools are quite useful in predicting the overall
convective heat transfer. When the pellets are much less
flow patterns and heat transfer in the molten bath. However,
dense than the surrounding liquid bath, they then float
in many pyrometallurgical processes reactive solids are
rapidly to the surface at speeds dependent on the density
added to the molten bath in the form of pellets or briquettes.
difference. The conductive and convective heat transfer is
Heat is transferred from the bath into the porous solids and
shown by the change in colour of the pellet surfaces
reaction proceeds with possible metal, slag and gaseous
(from blue to light green and towards yellow) as well as that
products being formed depending on the specific chemistry
of the surrounding fluid (from red to yellow in certain
of the system. For example, in the carbothermic reduction of
regions) as the cold pellets heat up and cool the surrounding
iron oxide pellets in slag, molten oxides, carburised iron
fluid. These simple tests show that the method produces
droplets and a carbon monoxide rich gas are formed
qualitatively sensible predictions for pellet motion and
(Street et al., 1998). It is computationally prohibitive to
heating.
simulate the individual pellets using grid-based CFD
methods (Johannessen et al., 2001). For grid-based methods Table 1 Properties for pellets and fluid in Figure 15
one would also need to incorporate complicated mesh
deformations to account for the interaction between the fluid Fluid Pellet
and solid pellets. Due to its grid-free nature SPH has the Conductivity (W/m/K) 53.0 19.0
following advantages over other CFD tools:
Specific heat (J/kg/K) 241.0 1434.0
• free surface flows can be handled easily and accurately Density (kg/m3) 5860.0 5860.0
1860.0
• solid/fluid interactions with complicated motions
of the solid parts can be simulated with few additional
computational penalties Figure 15 Motion of pellets in fluid with coupled heat transfer.
Column 1: Neutrally buoyant and Column 2: Strongly
• complicated physics can be added easily (examples buoyant pellets
include non-Newtonian rheology, solidification,
freezing, solid stress and history dependence of fluid
variables)
Since SPH has these inherent advantages we have extended
the method to include solid reactive pellets in a molten
phase bath for pyrometallurgical simulation. We summarise
the key changes and demonstrate the additional physics
(solids, gas generation and gas transport) using a series of
simple examples.

7.1 Solid pellets in a fluid with buoyancy


Solid pellets are represented by collections of SPH particles
as in the previous section. Buoyancy effects due to the
density difference between the pellets and fluid are
automatically dealt with by SPH. Thermal induced
buoyancy effects are small compared to these and are
omitted. Equations for heat transfer and species transport
are solved using fluid and solid particles with appropriate
material properties. The coupled thermal to flow and
7.2 Gas generation and gas transport in liquid
buoyancy effects described in this section is an extension to The model developed here is intended to apply to
the validated pellet motion described in the last section. reasonable volumes of small bubbles in a liquid.
Figure 15 compares the motion of pellets in the fluid A reasonable representation of such two-phase flow is an
with and without density differences between the fluid and inter-penetrating fluids model. Two separate phases,
pellets and with heat transfer included. Column 1 shows treated as continua, have their own momentum and mass
results when there is no density difference. The liquid conservation laws and interact through a drag force that
bath and pellets are coloured according to the temperature each exerts on the other. This is an extremely common
with the initially cold pellets cooling adjacent hot liquid. approach in traditional Eulerian CFD modelling of
Column 2 shows results when pellets are strongly buoyant. multiphase systems. We now implement this in an SPH
Table 1 gives the properties of the fluid and pellet for cases method. This model is subject to the same limitations of
Smooth particle hydrodynamics: status and future potential 83

adequacy of knowledge of bubble size distribution, bubble recirculation patterns. This results predominantly from the
loading and bubble shape and aggregation as it has in imposed central updraft. When the pellets are less dense
conventional CFD. than the liquid bath (as is the case in the right column of
We use the subscript l for the liquid phase and g for the each time frame), then the recirculation is much less
gas phase. A key unknown is the gas volume fraction ε. noticeable. The buoyant pellets float rapidly and cluster near
We assume the density of the gas is zero so there is no the surface because they are positively buoyant. Note that
acceleration term for the gas momentum equation. the gas motion around and above each pellet creates a
The density of the liquid is ρ. As usual v represents the buoyant plume that tends to entrain fluid which in turn
velocity, while η is the viscous stress in the liquid (assumed pushes the pellet upwards. So the natural buoyancy of the
unaffected by the gas). The model uses the speed of sound pellets is enhanced by the generation of buoyant gas plumes
in liquid, c, the liquid viscosity µ, and the bubble diameter, from the reacting pellets.
d. We also assume that the gas is incompressible since
hydrostatic pressure will cause only small density Figure 16 Gas generation and its transport in two dimensions.
variations. The gas conservation law is then just an Column 1 in each time frame: neutrally buoyant
pellets a and column 2 in each time frame: pellets are
advection equation: less dense than the liquid bath
ε1 + v g ⋅ ∇ε + ε∇ ⋅ v g = 0. (9)

The conservation law for the liquid can be written as:


1 dP
ε l + v l ⋅∇ε = + ∇ ⋅ vl . (10)
ρ c 2 dt
The terms on the left represent the change in gas volume for
which there is an equal change in liquid volume fraction.
The momentum equation for the gas can be simplified to:
H ( v g − v l ) + ∇P = 0 (11)

here H is the inter-phase drag term and P the pressure.


