Jasper Van de Vyver Numerical simulation of evaporating diesel sprays Academic year 2016-2017 Faculty of Engineering and Architecture Chair: Prof. dr. ir. Jan Vierendeels Department of Flow, Heat and Combustion Mechanics Master of Science in Electromechanical Engineering Master's dissertation submitted in order to obtain the academic degree of Counsellor: Gilles Decan Supervisors: Prof. dr. ir. Jan Vierendeels, Prof. dr. ir. Sebastian Verhelst
110
Embed
Numerical simulation of evaporating diesel sprays Jasper ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The Clean Thesis StyleNumerical simulation of evaporating diesel
sprays
Academic year 2016-2017 Faculty of Engineering and Architecture
Chair: Prof. dr. ir. Jan Vierendeels Department of Flow, Heat and
Combustion Mechanics
Master of Science in Electromechanical Engineering Master's
dissertation submitted in order to obtain the academic degree
of
Counsellor: Gilles Decan Supervisors: Prof. dr. ir. Jan
Vierendeels, Prof. dr. ir. Sebastian Verhelst
Jasper Van de Vyver
Numerical simulation of evaporating diesel sprays
Academic year 2016-2017 Faculty of Engineering and Architecture
Chair: Prof. dr. ir. Jan Vierendeels Department of Flow, Heat and
Combustion Mechanics
Master of Science in Electromechanical Engineering Master's
dissertation submitted in order to obtain the academic degree
of
Counsellor: Gilles Decan Supervisors: Prof. dr. ir. Jan
Vierendeels, Prof. dr. ir. Sebastian Verhelst
Acknowledgement
I would like to express my sincere gratitude towards everyone who
made the realisation of this master’s thesis possible. First off, I
would like to thank professor Vierendeels and professor Verhelst
for their assistance during fortnightly meetings. Without their
suggestions, constructive feedback and insights in the matter, this
master’s thesis would not be what it has become. Furthermore, a
special mention goes to my counsellor Gilles Decan whose work
provided the basis for this master’s thesis. He was always
available to answer any questions, even during his research stay at
Politecnico di Milano. His extensive feedback on my writing is
greatly valued as well.
I would also like to thank Dr. Tarek Beji and others who were
sporadically present at my thesis meetings. The department of flow,
heat and combustion mechanics deserves to be mentioned for
providing me with the necessary software and computing power for my
spray simulations. Yves Maenhout was always there to fix any
occasional IT-related problem. The people who contributed to the
ECN should also not be forgotten. Without their experimental data,
this thesis would have looked very different.
I am also very grateful to my parents, grandparents and sister.
They have always supported me both mentally and financially during
my studies and during the writing of this thesis. Last but not
least, my thanks go out to my friends and fellow engineering
students whom I could turn to for serious and less serious
talks.
iii
Permission for usage
The author gives permission to make this master dissertation
available for consultation and to copy parts of this master
dissertation for personal use. In the case of any other use, the
copyright terms have to be respected, in particular with regard to
the obligation to state expressly the source when quoting results
from this master dissertation.
Jasper Van de Vyver, 02/06/2017
iv
Jasper Van de Vyver
Supervisors: Prof. dr. ir. Jan Vierendeels, Prof. dr. ir. Sebastian
Verhelst Counsellor: Gilles Decan
Master’s dissertation submitted in order to obtain the academic
degree of Master of Science in Electromechanical Engineering
Department of Flow, Heat and Combustion Mechanics Chair: Prof. dr.
ir. Jan Vierendeels Faculty of Engineering and Architecture
Academic year 2016-2017
Abstract and extended abstract
v
Numerical simulation of evaporating diesel sprays Jasper Van de
Vyver Ghent University, Belgium
Supervisor: Prof. dr. ir. Jan Vierendeels, Prof. dr. ir. Sebastian
Verhelst
Counsellor: ir. Gilles Decan
ABSTRACT
Within the context of more stringent emission legislations,
researchers are searching for ways to better understand pollutant
formations in internal combustion engines. One of the primary
factors in pollutant formation in compres- sion ignition engines is
the spray formation, since it determines the degree to which the
fuel mixes with the air. Spray formation, however, is still a
complex and not fully understood multiphase phenomenon. The
increase in accuracy of measurement techniques and the increase in
computing power go hand in hand to develop the knowledge about this
topic.
The main objective of this research is to provide a rec- ommended
approach to simulate diesel sprays in a com- putational fluid
dynamics (CFD) software package within a reasonable calculation
time. Therefore, the commercial CFD package ANSYS Fluent was used
to simulate ECN spray A, a widely tested n-dodecane spray with pub-
licly available experimental data. A Eulerian-Lagrangian phase
model was used together with various spray sub- models governing
the evaporation, breakup, collision and drag forces of the fuel
droplets. As breakup model, the KH-RT model was found to be the
most suitable. The sensitivity to fluid properties as well as to
parameters of the various spray submodels and turbulence models was
identified. Based on this sensitivity analysis, the values of the
surface tension, liquid density and the thermal conductivity of the
gas mixture were made temperature- dependent. Additionally, the
model’s capabilities to predict parametric variations of ECN spray
A were explored. All trends were reproduced with a fairly good
agreement, but the model is unable to take into account different
conditions in the nozzle. Future research will still have to
investigate if this approach can be successfully extended to
reacting conditions. It also remains to be seen that
this approach scales up well to the dimensions of marine diesel
engines, as well as dual fuel engines with premixed natural
gas.
INTRODUCTION
Recently, the International Maritime Organisation (IMO) introduced
the IMO NOx Tier III requirement [1], which severely limits the
allowable NOx emissions for marine diesel engines in global
shipping. Compared to the last iteration, the limit has gone down
by about 75%. This means that engine tuning alone is not sufficient
any more to meet these new standards. A possible short-term
solution is to install dual fuel engines. This technology can be
retrofitted to existing diesel engines by adding a natural gas tank
and a natural gas port fuel injection system. The natural gas-air
mixture is sucked into the cylinder before being compressed and
ignited by a small pilot injection of diesel. In this way, the
engine mainly runs on natural gas and the NOx emissions are
significantly improved.
Dual fuel operation has a complex combustion behaviour. This leads
to quite a few unsolved issues primarily due to a lack of
optimisation of dual fuel engines at low loads. A lot of
experimental research is being done on dual fuel engines but also
more fundamental research has to be done to fully optimise and
understand this kind of combustion. To this end, Computational
Fluid Dynamics (CFD) can be a useful tool. The in-cylinder flow
field, spray formation, combustion and heat transfer can all be
modelled to have a better understanding of what happens inside the
cylinder for different operating conditions. How- ever, a CFD
simulation of an entire dual fuel engine is a very daunting task in
terms of modelling and calculation time. The flow in an engine is
highly reactive and turbulent and moves in a geometrically complex,
moving mesh. Furthermore, the different processes in an engine
have
1
a large variety of characteristic time constants. In an effort to
reach this goal, various processes are studied separately. In this
research, there will be focussed on diesel spray breakup at
engine-like ambient conditions, in particular automotive
engine-like ambient conditions rather than marine diesel engines.
The choice for a spray representative of an automotive engine is
linked to the extensive experimental database of ECN [2] and the
amount of research done on ECN spray A. A diesel spray could be
simulated in a cylinder, but the cyclic variations and the
uncertainty on the boundary conditions ask for a different
approach. Most of the time, research on spray breakup is done in
constant volume combustion chambers. They have the added benefit of
being optically accessible for measuring the spray
characteristics.
LIQUID PHASE REPRESENTATION The Eulerian- Lagrangian approach is
frequently used by many authors[3, 4, 5, 6, 7]. It is also called a
discrete phase model or DPM. In this approach, droplets of similar
properties are grouped in discrete particles and are tracked along
the mesh. Additional submodels govern the evaporation, breakup,
drag forces, coalescence and shedding of those particles. In this
way, they interact with the continuous gas phase. Each time step,
the continuous phase flow field is calculated. This flow field
determines the trajectory of the discrete phase particles and
these, in turn, determine the new continuous phase source terms.
This process is then repeated every time step. One of the main
assumptions of this approach is that the volume fraction of liquid
droplets in a cell stays low. This creates a possible source for
errors in the near-nozzle region and imposes a lower limit on the
used cell sizes. Lucchini et al. [5] remark that the grid size
generally adopted is much larger than the nozzle diameter (about
2-5 times). The need for a relatively coarser mesh is at the same
time an advantage. It provides an efficient way of representing the
small droplets (in the order of 0.1 µm) naturally present in
high-pressure fuel sprays. This makes it a frequently used
approach.
If a high accuracy in the near-nozzle region is desirable, there is
a tendency to use a Eulerian-Eulerian approach. Here, both the gas
and liquid phase have a continu- ous representation. The interface
between the two is reconstructed based on an extra scalar quantity
that is transported over the mesh. In the Level-Set method, this
scalar quantity is the distance to the interface. The interface can
then be reconstructed by connecting the cells for which this
quantity is 0. Herrmann [8] used this method to study the primary
atomisation of sprays. He determined that at least 6 grid points
are needed to resolve a droplet to get a grid independent result.
In other words, at least 63 cells are needed to properly represent
a droplet. When comparing the size of droplets in diesel sprays to
the dimensions of a combustion chamber, it becomes clear that
meshes of millions of cells are needed for this approach. This does
not allow to obtain a solution within a reasonable calculation time
on a cluster with a
dozen cores.
DISCRETISATION The design of the mesh depends on the choice of a
Lagrangian or a Eulerian approach for the liquid phase. In a
Lagrangian approach, the smallest grid size will seldom go below
the nozzle diameter, while for a Eulerian approach the smallest
grid size is generally a fraction of the nozzle diameter. It is
clear that for optimal use of the mesh and for the most accurate
results, the region near the nozzle and around the spray axis
should be the most refined, as it is in these regions that biggest
velocity gradients occur. The degree of refinement of the mesh is
also related to the time step. One can imagine that if the flow
travels more than one cell per time step the accuracy can suffer.
