Chalmers University of Technology

CFD with OpenSource software, assignment 3

Description and implementation of

particle injection in OpenFOAM
Developed for OpenFOAM-1.7x

Peer reviewed by:

Author:
Anton Persson

Jelena Andric
Erwin Adi Hartono

November 5, 2010

Contents
1 Introduction and background
2 Particle injection in dieselSpray
2.1 Particles and injectors . . . . .
2.2 The case directory . . . . . . .
2.3 Data processing . . . . . . . . .
2.3.1 unitInjector . . . . . . .
2.3.2 hollowCone . . . . . . .
2.3.3 sprayInject . . . . . . .
2.4 Injection example . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

3
3
4
6
6
11
14
15

3 A simple particle injector

3.1 injectorSolidParticleFoam . . . . . . . . .
3.1.1 Directory and compilation settings
3.1.2 injectorSolidParticleFoam.C . . . .
3.1.3 injectorSolidParticleCloud.C . . .
3.1.4 createFields.H . . . . . . . . . . .
3.2 Example case . . . . . . . . . . . . . . . .
3.3 Two injectors . . . . . . . . . . . . . . . .
3.4 Evenly distributed injection . . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

19
19
19
20
21
21
22
24
25

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

Chapter 1

Introduction and background

This document describes the implementation of particle injection in the dieselSpray class, which is
used in icoLagrangianFoam and dieselFoam. Also an implementation of a particle injector in the
solidParticle class will be done.
In dieselSpray there are several different tools for introducing particles in the fluid flow that is
being solved. These tools are properly called injectors. The injectors introduce particles in the
domain, according to settings specified by the user. In the solidParticle class there is no injector
tool. To be able to implement injection of particles in solidParticle, a deeper understanding of injectors is needed. This will be aquired by the description of the injection process that is made in
this document.

Chapter 2

Particle injection in dieselSpray

2.1

Particles and injectors

There are some frequently used and useful expressions when treating particles modeled in a fluid
flow. A term used in connection to computational descriptions is parcel. A parcel is really what
is used in numerical simulations and is a representation of a gathering of real particles. This consruction is plainly made because that it is almost always too computational demanding to simulate
all the real particles. To capture the behaviour of the real particles, some real case properties are
defined, typically the massflow. In connection with dieselSpray the spray corresponds to the entire
cloud containing the fuel injected. The solidParticle class contains no parcel treatment, only real
particles.
In the dieselSpray class there are two objects that mainly determine how the injection of parcels will
appear. These objects are the injector type and the injector model. The type determines the main
characteristics of the injection based on specified physical properties as velocity, pressure, mass flow,
injection region, number of parcels, mass of a parcel etc. Different injector types treat and process
given information in different ways. For example, in the type commonRailInjector the velocity profile
U , is calculated from
s
2(Pinj Pamb )
U=
,
(2.1)

while in unitInjector the corresponding profile is based on

U=

CD A

(2.2)

In Eq. 2.1, Pinj is the pressure at the current injection position, Pamb the ambient pressure and
(also in Eq.2.2) the density of the injected parcel. In Eq.2.2, m
is the mass flow of injected parcels,
CD the particle drag coefficient and A the area of the injection region.
The models in turn describes the shape of the injection region, i.e. spray appearance. An example is how in the Chomiak injector, the angle , at which a parcel is injected is dependent on the
parcel size, see Eq. 2.3. In Eq. 2.3, max is the maximum allowed injection angle, d the diameter of
the current parcel and dmin and dmax the maximum and minimum diameter of all injected parcels.
=

d dmax
max
dmin dmax

(2.3)

The result in this example is that the large parcels go straight and the smaller are injected at a
larger spray angle.

2.2. THE CASE DIRECTORY

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

To aquire the understanding of how the injectors work, it is here choosen to focus on the injector type unitInjector and the model hollowConeInjector.

2.2

The case directory

For understanding of the setup of a particle injector, let us take a closer look at the aachenBomb
tutorial located in $FOAM_TUTORIALS/dieselFoam/aachenBomb. The directory structure is as follows
|-|
|
|
|
|
|
|
|
|
|
|
|-|
|
|
|
|
|-|
|
|
|
|
|
|
|
|
|
--

0
|-- epsilon
|-- ft
|-- fu
|-- k
|-- N2
|-- O2
|-- p
|-- spray
|-- T
|-- U
-- Ydefault
chemkin
|-- chem.inp
|-- chem.inp.1
|-- chem.inp_15
|-- chem.inp.full
-- therm.dat
constant
|-- chemistryProperties
|-- combustionProperties
|-- environmentalProperties
|-- injectorProperties
|-- polyMesh
|
|-- blockMeshDict
|
-- boundary
|-- RASProperties
|-- sprayProperties
-- thermophysicalProperties
system
|-- controlDict
|-- fvSchemes
-- fvSolution

The files strongly connected to the injection of particles are /constant/injectorProperties and
/constant/sprayProperties. A closer look in constant/injectorProperties shows the properties which can be defined for both the unitInjector and commonRail types. The type defined to be
used in this case is though unitInjector, see below.
{
injectorType

unitInjector;

unitInjectorProps
{
4

2.2. THE CASE DIRECTORY

position
direction
diameter
Cd
mass
nParcels

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

(0 0.0995 0);
(0 -1 0);
0.00019;
0.9;
6e-06;
5000;