In this preliminary implementation there was no attempt
made to model the effect of bubble size. Instead a constant
drag factor H was assumed for the calculations which gave a
superficial bubble rise velocity of around 1 m/s.
The momentum equation for the liquid can be
written as:
dvl
ρ + ρ v l ⋅ ∇v l = (1 − ε )( ρ g − ∇P ) + ∇ ⋅ τ l . (12)
dt
In order to assess the functionality of the gas generation
model explained above and buoyancy effects, further
simulations were carried out in 2D and 3D. In Figures 16
and 17, results are shown in 2D and 3D for cases neutrally
and positively buoyant pellets. The aim here is to examine Figure 17 shows an analogous rectangular tank in 3D.
gas generation from the pellet at a specified rate and its The colour on the pellet surface again represents the volume
transport through the liquid bath. To mimic the fluid motion fraction of gas. In the liquid bath, the gas distribution is
generated by stirring generally done in pyrometallurgical represented by spheres whose size and colour show
applications to improve mixing, there is a central updraft increasing gas fraction. The larger the sphere, the higher is
generated during these simulations. the local volume fraction of gas. The flow behaviour
In Figure 16, the colour represents the volume fraction predicted in 3D is broadly similar to the 2D result. The two
of gas. It shows the generation of gas from the pellets and columns of Figure 17 show the influence of density on
its transport through the liquid bath. The gas motion through pellet motion with pellets. In column 1 the pellets have a
the liquid bath and the updraft causes pellet motion even lower density than the fluid whilst in column 2 they have
when the pellet and the liquid bath have the same density the same density as the fluid. When the pellets are less
(see the left column of each time frame). The pellet motions dense than the liquid bath, they rise more rapidly to the top
in the left column of each frame show two weak of the bath than for the neutrally buoyant case.
84 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

Figure 17 Gas generation and transport in 3D. Column 1: Pellet Of course, more complex rate relationships are expected for
less dense than liquid bath; Column 2: Pellet and liquid real systems. The liquid bath starts at 1600οC. The gas
bath with same density
motion is by buoyant convection with drag as described in
the previous section. Some approximations concerning
temperature diffusion had to be made because the thermal
boundary layer associated with reaction is very thin and is
averaged out over a particle. The thermal conductivity of the
pellets was 2250 J/m2/K, and their starting temperature was
just below the ignition temperature.
Figure 18 shows the volume fraction of gas from the
chemical reaction (left column) and temperature (right
column). The pellets have different densities with some
being lower, some equal to (neutrally buoyant) and some
higher than the fluid density. As expected, the heavier
pellets sink to the bottom whilst lighter pellets float to
the top of the fluid. Neutrally buoyant pellets also rise up to
the top of the fluid surface albeit at a slower rate due to the
additional buoyancy imparted by the gas plumes generated
from the pellets. The reaction temperature does not vary far
from 1500oC for the first second, with a high influx of heat.
The pellets were purposely given arbitrary shapes.
An observation resulting from the present fairly realistic
implementation is that the gas is preferentially emitted from
protruding parts of the pellet or sections of the pellet that are
more exposed to the surrounding hot fluid. These are the
parts that heat most rapidly. In future work, the pellets
will be allowed to erode rapidly as the ongoing chemical
These observations suggest that SPH is able to qualitatively reactions consume the source materials in the pellet.
predict gas generation from pellets with good representation This behaviour is straight forward to model with a particle
of buoyancy effects and gas transport through the liquid method such as SPH.
bath, including fluid convection generated by gas plumes
rising from the pellets. Figure 18 Gas generation and transport in 3D. Column 1: Pellet
less dense than liquid bath; Column 2: Pellet and liquid
bath with same density
7.3 Chemically reactive pellets
In the previous section gas was generated from the pellets
using a prescribed rule. In this section the implementation of
actual gas generated through a chemical reaction with a
known heat of reaction (∆Hr) is presented. The densities
will be given in moles/litre. The following equations apply
for reaction of the pellet:
∂θ ∂C ∂T
ρg = r, = r , ρ cv = − DHr (13)
∂t ∂t ∂t
where ρg is the density of a gas liberated from the reaction
(mol/l), θ is the fractional gas content of the liquid (vol/vol),
C is the concentration of the product formed from the
reaction (mol/l), r is the rate of progress of the reaction
(mol/sec/m3), ρ is the density of the current material
(for the heat equation) and cv is the specific heat (J/mol/K).
Note that conductivity within the pellet has already been
accounted for in the SPH implementation. We have
assumed the following kinetic dependence of the reaction
to reflect a specific ignition temperature for this
pyrometallurgical system:
0, T < 1500∞C Whilst, the model developed contains many assumptions,
r= . (14)
5 | T − 1500 |, T > 1500∞C the results obtained to date suggest that SPH is likely
Smooth particle hydrodynamics: status and future potential 85

to be a powerful technique for simulating complex Figure 19 Slurry flow in a SAG mill, (top) flow within the
pyrometallurgical systems where simultaneous heat and grinding chamber, and (bottom) flow through the
discharge grate into the pulp lifter chamber
mass transfer with multiple phases (particularly solids) is
important. The next stage of development will see the
incorporation of specific chemistry and a more realistic
treatment of bubble size on convective flow in the system.

