Eleventh International Conference on CFD in the Minerals and Process Industries

CSIRO, Melbourne, Australia

7-9 December 2015

A COMPARISON OF CFD SOFTWARE PACKAGES’

ABILITY TO MODEL A SUBMERGED JET

Alasdair MACKENZIE1*, A LOPEZ1, K RITOS1, M T STICKLAND2, W M DEMPSTER2

1 Weir Advanced Research Centre, University of Strathclyde, 99 George Street, Glasgow, SCOTLAND, G1 1RD
2 Department of Mechanical and Aerospace Engineering, University of Strathclyde
*Corresponding author E-mail address: alasdair.mackenzie.100@strath.ac.uk

Shirazi, & Rybicki, 2007), where a submerged liquid jet

ABSTRACT impingement test, commonly used in erosion testing was
studied using laser Doppler velocimetry (LDV), to measure
Particle erosion from slurry flow is a common problem in
velocity profiles and validate CFD simulations. Since then,
many industrial applications, including the mining and the
the main focus of study has concentrated on the particulate
oil and gas industry. Erosion modelling is a known complex
behaviour and the erosion process. This paper revisits the
problem, and consists of three equally important parts; fluid
fluid flow modelling, and investigates the effectiveness of
flow modelling, particulate flow modelling, and erosion
a number of commercially available CFD codes to predict
modelling. It is the first of these, the fluid flow, which is
the submerged jets. ANSYS Fluent 15, Star-CCM+ 10.02
analysed here. The paper compares three different
and OpenFOAM 2.3.x are compared to experimental data
computational fluid dynamics (CFD) software packages,
of a submerged jet impingement test constructed as part of
ANSYS Fluent, Star-CCM+ and OpenFOAM, on their
this study.
ability to model the fluid phase in the submerged jet
Some surprising results emerge on the ability of available
impingement test. The computational results were verified
models to reproduce the experimental findings. The
by results from an experimental rig where the fluid flow
experimental set up is discussed and results compared.
was measured by particle image velocimetry (PIV). Despite
the apparent simplicity of the jet impingement test, this
paper highlights the difficulties of capturing the
experimental results with computational methods. SETUP
Experiment
The setup for the experiment can be seen in Figure 1 below.
NOMENCLATURE Particles are injected to the header tank with a mass
CFD computational fluid dynamics concentration of <5%, and are well mixed. Due to the
fps frames per second pressure caused by height difference, they follow the flow
GUI graphical user interface and impinge on the sample surface. Particles of diameter
LDV laser Doppler velocimetry 20µm, with a specific gravity (SG) of 1 are used to track
PIV particle image velocimetry fluid velocities. The small particle diameter and equivalent
r radius density result in the particles following the flow (low
SG specific gravity Stokes number).
SST shear stress transport

INTRODUCTION
The study of wear due to particulate erosion is an
established research field because of its complexity and the
desire to increase the operational life of particle processing
equipment, e.g. slurry pumps, crushers, cyclones etc. The
current development of CFD techniques and their ability to
accurately predict flow behaviour promises improved wear
prediction since the particulate behaviour close to the wall
can be predicted. A technique, recently explored by a
number of authors, (Gnanavelu, Kapur, Neville, Flores, &
Ghorbani, 2011; A. Mansouri et al., 2015) proposed that a
wear model, i.e. a relationship between material removal
and particulate behaviour (usually particle velocity and
angle) can be developed using CFD. This methodology is
usually explored using a fluid jet impact test and the authors Figure 1: Experimental setup
show this as a promising approach to improving erosion
prediction. However, only limited validation work on this The camera was set to 500 fps (frames per second), with the
aspect of the problem has been carried out. One frequently image focused on the light sheet and particles. The frame
cited study is by Zhang et al. (Zhang, Reuterfors, McLaury, straddling technique was used to capture the particle