X
(
1.0
);
massFlowRateProfile
(
(0 0.1272)
(4.16667e-05 6.1634)
(8.33333e-05 9.4778)
(0.000125 9.5806)
(0.000166667 9.4184)
.
.
.
(0.00120833 5.1737)
(0.00125 3.9213)
);
temperatureProfile
(
(0.0
320.0)
(0.00125 320.0)
);

}
The unitInjectorProps section defines properties of the particles and is quite straight on. The
parameter position determines the center of the (circular) region where parcels are injected and d
the diameter of this disc (from now on called injection disc). The amount of the particle specie in
kg to be injected is mass, Cd the drag coefficient and nParcels the total number of parcels to be
injected. X defines the mass fraction of the particle specie. This corresponds to the first specie that
is defined in chemkin/chem.inp. Next, massFlowRateProfile sets the mass flow at a certain time,
(ti m).
For temperatureProfile the same setup is used, i.e. (ti T ). The temperature and mass flow
properties together indicate that the last parcel will be injected at t = 0.00125.[1][4] Consequently
the mass flow is varying during the injection while the temperature is constant. It is important
that pre-calculations of the injection parameters have been done properly, at least to get exactly
the desired output of the injection. For example the number of parcels, mass to be injected and the
mass flow rate profile should harmonise.
In constant/sprayProperties which injector model that is used is specified at
injectorModel

hollowConeInjector;

The injection model specifies the size, velocity and direction of the parcels injected at the injection
disc. Further down in the dictionary, coefficients for the hollow cone injector can be set:
hollowConeInjectorCoeffs
5

2.3. DATA PROCESSING

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

{
dropletPDF
{
pdfType
RosinRammler;
RosinRammlerPDF
{
minValue
1e-06;
maxValue
0.00015;
d
0.00015;
n
3;
}
exponentialPDF
{
minValue
maxValue
lambda
}

0.0001;
0.001;
10000;

}
innerConeAngle
outerConeAngle

( 0 );
( 20 );

}
PDF is here short for Probability Density Function. As seen, Rosin Rammler is the one used here.
Actually Rosin Rammler gives a distribution governed by a CDF, Cumulative Distribution Function.
The CDF of Rosin Rammler is defined in Eq. 2.4 where F is the mass fraction of particles with
diameter smaller than x, d the mean particle diameter and n a factor for setting the spread of particle
size. The value of x varies in the range set by minValue and maxValue in the dictionary. Rosin
Rammler distribution is commonly used for generating particle size distribution of grinding, milling
and crushing operations.[2] Here it is used to create a size distribution of the simulated parcels.
F (x, n, d) = 1 ex/d

(2.4)

When using the hollowcone injector, a spray shaped by a hollow cone will be created at the injection
region. The inner and outer cone angle defines how this hollow cone should be created. If the inner
cone angle, as in this case is set to zero, the cone will be solid. How this can look will be shown
later.

2.3

Data processing

In this section a description of how the user provided data is processed by the unitInjector type and
the hollowCone model. A short look into sprayInject will also be made.

2.3.1

unitInjector

The directory $FOAM_SRC/lagrangian/dieselSpray/injector/unitInjector includes the following files:

unitInjector.C unitInjector.dep unitInjector.H
The .dep file is created during the compilation of the .C file and includes all dependencies for the
injector. Meaning which files are needed, how they are to be registered and where on the users
machine they should be installed so that the unitInjector will work. The structure of a .C and .H
file is used through out all of OpenFOAM. Running the.H-header file requires some other files to be
included, which is done in the beginning of the file by:
6

2.3. DATA PROCESSING

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

#include "injectorType.H"
#include "vector.H"
Here vector.H is simply needed for being able to manage vectors in the code and injectorType.H
includes some definitions needed for all injector types. Including of .H files are almost always needed
and in them, as well as in this one there is a ifndef command that checks if the file has already been
included by some other file. This function can save a lot of time when running large compilations.
The main task done in the header file is that unitInjector is defined with its both private and public
attributes. There are both, private and public attributes of data and member functions. Among the
public attributes there also are constructors and a destructor. Worth noticing in the .H file is that
a new type is defined:
typedef VectorSpace<Vector<scalar>, scalar, 2> pair;
This is done for the solver to be able to handle the type that for example massFlowRateProfile
(see Sect. 2.2) is. With the declarations made in the .H file, there is no need of declaring them in
the .C file again, since the header file is included in the .C file.
In unitInjector.C functions and constructors carries out calculations needed for the injection process. Included in this file is unitInjector.H, addToRunTimeSelectionTable.H, Random.H and
mathematicalConstants.H. That unitInjector.H and some mathematical constants are needed
is obvious. The addToRunTimeSelectionTable.H is used so that the unitInjector can be reached
as a injectorType in injectorProperties dictionary while Random.H is for being able for creating
random numbers. The constructors section begins with reading the properties defined in the file
injectorProperties (see below).
injectorType(t, dict),
propsDict_(dict.subDict(typeName + "Props")),
position_(propsDict_.lookup("position")),
direction_(propsDict_.lookup("direction")),
d_(readScalar(propsDict_.lookup("diameter"))),
Cd_(readScalar(propsDict_.lookup("Cd"))),
mass_(readScalar(propsDict_.lookup("mass"))),
nParcels_(readLabel(propsDict_.lookup("nParcels"))),
X_(propsDict_.lookup("X")),
massFlowRateProfile_(propsDict_.lookup("massFlowRateProfile")),
velocityProfile_(massFlowRateProfile_),
injectionPressureProfile_(massFlowRateProfile_),
CdProfile_(massFlowRateProfile_),
TProfile_(propsDict_.lookup("temperatureProfile")),
averageParcelMass_(mass_/nParcels_),
pressureIndependentVelocity_(true)
As seen the properties velocityProfile and injectionPressureProfile is read in as
massFlowRateProfile. This is due to that these properties are not defined in the injector properties
dictionary, but still need to have values for the injection to function. They will also be corrected to
values corresponding to their quantities later in the code. The average parcel mass is also calculated
in this section and the property pressureIndependentVelocity is set as true. That this is set to
true is directly coupled to the earlier mentioned fact that the pressure profile is not specified. Note
also that a variable CdProfile is set equal to the mass flow rate profile. The reason for this will
also be understood later.
After this unitInjector.C continues with controling that the user has set the start and end times
of massFlowRateProfile and temperatureProfile to the same. Also the code makes sure that the
time column in massFlowRateProfile is specified as time in s. These control and correct operations
are shown in code below.
7

