Predicting Flow Dynamics of an Entire

Engine Cooling System Using 3D CFD

Adam Johansson
Jonas Gunnarsson

Engineering Physics and Electrical Engineering, masters level

2017

Luleå University of Technology

Department of Computer Science, Electrical and Space Engineering
 Abstract

A combustion engine generates a lot of heat which need to be cooled to prevent dam-
ages to the engine and the surrounding parts. If the cooling system can not provide
enough cooling to keep the engine in a well defined range of temperatures performance
and durability will decrease and emissions increase. It is also important that the cooling
system do not over-cool the engine, since this may result in rough running, increased
engine friction and an overall negative performance.

The aim of this thesis work is to create a complete 3D digital model of the cooling sys-
tem for the first generation VED4 HP with CFD in STAR-CCM+. The simulated results are
compared to available experimental data for validation.

Today the entire system is being modeled with 1D CFD. One of the selected compo-
nents in the cooling system being model in 3D at Volvo Cars is the water jacket. The
3D CFD model depends on the 1D CFD model for the boundary conditions which is an
ineffective and time consuming process, sending data back and forth between the mod-
els when making changes. A 3D CFD model is not only more accurate than the 1D CFD
model, since it capture the 3D flow phenomenas but it also allows parts or areas to be
studied in detail.

A study of four different turbulence models is conducted on the water jacket and on
an arbitrary pipe in the cooling system. A mesh study is carried on the water jacket, the
same arbitrary pipe and on the thermostat, both for the opened and closed thermostat.
These studies are done with regard to pressure drop only.The study yields a low Reynolds
model with the k-² v2f turbulence model gave the best results.

There is a discrepancy between the simulated results and the experiments. Main reasons
to this may be the difference in the geometry used in this thesis for the digital model and
the geometry used for the experiments together with the inaccuracies in the experimen-
tal data. The overall deviation is larger for a case with closed thermostat than for a case
with an open thermostat. With the correct geometry and more accurate experimental
data the simulations should be a close representation of reality.

i
Preface
This thesis work represent an end of an era, a five year long journey to our Master of Science
in Engineering Physics at Luleå University of Technology. These five years have been tough
and challenging but also very pleasant. We would like to thank our classmates and friends for
the great time we have had during our studies.

This thesis work has been conducted at Volvo Cars in Torslanda, Göteborg at the department
of Exhaust and Cooling System. We would like to express our sincere gratitude to Volvo Cars
for giving us this opportunity and for the kindness everyone has shown us. We would like to
especially thank our supervisor at Volvo Cars, Fabian Hasselby, for all the help, guidance and
productive discussions.

ii
Nomenclature and abbreviations

x, y, z Spatial coordinates (m)

u, v, w Velocity components (m/s)
t Time (s)
ρ Density (kg/m3 )
g Gravitational acceleration (m/s2 )
p Pressure (Pa)
fi Specific force tensor (m/s2 )
U Free stream flow velocity (m/s)
Re Reynolds number (-)
fD Darcy-Weibach friction factor (-)
Q Volumetric flow rate (m3 /s )
Dh Hydraulic diameter (m)
uτ Friction velocity (m/s)
y+ Normalized spatial y coordinate (-)
δ Boundary layer thickness (m)
τ Shear stress (N/m2 )
δi j Kronecker delta (-)
µ Dynamic viscosity (Pa ⋅s )
ν Kinematic viscosity (m2 /s )
hL Head loss (m)
κ Thermal conductivity (W/(m ⋅K ))
k Specific turbulence kinetic energy (m2 /s 2 )
ε Turbulence dissipation rate (m2 /s 3 )
ω Specific turbulence dissipation rate (1/s)

CFD Computational Fluid Dynamics

CAD Computer-aided design
CAE Computer-aided engineering
RANS Reynolds Average Navier-Stokes
SST Shear Stress Transport
EB Elliptic Blending
VED4 Gen1 HP Volvo Engine Diesel - 4 Cylinder - First Generation - High Performance
EOC Engine Oil Cooler
ALM Advancing Layer Mesher

iii
Contents

Abstract i

Preface ii

Nomenclature iii

1 Introduction 1
1.1 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Method and limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Theory 4
2.1 Governing equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1.1 Conservation of mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1.2 Conservation of momentum . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1.3 Conservation of energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Turbulence modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.1 Reynolds-Average Navier-Stokes . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 k-ε model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 Near-wall behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4.1 Wall functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4.2 Near-wall modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.5 Boundary layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.6 Flow separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.7 Head loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.7.1 Pressure drop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.7.2 Friction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.8 Mesh quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.9 Porous Baffle Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Method/Setup 18
3.1 Study of turbulence models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.1 Water jacket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.2 Pipe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.3 Transient simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Thermostat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.1 Closed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.2 Open . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Study of mesh settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.1 Water jacket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.2 Pipe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.3 Thermostat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Full system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

iv
 3.5 Porous Baffle Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

4 Results and discussion 33

4.1 Turbulence model results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.1.1 Water jacket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.1.2 Pipe section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1.3 Boundary layer thickness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2 Study of mesh settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.2.1 Water jacket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.2.2 Pipe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2.3 Thermostat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3 Full system simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.1 Case 1/Open thermostat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.1.1 Case 1/Open thermostat, with major geometry difference be-
tween CAD and test rig experiment (EGR-TOC junction) . . . . . 53
4.3.2 Case 2, closed thermostat . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5 Conclusions 57
5.1 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

A Surface roughness 59

v
1 Introduction
An engine cooling system is designed to maintain the temperature of the engine in an optimal
condition to maximize the potential of the engine and ensure proper combustion. To prevent
the engine from overheating, the cooling system needs to be able to transfer the heat, mainly
generated in the combustion chamber (cylinders), away from the heat sources to the radiator.
If the cooling system can’t provide enough cooling to keep the engine in a well defined range
of temperature it will results in negative effects in performance, high emissions and durabil-
ity. It is also important for the cooling system not to over-cool the engine, otherwise it may
result in rough running, increasing the engine friction and overall negative performance. [5]

One component of the cooling system is a water jacket. This is a hollow mantle made of
aluminum surrounding the cylinders. The coolant flows inside the the water jacket and ab-
sorbs the radiated heat coming from the cylinders generated by the combustion. The coolant
flow continues to absorbs heat generated by other heat exchangers, such as engine oil cooler
(EOC) and transmission oil cooler (TOC) throughout the rest of the system. The coolant flow
going through the EOC is circulating through the water jacket and the main part of the flow
through the water jacket flows through a thermally controlled thermostat which decides the
amount of coolant flow going to the radiator and to a by-pass circuit going back into the water
jacket. The hot flow coming from the thermostat to the radiator is met by an air flow coming
through the grill and in under the hood. At the radiator the coolant flow exchange heat with
the air flow, cooling the coolant flow.

The geometry of the cooling system is complex as well as the coolant flow through it. Even
when experimental models are available it may be difficult to measure and visualize the flow
correctly. Understanding the flow inside the cooling system is a key to optimizing the design
leading to increasing cooling efficiency. To gain more understanding, simulation with com-
putation fluid dynamic (CFD) is carried out and by analyzing the results new information can
be obtain that would have been very difficult to gain by measurements. Another advantage
with CFD is that it allows any part or region to be studied in detail which can be used to gain
deeper knowledge about different flow phenomena.

1.1 Aim
The aim of this thesis work is to create a complete 3D digital model of the cooling system for
a Volvo diesel engine currently in production. This is done by evaluating the pressure drops
through full system simulations of the cooling system.

1.2 Method and limitations

To fulfill the aim, the work is divided into two studies to find an appropriate model to solve
for. This is done with focus on the pressure drops and mass flow distribution through the
system. The studies consist of:

1
 1. Study of turbulence models, where an investigation of appropriate turbulence models
and the influence of y + -values on the system and turbulence models is carried out.

2. Study of mesh settings, evaluating different mesh settings to see what is required to get
mesh independent results.

The two studies of turbulence models and mesh settings were focused on the water jacket
and a pipe. The mesh settings study was also performed on a thermostat to obtain a more
accurate mesh to cope with a more complex inner geometry.

After the two studies were performed, a full system with the water jacket and the surround-
ing pipe systems was considered in order to evaluate the accuracy of using 3D CFD when
solving for pressure drops. When doing the full system simulation, components like pumps,
radiators and the engine oil cooler (EOC) were not modeled but instead represented as pres-
sure drop functions, saving time and computational resources, see Section 2.9. To transform
the pipe system into a digital model, several actions were performed including cleaning and
adapting the pipes to be acceptable for usage as a calculation domain and a removal of un-
necessary and to complex geometry parts from the calculation domain. As to the models of
the flow focus is set on RANS turbulence models only, see Section 2.2 for explanation.

The geometry for the cooling system is complex with regions e.g. contracting, expanding,
changing shape, flow entering and exit pipe. To cope with this, simplifications are desirable
and in many cases necessary. Part of the work is to determine where and when these simpli-
fications can be done, and to investigate the effects on the results. A visual representation of
the full system of the cooling system can be seen in Fig. 1, while the water jacket used for this
cooling system is shown in Fig. 2.

Figure 1: Illustration of the full cooling system sketch used at test rig, courtesy of Volvo Cars.

2
 Figure 2: A VED4 HP Gen1 water jacket.

If the thesis is finished successfully, this work will lead to a simplification of the work process
of today, making the simulation of the cooling jacket more effective than before.

3
2 Theory

2.1 Governing equations

Fluid flow problems can be described by three physical principles, the conservation of mass,
momentum and energy. These will further described in the upcoming section. These physi-
cal principles are expressed in terms of mathematical equations and these are solved numer-
ically in a discretized domain for a given set of boundary conditions.

2.1.1 Conservation of mass

The law of conservation of mass is also known as the continuity equation. This law states
that for a closed system, mass cannot be created nor destroyed. The mass can be rearranged
in space and transformed into different forms, but the quantity will always remain the same.
In fluid dynamics it is more common to speak about mass flows rather than mass. In these
terms the continuity equation is the net inflow of mass into an infinitely small fluid element
equalling the rate of change of mass in the same element.

∂ρ ∂
+ (ρu j ) = 0, (2.1)
∂t ∂x j
where ρ is density, t time and u j the velocity tensor. [6]

2.1.2 Conservation of momentum

The law of conservation of momentum is also known as Navier-Stokes equations in fluid dy-
namics. This law states that for a closed system, momentum cannot be created nor destroyed.
The only way to change the momentum is to apply external forces described by Newton’s sec-
ond law. In fluid dynamic terms, the conservation of momentum is the rate of increase of
momentum in an infinitely small fluid element equalling the net influx of momentum added
together with body and surface forces acting on the same element.

∂ ∂
(ρu i ) + (ρu i u j + pδi j − τi j ) = ρ f i , i = 1, 2, 3 (2.2)
∂t ∂x j
∂u i ∂u j
where p is pressure, δi j is the Kronecker delta and τi j = µ ( + ) is the viscous stress
∂x j ∂x i
tensor. [6]

2.1.3 Conservation of energy

The law of conservation of energy states that for a closed system, energy cannot be created
nor destroyed. The energy can take different forms but the quantity will always be the same.
This law is commonly known as the first law of thermodynamics. When speaking in terms of
fluid dynamic the law is stated as the rate of change of energy inside an infinitely small fluid
element equals the sum of the net flux of heat into the element and the rate of work done on
the element due to body and surface forces.