Copyright © 2015 CSIRO Australia 1

movements, as standard cross correlation was not capable.
The frame rate, combined with the laser pulses (set to
100µs), was fast enough to provide enough movement of
the particles to carry out correlation. The particles need to
move more than 6-7 pixels, but less than the size of the
frame, to make post processing possible. FlowManager
(Dantec Dynamics), a commercial program was used to
analyse seed particle velocities and produce measured
velocity fields. The pump recirculated the water back to the
header tank, for further experiments. The flowrate was
calculated by measuring the increase in weight of the tank
over a known period of time. This was done multiple times
to reduce measurement errors, and the average flowrate was
found to be 3.2 kg/min, or 0.053kg/s.
Computational Fluid Dynamics
Two commercial CFD codes and an open source CFD code
were examined: ANSYS Fluent 15, Star-CCM+ 10.02 and Figure 3: Mesh used for all simulations- showing refinement
OpenFOAM 2.3.x. They were all set up with the same region
boundary conditions to replicate the experiment. Water had
properties: density of 998.2kg/m3, and viscosity of
The standard k-epsilon turbulence model was used for all
0.001003 kg/ms.
three software packages, with the scalable wall function
The measured flowrate from the experimental test was used
used in fluent (since y*<11). To try and match experimental
as the inlet boundary condition, and other boundary
data better, an additional k-omega shear stress transport
conditions can be seen in Figure 2. As shown in Figure 3, a
(SST) turbulence model was run in StarCCM, since various
refinement region around the area of interest was made, and
papers use this to model the jet impingement test
a large enough volume was left around the impingement
(Mansouri, Shirazi, & Mclaury, 2015; Nguyen, Poh, &
surface to ensure no recirculation took place. The nozzle
Zhang, 2014). Convergence was ensured, with each
was more than 10 diameters long, and the top surface was
residual being less than 10-4 before results were taken.
too far away so as not to interfere with the impingement
zone. The geometry was drawn according to the
experimental test as a half symmetric model, in order to
reduce computational time. RESULTS AND DISCUSSION
The PIV data were analysed on FlowManager using 100
images; thus giving 50 frames. A filter was applied to each
frame to mask the area of interest, then a vector range of +3
to -3m/s was added to remove any noise. The 50 frames
were then averaged, with the resulting vectors
superimposed in Figure 4 below. The vectors can be seen,
along with the plate, nozzle and light sheet. The laser sheet
can be seen lighting up the surface of the sample, and the
seeding particles throughout the liquid.

Figure 2: Geometry and boundary conditions used for CFD

simulations

Initially the mesh for each code was generated by their own
mesh generation application, and were shown to be mesh
independent. However, the StarCCM mesh was chosen as Figure 4: Average vector plot from FlowManager
the reference mesh, as it had the best quality and uses
polyhedral cells that reduce the artificial diffusivity of the
flow. The mesh had 3.2 million cells, and used 16 inflation A sample line 1mm above the plate was then drawn, and
layers near the walls with a growth ratio of 1.2, and total velocity vectors extracted. The same sample line was drawn
thickness of 0.6mm. There was also a refinement region in the CFD packages for comparison purposes. This height
around the impingement area, since this is where high is suitable for particle data collection that can be used for
gradients exist. erosion equations. Further away, the data wouldn’t be
Each solver then ran using this mesh, and results are relevant to surface impacts, and closer would yield lower
compared here. The y-plus value was checked on the velocities due to the boundary layer. Furthermore there is
nozzle wall and impingement surface, and it was kept below also the practical issue of sampling the flow with PIV at less
1 for all simulations. than 1mm away from the surface.

Copyright © 2015 CSIRO Australia 2

Figures (5-7) show a comparison of the measured and modelling approach. Fluent and OpenFOAM both capture
computed velocities. The x-axis shows distance from the the position of the peak velocity magnitude, which is
centre of the nozzle, going to the extents of the plate, important for erosion modelling.
12.5mm in each direction. As the below figures indicate, All software setups capture the radial components well,
there is general agreement between all of the k-epsilon with the k-omega SST giving slightly better results near the
solvers. All three CFD packages overpredict the velocity in outer edge of the plate.
the centre of the jet (less than r=2.5mm) and near the edge The k-omega SST model captures the radial component of
of the plate (more than r=7mm), whereas they all (apart velocity very well, and also manages to capture the peak
from the k-omega SST) underpredict the velocity at the two values of velocity magnitude. However it overpredicts the
peaks around r=5mm. axial velocity more than any other model, leading to it not
matching the PIV. All solvers struggle the most with the
stagnation region in the axial direction.
Figure 8 shows a contour comparison of the PIV and Fluent
1.6 velocity data. The jets have a similar shape, with the PIV
FlowManager
Ansys
1.4 StarCCM having a stronger axial flow component in the centre than
OpenFOAM
1.2 StarCCM SST the CFD. This could suggest that the CFD over diffuses the
Velocity (m/s)

1 flow, having a larger radial component further out from the

0.8 centre of the jet (as seen in Figure 5, after 10mm from the
0.6
centre of the jet).
0.4
0.2
0
-15 -10 -5 0 5 10 15
Distance from centre of nozzle (mm)