2.3. DATA PROCESSING

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

// check if time entries for soi and eoi match

if (mag(massFlowRateProfile_[0][0]-TProfile_[0][0]) > SMALL)
{
FatalErrorIn
(
"unitInjector::unitInjector(const time& t, const dictionary dict)"
)<< "start-times do not match for TemperatureProfile and "
<< " massFlowRateProfile." << nl << exit (FatalError);
}
if
(
mag(massFlowRateProfile_[massFlowRateProfile_.size()-1][0]
- TProfile_[TProfile_.size()-1][0])
> SMALL
)
{
FatalErrorIn
(
"unitInjector::unitInjector(const time& t, const dictionary dict)"
)<< "end-times do not match for TemperatureProfile and "
<< "massFlowRateProfile." << nl << exit(FatalError);
}
// convert CA to real time
forAll(massFlowRateProfile_, i)
{
massFlowRateProfile_[i][0] =
t.userTimeToTime(massFlowRateProfile_[i][0]);
velocityProfile_[i][0] = massFlowRateProfile_[i][0];
injectionPressureProfile_[i][0] = massFlowRateProfile_[i][0];
}
forAll(TProfile_, i)
{
TProfile_[i][0] = t.userTimeToTime(TProfile_[i][0]);
}
When this is done the mass flow rate profile is corrected, so that it is ensured that it matches the
total mass to be injected, see Eq. 2.5. Also, a profile for the drag coefficient is created, containing
the time entries of the mass flow rate profile and the value of Cd . Below Eq. 2.5, the code for this
procedure is shown.
mtot
(2.5)
m
prof,new = m
prof R tend
m
prof
t0
scalar integratedMFR = integrateTable(massFlowRateProfile_);
forAll(massFlowRateProfile_, i)
{
// correct the massFlowRateProfile to match the injected mass
massFlowRateProfile_[i][1] *= mass_/integratedMFR;
CdProfile_[i][0] = massFlowRateProfile_[i][0];
CdProfile_[i][1] = Cd_;
}
8

2.3. DATA PROCESSING

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

After this, the injection direction vector is normalized by that it is divided it with its own magnitude
(see below).
// Normalize the direction vector
direction_ /= mag(direction_);
To keep track of every parcel a coordinate system for each one needs to be defined, three directions
perpendicular to each other are needed. Since the direction vector for them already is created, two
vectors perpendicular to this is sufficient. This is created based on a randomly generated vector
consisting of numbers between 0 and 1. The random vector, Arnd gives through manipulation, a
vector, B tangent perpendicular to the direction vector, dir. B tangent is normalized and a third vector,
perpendicular to both this and the direction vector, is simply created through a cross product. See
steps in Eq. 2.6 below.
Arnd
B tangent
magV

=
=
=

randomly generated vector in [0, 1]

Arnd (Arnd dir) dir
magnitude(Atangent )

tanInj1
tanInj2

=
=

Atangent
magV

(2.6)

dir tanInj1

In the code the section corresponding to Eq. 2.6 is

void Foam::unitInjector::setTangentialVectors()
{
Random rndGen(label(0));
scalar magV = 0.0;
vector tangent;
while (magV < SMALL)
{
vector testThis = rndGen.vector01();
tangent = testThis - (testThis & direction_)*direction_;
magV = mag(tangent);
}
tangentialInjectionVector1_ = tangent/magV;
tangentialInjectionVector2_ = direction_ ^ tangentialInjectionVector1_;
}
The while loop is as seen to make sure that magV is SMALL, which it should be as its set to 0.0 in
the beginning of the calculation.
For making sure that the mass fraction X sums up to one, this is checked. If it does not, this
is fixed with the expression in Eq. 2.7 where a loop over i is done.
X(i) =

X(i)
sum(X)

(2.7)

This equation is constructed so that it will work for injections with several species. The code of this
is shown below.
// check molar fractions
scalar Xsum = 0.0;
forAll(X_, i)
{
9