This corresponds to a Courant number C of 1. Explicit time stepping
schemes even become unstable above a certain Cmax.
C = ut
x ≤ Cmax (1)
Table 1: Courant numbers used by different authors for their fuel
spray simulations
Author x [mm] t [ns] u [m/s] C Som [4] 0.25 500 ±575 ±1.15 Pei [9]
1 4000 595.3 2.381
Decan [7] 1 400 460 0.184 Lucchini [5] 0.25 500 N/A ≤0.15
Som et al.[4] and Lucchini et al.[5] used an adaptive mesh
refinement (AMR) technique, so the smallest cells do not
necessarily coincide with the cells with the fastest flow. The x
given in the table is the smallest possible grid size. u is an
estimate of the injection velocity of the spray. The velocity of
the entrained gas will be a bit lower. The Courant number is thus
the largest possible one. It can be concluded that common Courant
numbers are in the order of 10−1 to 1.
INJECTION One of the challenges of using a discrete phase model is
how to link the nozzle flow to an appropri- ate diameter
distribution at the injection. The Lagrangian way of tracking
particles does not allow to represent a fully liquid flow, such as
in the nozzle. Furthermore, the levels of turbulence and cavitation
in the nozzle greatly influence the speed and diameters of the
droplets. Increased levels of turbulence in the nozzle flow
destabilise the jet and in- crease its breakup rate. The same goes
for the cavitation. One of the most basic ways of modelling the
injection is imposing a distribution function of droplet sizes at
the nozzle exit. This model assumes that the primary breakup has
already occurred at the nozzle exit. Considering the quick primary
atomisation of high-pressure fuel sprays, this is not a very hefty
assumption. For example, Pei et al. [9] used a uniform distribution
and tuned its initial droplet diameter to obtain approximately
correct liquid
2
lengths. The Rosin-Rammler distribution function is also sometimes
used. Probably, injecting droplets of the same diameter as the
nozzle is the most prevalent method. This method is referred to as
the Blob method. The breakup of these big droplets is handled by
the secondary breakup models, dispersing the spray in much finer
droplets as can be seen in figure 1. Its simplicity while retaining
fairly accurate results is its main advantage and it is therefore
used by many authors [4, 10, 5, 6] in their fuel spray simulation
studies.
Figure 1: Breakup of blobs by secondary breakup models using the
concept of liquid core length, figure from ANSYS Fluent Theory
Guide [11]
The speed of the ejected droplets can be determined by the
conservation of mass if the mass flow rate m(t) through the nozzle
is known. If there is no cavitation in the nozzle, the mean
velocity u(t) of the ejected droplets is given by m(t)/(ρlA). A is
the cross-sectional area of the nozzle (assumed to be constant) and
ρl the liquid density. If the mass flow rate is not known,
Bernoulli’s equation can be used to determine the maximum
theoretical speed from the pressure difference p over the nozzle.
The real speed will be lower, as friction losses are not taken into
account here. The real speed is often compared to the theoretical
one and their ratio is called the discharge coefficient Cd. The
value can be calculated by estimating the friction losses or
sometimes it is given for a particular nozzle.
Cd = m
ρlAutheor , utheor =
√ 2p
ρl (2)
There still exist plenty of other models, some of which are
commonly used but not available in ANSYS Fluent. For example, there
is the Huh-Gosman model as used by various researchers [4, 5, 10]
of which the first two used a modified version of the Huh-Gosman
model called the Bianchi model. They both take into account the
aerodynamically induced breakup by the Kelvin-Helmholtz instability
and the turbulence induced breakup by intro- ducing surface
perturbations linked to the turbulent length scales of the flow in
the nozzle. Other models also include the effect of cavitation,
such as the KH-ACT model, developed by Som et al. [12].
SECONDARY BREAKUP MODELS The Wave, Kelvin- Helmholtz or KH model is
a popular secondary breakup
model for high-Weber-number sprays, such as high- pressure fuel
sprays. Indeed, the high pressures lead to high injection
velocities u and consequently also a high Weber number We =
ρlu
2d/σ. This model has been developed by Reitz and Diwakar [13] in
1987 and it has since been used by many people. For example, Som et
al. [4] and Montanaro et al. [10] used it in their works. The
Kelvin-Helmholtz instability describes the unstable behaviour of
two fluids when there exists a relative speed difference between
them. If this speed difference exceeds a critical value, the wave
becomes unstable. It is an important breakup mechanism at the
surface of fuel droplets.
A common extension to the Wave model is to include the effect of
the Rayleigh-Taylor instability. The model combining these two
effects is called the KH-RT model. The RT instability is an
unstable flow of two fluids, one of which is denser than the other.
The less dense fluid pushes the heavier one and creates
instabilities while doing so. This happens when a denser fluid sits
atop a less dense fluid or when a less dense fluid is accelerated
into a denser fluid. The latter is an important effect in the
dilute region of the spray and the KH-RT model is therefore used by
many researchers [4, 10, 14, 6].
COLLISION MODEL In a Lagrangian Particle Tracking approach, there
is a possibility that particles collide. The outcome of such a
collision can be that the particles coalesce and continue their way
as one or that they bounce and continue their journey on a
different path. Although high-pressure fuel sprays can have high
collision rates, collision models are not used very often. Several
authors [4, 9, 5, 7] declare not having used a collision algorithm,
mainly because of its limited influence on the evaporation rate of
a spray.
TURBULENT DISPERSION Normally, the trajectory of Lagrangian
particles in a flow field is calculated based on the mean flow
velocities (u, v, w). However, in a real flow, the turbulent
fluctuations create an additional dispersion of the particles. It
would be better to calculate the trajectories based on the
instantaneous flow velocities (u + u′, v + v′, w + w′), but these
are not available in RANS models. In an effort to include this
dispersion effect in spray simulations, some researchers [4, 9, 10]
used turbulent dispersion models in their spray simulations.
Lucchini et al. [5] chose not to include this effect in order to
reduce the sensitivity of the results to the turbulence model. All
those who did use a turbulent dispersion model chose for the
stochastic tracking approach.
SIMULATION SETUP
ECN SPRAY A As already mentioned in the introduc- tion, the choice
for the studied case fell on ECN spray A. It is a spray
representative for an automotive diesel engine with moderate
exhaust gas recirculation. Liquid
3
n-dodecane at 363 K and 150 MPa is injected through a nozzle with a
diameter of 0.090 mm into a constant volume combustion chamber. The
ambient gas in the combustion chamber has a density ρa of 22.8
kg/m³, a temperature Ta of 900 K and a pressure pa near 6.0 MPa.
More specific boundary conditions can be found on the website of
ECN [2]. In the 3rd ECN workshop [15], a fixed set of parametric
variations of ECN spray A were defined. These will be used to
validate the model. The ambient pressure pa, injection velocity u
and mass flow rate m were calculated with equation 2 and the ideal
gas law with the gas constant of nitrogen R = 297 J/(kg.K). For the
liquid density ρl, a value between 701 and 706 kg/m³ was used to
account for the density change due to the ambient pressure
variation.
Table 2: Ambient temperature, density and injection pressure of the
parametric variations of ECN spray A. The ambient pressure,
injection speed and mass flow rate are calculated based on Cd =
0.89.
Ta[K] ρa[kg/m³] pinj [MPa] pa[MPa] u[m/s] m[g/s] 900 22.8 150 6.09
569 2.55 900 22.8 100 6.09 460 2.06 900 22.8 50 6.09 314 1.41 900
7.6 150 2.03 578 2.58 900 15.2 150 4.06 574 2.56 700 22.8 150 4.74
572 2.56 1000 22.8 150 6.77 567 2.54 1100 22.8 150 7.45 566 2.54
700 22.8 50 4.74 319 1.43 440 22.8 150 2.98 576 2.57 303 22.8 150
2.05 578 2.58
MESH AND TIME STEP These experiments are carried out in constant
volume combustion chambers. The one at Sandia National Laboratories
is responsible for a sizeable part of the experimental data
available on the website of ECN. Therefore, its geometry will serve
as an inspiration for the calculation domain. It has a
cubical-shaped combustion chamber with a characteristic length of
108 mm. Last year, Gilles Decan [7] already made a mesh for
simulations of sprays in the GUCCI (Ghent University Combustion
Chamber I). A rescaled version of his mesh was used in this work to
simulate ECN spray A.
Because of symmetry, the mesh represents a quarter of the
combustion chamber. The zone around the spray axis is a structured
mesh with a conical shape. The dimensions of the cells near the
nozzle are 0.675 mm in the axial direction, 0.135 mm in the radial
direction and 6in the circumferential direction. This results in a
total cell count of 138 754 cells. Mesh dependency was checked by
refining the structured part of the mesh to a total cell count of
432 194. It barely altered the vapour penetration results. This is
the maximum distance from the nozzle outlet to where the fuel
vapour mass fraction is 0.1%. The liquid penetration can be defined
in a similar
way, but for a liquid fuel volume fraction of 0.1%.[16] The liquid
penetration increased by about 10%, but this can be called a good
result considering the difficulty in obtaining a grid-independent
solution in spray simulations.[5] When a droplet exchanges momentum
with a very small cell, the velocity of the continuous gas phase
shoots up quickly. This leads to a lower relative velocity between
the gas and liquid phase, and the spray penetrates further. Time
steps smaller or equal to 0.4 µs produced quasi- identical results.
This corresponds to a maximum Courant number of 0.337.
Figure 2: Quarter cubic mesh as created by Gilles Decan[7]
BREAKUP MODEL As a breakup model, the KH-RT model was used. It has
two competing breakup mecha- nisms. B1 can be lowered to increase
the breakup rate due to the KH instability and CRT can be lowered
to increase the breakup rate due to the RT instability. In both
cases, the liquid penetration decreases due to the increased total
surface area of the smaller droplets. After comparing with the
experiments, a B1 of 1.73 and a CRT of 0.3 gave a good agreement
with the liquid penetration of the baseline condition of ECN spray
A. These values were used to calculate all the parametric
variations. Some cases were also calculated using a B1 of 0.5 and a
CRT of 0.5. For the baseline condition, this resulted in almost the
same liquid length and thus increasing the relative importance of
KH breakup. The results for the parametric variations look
promising as explained in the results.