Figure 5: Velocity magnitude 1mm above plate

1.5

0.5
Velocity (m/s)

-0.5

Figure 8: Velocity magnitude contour comparison

-1
FlowManager
Ansys
-1.5 StarCCM
OpenFOAM
StarCCM SST
Discussion
-2 The above results show that a ‘simple case’ is not as easily
-15 -10 -5 0 5 10 15
modelled as one might think. Current erosion prediction
Distance from centre of nozzle (mm)
approaches require near wall accuracy of CFD models. The
Figure 6: Radial velocity 1mm above plate general trends are captured, however with all codes there is
a discrepancy compared to the experimental velocity
magnitudes by up to 40%. This is a significant difference,
0.5
bearing in mind that the mass removal from particle impact
is dependent on the velocity magnitude to the power of
0 between 2 and 3, and impact angle. The CFD velocity
graphs do follow the same shape as the experimental
though, implying that erosion location should be accurate.
Velocity (m/s)

-0.5
Figure 9 and Figure 10 taken from Zhang et al. show the
results for radial and axial fluid velocity distribution of
-1 water released from the nozzle with average exit velocity of
12m/s. Their work was carried out using Fluent 6, and used
FlowManager the second order Reynolds stress turbulence model. The
Ansys
-1.5
StarCCM measured values of velocity were determined from point
OpenFOAM
StarCCM SST measurements using LDV at multiple positions on a grid.
-2 Although the turbulence model and boundary conditions
-15 -10 -5 0 5 10 15 are different, the results are similar in that the closer to the
Distance from centre of nozzle (mm)
plate, the less able the CFD becomes to predict velocities
Figure 7: Axial velocity 1mm above plate accurately. The comparisons indicate that the
experimentally measured values of axial velocity are 20-
30% higher nearer the wall than predicted. While the radial
Fluent and OpenFOAM give very similar velocity profiles, velocities show even greater error, particularly at radial
whereas StarCCM is slightly different, mainly in the axial positions just beyond the nozzle radius. If a robust geometry
plane. This could be due to StarCCM’s different wall

Copyright © 2015 CSIRO Australia 3

independent erosion model is to be made, the near wall fluid measured for a submerged jet, impacting on a plate indicate
modelling has to be improved. that:

1. Using two equation turbulence models, the CFD

codes could not accurately predict the impacting
jet flow field, 1mm above the plate.

2. Stagnation regions are still difficult to model.

3. Further work is required to see if particle

trajectories predicted by CFD are affected by this
inability to model near wall flows. If so, the jet
impingement test on a flat plate may not be the
best way to implement erosion modelling.

Noting that current approaches to erosion modelling have

three requirements, fluid flow, particle flow, and erosion
modelling, the first of these is often assumed. This paper
demonstrates that errors in the fluid flow’s magnitude
predictions could be sizable, however the general shape is
captured.

REFERENCES
Figure 9: Radial fluid velocity, LDV data vs. CFD result (Zhang
et al., 2007)
Gnanavelu, A., Kapur, N., Neville, A., Flores, J. F., &
Ghorbani, N. (2011). A numerical investigation of a
geometry independent integrated method to predict
erosion rates in slurry erosion. Wear, 271(5-6),
712–719. doi:10.1016/j.wear.2010.12.040

Mansouri, A., Arabnejad, H., Shirazi, S. A., & McLaury,

B. S. (2015). A combined CFD/experimental
methodology for erosion prediction. Wear, 332-333,
1090–1097. doi:10.1016/j.wear.2014.11.025

Mansouri, A., Shirazi, S. A., & Mclaury, B. S. (2015).

Experimental and numerical investigation of the
effect of viscosity and particle size on erosion
damage caused by solid particles. ASME, 1–10
doi:10.1115/FEDSM2014-21613

Nguyen, V. B., Poh, H. J., & Zhang, Y.-W. (2014).

Predicting shot peening coverage using multiphase
computational fluid dynamics simulations. Powder
Figure 10: Axial fluid velocity, LDV data vs CFD result (Zhang Technology, 256, 100–112.
et al., 2007)
doi:10.1016/j.powtec.2014.01.097

Potential sources of error Zhang, Y., Reuterfors, E. P., McLaury, B. S., Shirazi, S.
A., & Rybicki, E. F. (2007). Comparison of
The position of the laser relative to the nozzle was
computed and measured particle velocities and
considered a source of error, however steps were taken to
erosion in water and air flows. Wear, 263(1-6),
ensure this was minimised. The laser, and thus the light
330–338. doi:10.1016/j.wear.2006.12.048
sheet, was first aligned with the centre of the jet, and then
moved to 1mm either side and experiments repeated to
determine the sensitivity to position. The results were
analysed, and the position that gave the highest axial
velocity was chosen: being an indication of the centre of the
jet.

CONCLUDING REMARKS
A comparison between two commercial CFD codes
(ANSYS Fluent, STARCCM) and an open source CFD
code (OpenFOAM) with the velocity fields experimentally

Copyright © 2015 CSIRO Australia 4