2.3. DATA PROCESSING

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

Xsum += X_[i];
}
if (mag(Xsum - 1.0) > SMALL)
{
WarningIn("unitInjector::unitInjector(const time& t, Istream& is)")
<< "X does not sum to 1.0, correcting molar fractions."
<< nl << endl;
forAll(X_, i)
{
X_[i] /= Xsum;
}
}
Also needed for the injection is how many parcels that should be injected every time step, this
is done next. The number of parcels are calculated from dividing the mass to be injected by the
average mass of a parcel. It is also rounded off to the closest integer, see Eq. 2.8.
nparcels = integer of (

minj
+ 0.49)
mave

(2.8)

The translation into code of Eq. 2.8 is

Foam::label Foam::unitInjector::nParcelsToInject
(
const scalar time0,
const scalar time1
) const
{
scalar mInj = mass_*(fractionOfInjection(time1)-fractionOfInjection(time0));
label nParcels = label(mInj/averageParcelMass_ + 0.49);
return nParcels;
}
The position of the injection of each parcel, based on the injection disc diameter and the discs
center position is then defined for the solver. These positions are distributed within the injection
disc according to Eq. 2.9 where Srnd is a scalar with the same behaviour as the components of Arnd
in Eq. 2.6.
rinj
angleinjP os
posparc

=
=
=

d Srnd
2 Srnd
pos + rinj (tanInj1 cos(injAngle ) + tanInj2 sin(injAngle ))

(2.9)

The paranthesis in the last calculation of Eq. 2.9 forms a vector in the plane perpendicular to the
direction vector, since both tanInj1 and tanInj2 are perpendicular to the injection direction. When
adding the position vector to this vector, the position for the injected parcel starting from origo is
obtained. If the case solved for is in two dimensions a conversion for this is also implemented in the
code. The control of whether it is a two dimensional case or not is made in the file spray.C which
is a central part of the dieselSpray class (located in $FOAM_SRC/lagrangian/dieselSpray/spray).
The calculation for the three dimensional case, i.e. Eq. 2.9 can be seen in code format below.
Foam::vector Foam::unitInjector::position
(
const label n,
const scalar time,
const bool twoD,
const scalar angleOfWedge,
10

2.3. DATA PROCESSING

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

const vector& axisOfSymmetry,

const vector& axisOfWedge,
const vector& axisOfWedgeNormal,
Random& rndGen
) const
// otherwise, disc injection
scalar iRadius = d_*rndGen.scalar01();
scalar iAngle = 2.0*mathematicalConstant::pi*rndGen.scalar01();
return
(
position_
+ iRadius
* (
tangentialInjectionVector1_*cos(iAngle)
+ tangentialInjectionVector2_*sin(iAngle)
)
);
As earlier mentioned the velocity and pressure profiles are corrected for the injection. This is done
at the end of this file, with the expressions seen in Eq. 2.10.
A
U
p

=
=
=

0.25 d2
m

Cd A

pref + 0.5v

(2.10)
2

Eq. 2.10 is looped over the time positions specified in massFlowRateProfile, so that the profiles
are specified at all times. Note that the calculation of the velocity profile can be recognised from
the example stated in Eq. 2.2. In OpenFoam code this looks as
void Foam::unitInjector::correctProfiles
(
const liquidMixture& fuel,
const scalar referencePressure
)
{
scalar A = 0.25*mathematicalConstant::pi*pow(d_, 2.0);
scalar pDummy = 1.0e+5;
forAll(velocityProfile_, i)
{
scalar time = velocityProfile_[i][0];
scalar rho = fuel.rho(pDummy, T(time), X_);
scalar v = massFlowRateProfile_[i][1]/(Cd_*rho*A);
velocityProfile_[i][1] = v;
injectionPressureProfile_[i][1] = referencePressure + 0.5*rho*v*v;
}
}

2.3.2

hollowCone

In Section 2.2 it was mentioned that the spray properties dictionary defined which injector model
that is to be used. Here hollowConeInjector is used and it is built up in the same way as the injector
type, with hollowCone.H and hollowCone.C. It is located in
$FOAM_SRC/lagrangian/dieselSpray/spraySubModels/injectorModel/hollowCone.

11

2.3. DATA PROCESSING

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

In hollowCone.H it is first checked if the file already has been included and then files needed for
the file itself are included. Needed files are: injectorModel.H, scalarList.H and pdf.H. Which
are giving definitions for injector models in general, providing a type definition of a scalar list and
allowing a number of PDFs to be used. With these files included, the class declaration with public
and private attributes can be done.
The .C file includes except the .H file, addToRunTimeSelectionTable and mathematicalConstants
which are needed for the same purposes as in unitInjector.C. The first thing done under constructors is that the data for pdfType (including its settings), innerAngle and outerAngle are obtained
from the spray properties dictionary. It is also checked that the settings are done in a correct way.
Then the reference pressure is set to the ambient pressure and the velocity profiles are corrected
using the construction of Eq. 2.10 as in unitInjector.C. All this is seen below.
// Construct from components
hollowConeInjector::hollowConeInjector
(
const dictionary& dict,
spray& sm
)
:
injectorModel(dict, sm),
hollowConeDict_(dict.subDict(typeName + "Coeffs")),
dropletPDF_
(
pdfs::pdf::New
(
hollowConeDict_.subDict("dropletPDF"),
sm.rndGen()
)
),
innerAngle_(hollowConeDict_.lookup("innerConeAngle")),
outerAngle_(hollowConeDict_.lookup("outerConeAngle"))
{
if (sm.injectors().size() != innerAngle_.size())
{
FatalError << "hollowConeInjector::hollowConeInjector"
<< "(const dictionary& dict, spray& sm)\n"
<< "Wrong number of entries in innerAngle"
<< abort(FatalError);
}
if (sm.injectors().size() != outerAngle_.size())
{
FatalError << "hollowConeInjector::hollowConeInjector"
<< "(const dictionary& dict, spray& sm)\n"
<< "Wrong number of entries in outerAngle"
<< abort(FatalError);
}
scalar referencePressure = sm.ambientPressure();
// correct velocityProfile
forAll(sm.injectors(), i)
{
12