8 Slurry flow in sag mills


Particle breakage is an essential component of mineral
processing. The aim is to reduce the particle sizes
from those supplied to the much smaller ones required for
further processing as efficiently as possible, producing
maximum mill throughput and minimising operating
costs. An important piece of grinding equipment is a
Semi-Autogenous Grinding (SAG) mill. These are huge
rotating cylinders (up to 14 m in diameter) rotating typically
at around 80% of the critical centrifuge speed, and which
can grind up to 100,000 tonnes of rock per day by tumbling
the rocks around within the mill. This can consume up to
20 MW of power. Unfortunately, such milling processes are
believed to use less than 5% of the supplied energy for
particle breakage. However, significant opportunities exist
to improve the performance and control of such mills
using information obtained by computational modelling. 9 Future challenges and opportunities
The rocks are best modelled using the Discrete Element SPH has proved to be a versatile and adaptable tool which
Method (DEM) (see Cleary, 2004 for details) but the slurry can be extended to a broad range of physics and enables
within the mill is well modelled using SPH. realistic simulation of many very complex industrial and
Figure 19 shows the flow of slurry within the grinding geophysical flows. But there are still several challenges
chamber of the SAG mill (in the absence of the rock and confronting the method. In this section we discuss these
grinding media). This is a challenging modelling problem issues and note some interesting potential opportunities.
since the speeds are high and the geometry is complex.
The mill rotation means that the slurry is distributed over a 9.1 Turbulence
wide part of the mill shell from the toe (lowest part) to the
shoulder, where the slurry can be seen flowing off the rising SPH is ostensibly a method for simulating laminar fluid
lifter bars. The lifters crashing into the slurry pool at the toe flows, but it has been used very successfully in many fields
generate significant wave motion and splashing. The end (Monaghan, 1992) for simulating high speed turbulent
wall lifters generate significant recirculation against the flows. This occurs because SPH broadly fits within the class
ends walls which could lead to acceleration wear of these of simulation techniques known as Large Eddy Simulations
surfaces. The lower frame of Figure 19 shows the discharge (LES). The spatial interpolation used in SPH (Equation (1))
of slurry through the fine grate that separates the grinding is formally the same as the spatial filtering used to construct
chamber from the pulp lifters that are responsible for LES methods. An LES is a method that resolves all the
pouring the slurry out through the discharge port around the length scales of the fluid motion down to the grid scale and
axis of the mill on the discharge size. This is a very then uses a sub-grid scale model (for mathematical closure)
challenging problem to model because of the very fine to represent the smaller scales that cannot be resolved
geometry of the grate and the complexity of the pulp lifters, directly in the simulation. In the same way, SPH models all
discharge cone, mill shell and of the finely fragmented, fast scales of motion down to the particle resolution and relies
moving fluid flow that is generated. SPH is easily able to on numerical dissipation on the particle scale to act as a
model this system because of both its free surface default (and certainly incorrect) sub-grid scale model whose
capabilities and the absence of advection problems primary purpose is to dissipate energy cascading down to
(because of its Lagrangian nature). the small scale from the larger ones. Without a suitable
86 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