TURBULENCE MODEL The realisable k-ε model was chosen by comparing
the results of a wide range of turbulence models at standard
settings. The choice has a profound effect on the entire spray
formation, as can be seen in figure 3. The k-ε models behaved
similarly except for the initial transient which was better
captured by the realisable k-ε model. Despite the fact that the k-ω
SST model should behave similarly as a k-ε model away from walls,
the k-ω SST model severely underestimated the turbulent kinetic
energy. This means that the spray penetrates further because it
entrains less hot, stagnant air. The reduced mixing also leads to
higher velocities, higher vapour mass fractions and lower
temperatures on the spray axis, resulting in a lower evaporation
rate and
4
an increased liquid penetration. On the other hand, the
Spalart-Allmaras model heavily overestimated the turbu- lent
viscosity ratio. It entrains so much air that it almost spreads out
more quickly in the radial direction as in the axial direction. The
Spalart-Allmaras model was never designed to simulate jet-like
flows, so these inaccurate results were to be expected. For the
realisable k-ε model, the vapour penetration could be brought
within the limits of the experimental error by slightly decreasing
C2ε from 1.9 to 1.77. C2ε determines the rate at which the
turbulent dissipation ratio ε decays. So for lower values of C2ε, ε
is higher, resulting in a lower turbulent kinetic energy and
increased vapour penetration.
0 0.5 1 1.5 time ASOI [ms]
0
20
40
60
80
100
120
k- Standard k- RNG k- Realisable Spalart-Allmaras k- SST
experiment
Figure 3: Evolution of the vapour penetration in time
COLLISION MODEL As expected, switching off the collision model did
not alter the simulations results a lot, nor did it speed up the
calculation time. Due to its positive influence on the convergence
of the continuity equation, the O’Rourke collision model was left
enabled.
TURBULENT DISPERSION Turbulent dispersion makes it easier for
droplets to disperse away from the spray axis. This enhances the
evaporation because more droplets end up in hotter neighbouring
cells with a lower vapour mass fraction. When turbulent dispersion
was disabled while keeping all other settings the same, the liquid
length was almost doubled. It also slightly affected the vapour
penetration, especially in the early stages of the spray (0.1 ms -
0.3 ms) when the droplets are still close to the vapour boundary.
The results with turbulent dispersion (discrete random walk model)
were closer to the experiments, so it was used in all further
calculations.
SENSITIVITY ANALYSIS A sensitivity analysis was per- formed by
changing various fluid properties, initial condi- tions and other
parameters to see which ones influenced the result the most so that
extra care could be taken to provide accurate input. By looking at
the equations of the predicted wavelengths of the KH-RT model, the
wavelengths only scale approximately to the square root
of the surface tension. However, at 60 bar, the surface tension
drops quickly when it approaches its critical tem- perature of 659
K. [17] For this reason, surface tension values should be made
temperature dependent. The viscosity only affects the KH breakup,
so it had a rather small effect. The density of liquid n-dodecane
varies between 704 kg/m³ and 436 kg/m³ in the temperature range 363
K - 663 K at a pressure of 60 bar. Drag and breakup will thus be
more accurately predicted when a temperature-dependent density is
used. The thermal conductivity of the air-fuel vapour mixture also
influenced the liquid penetration quite a lot. A decrease of the
thermal conductivity by a factor 4.5 increased the liquid
penetration by 50%.
DISCUSSION OF RESULTS
The model as described in the last section was used to calculate
the parametric variations of ECN spray A. The model constants and
fluid properties remained un- changed, only the appropriate
boundary conditions from table 2 were applied. The vapour
penetration was well reproduced. The liquid penetration showed all
the right trends but was sometimes quantitatively a bit off. Addi-
tionally, quite a few cases were lacking experimental data. Figure
4 shows the evolution of the vapour penetration for the first 3
cases in the table, i.e. a variation in injection pressure. The
fuel vapour penetrates further into the domain for increasing
injection pressures due to the increased momentum of the fuel
leaving the nozzle. The simulation of 150 MPa was tuned to fit an
experiment at 152.7 MPa. The simulation at 50 MPa corresponds to an
experiment at 53.9 MPa and the vapour penetration was mostly within
the experimental error. The simulation at 100 MPa, however,
corresponds to an experiment at 94.5 MPa which explains to a large
extent why the vapour penetration is overestimated for that
case.
0 0.5 1 1.5 time ASOI [ms]
0
10
20
30
40
50
60
Figure 4: Vapour penetration for different injection pres-
sures
Figure 5 shows the liquid penetration for the same range
5
of injection pressures. Experimentally, the liquid penetra- tion
stays almost constant due to two balancing effects. For lower
injection pressures, the injection speed is lower, but the breakup
and evaporation are also less intense. The latter effect seems more
important in the simulation as in the experiments. A possible
explanation is that in the simulation at lower pressures, the
droplets them- selves cannot create additional turbulence that
improves the evaporation. In the simulation, the droplets act as
momentum sources and turbulence is created by shear in the flow of
the continuous gas phase, whereas in reality droplets can directly
produce turbulent eddies. In figure 6, liquid penetration values
for varying ambient density are presented. For decreasing ambient
densities, the liquid penetration increases because there is less
air hindering the movement of the spray. Drag forces and KH breakup
decrease, but RT breakup increases. This could explain why the
simulation is less sensitive to ambient density changes.
50 100 150 Injection Pressure [MPa]
0
2
4
6
8
10
12
7.6 15.2 22.6 Ambient density [kg/m³]
0
5
10
15
20
25
Figure 6: Liquid penetration for varying ambient densities
In figure 7, liquid penetration values for different ambient
temperatures are shown. Here, the simulation also follows the right
trend, while being less sensitive to changes in ambient temperature
compared to the experiments.
Introducing more temperature-dependent fluid properties such as the
viscosity should increase the sensitivity of the model to changes
in ambient temperature. Also, KH breakup and RT breakup do not
respond in the same way to variations in ambient temperature and
density. When the relative importance of KH breakup in the KH- RT
model was increased, the liquid penetration changes due to
variations in ambient density and temperature were more accurately
followed. For the case with the lower ambient density of 7.6 kg/m³
the liquid penetration increased from 13.7 mm to 18.4 mm, and for
the case with the lower ambient temperature of 700 K the liquid
penetration increased from 11.8 mm to 13.2 mm.
700 900 1100 Ambient temperature [K]
0
2
4
6
8
10
12
14
16
18
CONCLUSION
The Eulerian-Lagrangian approach to spray modelling can accurately
predict global quantities such as the vapour penetration and spray
angle. These are quite insensitive to the breakup and are mainly
determined by the total amount of momentum transferred to the gas
phase. If local quantities near the nozzle are important, Eulerian
phase models are better suited. The accuracy of La- grangian phase
models is determined by a rather complex interaction between
various spray submodels. Some know-how is needed to tune various
model constants in order to account for varying conditions in the
nozzle. Although models exist which take into account the cavi-
tation and turbulence in the nozzle [12], they are not yet
available in a lot of CFD software packages. Additionally, it is
hard to obtain a fully grid-independent solution for liquid-
related quantities.
Experimental data for the parametric variations of ECN spray A is
far from complete and additional experiments would benefit the
modelling community. It also remains to be seen that the best
approach developed for the ECN sprays scales up well to the
dimensions of marine diesel engines. Later, this knowledge needs to
be translated from marine diesel engines to dual fuel engines where
the most notable difference is the presence of premixed natural
gas.
6
Contents
1 Introduction 1 1.1 IMO Tier 3 NOx limits . . . . . . . . . . . .
. . . . . . . . . . . . . . 1 1.2 Dual fuel internal combustion
engines . . . . . . . . . . . . . . . . . 1
1.2.1 Working principle . . . . . . . . . . . . . . . . . . . . . .
. . . 2 1.2.2 Advantages, disadvantages and current problems . . .
. . . . 3
2 Computational Fluid Dynamics 5 2.1 Governing equations . . . . .
. . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Mass equation . . . . . . . . . . . . . . . . . . . . . . . .
. . 6 2.1.2 Navier-Stokes equations . . . . . . . . . . . . . . . .
. . . . . 6 2.1.3 Energy equation . . . . . . . . . . . . . . . . .
. . . . . . . . 7 2.1.4 Species equations . . . . . . . . . . . . .
. . . . . . . . . . . . 7
2.2 RANS turbulence models . . . . . . . . . . . . . . . . . . . .
. . . . . 7 2.2.1 Spalart-Allmaras model . . . . . . . . . . . . .
. . . . . . . . 8 2.2.2 k-ε models . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 9 2.2.3 k-ω SST model . . . . . . . . . .
. . . . . . . . . . . . . . . . 10 2.2.4 Reynolds Stress model . .
. . . . . . . . . . . . . . . . . . . . 10
3 Literature review 11 3.1 CFD strategies in spray formation . . .
. . . . . . . . . . . . . . . . . 11
3.1.1 Eulerian-Lagrangian approach . . . . . . . . . . . . . . . .
. . 11 3.1.2 Eulerian-Eulerian approach . . . . . . . . . . . . . .
. . . . . 12 3.1.3 Hybrid approach . . . . . . . . . . . . . . . .
. . . . . . . . . 13
3.2 Common practice in the simulation of fuel sprays . . . . . . .
. . . . 13 3.2.1 Discretisation, Mesh and Time step . . . . . . . .
. . . . . . . 13 3.2.2 Turbulence model . . . . . . . . . . . . . .
. . . . . . . . . . . 16 3.2.3 Injection and primary breakup model
. . . . . . . . . . . . . . 17 3.2.4 Secondary Breakup model . . .