4
 ∂ ∂
(ρe 0 ) + (ρu j e 0 + u j p − u i τi j + q j ) = 0 (2.3)
∂t ∂x j
∂T
where e 0 is the internal energy and q i = −κ is the heat flux with κ = κ(T ) as the thermal
∂x i
conductivity. [6]

2.2 Turbulence modeling

Turbulence is a three dimensional (3D), unsteady and chaotic property of the flow which oc-
cur at high Reynolds number. Turbulence is governed by disturbances of the laminar flow
and occurs over a large spectrum of both length and time scales. These disturbances can be
thought of as series of 3D eddies, which are swirling motions being the size of the local tur-
bulent length scale, of a variety of sizes that are in constant interaction with each other, these
eddies are being transported by the mean flow [7]. Due to the large spectra of sizes of the ed-
dies, they overlap, the larger eddies is transferring its energy to the smaller ones creating an
energy cascade. This leads to eddies decreasing in size and only exist for a short time before
they are dissipated into heat by molecular viscosity. [8]

The most accurate way to model turbulence is to solve everything directly, this is called direct
numerical simulation (DNS). In order to use DNS the flow domain must be highly resolved in
order to capture even the smallest eddies. For engineering purposes this leads to enormous
meshes (since they have be to in order of Re 9/4 ) and time steps, this can’t be done by today’s
computational resources. Due to the short comings with DNS several methods for modeling
the flow have been developed, Reynolds Average Navier Stokes (RANS) being the most com-
mon. Instead of solving all the quantities RANS only solves the averaged ones. The effect
of the turbulence scales are modelled with a turbulence model, saving a lot of computation
power but at the same time loses accuracy. [9]

Another way to model turbulence is Large Eddy Simulation (LES). In LES the large eddies
are solved directly and the small ones are modelled. This makes a significantly reduction in
computational power compared to DNS due to the decreasing amount of direct solving and
increasing accuracy compared to RANS. The increasing accuracy comes from the amount of
turbulent energy stored in the large eddies, which are now directly solved instead of being
modelled. The large eddies are also responsible for most of the turbulent momentum and
turbulent mixing. In addition to this, the smaller eddies tend to be more isotropic and homo-
geneous, hence easier to model than a large variety of scales. [9]

5
Figure 3: How different turbulence models work. Inspired by Turbulence models and their
applications presented by T.S.D.Karthik.

2.2.1 Reynolds-Average Navier-Stokes

The RANS equations are directly derived from the governing equations, Eq.(2.1)-(2.3). The
variables are decomposed, by Reynolds decomposition, into a time averaged and fluctuation
component

φ = φ + φ′ . (2.4)
Since this report only deals with incompressible flows, the incompressible version of RANS
will be used in this section. Substituting Eq.(2.4) for all variables into Eqs.(2.1)-(2.3) yields the
RANS equaitons

∂u j
= 0, (2.5)
∂x j

∂u i ∂ ∂p ∂ ∂u i ∂u j
ρ +ρ (u i u j ) = − + (µ ( + ) − ρu i′ u ′j ) + ρ f i . (2.6)
∂t ∂x j ∂x i ∂x j ∂x j ∂x i
Using Eq.(2.5) in Eq.(2.6) results in the final form of the RANS equations

∂u i ∂u i 1 ∂p ∂2 u i ∂u i′ u ′j
+uj =− +ν − + f i, (2.7)
∂t ∂x j ρ ∂x i ∂x j ∂x j ∂x j

where R i j = ρu i′ u ′j is known as the Reynold stress tensor. By assuming that the entire turbu-
lence spectrum can be modelled by decomposing the turbulence into a time average and a
fluctuating part six new unknowns are introduced, deriving from Reynolds stress tensor. The
number of equations remains the same and therefore six new equations are needed in order
to close the system.

6
This can be done my introducing turbulent viscosity, first done by Boussinesq hence called
the Boussinesq approximation which relates the Reynolds stresses to the mean velocity gra-
dients

∂u i ∂u j 2
u i′ u ′j = −νt ( + ) + kδi j , (2.8)
∂x j ∂x i 3

where νt is the turbulent eddy viscosity and k = 12 u i′ u i′ is the turbulent kinetic energy.
[10]

2.3 k-ε model

The k-ε model is a semi-empirical model which uses the turbulent energy k and its dissipa-
tion rate ε to model the eddy viscosity, being the most commonly used models for industrial
purposes. It is only valid for fully turbulent flows due to assumptions in the derivation of the
equations and that the effects of the molecular viscosity are negligible[12]. In the near-wall
regions the turbulent flow is heavily disturbed by the wall and the Reynolds number becomes
small and the viscous effects dominates the flow. It is well known that k-ε models have diffi-
culties handling large gradients in the near-wall region.

The turbulent eddy viscosity

k2
νt = C µ . (2.9)
ε
The equation for the turbulent energy, k

∂k ∂k ∂u i ∂ νt ∂k
+uj = −R i j −ε+ ((ν + ) ). (2.10)
∂t ∂x i ∂x j ∂x j σk ∂x j

The equation for the dissipation rate, ε

∂ε ∂ε ε ∂u i ε2 ∂ νt ∂ε
+uj = −C ε1 R i j − C ε2 + ((ν + ) ). (2.11)
∂t ∂x i k ∂x j k ∂x j σε ∂x j

The closure coefficients are

C ε1 = 1.44, C ε2 = 1.92, C µ = 0.09, σk = 1.0, σε = 1.3. (2.12)

2.4 Near-wall behaviour

2.4.1 Wall functions

One way of overcome the near-wall difficulties with the k-ε model is to use wall functions.
The idea is to place the first node inside the log-layer and make assumptions on how the
velocity profiles behaves inside the viscous sublayer. This is of course less demanding of
computational resources but a fair amount of information is lost. Inside the boundary layer
the dimensionless velocity u + = u /u ∗ varies with the local velocity.

7
 1
u+ = ln(E y + ) y + ≥ 11.225
κ , (2.13)
u+ = y + y + < 11.225
where κ is von Karmans constant, E is a log-layer constant and y + is a dimensionless wall
distance.
yu ∗
y+ = , (2.14)
ν
where u ∗ is the friction velocity, see Eq. (2.22).

Wall functions need to be provided with any turbulence variable during the simulation, most
commonly the turbulent kinetic energy, k, and its dissipation rate, ε. When turbulence in
the fully turbulent near-wall region is in equilibrium, it may be assumed that the production
and dissipation of the turbulent kinetic energy are in balance. [13] The turbulent production
can be found in Eq.(2.10) as the first term on the right hand side. Inside the log-layer region,
τw = −ρu i′ u ′j , using this together with the Boussinesq approximation, Eq.(2.8) the dissipation
rate in the near wall cell can be calculated as

∂u i u∗ u ∗3
ε = G k = Ri j = τw = . (2.15)
∂x j κy + κy +
Using the shear stress approximation τw = µt κy
u∗
+ together with Eq.(2.15), Eq.(2.9) and Eq.(2.22)
an expression for the turbulent kinetic energy in the near wall cell is received

u2
k = √∗ . (2.16)
Cµ
[10]

2.4.2 Near-wall modeling

In near-wall modeling the entire boundary layer, all the way into the wall is resolved with a
fine mesh. When using this method the turbulence models need modification in order to ac-
count for the viscous effects near the wall. The modifications is done by including damping
functions, which works in the viscous layer and part of the buffer layer [14]. The most impor-
tant damping function is the one damping the turbulent eddy viscosity νt , which is done by
adding a damping function f µ to C µ

k2
νt = f µC µ
, (2.17)
ε
where in the case of the Launder Sharma model the dampening function is expressed as

−3.4
f µ = exp [ ]. (2.18)
(1 + R T /50)2
[11]

8
2.5 Boundary layer
The boundary layer theory was developed in the early twentieth-century by Ludwig Prandtl.
He introduced the idea that due to friction, the fluid closest to the surface would actually stick
to the surface, i.e. a no-slip condition. It means the relative velocity between the boundary
and the flow particles closest to it. Further away from the boundary, the velocity increases
through the boundary layer to connect with the free stream velocity U . This forced the vis-
cosity of the fluid to be considered for the description of the near-surface flow, in contrast
to the free stream flow where the viscosity can be neglected. Frictional effects are only expe-
rienced in a thin layer close to the surface called the boundary layer. Outside the boundary
layer the friction has very little effect and therefore the assumption of neglecting the viscosity
in high Reynolds flow are still valid far from the surface. [1]

The boundary layer is important when describing the flow because the velocity gradient of
the streamwise flow u in the normal direction y is much higher inside the boundary layer
(BL) than in the free stream (FS),

∂u ∂u
∣ > ∣ . (2.19)
∂y BL ∂y FS
∂u
In general, the velocity gradient is defined as zero in the free stream, ∣
∂y FS
= 0. This yields for
exterior flows. Interior flows as in pipes has often a different velocity profile. The viscosity
(µ, ν) give rise to shear stresses in regions with a larger velocity gradient, i.e. the boundary
layer. [1] The shear stress τ can be expressed as

∂u
τ( y ) = µ . (2.20)
∂y
For a very small velocity gradient, i.e. the free stream, the shear stress disappear which ne-
glects the effect from viscosity. The wall shear stress is the shear stress in Eq. (2.20) at the
surface of the boundary

∂u
τw = µ ∣ . (2.21)
∂y y =0

A near wall velocity, also known as the friction velocity, can be related to the wall shear stress
with the density which is used e.g. when scaling to the dimensionless y + -value, [1]
√
τw
uτ = = 0. (2.22)
ρ
For parallel flows over a flat plate, the flow is said to occur in the boundary layer when the
streamwise velocity close to the surface is below or equals 99 % of the free stream velocity,
u ( y ) ≤ 0.99U . The relative height of the boundary layer depends on the length between the
origin of the boundary layer and the point of interest along the surface, and in which bound-
ary layer zone (laminar or turbulent boundary layer) the flow occurs. [1]

9
Figure 4: Each boundary layer zone with its characterised flow field along with the boundary
layer thickness, δ. Inspired by COMSOL.

The boundary layer can be divided into 3 separated regions, laminar, transition and turbu-
lent. The laminar boundary layer defines the region with smooth flow while the turbulent
region defines where the flow has taken an unorderly characteristic behaviour with relating
vortices and eddies. For a flat plate, the boundary layer regions is defined from the Reynold’s
number, with Re < 105 and Re > 3 ⋅ 106 for the laminar and turbulent region, respectively.
The boundary layer flow which ranges between the laminar and turbulent boundary layer is
called the transition region. A graphical representation of the boundary layer flow in each
region can be seen in Fig. 4. In general the turbulent boundary layer has larger velocity gra-
dients the laminar flow, resulting in a steeper velocity increase close to the boundary. [1]

The Navier-Stokes equations in the boundary layer, boundary layer equations, can be sim-
plified from the original equations in Eq. (2.1)-(2.2) by doing scaling analysis. The order of
the parameters in the equation for conservation of mass are

∂ 1 ∂ 1
∼ , ∼ , u ∼U, p ∼ ρU 2 . (2.23)
∂x L ∂y δ
Here, [x, u ] corresponds to the coordinate and velocity component in the direction of the
streamwise flow and [ y, v ] as the coordinate and velocity component perpendicular to stream-
wise flow, respectively. Assuming a two-dimensional [u, v ] incompressible flow without grav-
itational forces, (ρ = constant, w = 0, f i = 0 ), and a thin boundary layer L ≫ δ, results in

Uδ
v =− ∼ 0. (2.24)
L
Evaluating the equation for conservation of momentum for the spatial direction of y (i = 2)
with the scaling analysis gives

10
 ∂p ∂p dp
= 0, Ð→ = , (2.25)
∂y ∂x dx
so that the pressure is only dependent on the x coordinate. Evaluating for the spatial direction
of x (i = 1) with the scaling analysis gives

∂u ∂u 1 dp ∂2 u
u +v =− +ν , (2.26)
∂x ∂y ρ dx ∂y 2
which also is the sole momentum equation solved in the boundary layers. [1]

2.6 Flow separation

With flows in complex geometries, pipe bends, sharp turns/corners and sudden contractions
or enlargements, flow phenomena called flow separation may occur. The flow separation is a
lift-off between the boundary layer of the fluid and the surface of the boundary and instead
it creates a recirculation zone, often referred to as a wake, with recirculating flow in forms of
vortices and eddies. The existence of flow separation comes from the friction arising from the
no-slip boundary condition for walls and other non-moving surfaces and due to an adverse
pressure gradient

∂p
> 0, (2.27)
∂x
along the boundary in the streamwise direction. The friction decreases the velocity and there-
fore the kinetic energy of the flow, making the flow to easier separated from the boundary. If
the streamwise pressure gradient increases (adverse) along the boundary, the flow will be
more receptive to separated. For a large enough adverse pressure gradient, a negative veloc-
ity gradient

∂u
< 0, (2.28)
∂y
will start to decrease the streamwise velocity u near the boundary and finally reverse the flow
if u ( y ) < 0. The point of separation occur directly at the switching point when u ( y ) = 0.
Depending in the type of boundary layer (laminar, turbulent), the flow will be more or less
receptive to separate. A turbulent boundary layer more resistant of flow separation. [1]

11
Figure 5: The velocity profile along a boundary with an adverse pressure gradient causing
flow separation with reversed flow. Inspired by Aerospace Engineering.

