Sunteți pe pagina 1din 18

Comparative Analysis of Three Simulation Models

Applied on a Motored Internal Combustion Engine


Efthimios Pariotis
a
, George Kosmadakis
b
and Constantine Rakopoulos
c
a Laboratory of Naval Propulsion Systems, Section of Naval Architecture and Marine Engineering,
Department of Naval Sciences, Hellenic Naval Academy, End of Hatzikyriakou Ave., Hatzikyriakio, 18539
Piraeus, Greece, pariotis@snd.edu.gr, CA
b Internal Combustion Engines Laboratory, Thermal Engineering Department, School of Mechanical
Engineering, National Technical University of Athens, 9 Heroon Polytechniou St., Zografou Campus,
15780 Athens, Greece, gkosmad@central.ntua.gr
c Internal Combustion Engines Laboratory, Thermal Engineering Department, School of Mechanical
Engineering, National Technical University of Athens, 9 Heroon Polytechniou St., Zografou Campus,
15780 Athens, Greece, cdrakops@central.ntua.gr
Abstract:
The motivation of the present work is to comparatively evaluate the results obtained using three
computational models of increasing complexity, for the simulation of the closed part of the cycle of
an internal combustion engine running under motoring conditions. The first model is a single-zone
thermodynamic model, the second one is a hybrid quasi-dimensional model and the third one is a
computational fluid dynamics (CFD) model. All of them have been originally developed by the
authors, and each of these models has certain limitations due to their inherent features, while their
usage has a different scope. The three models have been applied to simulate the closed part of
the cycle of a high-speed diesel engine with a bowl in piston design, running under motoring
conditions at four engine speeds, in the range of 1200 to 3000 rpm.
From the analysis conducted, it is concluded that the single-zone model calculates with
reasonable accuracy the in-cylinder pressure, while it rather underestimates the peak in-cylinder
mean gas temperature near the top dead centre (TDC), compared to the other two models,
needing only 1 sec on a personal computer (PC), to simulate each case. On the other hand the
computational time required by the quasi-dimensional model is approximately 6 minutes to
simulate the closed part of the engine cycle, whereas the CFD model needs up to 20 hours (all
tested on a personal computer with a CPU running at 2.2 GHz). At the same time the quasi-
dimensional model, provides information concerning the local in-cylinder temperature distribution
and describes qualitatively correctly how the cylinder design affects the in-cylinder flow and
temperature field, as revealed by comparing its results with the corresponding ones obtained by
the more accurate and time consuming CFD model.
Highlighting the special features of each type of simulation model examined on a comparative
basis, provides useful information for the engineer to decide which type of model is best for each
application, taking into account not only the accuracy but also the computational time.
Keywords:
Computational fluid dynamics, Internal combustion engine, Quasi-dimensional, Single-zone,
Motoring.
1. Introduction
Engine simulation models are valuable tools for the engineers who are working on the automotive
industry or belong to the research community, aiming to design engines that comply with the strict
emission legislation and offer high performance. During the last decades various simulation models
for the internal combustion engines have been proposed. In general there are two main categories of
simulation models: the phenomenological and the computational fluid dynamic (CFD) ones. The main
difference is that in the phenomenological models, empirical or semi-empirical relations are used to
describe the physical and chemical processes taking place inside the cylinder, while in CFD models,
special iterative techniques are used to solve the conservation equations of mass, momentum, energy
and species, coupled to a turbulence model at the discretised computational domain which covers the
whole cylinder volume. The phenomenological models are further divided depending on their
complexity to one or two dimensional or multi-zone models. Recently, a new sub-category of