sub-grid scale model and correct transport of energy back pressure from the gas can affect the fluid filling and
between different scales, SPH is not a true turbulence where the trapping of gas is a key defect to be predicted.
method, but it is still often able to make very good
predictions for high speed turbulent flows just by reasonably 9.3 Rheology
representing the of the energy transfer in the larger length
scales (without having the fine details correct). SPH has predominantly been used to model single phase
Very few papers on turbulence modelling in SPH have Newtonian fluids, but there are significant opportunities
been reported in the literature. An SPH version of the alpha to use its Lagrangian and free surface capabilities for
turbulence model devised by Holm (1999) was formulated multi-phase and non-Newtonian flow problems. There has
for compressible flow by Monaghan (2002). Lo and Shao been some usage of SPH to model viscoelastic fluids and a
(2002) demonstrated the use of a Smagorinsky type sub-grid larger amount of work using SPH for elastic, elastoplastic
scale turbulence model for high velocity applications using and brittle elastic materials. There are opportunities to
an incompressible SPH formulation. Violeau et al. (2001) extend these to more complex rheological models.
demonstrated two ways of modelling turbulence in the SPH In particular the ability of SPH particles (which represent
framework using an eddy viscosity assumption as well as a volumes of real material) to store and carry history
Langevin or stochastic approach. A simple Poiseuille flow information (such as shear rate or strain rate history) offers
was used to test their models resulting in good agreement the possibility of simplifying the implementation of many
for velocity profiles. Despite these developments the non-Newtonian models. Of particular importance are
relationship between SPH and turbulence is far from settled, shear thinning or thixotropic rheologies which have broad
with particular questions arising about the ability of energy application from magnesium die casting, to slurries in mills
to correctly move up from the smaller scales to generate to blood flow. An important and early model should be that
coherent mesoscale structures. of a Bingham fluid so that materials with a yield stress can
be represented.
The other rheological area is the use of inter-penetrating
9.2 Gas-Liquid Interaction (large density
fluid models (such as the algebraic slip model) for
differences)
modelling multiphase flows where the particulate are too
The simulation of the interaction of gas and liquids is small to be modelled as discrete particles. This is a rich area
currently a major challenge for the SPH method. In cases in conventional CFD but there are opportunities with
with large density ratios, the question arises as to how the flows of such material where the free surface behaviour is
interface should be simulated. Large density differences complex or the flow speeds are very high. Here the
typically present major challenges for most CFD methods Lagrangian nature of SPH has specific advantages.
placing large strains on the pressure solvers, but this issue is Examples would range from sedimentation processes in
stronger for SPH because it is quasi-compressible. Consider river systems to pyroclastic flows. An example of using
a very simple arrangement of a square box half full of water such a model to examine fine particulate separation in a
with air above. The standard continuity equation smooths slurry using a spiral separator is given in Prakash et al.
the divergence of velocity across particles on both sides of (2005c).
the material interface. A key effect of this smoothing is
that the interface is smoothed out and the sounds waves 9.4 Solidification and phase change
travel easily from one material to the other. The critical
problem is that a 1% amplitude sound wave in a dense Since an SPH particle represents a specific volume of fluid,
liquid, after conserving energy and momentum, has an then the properties and behaviour of this particle can be
amplitude equal to the density ratio once it has travelled into changed in response to changes in its thermodynamic state.
the gas. For density ratios of more than 100 : 1 this leads to Shrinkage during solidification is trivially implemented in
catastrophic explosions in the lighter material as the sounds SPH by changing the reference density ρ0 of the particle in
waves move across the interface. What should happen is the equation of state (5) according to its temperature.
that the sound wave should predominantly reflect from the As a metal cools, its density increases leading to a decreased
free surface and most of the acoustic wave energy should volume for the particle (since its mass is constant).
remain in the reflected wave in the liquid. A small fraction This automatically generates low pressures that then suck in
of the wave energy is transmitted into the gas and continues adjacent particles leading to highly realistic predictions of
on. The current continuity equation smooths the interface shrinkage.
out and gives a 100% transmission of the waves and A second example of the power of SPH in solidification
prevents the required reflections that enable the system to be is the ability to change the governing equation that applies
stable. Development of modifications to SPH to allow to a particle in response to changes in its thermodynamic
solution of such problems would greatly extend the range of state. A liquid metal particle can participate with its
applications to which it can be deployed and would allow neighbours in the solution of the Navier-Stokes equation,
better and more accurate solutions to be calculated for but once its temperature falls below its solidus (freezing)
applications such as high pressure die casting where the temperature, one can freeze the particle in place and no
Smooth particle hydrodynamics: status and future potential 87

longer allow it to move. This is again a very simple 9.7 New applications
operation but enables very realistic simulation of feeding in
Two highly prospective application areas where SPH
low pressure die casting. A further extension of this idea
appears to have significant potential are:
would be to change the governing equation from the
Navier-Stokes equation to either a viscoelastic or elastic • Biomedical problems, including biomedical
solid equation. This allows SPH to grow a solid or devices, blood flow, heart modelling (with complex
semi-solid skin dynamically from a liquid in way that is heart muscle interaction with the flow of a
fully coupled to the fluid flow. This should allow simulation non-Newtonian fluid), tissue engineering, bioreactors,
of growth and interaction of the solid regions of castings joints, cartilage degradation, shock wave and impact
with the fluid flow in a way that enables prediction of on human physiology and many more. These difficult
deformation of components during casting. This would also but important applications have strong requirements
allow prediction, from first principles, of the growth of air for complex free surface motion, fully coupled motion
gaps as the solidifying metal pulls away from the die walls of different objects with different rheologies and
leading sharp decreases in the heat transfer to the die and potentially different governing equations and the
radically affecting the prediction of the subsequent ability to easily track history.
solidification processes. Similar advantages exist for any
other process involving phase change such a curing of • Turbomachinery, gears, rotating machinery.
polymers and plastics. Conventional CFD in using meshes or grids is highly
constrained in the problems that can be solved if
the boundary geometry performs non-trivial motion.
9.5 Fluid-structure interaction There are important special cases such as mixing tanks
Typically, the boundaries used for simulation with SPH are where a slip mesh can be used so that an inner mesh
rigid. However, since SPH can also simulate elastic solids is attached to a central impeller and an outer mesh is
by solving the governing equation for deviatoric stress, attached to the tank, but in general this is not possible.
it can also be used easily and naturally to simulate The motion does not need to be particularly complex
fluid-structure interactions. The difference in the natural in order to make simulation difficult to impossible with
timestep for the fluids and deforming solids coming from a mesh based method. The simulation of the ingot
the respective sound speeds means that the solid is much casting application (Prakash et al., 2005a) was
stiffer and some special treatment may be required to performed using SPH predominantly because
prevent the computational time becoming too large. conventional methods can not handle a combination
The ability to model fluid-structure interactions suggests of one set of objects translating, with one rotating
that SPH could be used to evaluate and optimise the use of and another being stationary. Similarly, the flow of
liquid sloshing for applications such as active control of plasma in a lobe pump was simulated by SPH because
earthquake induced vibration of structures thus minimising of the ease of simulating the movement of the twin
loss of life and property during an earthquake (Guzel et al., impellers (Prakash et al., 2003). The general area of
2005). Other important applications in this area include turbomachinery is large and diverse and has
simulating the interaction of breaking waves on off-shore oil applications in the biotech, environmental and chemical
platforms and floating/tethered production vessels in order industries. SPH has strong advantages in simulating
to understand the nature and location of damage to these these types of systems.
structures leading to better design strategies.