In Fig. 5, the effect of a adverse pressure gradient on the velocity profile is shown. As seen, the
boundary layer thickens after the point of separation along with the height of the recircula-
tion zone. A separation line can be visualized at u ( y ) = 0 along the streamwise flow direction
x, marking the edge of the recirculation zone with its origin at the point of separation.

2.7 Head loss

When dealing with separation and secondary flows, the friction of the flow has an important
effect on the flow characteristics. The total energy in a system is then no longer constant
which introduces the concept of head loss. The head loss defines the loss of energy in a
system as a hydraulic height with the physical dimension of length. Head loss is caused by
friction and by velocity changes originating from components e.g. bends, valves, inlets and
contractions, with an absence of pumps and motors/turbins. One of the most studied cases
with head losses is fluid flow in pipes and ducts. For both pipes and ducts, the head loss can
be described with Bernoulli’s equation for viscous, one dimensional flow

U12 p1 U22 p2
+ + z1 = + + z 2 + ∆hL . (2.29)
2g ρg 2g ρg
The scalar quantity z represents the elevation above a reference plane, ∆h L represents the
U2 p
total head loss, and the terms , and z as the velocity head, pressure head and eleva-
2g ρg
tion head of the system, respectively. The pressure head and elevation head together is often
referred to as hydraulic head loss. The head loss of the system equals the difference of total
head ∆h L = h L 1 − h L 2 , where each head represents each side of Eq. (2.29)

12
 U12 p1
hL1 = + + z1 ,
2g ρg
(2.30)
U22 p2
hL2 = + + z2 .
2g ρg

It should be noted that h L 1 − h L 2 > 0 at all times since ∆h L symbolizes an energy loss off the
system. The head loss can be expressed as a sum of addition ∆h a and removal ∆h r of head
loss, head loss due to geometry shaping ∆h K and the friction head loss ∆h f as

∆h L = ∆h f + ∆h K + ∆h r − ∆h a . (2.31)
The difference in addition of head ∆h a is due to pumps. Pumps supply energy to the flow and
therefore increasing the pressure. The difference in the removal of head ∆h r is due to motors
and turbines, which both works in a similar way as pumps but they are instead decreasing the
pressure of the flow. The minor head losses is denoted ∆h K and the major head losses due to
friction as ∆h f . It should however be clear that minor head losses do not necessary have to
be smaller than major losses but that it is dependent on the structure of the flow domain. For
major head losses, a friction slope (of the energy gradient line) S f is defined for pipe flows as

1 U2
S f = fD , (2.32)
D h 2g
where S f can be treated as the friction head loss ∆h f per unit pipe length L. Now Eq. (2.32)
can be expressed for the friction head loss instead as

L U2
∆h f = f D , (2.33)
D h 2g
where f D is the Darcy-Weisbach friction coefficient, D h is the hydraulic diameter, L is the pipe
length, U flow velocity and g the gravitational acceleration. The hydraulic diameter can be
expressed as a function of the cross-section area A of the pipe or duct and a wetted perimeter
p w as

4A
Dh = , (2.34)
pw
where p w corresponds to the length of the contact-surface between the pipe-wall and the
fluid, in this case the perimeter of the pipe’s cross-section. For a pipe with circular cross-
section yields that the hydraulic diameter is the cross-section diameter of the pipe D h = D.
Minor head losses is defined from the velocity head together with a loss coefficient K relating
the type of component to a head loss

U2
∆h K = K . (2.35)
2g

13
The loss coefficient K is known and tabulated for multiple components. The total head loss
in a system, without motors/turbines, for multiple pipe sections i , pipe components j and
pumps k yield

L i Ui2 U j2
∆h L = ∑ h f i + ∑ h K j + ∑ h ak = ∑ ( f D i + + h ak ) , (2.36)
i j k i , j ,k D hi 2g 2g

similar to the system of the thesis work. [1]

2.7.1 Pressure drop

Pressure drop is related to the total head loss from Eq. (2.29) as

p1 − p2 U22 − U12 ∆U 2
= + z 2 − z 1 + ∆hL = + ∆z + ∆hL . (2.37)
ρg 2g 2g
For a system only consisting of major head losses due to friction, Eq. (2.37) is reduced to

p1 − p2 ∆U 2
= + ∆z + ∆h f , (2.38)
ρg 2g
when used together with Eq. (2.31) as where ∆p ≡ p 1 − p 2 is the total pressure drop. For a
horizontal pipe with constant diameter yields ∆z = ∆U = 0, therefore Eq. (2.38) can simply
be written as ∆p s = ρg ∆h f , which corresponds to the static pressure loss .

For a complex, general pipe system with sections and components as defined in Eq. (2.36),
the pressure drop is calculated as
ρ
∆p = (∆U 2 + 2g (∆z + ∆hL )) , (2.39)
2
where h L is given by Eq. (2.36).

For laminar flow the pressure drop can further be connected to the mass flow Q as

∆pπD 2
Q= , (2.40)
128µL
known as the Poiseuille’s law. Noted that the flow (Hagen-Poiseuille flow) is for at straight
pipe with constant diameter. [1]

2.7.2 Friction

The Darcy-Weisbach friction factor, named after Henry Darcy and Julius Weisbach, is de-
scribed as a function of Reynold’s number Re and roughness ratio Dε , with ε as the roughness
height,

ε 8τw
f D (Re, )= . (2.41)
D ρU 2

14
 32µLU
For laminar flows yields ∆p = , which in combination with (2.33) and the relation ∆h =
D2
ρg ∆p, stated in Section 2.7.1, gives the friction fator as a funtion of the Reynold’s number

64µ 64
f = = . (2.42)
ρDU R
For turbulent flows, the friction factor is described as an implicit relation, known as the Cole-
brook equation,

1 ε/D 2.51
√ = −2.0 log ( + √ ). (2.43)
f 3.7 Re f
Because the implicit relation demands an iterative solution, an approximate explicit relation
was given by S. E. Haaland as

ε /D
1.11
1 6.9
√ = −1.8 log (( ) + ). (2.44)
f 3.7 Re

The relation of Darcy-Weisbach friction factor against a range of Reynold’s numbers and the
roughness factors is demonstrated in the Moody-chart, see Fig. 6. [1]

Figure 6: A Moody-chart displaying the relation between the friction factor, Reynold’s number
and roughness factor. Courtesy of S. Beck and R. Collins, University of Sheffield.

15
2.8 Mesh quality
Judging the quality of a mesh can be done by considering different properties of the mesh.
Five different measurements of the mesh that can be used to validate the mesh quality are
Aspect Ratio, Cell Quality, Face Validity, Skewness Angle and Volume Change.

Aspect Ratio is a measurement of the proportion between the faces of a cell. Cells are consid-
ered bad when the aspect ratio is at 0.01 or lower. [4]

The Cell Quality is a measurement of the relative geometric distribution of the cell centroids
of the face neighbor cells and also of the orientation of the cell faces. A cell is considered
perfect when its cell quality is of 1.0. A face quality less than 1.0e-5 is considered bad, but in
this thesis work, a face quality less than 0.01 is considered bad. [4]

The Face Validity is an area-weighted measure of the correctness of the face normals rela-
tive to their attached cell centroid. In a good quality cell, the face normals point outwards,
away from the cell centroid. In a cell with bad face validity, one or more of the face normals
point inwards, towards the cell centroid. A face validity of 1.0 means that all face normals are
correctly pointing away from the cell centroid. Values below 1.0 mean that some of the cell
faces have normals pointing inward towards the cell centroid, indicating some form of con-
cavity. Values below 0.5 signify a negative volume cell, but in this case values less than 0.01
are considered bad. [4]

The Skewness Angle θ is the angle between the face normal vector and the vector connecting
the two cell centroids, ds. Cells with a skewness angle greater than 85○ are considered bad
cells. Skewness angles of 90○ or greater, which can occur in concave cells where the centroid
of both cells lies on the same side of the boundary face, typically result in solver convergence
issues. Problems result because the diffusion term formulation for transported scalar vari-
ables contains the dot product in the denominator, and this dot product is zero when the
angle is 90○ . Even though STAR-CCM+ takes care to avoid such a divide-by-zero error, the ac-
curacy of the diffusion calculation is reduced and there is a loss of robustness in such cases.
[4]

The Volume Change metric describes the ratio of the volume of a cell to that of its largest
neighbor and a value of 1.0 indicates that the cell has a volume equal to or higher than its
neighbors. Cells with a volume change of 0.01 or lower are considered bad cells. [4]

16
 (a) Illustration of Cell Quality. (b) Illustration of Face Validity.

(c) Illustration of Skewness Angle (d) Illustration of Volume Change.

Figure 7: Courtesy of CD Adapco.

2.9 Porous Baffle Interface

The porous baffle interface is used to model a pressure drop without a geometrically model
the component causing the pressure drop. The flow characteristics are transferred from one
interface to the other, including a pressure drop described by the interface. Physically, a
porous baffle represents a porous membrane through which fluid passes and experiences
a pressure drop. The pressure drop through the interface is expressed as

∆p
= (α∣v n ∣ + β) v n (2.45)
ρ
where v n is a superficial velocity, α and β are constants describing the porosity of the inter-
face. [4]

17
3 Method/Setup
The simulations where made solely in STAR-CCM+. The geometry works and simplifications
were mainly done in ANSA together with SpaceClaim. Boundary conditions and other input
data for the simulations will be taken from the corresponding work-sheet of [2], a report of a
performed test-rig experiment. Assumptions and physics chosen for all simulations were:

1. Fluid: Water (25○ C). 4. Gravity.

2. Incompressible flow. 5. Turbulent.

3. Segregated flow. 6. Adiabatic process (no heat transfers).

An approximation used in this thesis work is to neglect the difference between mass flow
[kg/s] and volumetric flow [l/s] because the difference is approximate 0.3%. The majority of
the flow input data are given in [l/s].

The coolant were replaced with water of room temperature, sharing similar viscosity. The dif-
ference in viscosity for water for one degree is larger than the difference between water and
the coolant fluid. This is also the case for the test-rig report. The fluid velocities are expected
to be low relative to the criteria for compressible flows, therefore incompressibility and seg-
regated flows were chosen. Heat transfers were also neglected, merely to limit the work load
of the thesis work. By default in STAR-CCM+, a turbulent boundary layers was modeled for.
The simulations will mainly be focus towards stationary simulations, even though transient
simulation will occur when using some of the turbulence models. The results of the simula-
tion will be focused on pressures, pressure drops and mass distributions with the main focus
on the system pressure drop. Additional results such as visual comparisons of the velocities
fields will be conducted.