2.3. DATA PROCESSING

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

sm.injectors()[i].properties()->correctProfiles(sm.fuels(), referencePressure);
}
}
After this, the injection direction is calculated based on the inner and outer angle. In the unitInjector section (2.3.1) it was described how the injection positions of the parcels where set using the
calculation in 2.9. In the injector model, in this case hollowConeInjector, the initial direction of the
injected particles are determined from a similar calculation, see Eq. 2.11.
angle

dircorr

dirP arc
dirP arcnorm

=
=
=
=
=
=
=

innerAngle + Srnd (outerAngle innerAngle)

sin(angle/360)
cos(angle/360)
2 Srnd
(tanInj1 cos() + tanInj2 sin()
dircorr dir + n

(2.11)

dirP arc
magnitude(dirP arc)

In Eq. 2.11 angle is a random generated angle with a value between the inner and outer angle. A
direction vector (dirP arc) and a normal vector (
n) to this is created based on the original direction,
the tangentialvectors (dir, tanInj1 and tanInj2 ) and angle. The tangentialvectors are obtained using
Eq. 2.6. The result is a vector starting from origo and defining the direction of the injected parcel.
As in unitInjector.C, there is a option that calculates the direction if the case is two-dimensional.
In code, the calculation made in Eq. 2.11 looks as
vector hollowConeInjector::direction
(
const label n,
const label hole,
const scalar time,
const scalar d
) const
{
scalar angle = innerAngle_[n] + rndGen_.scalar01()*(outerAngle_[n]-innerAngle_[n]);
scalar alpha = sin(angle*mathematicalConstant::pi/360.0);
scalar dcorr = cos(angle*mathematicalConstant::pi/360.0);
scalar beta = 2.0*mathematicalConstant::pi*rndGen_.scalar01();
// randomly distributed vector normal to the injection vector
vector normal = vector::zero;
normal = alpha*
(
injectors_[n].properties()->tan1(hole)*cos(beta) +
injectors_[n].properties()->tan2(hole)*sin(beta)
);
// set the direction of injection by adding the normal vector
vector dir = dcorr*injectors_[n].properties()->direction(hole, time) + normal;
dir /= mag(dir);
return dir;
}

13

2.3. DATA PROCESSING

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

The velocity profile calculated by the injector type is then reconstructed. If the injector type has the
pressureIndependentVelocity property set to true the velocity will simply be fetched (see code
below), which is the case for the unitInjector type.
scalar hollowConeInjector::velocity
(
const label i,
const scalar time
) const
{
if (it.pressureIndependentVelocity())
{
return it.getTableValue(it.velocityProfile(), time);
}
The getTableValue function, as the name suggest, gets the value of the current position in velocity
profile and it is set to call the injector properties created by the injector type. If the hollowCone injector is used together with a injector where this property is false, a calculation based on
the pressure profile will be carried out to construct a velocity profile, see below (continuation of
hollowConeInjector::velocity).
else
{
scalar
scalar
scalar
scalar
return

Pref = sm_.ambientPressure();
Pinj = it.getTableValue(it.injectionPressureProfile(), time);
rho = sm_.fuels().rho(Pinj, it.T(time), it.X());
dp = max(0.0, Pinj - Pref);
sqrt(2.0*dp/rho);

}
}
At the end of hollowCone.C the average velocity is calculated by taking the mean of the velocity
over the injection time. See Eq. 2.12 and following code.
Umean =

sum(U )
no. of timesteps of injection

(2.12)

scalar hollowConeInjector::averageVelocity
(
const label i
) const
{
const injectorType& it = sm_.injectors()[i].properties();
scalar dt = it.teoi() - it.tsoi();
return it.integrateTable(it.velocityProfile())/dt;
}
The velocity profiles are here summed up by integrateTable.

2.3.3

sprayInject

When the injector type and injector model have defined all necessary parameters for the injection
they just need to be initialized in the code. Of course, this is coupled to many computations during
a simulation. But to get a better understanding of how all calculations made in the two previous sections are used, a segment of the file sprayInject.C (located in $FOAM_SRC/lagrangian/dieselSpray/spray
is shown below.