. . . . . . . . . . . . . . . . . 21 3.2.5 Collision model . . . .
. . . . . . . . . . . . . . . . . . . . . . 25 3.2.6 Droplet drag
model . . . . . . . . . . . . . . . . . . . . . . . . 26 3.2.7
Evaporation model . . . . . . . . . . . . . . . . . . . . . . . .
28 3.2.8 Turbulent dispersion . . . . . . . . . . . . . . . . . . .
. . . . 30
3.3 Goal statement . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . 32
xii
4 Simulation setup and model validation 34 4.1 ECN spray A . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 4.2
Discretisation . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . 37
4.2.1 Dimensions and structure of the mesh . . . . . . . . . . . .
. 37 4.2.2 Injection . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . 40 4.2.3 Pressure-Velocity Coupling . . . . . . . . .
. . . . . . . . . . . 48 4.2.4 Mesh dependence . . . . . . . . . .
. . . . . . . . . . . . . . 48 4.2.5 Time step dependence . . . . .
. . . . . . . . . . . . . . . . . 50
4.3 Choice of turbulence model . . . . . . . . . . . . . . . . . .
. . . . . 52 4.4 Influence of fluid properties and initial
conditions . . . . . . . . . . . 57
4.4.1 Sensitivity analysis . . . . . . . . . . . . . . . . . . . .
. . . . 57 4.4.2 Temperature-dependent fluid properties . . . . . .
. . . . . . 59
4.5 Proposed methodology for fuel spray simulations . . . . . . . .
. . . 61
5 Discussion of results 64 5.1 Spray penetration . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . 64
5.1.1 Evolution of the vapour boundary . . . . . . . . . . . . . .
. . 64 5.1.2 Vapour mass fraction distribution . . . . . . . . . .
. . . . . . 66 5.1.3 Streamlines and velocity field . . . . . . . .
. . . . . . . . . . 67 5.1.4 Momentum . . . . . . . . . . . . . . .
. . . . . . . . . . . . . 68 5.1.5 Temperature . . . . . . . . . .
. . . . . . . . . . . . . . . . . 69
5.2 Near-nozzle region . . . . . . . . . . . . . . . . . . . . . .
. . . . . . 70 5.2.1 Volume fraction of liquid n-dodecane . . . . .
. . . . . . . . . 70 5.2.2 Droplet size distribution . . . . . . .
. . . . . . . . . . . . . . 71 5.2.3 Breakup mechanisms . . . . . .
. . . . . . . . . . . . . . . . . 72
5.3 Parametric variations . . . . . . . . . . . . . . . . . . . . .
. . . . . . 73 5.3.1 Injection pressure . . . . . . . . . . . . . .
. . . . . . . . . . . 73 5.3.2 Ambient density . . . . . . . . . .
. . . . . . . . . . . . . . . 75 5.3.3 Ambient temperature . . . .
. . . . . . . . . . . . . . . . . . . 78 5.3.4 Combination of
ambient density and temperature . . . . . . . 80
6 Transition to reacting conditions 83
7 Conclusion 85
List of Figures
1.1 Regulations regarding NOx emissions issued by the IMO (figure
(a) by dieselnet.org and figure (b) from IMO’s Annex VI . . . . . .
. . . . . . 1
1.2 Working principle of a dual fuel engine (figure from
targettrainingcentre.nl) . . . . . . . . . . . . . . . . . . . . .
. . . . . 2
3.1 Left: Mesh is fine enough to represent the droplet Right:
Droplets smaller than the cell size cannot be represented [29] . .
. . . . . . . . 12
3.2 Adaptive mesh of Argonne National Laboratory at start of
injection, used in the first ECN Workshop [31] . . . . . . . . . .
. . . . . . . . . 14
3.3 Breakup of blobs by secondary breakup models using the concept
of liquid core length, figure from ANSYS Fluent Theory Guide [11] .
. . . 19
3.4 Kelvin-Helmholtz instability: related to the existence of a
difference in speed between two fluids [29] . . . . . . . . . . . .
. . . . . . . . . . . 21
3.5 Comparison of the creation of child droplets due to the KH
instability and RT instability [34] . . . . . . . . . . . . . . . .
. . . . . . . . . . . 23
4.1 The constant volume combustion chamber at Sandia National
Laboratories 36 4.2 Quarter cubic mesh as created by Gilles Decan
[7] . . . . . . . . . . . . 38 4.3 Comparison of the liquid
penetration in a 3D quarter cubic mesh and a
2D axisymmetric mesh . . . . . . . . . . . . . . . . . . . . . . .
. . . . 39 4.4 Injection rate as given by the tool on the website
of UPV compared to
the mass flow rate obtained by applying the principle of Bernoulli
and a Cd of 0.89 . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . 40
4.5 Influence on the liquid penetration of the B1 breakup constant
in the KH-RT model (CRT = 1) . . . . . . . . . . . . . . . . . . .
. . . . . . . 44
4.6 Influence on the liquid penetration of the CRT breakup constant
in the KH-RT model (B1 = 1.73) . . . . . . . . . . . . . . . . . .
. . . . . . . 44
4.7 Effect of node based averaging on the liquid penetration . . .
. . . . . 45 4.8 Comparison between the Blob method (90 µm at 1
particle per parcel)
and a uniform injection of 1 µm droplets at 500 particles per
parcel . . 46 4.9 Effect of Turbulent dispersion on the vapour
boundary at 0.2 ms . . . . 48 4.10 Liquid penetration for different
levels of refinement of the mesh . . . . 50 4.11 Vapour boundary at
1.44 ms for different levels of refinement of the mesh 50 4.12
Liquid penetration for different time steps . . . . . . . . . . . .
. . . . 51 4.13 Vapour penetration for different time steps . . . .
. . . . . . . . . . . . 52
xiv
4.14 Effect of the turbulence model on the liquid penetration . . .
. . . . . 54 4.15 Effect of the turbulence model on various
properties . . . . . . . . . . 56 4.16 Liquid penetration and
vapour boundary for different values of C2ε . . 57 4.17 Influence
of various fluid properties and initial conditions on the
liquid
penetration . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . 59
5.1 Vapour and liquid boundary at 0.02 ms, 0.04 ms, 0.1 ms, 0.24
ms, 0.5 ms, 0.8 ms, 1.08 ms and 1.5 ms: blue line = vapour boundary
of the simulation, black dotted line = experimental vapour
boundary, edge of blue surface = liquid boundary of the simulation
. . . . . . . . . . . . 65
5.2 Vapour mass fraction distribution at 1.44 ms ASOI . . . . . . .
. . . . 66 5.3 Pathlines coloured by velocity magnitude at 1.44 ms
ASOI . . . . . . . 67 5.4 Momentum flow rate through planes
orthogonal to the spray axis placed
at different distances from the nozzle at 1.44 ms ASOI . . . . . .
. . . 68 5.5 Temperature of the gas phase at 1.44 ms ASOI . . . . .
. . . . . . . . . 69 5.6 Temperatures of the discrete phase
droplets at 1.44 ms ASOI . . . . . . 70 5.7 Volume fraction of
n-dodecane droplets in the cells near the nozzle at
1.44 ms ASOI . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . 70 5.8 Droplets coloured by particle diameter with a
scaled diameter
representation (top) and a constant diameter representation
(bottom) 72 5.9 Comparison of the WAVE breakup times (top) and
KH-RT breakup times
(bottom) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . 73 5.10 Vapour penetration for different injection
pressures . . . . . . . . . . . 73 5.11 Vapour boundary for
different injection pressures at 1.44 ms ASOI . . . 74 5.13
Turbulent viscosity ratio and number of discrete phase particles
along
the spray axis at 1.44 ms ASOI . . . . . . . . . . . . . . . . . .
. . . . 75 5.12 Comparison of the liquid penetration at different
injection pressures for
simulation and experiment . . . . . . . . . . . . . . . . . . . . .
. . . . 75 5.14 Comparison of the liquid penetration at different
ambient densities for
simulation and experiment . . . . . . . . . . . . . . . . . . . . .
. . . . 76 5.15 Vapour penetration evolution and the vapour
boundary at 1.44 ms ASOI
for different ambient densities . . . . . . . . . . . . . . . . . .
. . . . . 77 5.16 Velocity of the continuous gas phase and the
discrete phase droplets
along the spray axis at 1.44 ms ASOI for different ambient
densities . . 77 5.17 Turbulent viscosity ratio and density of the
gas phase along the spray
axis at 1.44 ms ASOI . . . . . . . . . . . . . . . . . . . . . . .
. . . . . 77 5.18 Comparison of the liquid penetration at different
ambient temperatures
for simulation and experiment . . . . . . . . . . . . . . . . . . .
. . . . 78 5.19 Temperature of the gas phase and the fuel droplets
along the spray axis
at 1.44 ms ASOI . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . 79 5.20 Evolution of the vapour penetration for different
ambient temperatures 80
xv
5.21 Evolution of the static pressure change in the combustion
chamber for different ambient temperatures . . . . . . . . . . . .
. . . . . . . . . . 80
5.22 Comparison of the liquid penetration at different ambient
densities for simulation and experiment . . . . . . . . . . . . . .
. . . . . . . . . . . 81
5.23 Vapour penetration evolution and the vapour boundary at 1.44
ms ASOI for different ambient densities and temperatures . . . . .
. . . . . . . 82
6.1 Contours of constant temperature for ECN spray A in reacting
conditions 84 6.2 Contours of constant OH mass fraction for ECN
spray A in reacting
conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . 84
List of Tables
1.1 Pros and Cons of Dual fuel engines as listed by Decan et al.
[7] . . . . 3
3.1 Courant numbers used by different authors for fuel spray
simulations . 15
4.1 Most important specifications of the ECN Spray A baseline
operating condition [2] . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . 35
4.2 Parametric variations of ECN spray A with the first one being
the baseline condition [15] . . . . . . . . . . . . . . . . . . . .