When using turbulence models solving the flow down to the boundary, it is important that
the boundary layers of the flow are depicted by the inflation layer of the mesh. This is to en-
sure that the high velocity gradients that occur in the normal direction of the boundary are
captured. Evaluating the prism layers was done visually with figures representing the velocity
field, where the focus is on the velocity gradients.

The meshes used in thesis work were limited to only polyhedral elements due to its supe-
rior advantages for the cases in this report, for both the simulation times and mesh sizes,
when solving for the flow. [3]

To make the the work process more efficient, both the Study of turbulence models and Study
of mesh settings were performed separately for the water jacket and an arbitrary selected pipe
found in the system. This made the process faster and the difference in the geometry made
it necessary to study them separately. In this case it is the pipe between the water jacket and
the engine oil cooler (EOC).

18
3.1 Study of turbulence models
3.1.1 Water jacket

The water jacket in Fig. 2 was used as the base geometry. The geometry was modified with
extrusions for the inlet and outlets to get a more developed flow into the water jacket and a
more realistic velocity profile and an attempt of getting rid of reversible flows that potentially
occur at the outlets.

When evaluating the influence of the turbulence models and y + -values, the water jacket was
rebuild to try to have as little interfering influences as possible, reducing the number of pos-
sibilities affecting the flow when comparing the results from the different turbulence models
with each other. For that reason, both the EOC outlet and Climate outlet were closed, having
only the thermostat left as an outlet to the water jacket. When closing two outlets, the fluid
flow inside the water jacket changed character, creating zones with very low flow velocities far
away from the inlet and outlet. To cope with that and meanwhile optimizing the y + -values
around certain values, the geometry had to be divided into different parts. By this approach
the opportunity to control the surface meshes increases. This is represented in Fig. 8. Each
color corresponds to a specific part. Also note the extrusions for the inlet and the thermostat
outlet.

(a) Frontal (inlet) view.

(b) Backside (thermostat outlet) view.

Figure 8: Graphical representation of the different part identification surfaces for the VED4
HP Gen1 water jacket.

19
 All cases were done for water with a mass flow condition at the inlet. Boundary conditions in
these simulations are:

1. Inlet: Mass flow of 2 kg/s.

2. Outlet to thermostat: Mass flow split of 1.

Note that since no boundary condition is used to control the pressure, the magnitude of the
pressures throughout the water jacket are balanced around the initial condition rather than
a real life pressures. This will not impact the pressure or pressure drop of the system, which
is where the main focus us through these simulation.

Four different turbulence models were chosen to be examined, the k-ω SST model and three
k-ε models, the realizable (real.), v2f and elliptic blending (EB). All different turbulence mod-
els with corresponding mesh setup is presented in Table 1, solved with a stationary solver and
mass flow inlet of 2 kg/s, see Table 1. The distribution of the resulting y + values is shown in
Table 2.

Note that some y + naming in the tabular throughout the report has an ending of −R. This
symbolize when a mesh has been remade from a previous mesh setup, for example to change
the number of elements or to improve convergence.

Table 1: Setup for different cases of turbulence models and y + -values.

y+ No of Prism layers
Turbulence Models Wall Treatment average Elements Layers Total Thickness Stretching
+
Realizable & EB High y 30-R 4 258 743 1 1.20-2.40 mm -
Realizable Two-layer − All y + 30-R 4 258 743 1 1.20-2.40 mm -
Realizable Two-layer − All y + 30 1 226 024 1 1.20-2.40 mm -
k-ε
Realizable Two-layer − All y + 2 3 395 970 6-7 0.666 mm 1.0-1.4
Realizable Two-layer − All y + 1-R 15 249 990 10 0.666 mm 1.1-1.4
Realizable Two-layer − All y + 1 4 634 764 10 0.666 mm 1.1-1.4
Realizable Two-layer − All y + 0.5 4 717 889 10 0.666 mm 1.3-1.6
v2f & EB All y + 2 3 395 970 6-7 0.666 mm 1.0-1.4
k-ε v2f & EB All y + 1-R 15 249 990 10 0.666 mm 1.1-1.4
v2f & EB All y + 1 4 634 764 10 0.666 mm 1.1-1.4
v2f & EB All y + 0.5 4 717 889 10 0.666 mm 1.3-1.6
SST All y + 2 3 395 970 6-7 0.666 mm 1.0-1.4
k-ω SST All y + 1-R 15 249 990 10 0.666 mm 1.1-1.4
SST All y + 1 4 634 764 10 0.666 mm 1.1-1.4
SST All y + 0.5 4 717 889 10 0.666 mm 1.3-1.6

20
 Table 2: The distribution of the y + value over the water jacket’s total surface.
Gray-coloured cells equals values > 30%.

y + interval
0-0.5 0.5-1 1-1.5 1.5-2.5 2.5-7.5 7.5-30 30-100 100-Max
y +2 2.99% 11.86% 16.29% 33.20% 35.11% 0.55%
v2f y +1 15.60% 33.89% 26.01% 21.35% 3.15% 0.00%
y + 0.5 65.74% 30.40% 3.53% 0.33% 0.00%
y +2 1.96% 9.26% 17.25% 33.63% 37.17% 0.74%
SST y +1 13.95% 36.38% 24.75% 21.03% 3.88% 0.00%
y + 0.5 67.10% 29.44% 2.99% 0.46% 0.00%
y + 30-R 1.54% 15.80% 53.46% 29.20%
y + 30 1.47% 16.11% 54.58% 27.83%
y +2 3.06% 13.15% 18.77% 30.91% 33.54% 0.58%
Realizable
y +1 18.79% 35.99% 22.37% 19.75% 3.11% 0.00%
y + 0.5 70.16% 26.56% 2.88% 0.40% 0.00%

The turbulence models are also compared to each other graphically for two chosen section
planes. The red-colored plane (further away from the inlet) is referred as ’Plane 1’ while the
blue-colored plane (closer to the inlet) is referred to as ’Plane 2’.

Figure 9: The placement of two sectional planes in the water jacket.

21
The boundary layer evaluation is done for one section plane in the water jacket. The plane
section is placed where higher velocities are expected, in this case closer to to the inlet, to see
if the inflation layers captures the velocity changes. The placements of the section plane can
be seen in Fig. 9.

3.1.2 Pipe

Figure 10: Figure of the pipe with PIDs.

22
 Table 3: The different cases used for turbulence models, y + values and mesh grid sizes for the
mass flow of 1 kg/s.

Turbulence Models y+ No of Elements Solver

Realizable 30 186 977 Stationary
Realizable 5 327 086 Stationary
k-ε
Realizable 2 353 930 Stationary
Realizable 1 382 347 Stationary
v2f & EB 30 186 977 Stationary
v2f & EB 2 353 930 Stationary
v2f & EB 1 382 347 Stationary
v2f & EB 0.5 461 314 Stationary
k-ε
v2f & EB 30 186 977 Transient
v2f & EB 2 353 930 Transient
v2f & EB 1 382 347 Transient
v2f & EB 0.5-R 856 087 Transient
SST 2 353 930 Stationary
SST 1 382 347 Stationary
SST 0.5 461 314 Stationary
k-ω
SST 2 353 930 Transient
SST 1 382 347 Transient
SST 0.5-R 856 087 Transient

Table 4: The distribution of the y + value over the water jacket’s total surface. Gray-colored
cells equals values > 30%.

y + interval
0-0.5 0.5-1 1-1.5 1.5-2.5 2.5-7.5 7.5-30 30-100 100-Max
y + 30 0.39% 4.70% 93.25% 2.05%
y +2 0.34% 1.75% 2.25% 88.74% 6.92% 0.00%
v2f
y +1 2.62% 11.80% 77.24% 8.34% 0.00%
y + 0.5 77.75% 26.79% 0.46% 0.00% 0.00%
y +2 0.39% 2.24% 3.53% 82.33% 11.51% 0.00%
SST y +1 3.28% 15.84% 74.15% 6.72% 0.00%
y + 0.5 72.80% 26.39% 0.79% 0.46% 0.00%
y + 30 0.00% 8.68% 80.72% 10.60%
y +2 0.53% 1.53% 2.85% 85.79% 9.30% 0.00%
Realizable y +1 2.58% 13.61% 75.79% 8.03% 0.00%
y + 0.5 70.16% 26.56% 2.88% 0.40% 0.00%

23
 (a) Location of the section plane. (b) Location of the bend.

Figure 11: Location of the planes regarding the pipe.

3.1.3 Transient simulations

For the transient simulations a Courant number less or equal to one is chosen to assure sta-
bility. This is done by choosing u as approximately the highest velocity in the domain.

u∆t
C= (3.1)
∆x
Where u is velocity, ∆t is the time step and ∆x is the grid size, calculated as

N 1 /3
1
∆x = ( ∑(Vi )) (3.2)
N i =1

where N is the number of cells and Vi is the volume of the it h cell. With 505 837 cells and a
volume of 1.17e-4m3 with a maximum velocity of 2̃0 m/s gives a time step of 3.2e-5 s.

24
3.2 Thermostat
3.2.1 Closed

(a) Thermostat exterior before simplifications. (b) Thermostat exterior after simplifications.

(c) Thermostat interior before simplifications. (d) Thermostat interior after simplifications.

Figure 12: Simplifications of the closed thermostat.

25
When the engine doesn’t need to be cooled the thermostat is closed, all the coolant flows
through the by-pass and circulates through the water jacket until the engine needs to be
cooled. In Fig.12a the pipe to the left leads to the by-pass circuit and the pipe at the top leads
to the radiator. The simplifications at the outlet to the radiator circuit are visible in Fig.12b
and Fig.12d. The detailed geometry at the outlet will lead to difficulties with meshing due
to small cells and are therefor simplified. The main part of the flow will go directly from the
inlet to the by-pass outlet resulting in a low velocity zone at the radiator outlet. The spring
has been shortened at both the top and the bottom in order to avoid small cells which can
lead to difficulties when meshing. Several minor simplifications had been made for the same
reasons.

3.2.2 Open

Figure 13: The opened thermostat after the simplifications.

When the engine need a lot of cooling the thermostat is fully opened, in this case the plate at
the top of the thermostat is lowered 13mm and let the flow pass through the radiator circuit.
At the same time as the thermostat is opened the by-pass is closed. The same simplifications
applied to the closed thermostat is also applied on the open thermostat, except those regard-
ing the plate. When the thermostat is opened the spring is compressed and becomes more
dense. The spring is the model is not compressed but the modeled spring is a poor represen-

26
 tation of the real spring, which has smaller wire diameter and have more turns. Due to these
differences the modeled spring should be sufficient to represent a compressed spring.

3.3 Study of mesh settings

3.3.1 Water jacket

When examine the mesh settings, both the outlet towards the climate circuit as well as the
EOC circuit were open the get a similar flow patten as for a full system simulation. Even
though the split ratios aren’t the same for different mass flows into the water jacket, the dif-
ferences are very small and therefore neglected. As for the inlet, four different mass flows
were evaluated.

1. Inlet: Mass flow of 1, 2, 3 and 4 kg/s.

2. Outlet to climate: Mass flow split of 0.08362.

3. Outlet to EOC: Mass flow split of 0.14946.

4. Outlet to thermostat: Mass flow split of 0.76692.

From the results of the Study of turbulence models, the low Reynold’s model of v2f was chosen
for a y + -value approximate to 0.5.

Table 5: Mesh setups for an average of the y + -value approximate to 0.5.