simulation model has been proposed [1-3], signed between the multi-zone phenomenological and
CFD models, named hybrid quasi-dimensional model, which combines features of the multi-zones
and the CFD models. This type of model, offers a more fundamental description of the physical
processes taking place inside the combustion chamber compared to the multi-zone models, while it is
less accurate compared to the more detailed and more CPU demanding CFD models, since the
momentum equations are not taken into consideration for the calculation of the local flow-field.
It is clear that depending on the specific needs of the research, the engineer has to decide which type
of model should best fit his needs, in order to obtain the required results for his application and at the
same time spend the lowest possible computation resources. Although the differences between a
single zone and a CFD model are wide and their purpose of usage might be clear, sometimes a quasi-
dimensional model can substitute in a certain extent the usage of a CFD model and this can be
translated in gaining significant amount of effort and time.
The present authors have developed in the past and validated against experimental or published data
the quasi-dimensional and the CFD model [4-9], while the single-zone (1-D) model has been used to
support the development of various sub-models incorporated into the CFD model. The main scope of
the present work is to compare the results obtained using a 1-D, a quasi-dimensional and a CFD
model, applied to simulate a high speed diesel engine with a bowl in piston running under motoring
conditions. The study was intentionally restricted to the motoring conditions to understand how the
complexity of the model affects the description of the physical processes. When combustion takes
place, each type of model follows totally different approach to simulate the air-fuel mixing and the
chemical processes involved, which makes the comparison of the models more complex.
Apart from estimating the mean in-cylinder thermodynamic properties, the quasi-dimensional model
has the advantage of predicting the spatial distribution of temperature inside the whole cylinder
volume at each computational step, following a procedure resembling to the one used in pure CFD
codes, with a major difference: in the hybrid quasi-dimensional model, a phenomenological sub-
model is used for the estimation of the three dimensional velocity vector at the cells faces of the
computational domain, avoiding the solution of the momentum equations, as done in pure CFD codes.
This leads to significant CPU power saving, however the predicted velocity field is only a rough
estimation (although qualitatively correct), being inferior compared to the one provided by the CFD
code. The predicted velocities are used to calculate the convention terms of the energy conservation
equation, affecting the spatial temperature distribution. Comparing the estimated temperature fields
using the quasi-dimensional and the CFD model at certain time instants (corresponding to 160 and
220 CA degs ABDC, at 2000 rpm engine speed) it is observed that the quasi-dimensional model
manages to predict qualitatively correct the spatial temperature distribution. It is important to notice,
that using the quasi-dimensional model, it is possible to capture qualitatively the effect of piston bowl
geometry on in-cylinder temperature and velocity field.
2. Experimental facilities
The experimental tests were conducted on a Ricardo/Cussons Hydra, single cylinder, four-stroke,
water-cooled, high-speed, experimental standard engine located at the authors laboratory. The fully
automated test bed includes facilities to monitor and control engine variables such as engine speed,
load, static injection timing, water, lubricating oil and exhaust gas temperatures, fuel and air flows
etc. The engine is coupled to a McClure DC motoring dynamometer, having load absorbing and
motoring capabilities. The test bed control and engine operating conditions adjustment is achieved
through an electronic console. Electrically driven pumps assure the circulation of coolant and
lubricating-oil, while heaters are used to maintain their temperatures during warm-up conditions.