. . . . . . . 35
4.3 Cell count and dimensions of cells in the different zones of
the mesh . 38 4.4 Ambient temperature, density and pressure, liquid
density, injection
velocity and mass flow rate calculated based on Cd = 0.89 for the
parametric variations of ECN spray A . . . . . . . . . . . . . . .
. . . . 41
4.5 Standard breakup constants of the KH-RT model compared to the
tuned breakup constants . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . 42
4.6 Change in values of various fluid properties and initial
conditions for the sensitivity analysis . . . . . . . . . . . . . .
. . . . . . . . . . . . . 58
4.7 Isobaric data of the density for dodecane at 60 bar [17] . . .
. . . . . . 59 4.8 Temperature-dependent surface tension data for
saturated dodecane [17] 60 4.9 Other properties of liquid dodecane
. . . . . . . . . . . . . . . . . . . . 60
5.1 Droplet size statistics at 1.44 ms ASOI . . . . . . . . . . . .
. . . . . . 71 5.2 Ambient temperature, density and pressure,
liquid density, injection
velocity and mass flow rate calculated based on Cd = 0.89 for some
extra variations of ECN spray A . . . . . . . . . . . . . . . . . .
. . . . 81
xvi
Abbreviations and symbols
Symbol Unit Description
CFD Computational Fluid Dynamics x or x1 m (distance to the nozzle
along the) axial direction y or x2 m (distance to the nozzle along
the) radial
direction z or x3 m (distance to the nozzle along the)
direction
perpendicular to x and y u or u1 m/s velocity along the x-direction
v or u2 m/s velocity along the y-direction w or u3 m/s velocity
along the z-direction ρ kg/m³ mass density ρl kg/m³ mass density of
the liquid phase ρg kg/m³ mass density of the gas phase k m²/s²
turbulent kinetic energy ε m²/s³ turbulent dissipation rate µ Pa.s
(molecular) dynamic viscosity ν m²/s (molecular) kinematic
viscosity µt Pa.s turbulent dynamic viscosity νt m²/s turbulent
kinematic viscosity σ N/m surface tension cp J/(kg.K) heat capacity
at constant pressure pinj Pa injection pressure pa Pa ambient
pressure Ta K ambient temperature ρa kg/m³ ambient density m kg/s
mass flow rate RANS Reynolds-Averaged Navier-Stokes LES Large Eddy
Simulation ASOI after start of injection KH Kelvin-Helmholtz RT
Rayleigh-Taylor B1 model constant of KH breakup model (or WAVE) CRT
model constant of RT breakup model LP m liquid penetration or
liquid length VP m vapour penetration ECN Engine Combustion
Network
xvii
1Introduction
1.1 IMO Tier 3 NOx limits
The International Maritime Organisation (IMO) is a specialised
agency of the United Nations. Their main goal is to create a set of
global regulations to improve the safety, security and economical
performance of international shipping. As of January 2016, they
introduced the IMO NOx Tier III requirement.[1] This new iteration
of their NOx limit for large marine diesel engines (≥ 130 kW)
severely limits the allowable NOx emissions in certain emission
control areas (ECAs) which are mostly found along the coasts of
North America and Europe, as shown in figure 1.1b. Compared to Tier
II, the limit has gone down by about 75% to 2.0 g/kWh for an engine
with a rated speed above 2000 rpm and gradually going up to 3.4
g/kWh for rated engine speeds below 130 rpm as can be seen in
figure 1.1a. This reflects the observation that at lower engine
speeds generally more NOx is produced due to the higher available
time for formation. Engine tuning alone is not sufficient any more
to meet these new standards.
(a)IMO NOx limits with respect to rated engine speed
(b)Existing and possible future ECAs
Figure 1.1: Regulations regarding NOx emissions issued by the IMO
(figure (a) by dieselnet.org and figure (b) from IMO’s Annex
VI
1.2 Dual fuel internal combustion engines
1
1.2.1 Working principle
To address the increasingly tougher emissions regulations,
different solutions have to be considered. Besides long-term
solutions, there is also a more immediate need to reduce emissions
and improve the performance of the modern-day engines. One such
solution is the use of dual fuel systems. It consists in mixing the
air in the intake manifold with a high octane number fuel, such as
alcohols or natural gas. This gaseous air-fuel mixture then gets
compressed to finally be ignited with a small amount of pilot fuel.
The pilot fuel has a low octane number, so it ignites spontaneously
near the end of the compression stroke and further increases the
heat and pressure in the cylinder to ignite the premixed air-fuel
mixture. Most commonly diesel is used for this purpose. The entire
process is schematically shown in figure 1.2.
Figure 1.2: Working principle of a dual fuel engine (figure from
targettrainingcentre.nl)
The gaseous fuel is often natural gas because of its abundant
availability and clean burning properties. Moreover, Yang et al.
[18] mention that biogas, syngas and hydrogen are also sometimes
used in dual fuel engines, but the high octane number of natural
gas makes it a more suitable fuel for use in compression ignition
engines due to their high compression ratio. The lower combustion
rate and flame temperature also has a positive influence on the
choice of natural gas. The goal is to mainly run on natural gas.
When less diesel fuel is injected, NOx and PM emissions generally
decrease but at a certain point combustion might become unstable.
This amount is often fixed at around 10% of the mass flow rate at
full load and stays constant at lower loads such as in the study of
Wagemakers et al.[19]. This means that at lower loads a bigger
relative amount of diesel is injected and less favourable emissions
are obtained. By adding the natural gas circuit, existing diesel
engines can be converted to a dual fuel engine at a relatively low
cost and thus lowering the NOx emissions considerably to meet the
demand of future emission standards. The engine also retains its
capability to run in full diesel operation. Depending on the price
of diesel
1.2 Dual fuel internal combustion engines 2
and heavy fuel oil, it can be economically preferable to operate
solely on diesel outside of the ECAs and switch to dual fuel
operation when entering the ECAs.
1.2.2 Advantages, disadvantages and current problems
The explanation of the working principle already mentioned some
advantages such as the flexible choice of fuel, the possibility of
a retrofit on an existing diesel engine and reduced emissions. Dual
fuel systems are most often used in marine diesel engines or other
heavy-duty diesel engines where power-to-weight ratios are less
important. Lately, the attention of the automotive sector has also
increased, considering it could be an alternative to exhaust
aftertreatment systems because it drastically reduces engine-out
emissions. In this context, more advantages and disadvantages can
be found. Decan et al. [7] listed the following pros and
cons.
Table 1.1: Pros and Cons of Dual fuel engines as listed by Decan et
al. [7]
Pros Cons - Allows retrofitting of current engines - Complexity of
combustion - Low-cost retrofit - Lack of optimisation - Possibility
of reduced PM and NOx emissions
- Possibility of irregular / incomplete combustion and increased HC
and CO emissions
- Substitute diesel with plentiful natural gas - Issues present in
gasoline engines appear - Similar performance to regular diesel
engine
- Higher specific fuel consumption
- No need for additional maintenance
It can be noticed that most of the cons are due to a lack of
optimisation of dual fuel engines, especially at low load.
Wagemakers et al.[19] explained that the higher emissions of
hydrocarbons and CO are mainly caused by the over-lean fuel mixture
at low loads. The lower temperatures and the lower amount of fuel
can lead to unburnt fuel that gets left behind in crevice volumes
in the cylinder and which later appears in the exhaust.
Additionally, problems like knock and short circuiting of fuel from
the inlet valve to the exhaust valve could appear when operating on
dual fuel. The higher specific fuel consumption is again mainly at
low loads.
Srinivasan et al.[20] already found out early on that increasing
the intake charge temperature is an effective strategy to improve
combustion stability and emissions. However, methods to achieve
this almost always need extra equipment such as throttle bodies,
turbochargers or exhaust gas recirculation (EGR) systems. Other
researchers, such as Yang et al.[18] looked at the ideal injection
pressures and timing and natural gas injection timing. They
concluded that delaying the natural
1.2 Dual fuel internal combustion engines 3
gas injection at low load creates a more stratified charge in the
cylinder and thus improves emissions in most cases. A lot of
experimental research is being done on dual fuel engines but also
more fundamental research has to be done to fully optimise and
understand this kind of combustion.
To this end, Computational Fluid Dynamics (CFD) can be a useful
tool. The in-cylinder flow field, spray formation, combustion and
heat transfer can all be modelled to have a better understanding of
what happens inside the cylinder for different operating
conditions. However, a lot of factors influence the accuracy of a
CFD calculation and everywhere models are used, modelling errors
can occur. So apart from knowing the sensitivity of the CFD
simulation to certain parameters, experimental validation of
results is still needed. On the other hand, the benefits of CFD
include the access to quantities which are otherwise unmeasurable
and the ease at which initial conditions and boundary conditions
can be changed.
1.2 Dual fuel internal combustion engines 4
2Computational Fluid Dynamics
This chapter will give a brief introduction to Computational Fluid
Dynamics or CFD. It will explain what its principles are while
covering the governing equations and some turbulence models. It
should provide a sufficient amount of background knowledge for
those who are not so familiar with CFD to be able to follow the
subsequent chapters. If the end of this chapter leaves you wanting
to read more about CFD, the book of Versteeg et al. [21] is a good
starting point. In their book, they define CFD as “the analysis of
systems involving fluid flow, heat transfer and associated
phenomena such as chemical reactions by means of computer-based
simulation”. In other words, it tries to predict flow patterns and
other related quantities such as the heat transfer, pressure
differences and their associated forces on bodies in the flow,
...
A traditional example is trying to calculate the drag and lift of
an aerofoil. Early attempts used the potential flow theory which
made some heavy assumptions to be able to obtain an algebraic
solution of the velocity field. It assumed the flow to be
irrotational and inviscid (i.e. the effects of viscosity are
neglected). Therefore, this theory is unable to represent vortices
and boundary layers. The rise of the available computing power in
the middle of the 20th century made it possible to numerically
solve for the velocity field instead of assuming an inviscid,
irrotational flow. As the computing power and algorithms
progressed, more accurate solutions and increasingly complex
geometries came within reach. Nowadays, CFD is used in a wide range
of industries, ranging from aerospace and automotive to
construction and the chemical process industry. It is even used in
the healthcare industry to calculate blood flows in veins.