Prism layers
Setup Type of Mesher No of Elements Layers Total Thickness Stretching
1 Standard 4 561 791 10 0.666 mm 1.4
2 Standard 5 325 312 12 0.666 mm 1.3
3 Standard 6 467 394 15 0.666 mm 1.2
4 Advancing Layer Mesher 4 602 169 10 0.666 mm 1.4

From the results from study of mesh settings, setup 4 was chosen from Table 5. That setup was
also used to create four additional meshes to cope with different mass flow inlet conditions,
this to keep the y + -values closer to 0.5. The setups for these four meshes can be seen in Table
6.

Table 6: Mesh setups for different mass flow inlet conditions for an average y + -value approximate to 0.5.

Prism layers
Mass flow [kg/s] No of Elements y + average Layers Total Thickness Stretching
1 4 212 973 0.508 9 0.666 mm 1.35
2 4 602 169 0.506 10 0.666 mm 1.4
3 4 987 706 0.502 11 0.666 mm 1.4
4 5 372 801 0.502 12 0.666 mm 1.385

27
 3.3.2 Pipe

Table 7: Statistics of mesh for the pipe.

Prism layers
Setup Type of Mesher No of Elements Layers Total Thickness [mm] Stretching
1 Standard 480 295 17 1.95 1.45
2 Standard 505 406 18 1.95 1.50
3 Standard 530 517 19 1.95 1.40
4 Standard 595 544 20 1.95 1.35
6 Advancing Layer Mesher 505 837 18 1.95 1.50

For all the cases a target size of 2.25 mm has been used, except for case 4 where the target
sized was decreased to 1.5 mm.

3.3.3 Thermostat

Table 8: Statistics of mesh for the closed thermostat.

Prism layers
Setup Type of Mesher No of Elements Layers Total Thickness [mm] Stretching
1 Standard 1 634 072 11 1 1.53
2 Standard 1 742 967 12 1 1.45
3 Standard 1 852 413 13 1 1.41
4 Standard 1 958 876 14 1 1.35
5 Standard 2 067 691 15 1 1.32
6 Standard 2 176 241 16 1 1.29

Table 9: Statistics of mesh for the open thermostat.

Prism layers
Setup Type of Mesher No of Elements Layers Total Thickness [mm] Stretching
1 Standard 2 220 667 11 1 1.44
2 Standard 2 369 568 12 1 1.40
3 Standard 2 511 819 13 1 1.33
4 Standard 2 659 824 14 1 1.30
5 Standard 2 805 735 15 1 1.27
6 Standard 2 950 386 16 1 1.24
7 Advancing Layer Mesher 2 184 380 11 1 1.44

28
 3.4 Full system
The full system simulations were performed for two cases on the original system shown in
Fig. 1. One case with an open thermostat and a second for a closed thermostat (open towards
to by-pass circuit). The simulations were fitted against two experimental cases performed at
the test rig. These two cases are represented in Table 10.

Table 10: Component setups for the two full system cases
with the main difference as an open or closed thermostat.

Climate valve Vacuum

Case Thermostat WH-rad FOH EGR-pump S/S-pump
towards valves
1 Open Without Without Off Off HVAC Open
2 Closed With Without On Off HVAC Open

The CAD geometry obtained for CFD simulation didn’t match the system sketch in Fig. 1.
Instead, the system sketch of the CAD geometry looks like Fig. 14. The test rig experiment was
performed before the cooling system model’s final version, therefor the difference between
test rig geometry and CAD geometry.

Figure 14: The system sketch illustrating the CAD geometry given by Volvo.

The differences in the geometry is the wheel-house radiator (WH-rad) has been removed, the
climate circuit is attached closer to the water jacket inlet rather than the EOC circuit and the
TOC and EGR circuits have switch places when attaching towards the pipe out of the front
radiator.

29
To be able to compare the results of the simulation with the test rig results, the geometries
has to have a better agreement to each other then they does in Fig. 1 and Fig. 14. There-
fore the CAD geometry were rebuild to better imitate the test rig system sketch. The two
chosen cases in Table 10 were both without the wheel-house radiator (WH-rad). For Case 1
the wheel-house radiator circuit was disabled and for Case 2 the wheel-house radiator circuit
was not included in the geometry due to the closed thermostat. The climate circuit attach
point were changed to the EOC circuit for both Case 1 and 2. Additionally for Case 1, the TOC
and EGR circuits were switched in the CAD geometry compared to the test rig geometry but
a decision to keep the original CAD geometry were done to decreases the preparation time
for the simulations. After evaluating the simulated results from that geometry, it was decided
to copy the layout of the test rig for the TOC-EGR junction as well for the CAD geometry, to
investigate if the deviation between the results could be decreased. This was not needed for
Case 2 due to these parts were excluded from the simulation domain for a closed thermostat.

The system sketches used for Case 1 and Case 2 can be seen in Fig. 15 and Fig. 16, respectively.

Figure 15: The system sketch illustrating the CAD geometry used in Case 1.

30
 Figure 16: The system sketch illustrating the CAD geometry used in Case 2.

When modeling the full system, none of the heat exchangers are included in the geometrical
model to reduce the simulation times and the work process. These are instead replaced with
porous baffle interfaces. In order to get the interfaces to work in STAR-CCM+, the normals of
the interfaces needs to be parallel and pointing at each other. This were not the case with the
CAD models which lead to the modeling of pipe bends to satisfy the need for parallel normal
vectors.

31
3.5 Porous Baffle Interface
The constants α and β are calculated from the second degree polynomial describing the pres-
sure drop divided by the density as a function of superficial velocity in Eq. (2.45). The velocity
is in this case calculated as v n = Q / A, where Q is the volumetric flow rate through the compo-
nent and A the cross section area of the pipe leading in to the component. The pressure drop
and the volumetric flow used to calculate the constants are obtained from 1D simulations,
except the data for the climate which comes from experimental data.

Table 11: Constants for the porous baffle interface used in the full system simulations.

Component Data values Thermostat α β

Climate Experiment Open 2.880 8.207
Climate Experiment Closed 4.276 3.022
EGR 1D Open 3.335 0.607
EGR pump 1D Open 4.037 -0.791
EOC 1D Open and Closed 2.780 1.850
Radiator 1D Open 3.288 1.024
TOC 1D Open 3.011 0.466
TSC 1D Open 3.464 1.694

For the case of a closed thermostat the flow distribution in the water jacket will be changed
and therefor a new set of constants for Climate is needed to describe the pressure drop over
the component of the climate circuit.

32
4 Results and discussion

4.1 Turbulence model results

The results of the study of turbulence models for the water jacket are presented in Section
4.1.1 and the pipe system in Section 4.1.2 at page 40. A verifications of the total prism layers
thickness are presented in Sec. 4.1.3 at page 42 to ensure that the turbulence models have the
condition of solving the flows close to boundaries.

Note that since no boundary condition is used to control the pressure, the magnitude of the
pressures throughout the water jacket are balanced around the initial condition rather than
a real life pressures. This will not influence the pressure or pressure drop of the system, being
the main focus of these simulation.

4.1.1 Water jacket

The results of the water jacket are presented in tables and figures of pressure, pressure drops
and mass flows at different positions. The pressure drops, as the main objective, are pre-
sented in as a graphical presentation in Fig. 17.

Figure 17: Pressure drop as a function of y + .

As can be seen in Fig. 17, the biggest differences occur between low and high Reynold’s mod-
els. For the low Reynold’s models, the turbulence models appear to separate themselves into
three groups, realizable, EB-SST and v2f. When decreasing the y + -value, approaching ∼ 0.5,

33
 the differences in pressure drop between EB, SST and v2f is no more than 182Pa. In Table 12,
the mean values and standard deviations (SD) from Fig. 17 are presented.

Table 12: Mean value and standard deviation (SD) of the pressure drop, ∇: sorted by descending.

Turbulence Models y + average No of Elements Mean [Pa] ∇ SD [Pa]

Realizable 2 3 395 970 16 921 3.0
Realizable 1 4 634 764 16 874 2.9
Realizable 0.5 4 717 889 16 803 1.1
SST 2 3 395 970 16 626 268.3
SST 1 4 634 764 16 592 208.4
EB 1 4 634 764 16 543 119.8
EB 0.5 4 717 889 16 461 96.0
SST 0.5 4 717 889 16 408 289.1
EB 2 3 395 970 16 375 84.9
v2f 0.5 4 717 889 16 279 5.8
v2f 1 4 634 764 16 196 31.0
v2f 2 3 395 970 16 067 42.7
High Y+ EB 30 1 226 024 14 422 128.6
Realizable 30 1 226 024 13 997 66.0

Looking at the standard deviation, the realizable and the lowest y + -value for the v2f models
proves to be very stable while EB and specifically the SST models show a more unstable be-
haviour. When evaluating the importance of the number of elements in the simulation, see
Table 13, it can be seen that increasing the number of elements slightly change the pressure
drop. That means that the number of elements is high enough for all cases.

34
 Table 13: Mean value and standard deviation of the pressure drop for original and remeshed
models along with differences between them.

Turb. Mod. y + average No of Elements Mean [Pa] SD [Pa] Difference

Realizable 30-R 4 258 743 14 033 83.1 +36 Pa | +0.26%
Realizable 30 1 226 024 13 997 66.0 -
Realizable 1-R 15 249 990 16 680 6.0 −194 Pa | −1.15%
Realizable 1 4 634 764 16 874 2.9 -
High Y+ EB 30-R 4 258 743 13 949 178.2 −473 Pa | −3.28%
High Y+ EB 30 1 226 024 14 422 128.6 -
EB 1-R 15 249 990 16 416 102.6 −127 Pa | −0.77%
EB 1 4 634 764 16 543 119.8 -
v2f 1-R 15 249 990 16 464 41.3 +268 Pa | +1.65%
v2f 1 4 634 764 16 196 31.0 -
SST 1-R 15 249 990 16 927 305.5 +335 Pa | +2.02%
SST 1 4 634 764 16 592 208.4 -

Additional information were collected to evaluate different mass flow distributions and pres-
sure throughout the water jacket. The pressure at the inlet and outlet is shown in Table 14.
Here, the pressure is not taken at the boundary conditions for the extrusions but in close con-
tact with the original water jacket geometry. This is done with the function Arbitrary section
(AS) in STAR-CCM+.

Table 14: Mean value and standard deviation of the total pressure and mass flow at the inlet
and outlet of the water jacket, excluding the pressure drop through the extrusions.

Inlet (AS) Outlet (AS)

Turbulence Models y + average No of Elements Mean [Pa] SD [Pa] Mean [Pa] SD [Pa]
Realizable 2 3 395 970 111 392 3.0 94 471 3.6
Realizable 1 4 634 764 111 388 0.2 94 514 2.9
Realizable 0.5 4 717 889 111 306 0.4 94 503 1.2
SST 2 3 395 970 111 409 249.2 94 783 178.8
SST 1 4 634 764 111 514 249.5 94 922 100.9
SST 0.5 4 717 889 111 419 178.3 95 011 204.8
EB 2 3 395 970 111 170 121.4 94 795 41.2
EB 1 4 634 764 111 334 128.9 94 791 11.0
EB 0.5 4 717 889 111 209 126.3 94 748 32.1
v2f 2 3 395 970 111 107 48.6 95 040 22.5
v2f 1 4 634 764 111 227 37.6 95 032 13.7
v2f 0.5 4 717 889 111 289 2.2 95 010 5.6

35
 Figure 18: Mass flow as a function of y + .

Table 15: Total mass flow through all drilling holes, ∇: sorted by descending.