10 Conclusions
9.6 Fluid-particulate interaction
SPH being a Lagrangian meshless method has several
An important area where the natural advantages of SPH
advantages over traditional grid or mesh based CFD
are likely to have a large impact is the modelling of
methods:
fluid-particulate interactions. This could be a pure SPH
implementation as was done with the mixing tank earlier in • Free surfaces are handled easily and naturally.
this paper, or it could involve the coupling of the SPH fluid The particles move to new locations and the shape
flow to Discrete Element Method style solid particles. of the free surface is just the surface of the particles
The key advantage that SPH has here is that it can calculate wherever they happen to be. There is no need for any
a high quality flow solution around the particulates form of explicit surface tracking such as the
(subject as always to adequate spatial resolution) but Volume of Fluid (VOF) method which is required for
avoiding the prohibitive cost of re-meshing that makes grid based methods. This provides an easy and
simulating these systems with Finite Element Methods so highly accurate representation of fluid surface
expensive. Such direct coupling of SPH to DEM for high behaviour, including wave propagation, wave breaking,
integrity fluid-particulate modelling is now not far from splashing and fragmentation. These occur without
being a reality for large scale industrial flows. introducing numerical diffusion at the free surface
88 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

and without a loss of mass. Small fragments Government’s Cooperative Research Centres scheme.
of the fluid retain their mass and momentum on The particulate-fluid mixing and the pyrometallurgy work
ballistic trajectories. In addition, complex topological were conducted within the Light Metals Flagship,
changes in the surface can be easily tracked in a National Research Flagship initiated by CSIRO whose
situations where VOF, level set and other surface support is gratefully acknowledged.
tracking methods for grid based techniques struggle, or
fail completely.
• As it is a Lagrangian method, it does not have the References
non-linear convective term in the momentum equation. Chen, J.K. Beraun, J.E. and Jih, C.J. (2001) ‘A corrective
Therefore SPH is very good at simulating momentum smoothed particle method for transient elastoplastic
dominated flows and avoids issues relating to the need dynamics’, Comp. Mechanics, Vol. 27, pp.177–187.
to use upwinding schemes. Cleary, P. (1996) ‘New implementation of viscosity: tests with
Couette flows’, SPH Technical Note 8, CSIRO DMS,
• SPH is a fully transient method and so is suitable for Technical Report DMS – C 96/32.
environmental flows that are not steady state. Cleary, P.W. (1997) ‘High pressure die casting simulations with
SPH: Isothermal single fluid flow’, Proc. Chemeca 97, Paper
• SPH can easily include complex additional physics FL3B CD Rom, 29th September–1st October, Rotorua, NZ.
(e.g., solidification, freezing, non-Newtonian rheology, Cleary, P.W. (1998) ‘Modelling confined multi-material heat and
solid stress, history dependence of fluid variables, oxide mass flows using SPH’, Applied Mathematical Modelling,
formation, solute and sediment transport). This arises Vol. 22, pp.981–993.
partially from the flexibility of the continuum Cleary, P.W. (2004) ‘Large scale industrial DEM modelling’,
modelling, partially from the fact that each element of Engineering Computations, Vol. 21, pp.69–204.
fluid or solid has a specific identity (i.e., the particle) Cleary, P.W. and Ha, J. (2000) ‘Three dimensional modelling
that stays with the material over time, (thereby of high pressure die casting’, Int J. Cast Metal Research,
enormously simplifying the tracking of history Vol. 12, pp.357–365.
dependent quantities), and also from the ability to Cleary, P.W. and Ha, J. (2001) ‘Modelling the high pressure
impose specific rule bases on the particles that can die casting process using SPH’, Material Forum, Vol. 25,
change the way in which the PDE and other governing pp.1–29.
equations are used. Cleary, P.W. and Ha, J.H. (2003) ‘Three-dimensional SPH
simulation of light metal components’, J. Light Metals,
• Very complex geometries can be handled relatively Vols. 2–3, pp.169–183.
easily, eliminating the need to create complex volume Cleary, P.W., Ha, J. and Ahuja, V. (2000) ‘High pressure die
meshes in which to solve the fluid flow, as is needed casting simulation using smoothed particle hydrodynamics’,
for grid and mesh based methods. This is particularly Int. J. Cast Metals Res, Vol. 12, pp.335–355.
important for large scale transient free surface flows Cleary, P., Ha, J., Alguine, V. and Nguyen, T. (2002) ‘Flow
where the volumes of space that would need to be modelling in casting process’, Applied Mathematical
meshed would be significantly larger than the volume Modelling, Vol. 26, pp.171–190.
of fluid that is of interest. Cleary, P.W., Ha, J., Mooney, J. and Ahuja, V. (1998) ‘Effect of
heat transfer and solidification on high pressure die casting’,
These advantages make SPH very well suited to solving a in Thompson, M.C. and Hourigan, K. (Eds.): Proc. 13th
range of economically important industrial flow problems Australasian Fluid Mechanics Conference, Monash
and also problems in geophysics and civil and coastal University, Melbourne, pp.679–682.
engineering. These advantages are not universal and it is Cleary, P.W., Ha, J., Prakash, M. and Nguyen, T. (2003) ‘SPH:
important to keep in mind that conventional grid/mesh A new way of modelling high pressure die casting’, Proc:
3rd Int. Conf. on CFD in Minerals and Process Industries,
based CFD is able to solve many traditional industrial flow
pp.421–426.
problems more easily and more rapidly than can be done
Cleary, P. and Monaghan, J.J. (1993) ‘Boundary interactions and
using SPH. However, for flows ranging from die casting transition to turbulence for standard CFD problems using
and mould filling to digital content generation and from SPH’, in Stewart, D., Gardner, H. and Singleton, D. (Eds.):
milling of minerals to tsunamis and volcanoes and many Computational Techniques and Applications: CTAC’93,
more, SPH offers predictive capability that is difficult to World Scientific, Singapore, pp.157–165.
match with conventional methods. Cleary, P.W. and Monaghan, J.J. (1999) ‘Conduction modelling
using smoothed particle hydrodynamics’, J. Computat. Phys.,
Vol. 148, pp.227–264.
Acknowledgements Cleary, P.W. and Prakash, M. (2004) ‘Discrete element modelling
and smooth particle hydrodynamics: potential in the
The support of CSIRO MMT and ODT Engineering are environmental sciences’, Phil. Trans. R. Soc. Lond. A,
gratefully acknowledged for the ingot casting modelling. Vol. 362, pp.2003–2030.
CSIRO and ODT Engineering are core participants in the Cleary, P., Prakash, M., Ha, J., Sinnott, M., Nguyen, T. and
CRC for Cast Metals Manufacturing (CAST), which was Grandfield, J. (2004) ‘Modelling of cast systems using
established under and is supported in part by the Australian smoothed particle hydrodynamics’, JOM, Vol. 3, pp.67–70.
Smooth particle hydrodynamics: status and future potential 89