In the present work the Hydra engine is used as a naturally aspirated, DI diesel engine having a re-
entrant (torroidal) bowl-in-piston combustion chamber. In Table 1 is shown a summary of the basic
engine data.
Table 1. Diesel engine specifications
Engine model and type Ricardo/Cussons Hydra, single cylinder, diesel, naturally
aspirated, four-stroke, water-cooled, high-speed
Bore 80.26 mm
Stroke 88.90 mm
Swept volume 0.4498 l
Connecting rod length 158 mm
Compression ratio 19.81:1
Squish height 0.82 mm
Bowl-in-piston volume 19.737 cm
3
Swirl ratio 3.57
Speed range 1000 4500 rpm
Maximum cylinder pressure 120 bar
Valve
timing
Inlet valve opening 8
o
CA BTDC
Inlet valve closure 42
o
CA ABDC
Exhaust valve opening 60
o
CA BBDC
Exhaust valve closure 12
o
CA ATDC
The measuring set-up comprises the following two transducers:
Tektronix TDC marker (magnetic pick-up)
Kistler 6125B piezoelectric transducer for measuring the cylinder pressure, flush mounted to the
cylinder head and connected to a Kistler 5008 charge amplifier.
The output signals of these two transducers, while being continuously monitored on a dual beam
Tektronix oscilloscope, are fed to the input of the data acquisition module, which is a Keithley KUSB
3102 ADC card connected to a Pentium Dual Core PC via USB interface. This specific type has a
maximum sampling rate of 100 ksamples/sec, with a 12-bit resolution for its 8 differential (or 16
single-ended) input, and 2 output analogue channels. Control of this high-speed acquisition system is
achieved by using the quickDAQ data collecting software, supplied by Keithley, together with the
KUSB 3102 ADC card. For the present study 20 cycles, with 1 CA resolution, were recorded for
every engine speed. The mean measured motored pressure history was computed, though the cyclic
variability was negligible, while the measurement error was estimated to be smaller than 1%.
3. Numerical models
In the current study three numerical models have been applied for the simulation of the motored
Diesel engine. These are namely a single-zone (1-D) model, a quasi-dimensional (Q-D) model, and a
computational fluid dynamics (CFD) model, which have been originally developed by the authors,
and have been applied in the past for the simulation of various engines and operating conditions.
These three models are briefly described next.
3.1. Single-zone model
The single-zone model is based on the first law of thermodynamics. It incorporates a simple
turbulence model based on the k- principle, according to [10,11]. This turbulence model is actually
coupled with the heat transfer model. For the current application the Annand heat transfer model has
been chosen [12], in which the Reynolds number includes the effective conductivity (k
eff
) and
viscosity (
eff
), and a characteristic velocity. The heat transfer surface includes the real surface of the
engine under investigation, which is the sum of the surface surrounding the main chamber and the
piston bowl. For the rest of the current study, this model will be referred to as 1-D model.
3.2. Quasi-dimensional model
The quasi-dimensional model used is a hybrid one which means that it uses phenomenological sub-
models to describe the various processes taking place inside the combustion chamber, combined with
methods used in CFD models to calculate the local values of the various properties. Specifically,
phenomenological sub-models are used to describe the following: heat transfer through the cylinder
walls, fuel injection rate, fuel spray penetration, evaporation, combustion, and pollutants formation. In
the present study, all the simulations are done in motoring conditions, thus the sub-models related to
fuel injection and combustion have been deactivated.
In contrast to what is done in most multi-zone models [13], the whole cylinder volume is divided in
computational cells which consist the computational domain and the local value of each characteristic
property of the field (specific enthalpy and temperature, velocity and species concentrations) is
calculated at each computational cell, by solving the general conservation equation of property (1).
( ) ( ) ( ) ( ) piston
piston