14

2.4. INJECTION EXAMPLE

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

forAll(injectors_, i)
{
if (mass > 0)
for(label j=0; j<Np; j++)
{
for(label n=0; n<nHoles; n++)
{
if (injectorCell >= 0)
{
} // if (injectorCell....
} // for(label n=0...
} // for(label j=0....
} // if (mass>0)...
} // forAll(injectors)...
This piece of code shows how the injection is done. The if statements above are for making sure that
injection really should take place and the for loops are there so that the injection of parcels is made
as many times as desired. The most outer loop repeats as many time as there are injectors (one in
the case examined here), the Np loop is for the number of parcels each time step (recognized from
earlier) and nHoles is for the number of holes used in the current injector type (one for unitInjector).
With the now obtained knowledge, a picture of the injection process can be made. A try to visualize the injection created by unitInjector and hollowCone (with innerAngle = 0, so solid) is done
in figure 2.1

Figure 2.1: Fundamental picture of injection using unitInjector and hollowCone. Size distribution governed
by Rosin Rammler, innerAngle = 0.

2.4

Injection example

To be able to further visualize how the injection process is carried out, snapshots of the aachenBomb tutorial (no ignition) simulation will be shown here. In the following figures, the standard
run where outerAngle = 20 is showed together with a case where outerAngle = 60. In both cases
innerAngle = 0. By studying the difference between them a interpretation of how the hollowCone
15

2.4. INJECTION EXAMPLE

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

injector shapes the spray should be aquired. In the figures, the plane to the left is placed at the
position of the injection disc and the clips are made to easy illustrate how the spray is spread in
the domain. The negative y-direction and thus the direction of gravity is perpendicular to the plane
created at the injection disc and pointing in the same direction as the top of the spray.

Figure 2.2: T ime = 0.0001. Injection just initialised, sprays almost similar. But a closer look gives that the
60 case spray is a bit wider.

16

2.4. INJECTION EXAMPLE

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

Figure 2.3: T ime = 0.00065. The difference between the cases grow larger. A cone shape can be identified,
starting from the injection disc

Figure 2.4: T ime = 0.00165. The difference continues to grow. The 60 case has clearly reached the outer
bondaries of the domain.

17

2.4. INJECTION EXAMPLE

CHAPTER 2. PARTICLE INJECTION IN DIESELSPRAY

Figure 2.5: T ime = 0.01. The last time step, both sprays have distributed evenly in the domains

18

Chapter 3

A simple particle injector

The solidParticle class offers simulation or at least computations for demo purposes of solid particles.
Here, a particle injector will be added in the class and a case will be solved with a modified version
of the simpleFoam solver.

3.1

injectorSolidParticleFoam

The modification will be compiled as a new class, injectorSolidParticleFoam. The first step is
to set up a new directory for compilation.

3.1.1

Directory and compilation settings

First of all, create a directory for the new class.

mkdir $WM_PROJECT_USER_DIR/applications/solvers/injectorSolidParticleFoam
cd $WM_PROJECT_USER_DIR/applications/solvers/injectorSolidParticleFoam
Now copy the needed files to the new directory.
cp -r $FOAM_SRC/lagrangian/solidParticle/* .
cp $FOAM_SOLVERS/incompressible/simpleFoam/* .
Rename the files corresponding to the new class name (the indentation below is so that it is possible
to easily copy the commands in all pdf-viewers).
mv solidParticle.C injectorSolidParticle.C
mv solidParticle.H injectorSolidParticle.H
mv solidParticleI.H injectorSolidParticleI.H
mv solidParticleIO.C injectorSolidParticleIO.C
mv solidParticleCloud.C injectorSolidParticleCloud.C
mv solidParticleCloud.H injectorSolidParticleCloud.H
mv solidParticleCloudI.H injectorSolidParticleCloudI.H
mv simpleFoam.C injectorSolidParticleFoam.C
Then make sure that the code within the files also agrees with the new names.
sed -i s/solid/injectorSolid/g injectorSolidParticle.C
sed -i s/solid/injectorSolid/g injectorSolidParticle.H
sed -i s/solid/injectorSolid/g injectorSolidParticleI.H
sed -i s/solid/injectorSolid/g injectorSolidParticleIO.C
sed -i s/solid/injectorSolid/g injectorSolidParticleCloud.C
sed -i s/solid/injectorSolid/g injectorSolidParticleCloud.H
sed -i s/solid/injectorSolid/g injectorSolidParticleCloudI.H
sed -i s/simpleFoam/injectorSolidParticleFoam/g injectorSolidParticleFoam.C
19

3.1. INJECTORSOLIDPARTICLEFOAM

CHAPTER 3. A SIMPLE PARTICLE INJECTOR

The last row above is not neccesary, but it is done so that it is easy to see in the beginning of the
file that the solver is injectorSolidParticleFoam and not simpleFoam. For correct compiling, the
files in /Make needs to be changed. Edit Make/files with whatever text editor you like. Make sure
it looks like:
injectorSolidParticle.C
injectorSolidParticleIO.C
injectorSolidParticleCloud.C
injectorSolidParticleFoam.C
EXE = $(FOAM_USER_APPBIN)/injectorSolidParticleFoam
Also change Make/options to
EXE_INC = \
-I$(LIB_SRC)/turbulenceModels \
-I$(LIB_SRC)/turbulenceModels/incompressible/RAS/RASModel \
-I$(LIB_SRC)/transportModels \
-I$(LIB_SRC)/transportModels/incompressible/singlePhaseTransportModel \
-I$(LIB_SRC)/finiteVolume/lnInclude \
-I$(LIB_SRC)/lagrangian/basic/lnInclude
EXE_LIBS = \
-lincompressibleRASModels \
-lincompressibleTransportModels \
-lfiniteVolume \
-llagrangian \
-lincompressibleTurbulenceModel
The settings in files are needed to define which files that are to be compiled and which command
the new class will be called with. In options, models for solving both the flow and the particle
tracks are included. The new class is now setup for the compilation. But before doing that, changes
to implement the particle injector are needed.