Pressure [Pa] Mass Flow [g/s]

Turbulence Models +
y average No of Elements Mean SD Mean ∇ SD
SST 0.5 4 717 889 100 606 301.7 52.03 1.316e-1
SST 2 3 395 970 100 826 121.8 51.82 5.671e-1
v2f 2 3 395 970 100 550 73.8 51.59 1.992e-1
High Y+ EB 30-R 4 258 743 100 993 78.6 47.59 5.856e-1
EB 2 3 395 970 100 430 64.7 50.75 2.360e-4
Realizable 30 1 226 024 101 137 34.9 50.73 3.391e-2
High Y+ EB 30 1 226 024 101 102 51.3 49.77 5.914e-2
SST 1 4 634 764 100 748 204.7 50.55 1.099
EB 1 4 634 764 100 468 110.5 51.42 4.388e-4
v2f 1 4 634 764 100 410 22.3 50.39 7.488e-2
Realizable 30-R 4 258 743 101 161 53.0 50.07 1.110e-1
v2f 0.5 4 717 889 100 500 5.6 49.89 4.554e-3
Realizable 1 4 634 764 100 669 0.4 49.64 7.384e-4
EB 0.5 4 717 889 100 686 45.3 49.06 2.993e-4
Realizable 2 3 395 970 100 605 1.2 48.96 1.533e-3
Realizable 0.5 4 717 889 100 404 0.7 47.97 7.115e-4

The pressure and mass flow at the drilling holes of the water jacket is presented is Fig. 18
and in Table 15 above. As seen in the figure, the difference between different y + -values has

36
 not the same behaviour as for Fig. 17. The mass flow does not seem to converge towards a
specific value when decreasing the y + -value.

For one of the water jacket’s long sides, the side of the inlet, four distinctive passages be-
tween the cylinder block and cylinder head exists. The passages is named in a numbering
order, with the passage closed to the water jacket inlet as 1 and with increasing number fur-
ther away from the inlet.

Table 16: Mean value and standard deviation of the total pressure and mass flow
for the passages between the cylinder block and cylinder head.

Passage 1 Passage 2
Total Pressure [Pa] Mass Flow [g/s] Total Pressure [Pa] Mass Flow [g/s]
Turb. Mod. y + average Mean SD Mean SD Mean SD Mean SD
SST 2 107 515 195.6 472.6 5.083 105 912 644.1 427.0 3.893
SST 1 107 795 236.1 471.9 4.019 106 288 636.5 427.8 5.670
SST 0.5 107 679 150.1 474.1 5.569 106 062 369.9 420.0 5.578
EB 2 107 043 85.9 449.1 1.085 105 469 162.0 443.7 7.944e-01
EB 1 107 174 114.0 455.0 1.727 106 069 319.6 447.6 1.994
EB 0.5 107 091 67.1 455.3 9.901e-01 105 589 186.2 451.0 2.392
v2f 2 106 928 46.9 456.6 1.127 105 705 135.5 440.5 2.992
v2f 1 107 001 36.7 462.1 6.564e-01 105 749 62.1 437.2 3.710e-01
v2f 0.5 107 080 2.2 461.5 1.627e-01 105 798 2.5 449.9 1.471e-01
Realizable 2 106 875 3.0 443.4 2.814e-03 104 984 2.7 448.5 7.092e-03
Realizable 1 106 910 0.2 446.9 5.180e-03 105 243 0.2 450.3 1.043e-02
Realizable 0.5 106 867 0.4 445.9 1.482e-03 105 006 0.3 463.8 2.917e-03

Passage 3 Passage 4
Total Pressure [Pa] Mass Flow [g/s] Total Pressure [Pa] Mass Flow [g/s]
Turb. Mod. y + average Mean SD Mean SD Mean SD Mean SD
SST 2 103 358 121.1 456.6 4.884 103 564 100.4 424.7 3.610
SST 1 103 519 119.9 466.2 3.887 103 668 105.5 418.1 1.072
SST 0.5 103 602 206.7 465.4 3.639 103 551 66.3 433.9 3.037
EB 2 104 378 306.3 462.6 1.883 103 686 23.8 429.2 1.221
EB 1 104 048 109.6 460.4 2.401 103 596 39.6 432.5 1.362
EB 0.5 104 446 208.1 466.1 1.759 103 608 29.5 432.0 9.733e-01
v2f 2 103 679 57.3 461.3 7.999e-01 103 652 32.9 420.7 3.033
v2f 1 103 737 15.0 464.6 4.773e-01 103 679 5.0 425.2 7.585e-01
v2f 0.5 103 698 1.0 465.5 8.939e-02 103 653 0.6 422.6 1.999e-01
Realizable 2 104 793 2.3 458.3 1.334e-02 103 515 1.3 433.7 1.383e-02
Realizable 1 104 380 0.2 460.1 6.193e-03 103 587 0.2 436.6 2.278e-02
Realizable 0.5 104 818 0.2 453.5 2.661e-03 103 491 0.2 438.3 1.315e-03

37
From Tables 15 and 16 it can be seen the there is no distinct correlation between the total
mass flow and pressure, but rather the chose of turbulence model seem to have the largest
impact. Theoretical, the pressure drop through the a system should be proportional towards
the mass flow, meaning the turbulence models has altered that relation in terms how the
solve the flows.

(a) SST. (b) EB.

(c) v2f. (d) Realizable.

Figure 19: Section planes of the velocity at ’Plane 1’.

38
 (a) SST. (b) EB.

(c) v2f. (d) Realizable.

Figure 20: Section plane of the velocity at ’Plane 2’.

Also a comparison of the velocity field is presented in Fig. 19 and Fig. 20, see figure 9 for the
placement of the section planes. Section plane 2 show a very similar velocity field between
the different turbulence models but moving slightly away from the inlet (section plane 1),
the turbulence models shows a drastic difference in the velocities. It can be seen that the

39
k-ω SST model clearly calculates higher velocities at the right side than neither of the three
k-ε models, even due the EB model separates itself from the other k-ε models with higher
velocities.

4.1.2 Pipe section

A figure of the pipe can be seen in Fig. 10, where the inlet is at the bottom of the figure and
the outlet at the top. The location of the bend can be seen in Fig. 11b.

Table 17: Pressure drop for a mass flow of 1 kg/s.

In/Out [kPa] Bend [kPa]

Turb. Mod. +
y average Mean ∇ SD Mean SD
SST 0.5 137.3 8.6 57.4 8.4
SST 2 133.9 7.4 56.7 7.2
EB 2 133.8 5.4 53.1 3.7
EB 0.5 133.5 4.0 51.3 3.3
EB 30 132.0 2.8 52.0 2.5
EB 1 131.3 3.6 50.5 3.0
v2f 0.5 128.9 2.1 49.5 1.8
SST 1 127.7 9.8 53.6 9.1
v2f 1 127.1 1.9 48.4 1.7
v2f 2 126.6 3.3 50.0 2.7
Realizable 2 126.0 - 49.4 -
Realizable 1 123.9 - 48.3 -
v2f 30 119.4 3.0 95.3 2.9
Realizable 30 118.0 - 48.7 -

From Table 17 and Fig. 26 it is seen that the realizable model predicts the lowest pressure
drop, which is the least complex model of the four. The high Reynolds model for the v2f
model is significantly lower than the low Reynolds model for the same turbulence model
which doesn’t make high Reynolds models useful for modeling pipes in this case. The SST
model predicts the highest pressure drop, except for y + ∼ 1 which is surprisingly low.

40
 Figure 21: Pressure drop for different turbulence models.

(a) SST turbulence model. (b) v2f turbulence model.

Figure 22: Comparison between steady state and transient simulations.

For all the turbulence models except k-ε realizable an oscillating behaviour were observed.
Transient simulations were carried out to investigate if the oscillating behaviour were a result

41
of a transient behaviour. The result can be seen in Fig. 22. The result for the transient sim-
ulations is similar to the steady simulations, the difference is the magnitude of the pressure
drop. The pressure drop for the transient simulations are much higher. The two different
turbulence models compared in Fig. 22 shows no real difference in the pressure drop.

(a) SST. (b) EB.

(c) v2f. (d) Realizable.

Figure 23: Cross section of the pipe showing the prism layers.

The location of the cross section of the pipe in Fig. 25b can be found in Fig. 11a. The SST
model predicts the highest velocity, followed by v2f, EB and realizable.

4.1.3 Boundary layer thickness

A qualitative evaluation of the boundary layer thickness for the water jacket gives, as seen in
all eight figures, Fig. 19-20, the velocity change between the innermost inflation layer and the
its neighbouring core element is small, meaning that the inflation layers is thick enough to
capture the velocity gradients in the boundary layer. If the velocity would have taken a much
higher step between the inflation layer and core element, it would mean that the boundary

42
 layer is thicker than the total height of the inflation layer which would result in a wrong de-
piction of the velocity profile at the boundary. As a result, the total thickness of the inflation
layers is appropriate to use in a mesh model.

A closer look at the inflation layers is presented in Fig. 24. As seen, the boundary layer
stretches until the ninth inflation layer (second last), giving some security with the last in-
flation layer. As should be noted, the height of the boundary layer will vary all over the water
jacket, but the largest velocity gradients can be predicted to exist closer to the inlet. The high-
est velocity in the water jacket is 5.70 m/s, so the difference to Fig. 24 is believed to be within
the error margin of the last (tenth) inflation layer. In Fig. 24c, a threshold mesh is displayed,
which means that the elements stay intact for the section plane, instead of cut through as in
Fig. 24b.

(a) Area of interest. (b) Section plane mesh. (c) Threshold mesh.

Figure 24: BL thickness for SST.

The boundary layer thickness of the pipe can be seen in Fig. 25.

43
 (a) Boundary layer thickness. (b) Close up figure of the boundary layer.

Figure 25: Cross section of the pipe after the bend, along the stream wise direction.

In Fig. 25a it can be seen that the transition between the prism layer cells at the ends and the
cells in the bulk region is smooth.

4.2 Study of mesh settings

4.2.1 Water jacket

The results of the meshes in Table 5, in terms of the y + distribution, the criteria of mesh
quality in Sec. 2.8 and residual convergence, can be seen in Tables 18, 19 and 20, respectively.

Table 18: Distribution of the y + -values and the pressure drop between the inlet and the outlet
towards the thermostat.

y + interval
Setup No of Elements +
y average 0-0.25 0.25-0.75 0.75-Max Pressure drop [Pa]
1 4 561 791 0.496 23.64% 56.12% 20.24% 14 549
2 5 325 312 0.469 25.88% 57.09% 17.03% 14 517
3 6 467 394 0.484 24.55% 56.75% 18.70% 14 674
4 4 602 169 0.506 22.99% 55.63% 21.37% 14 518

Table 19: Number of elements with bad mesh quality.

Cell Aspect Ratio Cell Quality Face Validity Skewness Angle Volume Change
Setup < 0.01 < 0.01 < 0.9 > 85 < 0.01
1 0 23 0 85 0
2 0 33 0 111 0
3 0 43 0 85 0
4 0 0 0 8 0

44
 Table 20: Residual convergence.

Setup Continuity X-mom. Y-mom. Z-mom. Tke Tdr V2 F

1 2.07e-7 1.84e-7 2.14e-7 1.74e-7 3.22e-8 6.07e-6 1.47e-8 9.80e-4
2 3.93e-7 5.65e-7 5.30e-7 5.11e-7 1.06e-7 4.61e-5 5.77e-8 4.74e-3
3 4.63e-7 3.88e-7 5.18e-7 5.36e-7 5.78e-8 3.85e-5 2.66e-8 2.30e-3
4 5.73e-7 5.80e-7 5.96e-7 5.44e-7 1.56e-7 5.13e-5 9.68e-8 4.37e-3

Comparing setup 1 through 3 gives that increasing the fines of the prism layer does not im-
prove the pressure drop, see Table 18, and the residual convergence as well as the mesh qual-
ity is very similar. The largest difference comes between setup 4 and the first three, where the
Advancing Layer Mesher improves significant the mesh quality and also improve the residual
convergence. What also should be taken into account is that the number of elements has
increased for setup 2 and 3, compared to 1 and 4, without improving the simulations.