z
u r v w
1 1 1
z t r r r z

r r
1 1 r z
S
r r r z


+ + +

_ _ _



, , ,
+ + +

(1)
where, u, v, and w are the components of the local velocity vector at each computational node, the
gas density,

is the diffusivity of the dependent variable and S



is

the volumetric source rate. At
motoring conditions, the dependent variable is the local gas enthalpy, and the source rate S

,
represents the contribution on the energy equation of the heat transfer through the cylinder walls
(according to the Annand heat transfer model) and the power due to the pressure variation.
Equation (1) is solved using the finite volume method, similarly to what is done in most CFD codes.
One important difference with the CFD models is the methodology followed for the calculation of the
velocity field, where a new phenomenological sub-model is used, to avoid solving the equation of
momentum conservation, which is highly CPU demanding, since the convective terms of the energy
conservation equation contain non-linear quantities, and the momentum equation (containing the local
pressure) and the continuity equation are intricately coupled.
Following this approach, the physical phenomenon of mass transfer between the computational cells
is simulated in a realistic way, based on the assumption that the in-cylinder pressure must be
practically spatially uniform at each crank angle. The velocity is estimated at the boundaries of each
computational cell, by calculating the mass that should be transferred through each cells faces from
the neighbouring cells to achieve a uniform pressure field inside the cylinder. The velocity field is
obtained following an iterative procedure. First, at each crank angle the energy conservation equation
is solved, and a spatial distribution of the temperature is obtained assuming that the velocity of the gas
relative to the grid is zero. Then the pressure distribution is obtained using the perfect gas state
equation. Given that the pressure must practically be uniform, an amount of mass dm
cell
is calculated
that should be transferred inside each computational cell through its boundaries from its neighbouring
cells to eliminate the pressure difference and make its pressure practically equal to the mean pressure
of the cylinder. However, the sum of the transferred mass for all computational cells at each crank
angle should be equal to zero, given that the total mass of the gas inside the cylinder is constant. In
this way the amount of mass dm
cell
transferred to each computational cell is estimated. This amount of
mass is distributed to each cells faces using an empirical method, based on the local pressure
difference at each direction. In this way, the velocities at the boundaries of all the computational cells
are determined. These velocities are used when solving the energy equation (1).
Following this procedure, the solution of the momentum equation is by-passed, and the in-cylinder
velocity field is estimated faster but less accurately compared to the pure CFD models. Taking into
account that the accuracy of a simulation model depends on the less accurate sub-model used, in the
proposed quasi-dimensional model, semi-empirical relations are used for the simulation of the heat
transfer, fuel injection rate, fuel spray penetration, evaporation, combustion, and pollutants formation,
as done in most multi-zone models.
According to the aforementioned methodology for the estimation of the in-cylinder velocity field, the
developed model describes in a more fundamental way the transport phenomena of the in-cylinder gas
compared to existing multi-zone phenomenological models, being able to capture the effect of
cylinder geometry (i.e. piston bowl design). Furthermore, it overcomes basic difficulties experienced
in existing multi-zone models, as far as air-fuel mixing is concerned (in firing conditions), where only
empirical relations are used. A more detailed description of the model can be found in previous
publications [1-3].
In the present work where motoring conditions are examined, the general transport equation is solved
only for enthalpy, to estimate the local value of temperature at each computational cell inside the
cylinder. For the rest of the current study, this model will be referred to as Q-D model.
3.3. Computational fluid dynamics model
The CFD code developed can simulate three-dimensional curvilinear domains using the finite volume
method in a collocated grid. It incorporates the RNG k- turbulence model with some slight
modifications to introduce the compressibility of a fluid in generalized coordinates, as described in
[14]. All the constants used in the RNG k- turbulence model and the enthalpy equation, are in
accordance with [15]. These constants retain their values at all cases examined in this study. The
following transport equations for the conservation of mass, momentum, energy and turbulence
properties are solved:
( )
( ) ( ) ( )
cr
S S u
t
+ + +

(2)
where S
cr
is a source term due to crevice regions [6] and the generalized variable

is replaced by the
appropriate variable, shown in Table 2.
Table 2. Variables that represent the generalized variable
Generalized variable Equation

1 Continuity 0
u umomentum
eff
v vmomentum
eff
w wmomentum
eff
h Enthalpy
eff
/
h
k Turbulent kinetic energy
eff
/
k
Dissipation of turbulent kinetic energy
eff
/