2.1 Governing equations
The differential equations that define fluid flow are often called
the transport equations because they express the conservation of
several quantities in infinitesimal domains. These quantities are
mass, the different momentum components (x and y in 2D and x, y and
z in 3D) and energy. All these equations have the same basic form.
The net change of the quantity in the volume is equal to the amount
of this quantity that enters the volume minus what leaves the
volume plus the net amount
5
of this quantity that gets produced in the volume. What leaves one
volume enters other volumes and in this way the quantities get
transported along the domain.
The user of a CFD program needs to divide or discretise the domain
into cells in a thoughtful manner, which is also called meshing.
This allows the transport equations to be discretised into a system
of algebraic equations. Most CFD programs use the finite volume
method for this purpose. The resulting system of algebraic
equations is then solved iteratively until a solution is
obtained.
2.1.1 Mass equation
The mass equation or the continuity equation is one of the basic
conservation equations used in any CFD software. If the
conservation of mass in an infinitesimal cube around a given point
is expressed, the following equation is obtained. It is valid at
any given point in an unsteady, compressible flow.
∂ρ
∂ρ
∂xi = 0 (2.2)
In these equations u or u1, v or u2 and w or u3 denote the
velocities along the x, y and z directions, respectively. When the
density ρ in a point is constant, equation 2.2 simplifies to the
divergence of the velocity vector being zero.
∂ui ∂xi
= 0 (2.3)
All flows which obey equation 2.3 are said to be
incompressible.
2.1.2 Navier-Stokes equations
The Navier-Stokes equations or the momentum equations use Newton’s
second law to state that the rate of momentum increase of a fluid
particle is equal to the sum of forces acting on it. This includes
forces due to a pressure difference, viscosity, gravity and other
external body forces acting on the volume. The Navier-Stokes
equations
2.1 Governing equations 6
given below are for a Newtonian fluid with a constant density ρ0
and a constant kinematic viscosity ν.
∂ui ∂t
+ fi body forces
2.1.3 Energy equation
∂
∂xj (−ujp− qj + uiτij) (2.5)
In this equation e0 is the total energy e+ ukuk 2 , qj is the heat
flux calculated by the
law of Fourier and τij is the viscous stress. For a Newtonian fluid
τij is defined as follows.
τij = µ
( ∂ui ∂xj
2.1.4 Species equations
∂
∂Yk ∂xi
) + ωk (2.7)
In which Dk is the diffusion coefficient and ωk the source term for
the species k.
2.2 RANS turbulence models
A lot of flows in engineering applications are turbulent in nature.
These chaotic 3D fluctuations make it hard to compare experiments
that are seemingly carried out under the same circumstances.
Therefore, one is often only interested in the mean values of a
flow. This is the statistical approach to turbulence. In order to
obtain these mean values, the Reynolds-Averaged Navier-Stokes
(RANS) equations are used. They are derived by splitting up the
velocity in its mean value ui and
2.2 RANS turbulence models 7
its fluctuating part u′i, substituting this in the Navier-Stokes
equations (equation 2.4) and time-averaging the entire equation.
The result are equations that are very similar to the Navier-Stokes
but with an added term.
∂ui ∂t
∂xj (u′iu′j) + fi, i = 1, 2, 3 (2.8)
And similarly for the continuity equation (equation 2.3)
∂ui ∂xi
= 0 (2.9)
−ρ0u′iu ′ j is defined as the Reynolds stress tensor Rij . It
represents the influence of
the fluctuating velocities on the mean flow field. To close the
system of equations, Rij has to be determined as a function of the
mean velocities, mean pressure and other available quantities. In
what follows, different RANS turbulence models will be introduced.
They each close the system of equations in a different way. The
models discussed here, except for the Reynolds Stress model, all
rely on the Boussinesq hypothesis. He proposed that the Reynolds
stress tensor is proportional to mean rates of deformation. In this
way, turbulence can be modelled as an additional turbulent
viscosity µt (or eddy viscosity) that adds to the effect of the
molecular viscosity.
Rij = −ρ0u′iu ′ j = µt
( ∂ui ∂xj
+ ∂uj ∂xi
3ρ0u′iu ′ i (2.10)
RANS turbulence models are not the only way of modelling
turbulence. Large eddy simulation and direct numerical simulation
are briefly mentioned in section 3.2.2.
2.2.1 Spalart-Allmaras model
The Spalart-Allmaras model [22] is one of the simplest and most
robust turbulence models because it uses only one additional
transport equation (equation 2.11). The variable ν is proportional
to the turbulent viscosity, so it can be used to directly determine
µt in equation 2.10. The model was developed with aerospace
applications in mind, so it is commonly used to predict the flow
over aerofoils. It produces quite accurate results in these
applications, but in other applications the results can be quite
inaccurate. The ANSYS Fluent User guide [23] even warns for
relatively large errors in free shear flows, such as round and
plane jets.
∂ν
2.2.2 k-ε models
The k-εmodel is the most used turbulence model for general
engineering applications in the industry. This rather old model
still is a good compromise between its ease of use on the one hand
and its performance on the other hand.[24] It uses two additional
transport equations. One for the turbulent kinetic energy k =
1
2u ′ iu ′ i and
one for the turbulent dissipation rate ε. The turbulent kinematic
viscosity is then based on the instantaneous values of these two
quantities.
νt = Cµ k2
ε (2.12)
where Cµ is a model constant. The two transport equations are given
here in vector notation for a fluid with a constant density and
viscosity. Once again, it can be noticed that the transported
quantities change by convection, diffusion, local production and
destruction.
∂k
k (C1εP − C2εε) (2.14)
In these equations u is the mean flow velocity vector, P the local
production term and the 5 C’s are model constants.
P = 1 2Cµ
) (2.15)
The model constants are determined by experimentally comparing the
results to basic flows. Commonly used values are:
Cµ = 0.09, CK = 1, Cε = 0.069, C1ε = 1.44, C2ε = 1.92 (2.16)
Although the k-ε model performs well in a lot of applications, it
still has a few shortcomings. Métais [24] states that the k-ε model
is known to produce poor results in flows where normal stresses
dominate the shear stresses, flows with strong streamline
curvatures or flows with high levels of anisotropy. Examples of
such flows include flows with a large adverse pressure gradient or
flows in a rotating frame. The k-ε model is also known for the
round jet / plane jet anomaly. Using the constants given above it
can predict the flow field of a plane jet quite accurately, but
when applied to an axisymmetric jet the spreading rate is severely
overestimated. Because of its age and popularity, a lot of people
have tried to improve the standard model. For example, the round
jet / plane jet anomaly was solved in the realisable k-ε model by
imposing some physical constraints on the solution. In the standard
model u2
2.2 RANS turbulence models 9
could become negative, so one of the adjustments was to make Cµ
locally dependent on the mean flow field to guarantee that u2 ≥ 0 (
= condition for realisability). Another commonly used k-ε model is
the RNG model, but as its principles do not help to further
understand this thesis, the interested reader is referred to the
literature.
2.2.3 k-ω SST model
Not all two equation turbulence models use k and ε as the
transported variables. The specific dissipation rate ω = ε/k can
also be used instead of the dissipation rate ε. The result is a
model that performs better near walls, but that is much more
sensitive to the free stream boundary conditions. The idea of
Menter [25] was to develop a hybrid model that switched between the
k-ε model in free stream conditions and the k-ω model near the
wall. In this manner, he got rid of the strong dependency of the
free stream boundary conditions while retaining the superior
performance of the k-ω model near the wall. In his SST (Shear
Stress Transport) model, he also accounted for the transport of
turbulent shear stress in the boundary layer by assuming the
principal turbulent shear stress to be proportional to the
turbulent kinetic energy. It is worth noting that when Menter
applied his model on some free shear flows, the results were almost
identical to the k-ε model.[25] Due to its versatile behaviour, the
k-ω SST model has been successfully used in applications with
wall-bounded flows, e.g., turbomachinery.
2.2.4 Reynolds Stress model
The Reynolds Stress model is a 7 equation turbulence model. It has
6 transport equations for each of the components of the Reynolds
stress tensor Rij = −ρ0u′iu
′ j
and a transport equation for ε similar to the one in the k-ε model
(equation 2.14) but with a modified expression for P. The Reynolds
Stress model does not use the eddy viscosity hypothesis (equation
2.10), so it is very well suited to handle anisotropic flows.
However, this comes at a rather large additional computational
cost. It has also proven itself to be less numerically stable than
the aforementioned two equation turbulence models.
2.2 RANS turbulence models 10
3Literature review
3.1 CFD strategies in spray formation
A CFD simulation of an entire dual fuel engine is a very daunting
task in terms of modelling and calculation time. The flow in an
engine is highly reactive and turbulent and moves in a
geometrically complex, moving mesh. Furthermore, the different
processes in an engine have a large variety of characteristic time
constants. In an effort to reach this goal, various processes are
studied separately. This thesis will focus on the diesel spray
breakup at engine-like ambient conditions, in particular automotive
engine-like ambient conditions rather than marine diesel
engine-like ambient conditions. The choice for a spray
representative of an automotive engine is linked to the extensive
experimental database of ECN [2] (more info in section 4.1) and the
amount of research done on ECN spray A. A diesel spray could be
simulated in a cylinder, but the cyclic variations and the
uncertainty on the boundary conditions ask for a different
approach. Most of the time, research on spray breakup is done in
constant volume combustion chambers. They have the added benefit of
being optical accessible for measuring spray characteristics. In
the next sections, different simulation approaches are
discussed.