3.1.2

injectorSolidParticleFoam.C

In the beginning of this file, among the include statements, make sure to include
#include "injectorSolidParticleCloud.H"
This is done so that the solver can call functions of injectorSolidParticleCloud.H, which will be
modified in the following section. Also, after #include "initContinuityErrs.H", state
#include "readGravitationalAcceleration.H"
This is done since the gravitational acceleration, g needs to be read when solving the particle tracks.
Directly before Info<< "\nStarting time loop\n" << endl; insert
injectorSolidParticleCloud particles(mesh);
so that the solver knows that particles are being included in the flow. To update the movement of
particles, just before runTime.write();, add
particles.move(g);
The first three of the above statements need to be included before this last statement. This is due
to that these make sure that everything needed for solving the particle tracks is included. The
calculation of all particle positions is then done at particles.move(g);.
Save and close the file. The solver should now work as desired, but the injection of the particles has
not yet been defined.
20

3.1. INJECTORSOLIDPARTICLEFOAM

3.1.3

CHAPTER 3. A SIMPLE PARTICLE INJECTOR

injectorSolidParticleCloud.C

For specification of the injection process, the following code is needed. This can be included anywhere
inside the Member function void Foam::solidParticleCloud::move(const dimensionedVector& g),
where the updating of particle positions is done. So to be sure to get it right, put it at the top, just
after the first curly bracket.[3]
// Injector 1
//Set injection position (z=0 if 2d)
scalar posy=0.015;
scalar posz=0;
scalar posx=-0.0203;
vector pos = vector(posx,posy,posz);
//Set initial velocity vector
vector vel=vector(0,0,0);
//Particle diameter
scalar d = 1e-3;
// Find cell at specified injection position and add particle here
label cellI=mesh_.findCell(pos);
if(cellI>=0) {
injectorSolidParticle* ptr= new injectorSolidParticle(*this,pos,cellI,d,vel);
Cloud<injectorSolidParticle>::addParticle(ptr);
}
As can be seen, properties like diameter, initial velocity and initial position of the injected particles
are specified here. The label cellI finds a cell located at the injection position. Then, in the if
statement, a particle with the specified diameter and velocity is added in this cell, at the injection
position. Save and close.

3.1.4

createFields.H

The solver injectorSolidParticleFoam will need to read the density of the fluid the particles are
introduced in. To be able to do this the following should be added at the top of createFields.H.
Info<< "\nReading transportProperties\n" << endl;
IOdictionary transportProperties
(
IOobject
(
"transportProperties",
runTime.constant(),
mesh,
IOobject::MUST_READ,
IOobject::NO_WRITE,
false
)
);
dimensionedScalar rhotP(transportProperties.lookup("rho"));
volScalarField rho
(
IOobject
(
"rho",
runTime.timeName(),
21

3.2. EXAMPLE CASE

CHAPTER 3. A SIMPLE PARTICLE INJECTOR

mesh,
IOobject::NO_READ,
IOobject::NO_WRITE
),
mesh,
dimensionedScalar("rho", rhotP.dimensions(), rhotP.value())
)
;
This is the final modification before compilation, save and close the file. Make sure to be in the
directory of the .C files that are to be compiled and do
wclean
wmake

3.2

Example case

The new class will now be implemented on the pitzDaily case. This case is appropriate since it
is possible to, in simple way demonstrate how the particles are influenced by the fluid flow. The
domain of the pitzDaily case can be seen in figure 3.1. [5]

Figure 3.1: Geometry of the pitzDaily tutorial case.

Start by copying and renaming the case.

cd $FOAM_RUN
cp -r $FOAM_TUTORIALS/incompressible/simpleFoam/pitzDaily ./pitzDailyInject
cd pitzDailyInject
In this example, the flow will be solved to steady state before the particles are introduced into the
flow. This technique is commonly used in CFD-simulations, even though there normally are more
sophisticated models for particle-particle, particle-fluid and particle-wall interaction. So, to solve
the flow do
blockMesh
simpleFoam
This will take a couple of minutes. Meanwhile, locate the attached folder files which include
boundary conditions and physical properties for particles in accordance with the solidParticle class.
Since the particle injector will be used here, the conditions are set up so that no additional particles
are injected at the simulation start. When simpleFoam has solved the flow, the case directory have
several time directorys, with 1000 as the latest. Do the following.
22

3.2. EXAMPLE CASE

CHAPTER 3. A SIMPLE PARTICLE INJECTOR

Copy the folder lagrangian to ./1000

Copy the file particleProperties to ./constant
Copy the file g to .constant.
Density is in particleProperties set to 1.2 kg/m3 and in g the gravity to 9.81 m/s2 . Then edit
system/controlDict and change the following parameters as
startTime
endTime
deltaT
writeInterval
timePrecision

1000;
1000.15;
0.003;
1;
8;

This means that the simulation only will run for additionaly 0.15 seconds and data for every time
step of 0.003 s will be saved. The time precision is changed, so that all numbers of these time steps
will be included when being saved.
Finally, for setting the fluids density, edit /constant/transportProperties and after
nu nu [ 0 2 -1 0 0 0 0 ] 1e-05; insert
rho rho [ 1 -3 0 0 0 0 0 ] 1;
This density value is what is read in the code inserted in createFields.H in Section 3.1.4. Necessary
settings for the simulation are now set. Run the simulation, convert it and postprocess by
injectorSolidParticleFoam
foamToVTK
paraview
In paraview click Open and open pitzDailyInject.vtk in the /VTK directory. Click Apply
and choose to display the domain with velocity, U and as Surface. See below.