The solution follows the PISO algorithm for the velocity-density-pressure coupling, according to the
methodology found in [16], which is more robust than the SIMPLE algorithm [17]. The spatial
discretization is based on the hybrid-differencing scheme, while the temporal one on the backward
second-order Euler scheme. The transport properties (specific heat capacity under constant pressure,
thermal conductivity and laminar viscosity) are calculated from least-squares fits proposed by NASA
[18]. The heat transfer model used has been developed by the authors and validated under motoring
and firing conditions [7,8], which provides more reliable calculations, especially during the
compression stroke. The expression of the local heat flux (q
w
) includes the compressible form of the
wall-functions and a pressure term (dP/dt) [7], while it is given by the following equation:
10.2384
0.4767Pr
1
40 ln
0.4767Pr
1
y ln
0.4767
1
117.31
Pr
1
0.4767
40 y
u

dt
dP
T
T
Tln c u
q
T
W
p T
W
+
1
]
1

,
_

+
,
_

,
_

+
+

,
_

+
+
(3)
where u
T
is the friction velocity, c
p
the heat capacity under constant pressure, T
W
the wall temperature,
the kinematic viscosity, y
+
the non-dimensional distance from the wall, and Pr the Prandtl number.
The CFD model incorporates also a phenomenological crevice model, in order to simulate the mass
transfer between the combustion chamber and the crevice regions [6], predicting with a higher
accuracy especially the peak cylinder pressure. The mesh generation model used belongs to the wide
family of algebraic grid generation methods, and has been originally developed by the authors. This
model is capable of creating 3D structured grids consisting of hexahedral cells. Additionally, mesh
layers are removed and added, during the compression and expansion stroke respectively, in order for
the cells dimension along the axial direction to keep an almost constant value. Further details of the
in-house CFD model, its evaluation and validation, together with the mesh generation model and the
crevice model used, can be found in previous published studies of the authors [4-9]. For the rest of the
current study, this model will be referred to as CFD model.
4. Results and discussion
4.1. Test cases examined / Numerical Data
The three numerical models have been applied to simulate the closed part of the engine cycle of a
diesel engine, with a bowl in piston configuration, running under motoring conditions, at 1200, 1500,
2000 and 3000 rpm engine speed. The measured cylinder pressure traces are used to evaluate the
predictive accuracy of each model, as far as the engine performance is concerned. Based on the fact
that the CFD code used has already been validated [4,6,7] at various types of engines, it is used as a
reference to evaluate the calculated results obtained using the other two models, regarding the in-
cylinder mean gas temperature and the wall heat losses during the closed part of engine cycle. The
same also applies, as far as the spatial distribution of temperature and the velocity field is concerned
calculated from the quasi-dimensional model, at two time instants corresponding to 160 and 220 CA
degs ABDC at 2000 rpm engine speed. Table 3 summarizes the test conditions examined in the
present study.
Table 3. Diesel engine test conditions (motored)
Engine operation Motored
Engine speed 1200, 1500, 2000, 3000 rpm
Ambient pressure / temperature 1 bar / 23
o
C
Lubricating oil temperature 80
o
C
Coolant (water) temperature 80
o
C
Some initial and boundary conditions used, which are common for all three models are listed in Table
4. The Q-D and the CFD model use a mesh, covering the entire cylinder volume, where the
conservation equations are solved. When applying the CFD model the computational mesh at the
beginning of the simulations consists of 30x30x40 grid lines along the x,y,z axis respectively (z-axis
corresponds to the axial direction). This mesh configuration is selected based on previous grid
independence studies [6].
Table 4. Initial/boundary conditions used
Swirl ratio at IVC 3.57
Air temperature at IVC 310 K
Wall temperature 373 K
Turbulent kinetic energy at IVC 0.8 x (mean piston speed)
2
On the other hand, when applying the quasi-dimensional model the area inside the cylinder is divided
into cylindrical computational cells, using a structured grid that contracts and expands to account for
the variation of the swept volume. Since in the present study only motoring conditions are examined,
there is no need to divide the cylinder volume into the circumferential direction, assuming that the
piston bowl is centrally located and symmetric. A schematic view of the combustion chamber
geometry assumed, together with the computational mesh used with the Q-D model is shown in Fig.
1. The number of cells used in the radial direction above the piston crown is equal to 16, and inside
the piston bowl is equal to 8, while in the axial direction the number of cells inside the piston bowl is
equal to 7 and in the outer region varies from 37 to 1 depending on the distance of the piston from the
cylinder top. With this configuration, mesh independent solutions are derived, and the dimensions of
the computational cells are similar with those when the CFD model is applied. During all the
simulations, the same computational time-step has been used, corresponding to 0.5
o
CA. This decision
is supported by a time-step independence study that has already been conducted in the past [6].