3.1.1 Eulerian-Lagrangian approach
Most authors [3, 4, 5, 6, 7] use a Eulerian-Lagrangian approach. It
is also called a discrete phase model or DPM. In this approach,
droplets of similar properties are grouped in discrete particles
and are tracked along the mesh. Additional models govern the
evaporation, coalescence and shedding of those particles. In this
way, they interact with the continuous gas phase. Each time step,
the continuous phase flow field is calculated. This flow field
determines the trajectory of the discrete phase particles and these
in turn determine the new continuous phase source terms. This
process is then repeated every time step. One of the main
assumptions of this approach is that the volume fraction of liquid
droplets in a cell stays low, generally less than 10 %. This
creates a possible source for errors in the near-nozzle region and
imposes a lower limit on the used cell sizes. Lucchini et al. [5]
remark that the grid size generally adopted is much larger than the
nozzle diameter (about 2-5 times). The need for a relatively
coarser mesh is at the same time an advantage.
11
It provides an efficient way of representing the small droplets (in
the order of 0.1 µm) naturally present in high-pressure fuel
sprays. This makes it a frequently used approach.
3.1.2 Eulerian-Eulerian approach
Some authors [26, 27, 28] use a Eulerian-Eulerian approach. Most of
the time a Volume of Fluid (VOF) method is used. The transport
equation 3.1 of the volume fraction α1 of phase 1 governs the
transport of the two phases, along with the conservation of
momentum. The transport equation expresses the conservation of mass
of phase 1.
∂α1 ∂t
= 0 (3.1)
with u,v and w the velocities according to the x, y and
z-direction. The solution is a scalar field of volume fractions.
The interface of the two phases is generally taken in the cells
where the volume fraction is 0.5 (red line in figure 3.1). A
general observation is the need of much smaller cell sizes. For
being able to represent a droplet in a VOF model, the cell size
must be smaller than the droplet.
Figure 3.1: Left: Mesh is fine enough to represent the droplet
Right: Droplets smaller than the cell size cannot be represented
[29]
As droplets in high-pressure diesel sprays can reach diameters
smaller than 1 µm, one can easily imagine the large increase in
cell count. Another model is the Level-Set method as used by
Herrmann [8], but this method also experiences the same issues as
VOF. The Level-Set method also expresses the conservation of a
scalar quantity, but instead of the volume fraction of phase 1 a
function that expresses the distance to the interface is used. The
interface can then be reconstructed by the level surface for which
the distance function is equal to 0. An advantage of the Level-Set
method is that it is easier to reconstruct a smooth interface, but
for very distorted surfaces the error on the conservation of mass
increases.[29] Herrmann [8] determined that at least 6 grid points
are needed to resolve a droplet to get a grid independent result.
In other words, at least 63 cells are needed to properly represent
a droplet. Ghiji [26] used a mesh of 20 million cells, a time step
in the order of nanoseconds and a computational cluster with 384
cores. Even then simulations could take up to one month. These
kinds of computational loads are out of the scope of this thesis,
so
3.1 CFD strategies in spray formation 12
Eulerian methods will not be the main focus. With the presently
available computing power, the main use of Eulerian methods is the
validation of breakup models or for simulations including the
needle and sac volume. The Eulerian approach solves the problem of
being dependent on a breakup model, but it does not guarantee a
mesh independent solution [26].
3.1.3 Hybrid approach
Sometimes, a Eulerian approach is used in the near-nozzle region
and a Lagrangian approach outside of the liquid core region, as
Bravo et al. [27] did in their high- fidelity diesel spray
simulation and Moriyoshi [30] in their study of a swirl-type
injector. The coupling between the two is then done by an algorithm
that detects small underresolved continuous liquid phase droplets
and convert them to discrete phase droplets. The strategy that
Bravo et al.[27] used, is to replace the continuous liquid phase
droplets (contained within fewer than 53 contiguous cells) by
discrete phase droplets of equal volume. The goal of this approach
is to increase the precision in the near-nozzle region while
retaining the advantages and disadvantages of the Lagrangian
approach. The computational load lies between both approaches but
it is still too big for the purpose of this thesis. Bravo et al.
[27] reached a cell count of 60 million cells.
3.2 Common practice in the simulation of fuel sprays
The result of a simulation in a CFD software package greatly
depends on the used models and the temporal and spatial
discretisation of the problem. This section will look at which
models are commonly used in the literature to model fuel sprays
using the Eulerian-Lagrangian approach. Special interest will be
given to the models that are available in ANSYS Fluent, the CFD
software package used in this thesis. One of the advantages of this
software package is the availability of a comprehensive user manual
and theory guide. So even while it is a commercial software
package, the equations on which the different models are based are
elaborately explained.
3.2.1 Discretisation, Mesh and Time step
As already mentioned in the previous section, the design of the
mesh depends on the choice of a Lagrangian or an Eulerian approach
for the liquid phase. In a Lagrangian approach, the smallest grid
size will seldom go below the nozzle diameter, while for a Eulerian
approach the smallest grid size is generally a fraction of the
nozzle diameter. It is clear that for optimal use of the mesh and
for the most
3.2 Common practice in the simulation of fuel sprays 13
accurate results, the region near the nozzle and around the spray
axis should be the most refined, as it is in these regions that
biggest velocity gradients occur.
Another peculiarity of meshes for fuel spray simulations is that a
big part of the mesh sees no notable flow, especially in the first
moments after injection. Therefore, adaptive mesh refinement (AMR)
or adaptive local mesh refinement (ALMR) is sometimes used.[4, 10,
31] During simulation, cells for which a certain criterion is met,
are divided into multiple cells (4 in 2D, 8 in 3D). The criterion
compares a quantity in the cell to a predefined value. If the
quantity becomes smaller or bigger than the predefined value, the
cell is split. Possible quantities for fuel spray simulations are
the velocity gradient , temperature gradient [31] or the fuel mass
fraction [4]. The user can also specify a predetermined maximum
number of levels of refinement or a maximum number of cells after
which cells are no longer split.
To ensure an accurate solution of the flow near the nozzle, a
common choice is to permanently refine this region, regardless of
the criterion. An example of this can be seen in figure 3.2. It
shows the mesh that Argonne National Laboratory used for its
contribution to the first ECN Workshop [31] at the start of
injection. As the spray penetrates further into the domain, the
refined region will grow, following the spray. In such a way, a
pronounced improvement in calculation time can be noticed.
Especially in the moments after the start of injection where the
total number of cells is considerably lower as for the equivalent
fixed grid. Lucchini et al. [5] report in their study an
improvement in calculation time by a factor of 5 to 12.
Figure 3.2: Adaptive mesh of Argonne National Laboratory at start
of injection, used in the first ECN Workshop [31]
Time steps in the order of 0.5 µs are most common for grids with
the smallest cell size around 1 mm [31]. Decreasing grid sizes
bring forth smaller time steps. This can be explained by looking at
the Courant–Friedrichs–Lewy (CFL) condition. This
3.2 Common practice in the simulation of fuel sprays 14
condition needs to be satisfied in order for a partial differential
equation problem to converge. In 1D, it takes the following
form.
C = ut x ≤ Cmax (3.2)
In which C is the Courant number, u is the speed of the flow, x the
length of the cell and t the time step. It only poses a strict
limit when the time steps are solved explicitly. In such cases,
Cmax is around 1 and depends on the spatial discretisation scheme.
Implicit algorithms are less prone to numerical instabilities, so
higher values of Cmax can be used.[21] Still, when decreasing the
cell size, the time step is usually reduced as well, irrespective
of the algorithm. In 3D, C is equal to the sum of the 1D cases
along the different dimensions, but in the context of fuel sprays,
the velocities perpendicular to the spray axis can be neglected
with respect to the velocity along the spray axis. So for fuel
sprays, equation 3.2 can still be used to determine the Courant
number. Some articles in the literature provide enough information
to calculate this number and in table 3.1 an overview is given of
some Courant numbers used by different authors for fuel spray
simulations with Lagrangian particle tracking methods.
Table 3.1: Courant numbers used by different authors for fuel spray
simulations
Author x [mm] t [ns] u [m/s] C Mesh P-V coupling
Som [4] 0.25 500 ±575 ±1.15 3D PISO-SIMPLE Pei [9] 1 4000 595.3
2.381 2D SIMPLE
Decan [7] 1 400 460 0.184 3D Coupled Lucchini [5] 0.25 500 N/A
≤0.15 3D PISO
Som et al.[4] and Lucchini et al.[5] used an adaptive mesh
refinement (AMR) technique, so the smallest cells do not
necessarily coincide with the cells with the fastest flow. The x
given in the table is the smallest possible grid size. u is an
estimate of the injection velocity of the spray. The velocity of
the entrained gas will be a bit lower. The Courant number is thus
the largest possible one. Lucchini [5] explicitly stated that the
AMR ensured a maximum Courant number of 0.15. Common Courant
numbers are in the order of 10−1 to 1. When using higher Courant
numbers there is a tendency to use the generally more stable
pressure-velocity coupling schemes, such as PISO and SIMPLE or even
the combination of both for superior stability, but at the cost of
added calculation time. When a good temporal discretisation is
needed, one could opt for lower Courant numbers in combination with
a coupled solver which is generally faster to converge but somewhat
less stable. Still, most authors use PISO [4, 26, 5], SIMPLE [3, 9]
or a combination of both.
3.2 Common practice in the simulation of fuel sprays 15
3.2.2 Turbulence model
A lot of fluid mechanics problems of engineering interest have to
handle turbulent flows which inherently have an unstable character.
This is especially true for the flows inside the cylinder of an
ICE. Due to the high velocities, the viscosity of the fluid is not
able to dampen fluctuations and large turbulent eddies are formed.