Again, click Open, browse into /lagrangian/defaultCloud and open defaultCloud.vtk. Click
Apply, go to the last time step with
and rescale data to range with
. By doing this,
it is ensured that the particle data can be reached in paraview since they are introduced in the
flow at after about twenty time steps. On defaultCloud insert a Glyph filter by marking it and
clicking
. Choose d for Scalars, Sphere for Glyph Type, Scale Mode as off, enable
edit of Set Scale Factor and set it to 0.005. Click Apply and . The parcels are injected as
specified and their path through the domain can be seen in figure 3.2.

23

3.3. TWO INJECTORS

CHAPTER 3. A SIMPLE PARTICLE INJECTOR

Figure 3.2: Solid particles in the case domain after the last time step.

3.3

Two injectors

With a slight modification of the setup in previous section, a second injector can be implemented
on the pitzDaily case. Go back to the injectorSolidParticleFoam class directory.
cd $WM_PROJECT_USER_DIR/applications/solvers/injectorSolidParticleFoam
Edit the file injectorSolidParticleCloud.C and after the previous specification of the first injector, add
// Injector 2
//Set injection position (z=0 if 2d)
scalar posx2=0.15;
scalar posz2=0;
scalar posy2=-0.0243;
vector pos2 = vector(posx2,posy2,posz2);
//Set initial velocity vector
vector vel2=vector(0,0,0);
//Particle diameter
scalar d2 = 1e-3;
// Find cell at specified injection position and add particle here
label cellI2=mesh_.findCell(pos);
if(cellI2>=0) {
injectorSolidParticle* ptr= new injectorSolidParticle(*this,pos2,cellI2,d2,vel2);
Cloud<injectorSolidParticle>::addParticle(ptr);
}
This injector will be placed at the lower wall of the domain. Save and close, clean and compile
(wclean and wmake) again.
Go to the case directory again.
24

3.4. EVENLY DISTRIBUTED INJECTION CHAPTER 3. A SIMPLE PARTICLE INJECTOR

cd $FOAM_RUN/pitzDailyInject
Previous knowledge tells that in order to be able to follow the entire track of the particles injected
at the second injector a longer runtime is needed. Edit .system/controlDict and change the end
time to
endTime

1000.42;

Before rerunning the case, delete the previous particle track results.
rm -rf 1000.*
rm -rf VTK
Solve with the recompiled solver, convert and postprocess as in Sect. 3.2. The result shows that the
second injector works and that its particles are injected into the recirculation region of the domain.
Towards the end of the simulation, it can though be seen that the particles tears away from this
region and flow out of the domain. It can also be seen that the behaviour of the particles is not
totally steady. Reasons for this could be the nature of the flow and/or how the particle behaviour
is modeled. To analyze this further is however outside the scope of this tutorial. The result after
the last time step can be seen in figure 3.3

Figure 3.3: Solid particles in the case domain after the last time step. Two injectors used

3.4

Evenly distributed injection

The injector implemented in previous sections can easily be modified so that a collection of particles
with a constant spacing is introduced. Go back to the directory of injectorSolidParticle.
cd $WM_PROJECT_USER_DIR/applications/solvers/injectorSolidParticleFoam
Up to now there has only been one particle injected per time step and injector. So, to inject several
particles per time step it should be sufficient to repeat the injection process as many times as the desired number of particles is. This can be achieved by a for loop. Open injectorSolidParticleCloud.C
and scroll down to the section where the injector is defined. Remove entirely the second injector
and before //Injector 1 insert
25

3.4. EVENLY DISTRIBUTED INJECTION CHAPTER 3. A SIMPLE PARTICLE INJECTOR

for (label i=0; i<=25; i++) {

To end the loop, dont forget to put a } directly after the end of the if statement at the end of
the injector. With this setup, 25 particles will be injected per time step. But to make them spread
over a certain area, an additionaly modification is needed. Change the y-pos to
scalar posy=-0.0248+0.002*i;
Also set the x-pos as
scalar posx=0.1;
Save and close. The injector will now be placed in the widest area of the domain and inject 25
particles with a spacing of 0.002 m. Compile the class, clear previous results, run and postprocess
the case as earlier. Results are seen in figure 3.4

Figure 3.4: Solid particles in the case domain after the last time step. Evenly distributed injector used

26

Bibliography
[1] Per Carlsson Tutorial dieselFoam. Gothenburg, Sweden 2009.
[2] http://en.wikipedia.org/wiki/Weibull_distribution
[3] Aurelia Vallier Tutorial icoLagrangianFoam/solidParticle Gothenburg, Sweden 2010
[4] http://openfoamwiki.net/index.php?title=Special:Search&ns0=1&redirs=1&search=
injector&limit=100&offset=0
[5] foam.sourceforge.net/doc/Guides-a4/ProgrammersGuide.pdf

27