D
cyl
d
bowl
r direction

d
i
r
e
c
t
i
o
n
Fig. 1. Computational domain used in the quasi-dimensional model at a vertical plane (r-z)
It should be mentioned that the two phenomenological models (1-D and Q-D) need to be calibrated at
one operating condition to match the calculated cylinder pressure traces with the measured one. This
is done by changing the exponential constant of the Reynolds number in the Annands equation which
is used to estimate the heat transfer through the cylinder walls. After having calibrated this constant, it
retains its value at all the operating conditions examined. On the contrary, the CFD model has no
calibration constants, since the wall-function formulation used to calculate the wall heat losses is
based on the one-dimensional energy equation applied on the boundary cells. This is an advantageous
feature of the CFD model, indicating that it is preferable for conducting fundamental research [5].
4.2. Models Validation Calculated Cylinder Pressure Traces
At first, the predictive ability of the three simulation models is examined as far as the cylinder
pressure trace is concerned, at four engine speeds i.e. 1200, 1500, 2000 and 3000 rpm, since the
accurate prediction of the cylinder pressure is a prerequisite for any simulation model used in a real
engine application. To this scope, the calculated cylinder pressure during the closed part of engine
cycle is compared with the corresponding measured one. As shown in Fig. 2, at all operating
conditions examined all the simulation models manage to predict well the cylinder pressure traces.
The higher discrepancies are observed near the TDC, where the maximum difference is
approximately 5 bar between the measured and the calculated peak cylinder pressure using the 1-D
model. On the other hand the quasi-dimensional model, predicts with high accuracy the measured
peak cylinder pressure at all cases examined, performing slightly better than the CFD model, with the
maximum difference observed at 3000 rpm engine speed, which is approximately equal to 2 bar
(although in this case, the predicted cylinder pressure is identical with the one obtained using the CFD
model).
1 4 0 1 6 0 1 8 0 2 0 0 2 2 0
C r a n k A n g l e [ d e g s a B D C ]
0
1 0
2 0
3 0
4 0
5 0
6 0
C
y
l
i
n
d
e
r

P
r
e
s
s
u
r
e

[
b
a
r
]
1 2 0 0 r p m
M e a s .
1 - D
Q - D
C F D
1 4 0 1 6 0 1 8 0 2 0 0 2 2 0
C r a n k A n g l e [ d e g s a B D C ]
0
1 0
2 0
3 0
4 0
5 0
6 0
C
y
l
i
n
d
e
r

P
r
e
s
s
u
r
e

[
b
a
r
]
1 5 0 0 r p m
M e a s .
1 - D
Q - D
C F D
1 4 0 1 6 0 1 8 0 2 0 0 2 2 0
C r a n k A n g l e [ d e g s a B D C ]
0
1 0
2 0
3 0
4 0
5 0
6 0
C
y
l
i
n
d
e
r

P
r
e
s
s
u
r
e

[
b
a
r
]
2 0 0 0 r p m
M e a s .
1 - D
Q - D
C F D
1 4 0 1 6 0 1 8 0 2 0 0 2 2 0
C r a n k A n g l e [ d e g s a B D C ]
0
1 0
2 0
3 0
4 0
5 0
6 0
C
y
l
i
n
d
e
r