These turbulent eddies are unstable and split up in smaller eddies
which in turn split up in even smaller eddies, carrying over their
kinetic energy to smaller and smaller length scales until they get
dissipated as heat by the viscous forces. To be able to represent
the smallest eddies in a mesh, the mesh needs to be made very fine
which is computationally very expensive. This strategy is called
Direct Numerical Simulation (DNS). Another strategy is to
explicitly resolve the large eddies, but model the turbulent eddies
that are smaller than the cell sizes (also called the subgrid
scales). This strategy called is Large Eddy Simulation (LES). The
computational load is still relatively high, but LES calculations
come within reach for a lot of applications and is becoming more
and more popular. For complex geometries, the computational load of
LES can be too high and that is why Reynolds-averaged Navier-Stokes
(RANS) models are still used for modelling turbulence. RANS models
only solve for the mean flow field and the effect of turbulence on
this flow field is entirely modelled. This gives a significant
reduction in computational load.
Direct Numerical Simulation
One option could be doing a Direct Numerical Simulation (DNS) of a
fuel spray, but the use of DNS in spray breakup has been hindered
by the available computing power. The high Reynolds number flow
implies a mesh so fine it would be impossible to calculate a
solution within a reasonable time frame, even with modern-day
supercomputers. Just for resolving all the turbulent length scales
of a low Reynolds number liquid jet, meshes of hundreds of millions
of cells are needed. So, simulating the breakup of a high-pressure
diesel spray is not feasible using DNS, although the result would
be largely independent of the various models currently used in
spray breakup.
Reynolds-averaged Navier-Stokes
The generally high computational load of DNS has led to the
widespread use of turbulence models. Modelling the turbulence
allows for much coarser meshes. The RANS equations were developed
for this purpose. The turbulent flow field is split up into a mean
part and a part due to the turbulent fluctuations. This is
more
3.2 Common practice in the simulation of fuel sprays 16
elaborately explained in section 2.2. There are different models to
determine the strength of the fluctuating part and its influence on
the mean flow field. One of the most popular models is the k-ε
model as explained in section 2.2.2. Th is due to its relative
simplicity and fairly accurate results in a lot of situations. RANS
equations in general also allow for the use of symmetry, reducing
the computational load even further. As only the mean flow field is
solved, there is no need for a full 3D simulation to resolve the
effect of the turbulence, which inherently creates a 3D flow. So
when using RANS for fuel spray simulations, a quarter cube 3D
domain or even a 2D axisymmetric domain could be used.
It is generally agreed upon that k-ε is also the most widely used
turbulence model for spray simulations. It is used by a lot of
authors in all its different varieties (standard [4, 9, 6],
realisable [3, 7] or RNG [4]). There is no clear preference to use
one over the other and usually, more than one turbulence model is
tried to see which one gives the best results. For the simulation
of turbulent gas jets, it is common practice to increase constant
C1 of the standard k-ε model. [15] Sutradhar also used this
strategy in his master’s thesis [6].
Large Eddy Simulation
If a more detailed solution is desirable, a Large Eddy Simulation
(LES) is a possible solution. When using LES, the large turbulent
flow structures are directly resolved and only the eddies with
turbulent length scales smaller than the filter size (proportional
to the grid size) are modelled. In this way, the modelling error
decreases with the grid size. The simulation now also provides
information about the turbulent fluctuations and gives a more
accurate distribution of the evaporated fuel. However, it comes at
an added computational cost. In general, LES requires a finer grid.
Additionally, to obtain the mean flow field several solutions need
to be averaged, whereas a RANS method directly provides the mean
flow field. LES is gaining momentum in the modelling community at
the expense of RANS modelling. High-fidelity simulations often use
LES in combination with an Eulerian method for the liquid phase, as
Bravo et al. [27] and Ghiji et al. [26] have done in their works.
This is not to say that LES is not useful within the context of
this thesis. Section 4.3 will investigate if LES can present an
added value for the simulation of the spray in this thesis.
3.2.3 Injection and primary breakup model
One of the challenges of using a discrete phase model is how to
link the nozzle flow to an appropriate diameter distribution at the
injection. The Lagrangian way
3.2 Common practice in the simulation of fuel sprays 17
of tracking particles does not allow to represent a fully liquid
flow, such as in the nozzle. Furthermore, the levels of turbulence
and cavitation in the nozzle greatly influence the speed and
diameters of the droplets. Increased levels of turbulence in the
nozzle flow destabilise the jet and increase its breakup rate. The
same goes for the cavitation. The high local speeds in the nozzle
can cause pressures below the vapour pressure of the fuel. When the
cavitation bubbles implode again at the nozzle exit, the jet
breakup is accelerated. Another effect of the cavitation is the
reduction of the effective cross-sectional flow area in the nozzle
and thus an increase in speed of the droplets leaving the nozzle.
The most accurate way to model the injection would be to use a
Eulerian (section 3.1.2) or a hybrid (section 3.1.3) model to
simulate the nozzle flow and the liquid core of the jet. However,
the computational cost of this method has hindered its use.
Usually, the domain of the simulation only extends to the nozzle
orifice. The nozzle and sac volume are then excluded from the
simulation. It’s hard to account for all of the aforementioned
influences in the boundary condition at the nozzle orifice.
Therefore, different approximations can be made which are explained
in the following sections.
Droplet diameter distribution functions
One of the most basic ways of modelling the injection is imposing a
distribution function of droplet sizes at the nozzle exit. This
model assumes that the primary breakup has already occurred at the
nozzle exit. Considering the quick primary atomisation of
high-pressure fuel sprays, this is not a very hefty assumption.
ANSYS Fluent supports the use of a uniform distribution and a
Rosin-Rammler distribution. For both the user has to specify a
standard diameter. Also for the Rosin-Rammler distribution the
spreading parameter has to be specified. According to the ANSYS
Fluent Theory guide [11], the Rosin-Rammler distribution function
is defined as follows.
1− Y = exp
)n] (3.3)
Y is the mass fraction of droplets for which the diameter is
smaller than D, d is the Rosin-Rammler diameter and n is
Rosin-Rammler exponent which determines the spread. The main
difficulty is determining these parameters, as the current
experimental measurement techniques do not allow to measure droplet
sizes in the dense spray near the nozzle. The parameters need to be
found iteratively by comparing the results of a simulation to
experimental measurements in the region
3.2 Common practice in the simulation of fuel sprays 18
away from the nozzle. Alternatively, there also exist correlations
to estimate d and n. The ANSYS Fluent Theory guide proposes the
following value for d.
d = 1.2726 (
133.0d8We−0.74 )(
(3.5)
and d the nozzle diameter, u the injection velocity, ρ the liquid
density and σ the surface tension of n-dodecane. There still are
other ways to find an appropriate distribution of droplets. For
example, Pei et al. [9] used a uniform distribution and tuned its
initial droplet diameter to obtain approximately correct liquid
lengths.
Blob model
Another more frequently used model is the Blob Method. It is based
on the assumption that spherical droplets with a diameter equal to
the nozzle diameter are introduced into the domain. The breakup of
these big droplets is from thereon handled by the secondary breakup
models, dispersing the spray in much finer droplets as can be seen
in figure 3.3. Its simplicity while retaining fairly accurate
results is its main advantage. It is therefore used by many authors
[4, 10, 5, 6] in their fuel spray simulation studies.
Figure 3.3: Breakup of blobs by secondary breakup models using the
concept of liquid core length, figure from ANSYS Fluent Theory
Guide [11]
The speed of the ejected droplets can be determined by the
conservation of mass if the mass flow rate m(t) through the nozzle
is known. If there is no cavitation in the nozzle, the mean
velocity u(t) of the ejected droplets is given by equation
3.6.
u(t) = m(t) ρlA
3.2 Common practice in the simulation of fuel sprays 19
A is the cross-sectional area of the nozzle (assumed to be
constant) and ρl the liquid density. If the mass flow rate is not
known, Bernoulli’s equation can be used to determine the maximum
theoretical speed from the pressure difference p over the nozzle.
The real speed will be lower, as friction losses are not taken into
account here.
utheoretical = √
(3.7)
The real speed is often compared to the theoretical one and their
ratio is called the discharge coefficient Cd. The value can be
determined by calculating the friction losses. Sometimes, it is
also given for a particular nozzle.
Cd = m
mtheoretical = ρlAu
ρlAutheoretical (3.8)
If there is cavitation, the effective cross-sectional area of the
flow will decrease and the speed of the droplets leaving the nozzle
will increase. Von Kuensberg Sarre et al. [32] have developed a
method to quantify this effect. It is based on the existence of a
vena contracta in the nozzle surrounded by evaporated fuel. They
calculate the cross-sectional area of the vena contracta Avena with
Nurick’s expression.
Avena = CcA, Cc = [( 1
D
]−0.5
(3.9)
Cc0 is a constant equal to 0.61, r is the radius of the edge at the
inlet of the nozzle and D is the diameter of the nozzle. Assuming a
steady flow, the conservation of mass gives the speed at the vena
contracta (uvena).
uvena = m
ρlCcA = u
Cc (3.10)
If the conservation of momentum between the vena contracta and the
nozzle exit is expressed, equation 3.11 is obtained. It assumes
that the cavitation keeps the pressure at the vapour pressure at
the vena contracta.
ueff = A
m (pvap − pexit) + uvena (3.11)
ueff is the effective speed of the droplets at the nozzle exit due
to cavitation, pvap the vapour pressure for the fuel at a given
temperature and pexit the static pressure at the nozzle exit. Using
the mass flow rate, a decreased effective cross-sectional area at
the nozzle exit can be calculated with ueff . The increased breakup
rate due to implosions of cavitation bubbles is not accounted for
in this model.
3.2 Common practice in the simulation of fuel sprays 20
Other models
There still exist plenty of other models, some of which are
commonly used but not available in ANSYS Fluent. For example, there
is the Huh-Gosman model as used by various researchers [4, 5, 10]
of which the first two used a modified version of the Huh-Gosman
model called the Bianchi model. They both take into account the
aerodynamically induced breakup by the Kelvin-Helmholtz instability
(section 3.2.4) and the turbulence induced breakup by introducing
surface perturbations linked to the turbulent length scales of the
flow in the nozzle. Other models also include the effect of
cavitation, such as the KH-ACT model, developed