Couchman, H.M.P., Thomas, P.A. and Pearce, F.R. (1995) Itoh, H., Takahama, J., Takahashi, M. and Miyamoto, K. (2000)
‘Hydra: an adaptive – mesh implementation of P3M–SPH’, ‘Hazard estimation of the possible pyroclastic flow disasters
Astrophysical Journal, Vol. 452, pp.797–813. using numerical simulation related to the 1994 activity at
Cummins, S.J. and Rudman, M. (1999) ‘An SPH projection Merapi Volcano’, J. Volcanol. Geotherm. Res., Vol. 100,
method’, J. Comp. Phys., Vol. 152, pp.584–607. pp.503–516.
Ellero, M., Kroger, M. and Hess, S. (2002) ‘Viscoelastic flows Johannessen, T., Larsen, J.H., Chorkendorff, I., Livbjerg, H. and
studied by smoothed particle dynamics’, Journal of Topsøe, H. (2001) ‘Catalyst dynamics: consequences for
Non-Newtonian Fluid Mechanics, Vol. 105, pp.35–51. classical kinetic descriptions of reactors’, Chemical
Engineering Journal, Vol. 82, pp.219–230.
Evrard, A.E. (1988) ‘Beyond N-body – 3D cosmological gas
dynamics’, Mon. Not. R. Astron., Vol. 235, pp.911–934. Klingelhöfer, F., Hort, M., Kümpel, H.J. and Schminke, H.U.
(1999) ‘Constraints on the formation of submarine lava flows
Frazao, S.S., Alcrudo, F. and Goutal N. (1999) ‘Dam-break test
from numerical model calculations’, J. Volcanology and
cases summary’, 4th CADAM meeting, Zaragoza, Spain,
Geothermal Res., Vol, 92, pp.215–229.
pp.9–25.
Libersky, L.D., Randles, P.W., Carney, T.C. and Dickinson, D.L.
Gingold, R.A. and Monaghan, J.J. (1977) ‘Smoothed particle
(1997) ‘Recent improvements in SPH modelling of
hydrodynamics, Theory and application to non-spherical
hypervelocity impact’, Int. Journal of Impact Engineering,
stars’, Mon. Not. Roy. Astron. Soc., Vol. 181, pp.375–389.
Vol. 20, pp.525–532.
Gomez-Gesteira, M., Cerqueiro, D., Crespo, C. and
Dalrymple, R.A. (2005) ‘Green water overtopping analysed Liu, M.B., Liu, G.R., Lam, K.Y. and Zong, Z. (2003a) ‘Smoothed
with a SPH model’, Ocean Engineering, Vol. 32, pp.223–238. particle hydrodynamics for numerical simulation of
underwater explosion’, Computational Mechanics, Vol. 30,
Grandfield, J., Cleary, P., Prakash, M., Sinnott, M., Oswald, K. pp.106–118.
and Alguine, V. (2003) ‘Mathematical modelling of ingot
caster filling systems’, in Whiteley, P. (Ed.): Proc: 8th Liu, M.B., Liu, G.R., Zong, Z. and Lam, K.Y. (2003b)
Australian Asian Pacific Conference on Aluminium Cast ‘Computer simulation of high explosive explosion using
House Technology, TMS 2003, pp.271–276. smoothed particle hydrodynamics methodology’, Computers
and Fluids, Vol. 32, pp.305–322.
Gray, J.P. and Monaghan, J.J. (2004) ‘Numerical modelling
of stress fields and fracture around magma chambers’, Lo, E.Y.M. and Shao, S. (2002) ‘Simulation of near-shore solitary
Journal of Volcanology and Geothermal Research, Vol. 135, wave mechanics by an incompressible SPH method’, Applied
pp.259–283. Ocean Research, Vol. 24, pp.275–286.
Gray, J.P., Monaghan, J.J. and Swift, R.P. (2001) ‘SPH elastic Meglicki, Z. (1994) ‘Verification and accuracy of
dynamics’, Computer Methods in Applied Mechanics and smoothed particle magnetohydrodynamics’, Comput. Phys.