Table 21: Distribution of the y + -values from Table 6

y + interval
+
Mass flow [kg/s] No Elements y average 0-0.25 0.25-0.75 0.75-1 1-Max
1 4 212 973 0.508 22.50% 56.06% 13.81% 7.62%
2 4 602 169 0.506 22.99% 55.63% 14.17% 7.20%
3 4 987 706 0.502 23.23% 55.65% 14.08% 7.03%
4 5 372 801 0.502 23.25% 55.61% 14.22% 6.92%

In Table 21 the distribution of y + -value for Table 6 and the resulting pressure drops between
the water jacket inlet and each of the three outlets in Table 22.

Table 22: Pressure drops between the inlet at respectively outlet over the water jacket. This is
done for the four mass flows in Table 6.

Pressure drop in total pressure [Pa] Pressure drop in working pressure [Pa]
Mass flow [kg/s] Inlet-Thermostat Inlet-Thermostat Inlet-EOC Inlet-Climate
1 4 085 3 038 2 734 1 952
2 14 518 10 729 10 419 6 974
3 30 469 22 189 23 271 14 265
4 52 151 37 877 41 362 23 963

45
 4.2.2 Pipe

Table 23: Statistics of mesh for the pipe.

y + interval
+
Setup No of Elements y average 0-0.25 0.25-0.75 0.75-Max
1 480 295 0.58 0.78% 90.97% 8.26%
2 505 406 0.56 1.60% 91.39% 6.42%
3 530 517 0.53 1.97% 91.61% 6.42%
4 595 544 0.50 2.34% 91.53% 6.13%
5 505 837 0.57 1.05% 91.81% 7.14%

Table 24: Mesh quality for the pipe.

Cell Aspect Ratio Cell Quality Face Validity Skewness Angle Volume Change Pressure drop [Pa]
Setup < 0.01 < 0.01 < 0.9 > 85 < 0.01 [Pa]
1 72 391 0 0 173 0 128 407
2 43 123 0 0 184 1 131 976
3 65 006 0 0 249 3 128 716
4 62 285 0 0 382 1 133 244
2 ALM 44 394 0 0 97 0 135 285

From Table 24 it can be seen that setup 2 has the lowest number of cells with low quality.
This setup is rerun with ALM to see if the mesh quality can be improved further. Comparing
setup 2, with and without ALM, it can been seen that the number of cells with high aspect
ratio has increased while the number of cells with high skewness angle and volume change
has decreased. The ALM works differently from the regular prism layer mesher, in e.g. narrow
gaps the prism layer mesher can reduce the prism layer thickness and remove layers in order
to have some normal cells between the prism layers. The ALM instead removes prism layers
without changing the first layer thickness which doesn’t change the y + value.

46
 4.2.3 Thermostat

Table 25: Statistics of mesh for the closed thermostat.

y + interval
+
Setup No of Elements y average 0-0.25 0.25-0.75 0.75-Max
1 1 634 072 0.50 60.56% 32.90% 6.54%
2 1 742 967 0.53 55.41% 37.72% 6.86%
3 1 852 413 0.48 60.97% 32.84% 6.19%
4 1 958 876 0.53 54.81% 38.10% 7.09%
5 2 067 691 0.51 57.55% 35.78% 6.66%
6 2 176 241 0.50 59.27% 34.11% 6.62%

Table 26: Mesh quality for the closed thermostat.

Cell Aspect Ratio Cell Quality Face Validity Skewness Angle Volume Change Pressure drop [Pa]
Setup < 0.01 < 0.01 < 0.9 > 85 < 0.01 [Pa]
1 2 16 3 475 57 128 165
2 1 35 9 408 35 127 876
3 3 70 9 529 47 131 714
4 1 84 9 416 51 133 349
5 1 116 11 471 49 131 918
6 2 113 16 499 48 131 177

In the case of a closed thermostat the inlet condition was a mass flow of 1.312 kg/s. In the
other cases there have been an improvement in the mesh quality when ALM have been used.
In this case with the closed thermostat it was not possible to use the ALM, it resulted in errors
in the meshing. If there had been more time this could possibly have been solved with a
extrusion at the inlet.

Table 27: Statistics of mesh for the open thermostat.

y + interval
Setup No of Elements y + average 0-0.25 0.25-0.75 0.75-Max
1 2 220 667 0.51 34.47% 52.75% 12.78%
2 2 369 568 0.45 40.92% 50.01% 9.07%
3 2 511 819 0.53 32.86% 53.14% 14.00%
4 2 659 824 0.49 36.46% 52.10% 11.44%
5 2 805 735 0.48 36.69% 52.24% 11.07%
6 2 950 386 0.50 35.45% 52.65% 11.90%
7 2 184 380 0.57 12.56% 69.95% 17.49%

47
 Table 28: Mesh quality for the open thermostat.

Cell Aspect Ratio Cell Quality Face Validity Skewness Angle Volume Change Pressure drop
Setup < 0.01 < 0.01 < 0.9 > 85 < 0.01 [Pa]
1 0 56 18 179 145 6 787
2 0 72 12 270 158 6 641
3 0 81 22 223 144 7 033
4 0 157 29 305 173 6 729
5 0 176 36 351 165 6 599
6 0 224 39 353 164 6 662
1 ALM 0 4 0 12 126 6 736

In the case of a open thermostat the inlet condition was a mass flow of 1.312 kg/s. From
Table 28 it can be seen that the pressure drop does not vary considerably when changing
the mesh. Setup 1 has lowest number of cells with low quality and were therefor rerun with
the ALM to see if the quality could be improved. The number of cells with low quality are
reduced, meaning a lot of the bad cells were within the inflation layers. When using the ALM
the average y + is increased from 0.51 to 0.57, seen in Table 27, but the the number of cells in
the middle y + interval as increased from 52.75 to 69.95%.

48
 4.3 Full system simulation
4.3.1 Case 1/Open thermostat

Table 29: Pressure magnitude in working pressure (including the reference pressure).

ṁ = 0.975 kg/s Inlet Outlet Thermostat out Climate EOC Thermostat in

Simulated 118 163 106 042 113 043 116 222 116 155 115 485
Experimental 119 439 106 042 109 602 116 493 115 807 -

ṁ = 2.08 kg/s Inlet Outlet Thermostat out Climate EOC Thermostat in

Simulated 156 719 107 417 134 393 147 132 146 422 145 281
Experimental 164 850 107 417 132 224 152 742 154 992 -

ṁ = 3.022 kg/s Inlet Outlet Thermostat out Climate EOC Thermostat in

Simulated 210 319 110 383 164 806 189 827 188 123 186 754
Experimental 227 146 110 383 163 317 201 438 206 506 -

Table 30: Pressure magnitude in working pressure (including the reference pressure).

ṁ = 0.975 kg/s p3 p4 p6 p7 p8 p9 p14 p16

Simulated 112 614 108 901 110 759 110 813 109 127 109 138 109 744 108 668
Experimental 114 670 109 988 111 722 114 208 108 591 110 813 109 422 106 240
Difference -1.8% -1.0% -0.9% -3.0% +0.5% -1.5% +0.3% +2.3%

ṁ = 2.08 kg/s p3 p4 p6 p7 p8 p9 p14 p16

Simulated 133 134 120 471 125 283 126 904 119 965 117 671 125 754 117 905
Experimental 136 921 125 428 130 833 130 584 120 422 120 200 132 258 117 248
Difference -2.8% -4.0% -4.2% -2.8% -0.4% -2.1% -4.9% +0.6

ṁ = 3.022 kg/s p3 p4 p6 p7 p8 p9 p14 p16

Simulated 162 070 138 351 145 967 149 771 135 995 136 428 150 713 131 550
Experimental 168 080 146 073 156 549 155 941 136 563 141 965 163 092 132 420
Difference -3.6% -5.3% -6.8% -4.0% -0.4% -3.9% -7.6% 0.7%

49
 Table 31: Mass flows through all relevant flow sensors from Fig. 14.

ṁ = 0.975 kg/s f1 f2 f5 f6 f7 Radiator Thermostat

Simulated 0.662 0.127 0.0937 0.114 0.0714 0.568 0.776
Experimental or 1D 0.618 0.148 0.0154 0.126 0.0836 0.602 0.744
Difference +7.1% -14.2% +508.4% -9.5% -14.6% -5.6% +4.3%

ṁ = 2.08 kg/s f1 f2 f5 f6 f7 Radiator Thermostat

Simulated 1.401 0.268 0.209 0.228 0.183 1.192 1.629
Experimental or 1D 1.333 0.313 0.206 0.254 0.181 1.126 1.587
Difference +5.1% -14.4% +1.5% -10.2% +1.1% +5.9% +2.6%

ṁ = 3.022 kg/s f1 f2 f5 f6 f7 Radiator Thermostat

Simulated 2.072 0.386 0.304 0.324 0.282 1.727 2.354
Experimental or 1D 1.927 0.455 0.305 0.364 0.276 1.622 2.291
Difference +7.5% -15.2% -0.3% -11.0% +2.2% +6.5% +2.7%

Table 32: Pressure drops over the porous baffle interfaces, see Table 11.

ṁ = 0.975 kg/s Climate EOC TOC Radiator TSC EGR EGR pump
Simulated 6 478 5 875 1 129 3 656 1 914 1 846 1 387
Experimental or 1D 7 755 6 982 -17 4 093 475 2 122 1 380
Difference -16.5% -15.9% − -10.7% +302.9% -13.0% +0.5%

ṁ = 2.08 kg/s Climate EOC TOC Radiator TSC EGR EGR pump
Simulated 21 379 21 227 4 876 13 934 7 011 6 534 5 875
Experimental or 1D 20 908 26 163 4 541 12 476 6 967 7 743 7 082
Difference +2.3% -18.9% +7.4% +11.7% +0.6% -15.6% -17.0%

ṁ = 3.022 kg/s Climate EOC TOC Radiator TSC EGR EGR pump
Simulated 39 115 41 296 9 932 27 938 13 475 12 641 12 404
Experimental or 1D 37 981 52 120 9 669 24 618 13 512 15 361 15 358
Difference +3.0% -19.2% +2.7% +13.5% -0.3% -17.7% -19.2%

50
Figure 26: Mass flow distribution for between the climate circuit and the outlet, for the open
and closed thermostat.

Figure 27: Comparison between earlier simulations, experimental data and simulations of
the water jacket in the case of an open thermostat for 4 kg/s.

51
Table 33: Mesh statistics for previous work and current simulations for the water jacket for a
mass flow of 4 kg/s.

y + interval
+
Case y average 0-2 2-7 7-30 30-max
Previous work 11.6 8.7% 37.3% 52.9% 1.1%
Simulations 0.50 99.98% 0.02% - -

The difference between the previous work at Volvo and the simulations done in this report for
the water jacket can partly be explained by the difference in turbulence models. In the ear-
lier simulations the k-ε realizable model was used while the v2f model has been used in this
report. Recalling the result from the turbulence study, Fig. 17 it is seen that the k-ε realizable
model gives a larger pressure drop than the v2f model. Another part of the explanation is their
mesh; the majority of the cells has a y+ value in the range 7-30, which should be avoided. It
should be mentioned that experimental data for the thermostat is measured inside the ther-
mostat while in the simulations the data is measured before the thermostat, one should be
careful comparing the simulated results with the experimental data for the thermostat.