Engineering, Vol. 190, No. 49, pp.6641–6662. Communications, Vol. 81, pp.91–104.
Guzel, B., Prakash, M., Turan, Ö.F. and Semercigil, S.E. (2005) Monaghan, J.J. (1992) ‘Smoothed particle hydrodynamics’, Ann.
‘A study of the mechanism of dissipation of energy with a Rev. Astron. Astrophys., Vol. 30, pp.543–574.
sloshing absorber’, Preparation for ASME 2005, Orlando, Monaghan, J.J. (1994) ‘Simulating free surface flows with SPH’,
Florida, November 5–11. J. Comp. Phys., Vol. 110, pp.399–406.
Ha, J. and Cleary, P. (2000) ‘Comparison of SPH simulations of Monaghan, J.J. (2002) ‘SPH compressible turbulence’, Mon. Not.
high pressure die casting with the experiments and VOF R. Astron. Soc., Vol. 335, pp.843–852.
simulations of Schmid and Klein’, Int. Journal of Cast Metals
Monaghan, J.J. (2005) ‘Smoothed particle hydrodynamics’, Rep.
Research, Vol. 12, pp.409–418.
Prog. Phys., Vol. 68, pp.1703–1759.
Ha, J., Cleary, P., Alguine, V. and Nguyen, T. (1999) ‘Simulation
of die filling in gravity die casting using SPH and Monaghan, J.J., Cas, R.A.F., Kos, A. and Hallworth, M, (1999)
MAGMAsoft.’, Proc. 2nd Int. Conf. on CFD in Minerals & ‘Gravity currents descending a ramp in a stratified tank’,
Processe Ind., Melbourne, Australia, pp.423–428. J. Fluid Mech., Vol. 379, pp.39–70.
Ha, J., Cleary, P.W., Prakash, M., Alguine, V., Nguyen, T. and Monaghan, J.J., Huppert, H.E. and Worster, M.G. (2005)
Scott, C. (2003) ‘SPH, MAGMAsoft and water analogue ‘Solidification using smoothed particle hydrodynamics’,
modelling of die casting of a servo piston’, Proc: 3rd Int Conf Journal of Computational Physics, Vol. 206, pp.684–705.
on CFD in Minerals and Process Industries, pp.587–592. Monaghan, J.J. and Kos, A. (1999) ‘Solitary waves on a Cretan
Ha, J., Cleary, P.W., Prakash, M. and Nguyen, T. (2004) ‘3D SPH beach’, J. Waterways, Ports, Coastal and Ocean Eng.,
simulations and validation of high pressure die casting of Vol. 125, pp.145–154.
automotive components’, Proc. Die Casting Conf. 2004, Monaghan, J.J. and Kos, A. (2000) ‘Scott Russell’s wave
Paper 21. generator’, Phys. of Fluids, Vol. A12, pp.622–630
Ha, J., Schuhmann, R., Alguine, V., Cleary, P. and Nguyen, T. Morris, J.P., Fox, P.J. and Zhu, Y. (1997) ‘Modelling low reynolds
(2000) ‘Real-time X-ray imaging and numerical simulation number incompressible flows using SPH’, J. Comput. Phys.,
of die filling in gravity die casting’, Proc. Modelling of Vol. 136, pp.214–226.
Casting, Welding and Advanced Solidification Processes IX, Parshikov, A.N., Medin, S.A., Loukashenko, I.I. and
pp.151–158. Milekhin, V.A. (2000) ‘Improvements in SPH method
Hernquist, L. and Katz, N. (1989) ‘TREESPH – A unification of by means of inter-particle contact algorithm and analysis of
SPH with the hierarchical tree method’, Astrophysical perforation tests at moderate projectile’, International Journal
Journal Supplement Series, Vol. 70, pp.419–446. of Impact Engineering, Vol. 24, pp.779–796.
90 P.W. Cleary, M. Prakash, J. Ha, N. Stokes and C. Scott