Comparing the mass flows from the original model with the model with the switched junc-
tion, Tables 31 with Table 34 it can be seen that the largest difference between the experimen-
tal and simulated flow is through f6, EGR flow. The percental difference has approximately
been halved. The second largest difference is for f5, TOC flow. The difference for the other
flows has been reduced, except for f2 which approximately the same after the change.

(a) The mass flow through f5 as a function of the (b) The mass flow through f7 as a function of the
inlet mass flow. inlet mass flow.

Figure 28: Comparison between experimental and simulated data for the mass flow through
TOC and climate circuit respectively.

In Table 31 the difference for f5 stands out with 553% for a mass flow ṁ = 0.975 kg/s, compar-

52
 ing this percentage for the higher mass flows ṁ = 2.08 kg/s and ṁ = 3.022 kg/s it can be seen
that the differences has decreased to approximately 8% which is more plausible. An explana-
tion for the large difference through f5 for the lowest mass flow can be found in Fig. 28a. The
first experimental values deviates from the expected linear behaviour, which can be seen for
the rest of the values. This is probably due to some errors in the measurement of the exper-
imental data. The same trend can be seen for f7, even if it is not as distinct. The difference
in the mass flow, in Table 31, through f7 = −18% for the lowest mass flow and for the higher
mass flows the percental difference has decreased to 2% − 3%. From Fig. 28b, no obvious
errors with the experimental data can be seen.

4.3.1.1 Case 1/Open thermostat, with major geometry difference between CAD and test
rig experiment (EGR-TOC junction)

Table 34: Mass flows through all relevant flow sensors from Fig. 14.

ṁ = 0.975 kg/s f1 f2 f5 f6 f7 Radiator Thermostat

Simulated 0.708 0.128 0.1006 0.0988 0.0684 0.576 0.776
Experimental or 1D 0.618 0.148 0.0154 0.1261 0.0836 0.602 0.744
Difference +14.6% -13.5% +553.2% -20.9% -18.2% -4.3% +4.3%

ṁ = 2.08 kg/s f1 f2 f5 f6 f7 Radiator Thermostat

Simulated 1.419 0.272 0.222 0.204 0.185 1.198 1.623
Experimental or 1D 1.333 0.313 0.206 0.254 0.181 1.126 1.587
Difference +6.5% -13.1% +7.8% -19.7% +2.2% +6.4% +2.3%

ṁ = 3.022 kg/s f1 f2 f5 f6 f7 Radiator Thermostat

Simulated 2.063 0.390 0.330 0.285 0.284 1.734 2.348
Experimental or 1D 1.927 0.455 0.305 0.364 0.276 1.622 2.291
Difference +7.1% -14.3% +8.2% -21.7% +2.9% +6.9% +2.5%

53
 Table 35: Pressure drops over the porous baffle interfaces, see Table 11.

ṁ = 0.975 kg/s Climate EOC TOC Radiator TSC EGR EGR pump
Simulated 6 462 5 897 1 275 3 733 2 131 1 422 1 310
Experimental or 1D 7 755 6 982 -17 4 093 475 2 122 1 380
Difference -16.7% -15.5% +740% -8.8% +348.6% -33.0% -5.1%

ṁ = 2.08 kg/s Climate EOC TOC Radiator TSC EGR EGR pump
Simulated 21 762 21 669 5 514 14 040 7 814 5 290 4 587
Experimental or 1D 20 908 26 163 4 541 12 476 6 967 7 743 7 082
Difference +% -% +% +% +% -% -%

ṁ = 3.022 kg/s Climate EOC TOC Radiator TSC EGR EGR pump
Simulated 40 178 41 797 11 634 28 166 15 576 9 957 9 493
Experimental or 1D 37 981 52 120 9 669 24 618 13 512 15 361 15 358
Difference +% -% +% +% +% -% -%

Comparing the mass flow distribution between Section 4.3.1 and 4.3.1.1, it can be seen that
the flow distribution improved for the majority of the flow sensors, which directly improves
the pressure drops over the porous baffle interfaces. Performing the CAD rebuild proved to
be necessary for gaining less deviating results from the experimental test rig data.

4.3.2 Case 2, closed thermostat

Table 36: Pressure magnitude in working pressure (including the reference pressure).

ṁ = 0.538 kg/s Inlet Outlet Climate EOC

Simulated 118 294 106 420 117 119 116 853
Experimental 118 169 106 420 119 462 116 953

ṁ = 1.011 kg/s Inlet Outlet Climate EOC

Simulated 145 781 107 064 141 572 140 346
Experimental 151 505 107 064 150 454 147 116

ṁ = 2.09 kg/s Inlet Outlet Climate EOC

Simulated 264 806 108 790 246 779 240 405
Experimental 309 655 108 790 295 078 278 827

54
Table 37: Pressure magnitude in working pressure (including the reference pressure).

ṁ = 0.538 kg/s p4 p5 p14

Simulated 109 321 108 721 111 069
Experimental 113 816 110 039 112 654
Difference -3.9% -1.2% -1.4%

ṁ = 1.011 kg/s p4 p5 p14

Simulated 117 444 114 122 123 487
Experimental 125 374 116 724 131 130
Difference -6.3% -2.2% -5.8%

ṁ = 2.09 kg/s p4 p5 p14

Simulated 152 027 134 758 177 607
Experimental 178 625 136 536 211 618
Difference -14.9% -1.3% -16.1%

Table 38: Mass flows through all relevant flow sensors from Fig. 14.

ṁ = 0.538 kg/s f2 f7 Thermostat

Simulated 0.1267 0.0990 0.312
Experimental 0.0790 0.1057 0.353
Difference +60.4% -6.3% -11.6%

ṁ = 1.011 kg/s f2 f7 Thermostat

Simulated 0.243 0.193 0.575
Experimental 0.153 0.206 0.652
Difference +58.8% -6.3% -11.8%

ṁ = 2.09 kg/s f2 f7 Thermostat

Simulated 0.506 0.410 1.174
Experimental 0.326 0.452 1.312
Difference +55.2% -9.3% -10.5%

55
 Table 39: Pressure drops over the porous baffle interfaces, see Table 11.

ṁ = 0.538 kg/s Climate EOC

Simulated 6 050 5 813
Experimental or 1D 6 900 2 461
Difference -12.3% +136%

ṁ = 1.011 kg/s Climate EOC

Simulated 18 084 17 677
Experimental or 1D 20 399 7 390
Difference -11.3% +139%

ṁ = 2.09 kg/s Climate EOC

Simulated 69 172 68 027
Experimental or 1D 82 886 28 176
Difference -16.5% +141%

The difference between the simulated and the experimental flow f2, seen in Table 38 is con-
nected with the difference in the pressure drops seen in Table 39. The pair of constants used
to describe this pressure drop was also used for the open thermostat, in that case the differ-
ence were not as high as for the closed thermostat.

56
5 Conclusions
The use of wall functions is not satisfactory in the water jacket. The method constantly un-
derpredicts the pressure drop, Table 12. For the pipe, wall functions gives similar results as
for the low Reynolds models when the elliptic blending turbulence modeled is used, Table 17.
The recommendation is not to use wall functions since the pressure drop in the water jacket
can not be modeled accurately, which is supported by theory, addressed in Section 2.4.1.

Using the SST and EB turbulence models gave oscillating results which can be seen in the
standard deviation in Table 12 and 17. The standard deviation can be reduced by choos-
ing the v2f model or the realizable model which doesn’t have any oscillating behaviour. The
v2f turbulence model is recommended because it is superior to the realizable model when
it comes to heat transfer. For low y + -values, close to 0.5, the difference in pressure drop be-
tween the v2f, SST and EB turbulence models are small, specially for the water jacket, see Fig.
17.

Comparing the results in Section 4.3.1 and 4.3.1.1 it can be seen that the quick fix in the ge-
ometry in Section 4.3.1 gave a clear improvement in the results, showing the importance of
accurate CAD geometry. It can also be seen that some of the experimental data could be
improved. The difference in the geometry used in thesis and the geometry used for the ex-
periments together with the inaccuracies in the experimental data make up the majority of
the difference between the experimental and simulated data. The overall deviation is larger
for the Case 2 (closed thermostat) than for Case 1 (open thermostat).

Mesh setup appropriate to use can be found in Table 5 (setup 4), Table 7, Table 8 and Ta-
ble 9 (setup 7) for the water jacket, pipe, closed and open thermostat, respectively.

Our completed conclusion is that more work has to be done before implementing the 3D
model. One option can be to combine the 3D model with 1D simulations.

5.1 Future work

A Richardson extrapolation should be carried out to rigorously investigate how many cells
are needed to acquire a tolerably discretization error. To get a more accurate model, surface
roughness should be included since all surfaces are in fact rough. Some attempts were made
to include surface roughness in the model but were left unfinished due to lack of time. The
results of the attempts can be seen in the Appendix, Section A. Heat transfer is of course of
interest for cooling systems and should thus be included in this model for future use.

Further validation of this 3D model can be to compare it to 1D simulations instead of ex-

perimental data.

57
References
[1] Cengel, Y. A. & Cimbala, J. M. (2010), Fluid Mechanics Fundamentals and Applications,
McGraw-Hill.

[2] Test report, No 312681, Volvo Car Corporation, 2013

[3] Peric, M. & Ferguson, S. The advantage of polyhedral meshes, www.cd-adapco.com

[4] CD Adapco STAR-CCM+ User Guide

[5] Robert N. Brady, (2010), Modern Diesel Engine, Prentice Hall.

[6] John D. Anderson, (1995), Computation Fluid Dynamics - The Basics with Applications,
McGraw-Hill.

[7] R.H. Nichols, Turbulence Models and Their Application to Complex Flows

[8] Ismail B. Celik, (1999), Introductory Turbulence Modeling

[9] Yang Zhiyin, (2014), Large-eddy simulation: Past, present and the future, Chinese Jour-
nal of Aeronautics, 28(1), 11-24

[10] Lars Davidsson, (2015), An Introduction to Turbulence

[11] Patel, V. C., Rodi, W. and Scheuerer. G. (1985). Turbulence models for near-wall and low
Reynolds number flows: A review, AIAA Journal, 23, 1308-1319

[12] Cuong Nguyen, (2005), Turbulence Modeling

[13] T. J. Craft, S. E. Gant, A. V. Gerasimov, H. Iacovides and B. E. Launder, Wall-function

Strategies for use in Turbulent Flow CFD

[14] Jonas Bredberg, (2000), On the Wall Boundary Condition for Turbulence Models

58
A Surface roughness
All of the simulations are done with smooth walls. This is primarily done to be able to com-
pare the simulated results with the experimental data. The real case is of course rough and
there is of interest to see how the surface roughness affects the pressure drop. The simula-
tions with surface roughness were carried on the pipe due the fast simulation time.

The surface roughness height, r , is the mean height of the surface roughness and it is dif-
ferent depending on the material. The roughness height has to be lower that the first layer
height.

From Fig. 29 it can be seen that for the fine mesh, the pressure drop does not depend on the
surface roughness. For the coarser mesh, using wall functions, the pressure drop is clearly
affected by the surface roughness. The reason the pressure drop is not affected for the fine
mesh is because the surface roughness height, r , change the behaviour in the log-law region
via E in Eq. 2.13. For low-Reynolds models the log-law is not valid and the flow is resolved by
the RANS equations.

Looking at Fig. 30 it can be seen that a transient simulation will not change the behaviour
of the surface roughness, it will however result in a larger pressure drop.

Figure 29: Pressure drop for the k-ε realizable turbulence model as a function of the wall
roughness height r .

59
 (a) Steady state simulation. (b) Transient simulation.

Figure 30: Comparision between steady state and transient pressure drop for the v2f turbu-
lence model as a function of the surface roughness height.

60