Prakash, M., Cleary, P.W., Grandfield, J., Rohan, P. and Price, D.J. and Monaghan, J.J. (2004b) ‘Smoothed particle
Nguyen, V. (2005a) ‘Optimisation of ingot casting wheel magnetohydrodynamics – II. Variational principles and
design using SPH simulations’, accepted at The 4th Int. Conf. variable smoothing-length terms’, Monthly Notices of the
on CFD in the Oil and Gas, Metallurgical and Process Royal Astronomical Society, Vol. 348, pp.139–152.
Industries, SINTEF/NTNU, Trondheim, Norway. Rabczuk, T. and Eibl, J. (2003) ‘Simulation of high velocity
Prakash, M., Cleary, P.W., Ha, J., Noui-Mehidi, M.N., concrete fragmentation using SPH/MLSPH’, Int. J. Numer.
Blackburn, H.M. and Brooks, G. (2005b) ‘Simulation of Methods Engng, Vol. 56, pp.1421–1444.
suspension of solids in a liquid mixing tank using SPH and Schmid, M. and Klein F. (1995) ‘Fluid flow in die cavities –
comparison with physical modelling experiments’, accepted experimental and numerical simulation’, NADCA 18,
at The 4th Int. Conf. on CFD in the Oil and Gas, Metall. and International Die Casting Congress and Exposition, Paper
Process Industries, SINTEF/NTNU, Trondheim, Norway. No. I-T95-034, October 2–5, Indianapolis, Indiana, pp.93–99.
Prakash, M., Cleary, P.W. and Stokes, A.N. (2005c) ‘A pseudo Street, S., Brooks, G.A., Reilly, L. and Worner, H.K. (1998)
two phase model to simulate slurry flow in spiral separators ‘Environment and other bath smelting processes for treating
using SPH’, accepted at The 4th Int. Conf. on CFD in the Oil organic and ferrous wastes’, JOM, Vol. 50, pp.43–47.
and Gas, Metall. and Process Industries, SINTEF/NTNU,
Trondheim, Norway. Sugino, T. and Yuu, S. (2002) ‘Numerical analysis of fine
powder flow using smoothed particle method and
Prakash, M., Debroux, F. and Cleary, P. (2001) experimental verification’, Chemical Engineering Science,
‘Three-dimensional modelling of dam-break induced flows Vol. 57, pp.227–237.
using smoothed particle hydrodynamics’, 14th Australasian
Fluid Mechanics Conference, Adelaide, Australia, Violeau, D., Piccon, S., Chabard, J.P. (2001) ‘Two attempts of
pp.379–382 turbulence modelling in smoothed particle hydrodynamics’,
8th International Symposium on Flow Modelling and
Prakash, M., Stokes, N., Bertolini, J., Tatford, O. and Gomme, P. Turbulence Measurements, Tokyo, Japan.
(2003) ‘SPH simulations of a lobe pump: prediction of
protein shear stress at different pump efficiencies’, Proc: 3rd Young, W-B. and Lai, C-L. (1997) ‘Analysis of the edge effect in
Int. Conf on CFD in Minerals and Process Industries, resin transfer molding’, Composites Part A, Vol. 28A,
pp.183–188. pp.817–822.
Price, D.J. and Monaghan, J.J. (2004a) ‘Smoothed particle Zoppou, C. and Roberts, S. (2000) ‘Numerical solution of the
magnetohydrodynamics: I. Algorithm and tests in one two-dimensional unsteady dam break’, App. Math. Modelling,
dimension’, Monthly Notices of the Royal Astronomical Vol. 24, pp.457–475.
Society, Vol. 348, pp.123–138.

View publication stats

You might also like