P
r
e
s
s
u
r
e

[
b
a
r
]
3 0 0 0 r p m
M e a s .
1 - D
Q - D
C F D
Fig. 2. Comparison of calculated pressure obtained from the three numerical models with the
measured one for various engine speeds.
4.3. Comparison between the calculated results of simulation models
After having validated that the three simulation models manage to predict the cylinder pressure traces
at all test cases examined, the investigation proceeds with the comparison of the calculated results of
the three simulation models, as far as the mean cylinder temperature, and the heat transfer through the
cylinder walls during the closed part of the engine cycle, at 1200, 1500, 2000 and 3000 rpm engine
speed, is concerned. Since no experimental data are available, the calculated results of the CFD model
will be used as a reference to evaluate the other two phenomenological models. This decision is based
on the fact that the CFD model calculates in a more fundamental and detailed way the physical
processes taking place inside the combustion chamber, which consequently affect the heat transfer
process. Besides, the validity of this model has already been confirmed by the authors in previous
publications [4,9].
4.3.1. Mean Cylinder Gas Temperature
In Fig. 3 is shown the mean in-cylinder gas temperature during the closed part of the engine cycle at
the four engine speeds examined, as calculated from the three simulation models.
1 2 0 1 4 0 1 6 0 1 8 0 2 0 0 2 2 0 2 4 0
C r a n k A n g l e [ d e g s a B D C ]
2 0 0
3 0 0
4 0 0
5 0 0
6 0 0
7 0 0
8 0 0
9 0 0
M
e
a
n

G
a
s

T
e
m
p
e
r
a
t
u
r
e

[
K
e
l
v
i
n
]
1 2 0 0 r p m
1 - D
Q - D
C F D
1 2 0 1 4 0 1 6 0 1 8 0 2 0 0 2 2 0 2 4 0
C r a n k A n g l e [ d e g s a B D C ]
2 0 0
3 0 0
4 0 0
5 0 0
6 0 0
7 0 0
8 0 0
9 0 0
M
e
a
n

G
a
s

T
e
m
p
e
r
a
t
u
r
e

[
K
e
l
v
i
n
]
1 5 0 0 r p m
1 - D
Q - D
C F D
1 2 0 1 4 0 1 6 0 1 8 0 2 0 0 2 2 0 2 4 0
C r a n k A n g l e [ d e g s a B D C ]
2 0 0
3 0 0
4 0 0
5 0 0
6 0 0
7 0 0
8 0 0
9 0 0
M
e
a
n

G
a
s

T
e
m
p
e
r
a
t
u
r
e

[
K
e
l
v
i
n
]
2 0 0 0 r p m
1 - D
Q - D
C F D
1 2 0 1 4 0 1 6 0 1 8 0 2 0 0 2 2 0 2 4 0
C r a n k A n g l e [ d e g s a B D C ]
2 0 0
3 0 0
4 0 0
5 0 0
6 0 0
7 0 0
8 0 0
9 0 0
M
e
a
n

G
a
s

T
e
m
p
e
r
a
t
u
r
e

[
K
e
l
v
i
n
]
3 0 0 0 r p m
1 - D
Q - D
C F D
Fig. 3. Calculated mean cylinder gas temperature using the three numerical models (1-D, Q-D and
CFD) for various engine speeds.
As observed, at all cases examined the predicted mean gas temperature at each CA deg. is higher
when using the CFD code. At all engine speeds examined, the difference between the peak mean gas
temperature of the cycle, as predicted by the CFD and the quasi-dimensional model, is almost the half
of the corresponding one, when comparing the CFD and the 1-D model (especially at higher engine
speeds), which is approximately equal to 50
o
C. It is noticed that the difference on the predicted peak
in-cylinder mean gas temperatures using the three simulation models remains practically constant at
all the range of engine speeds examined, with a slight increase at 3000 rpm. From the previous it can
be concluded that, although the prediction of the cylinder pressure traces is made with relatively
similar accuracy, the predicted mean gas temperature profiles vary in a wider extent when switching
between the three simulation models examined.
4.3.2. Wall heat losses
As far as the wall heat transfer is concerned, in Fig. 4 are shown the calculated values of the heat
transferred through the cylinder walls at each CA deg. using the three simulation models, at all engine
speeds examined. As observed, there is a general similarity on the predicted profiles between the CFD
and the quasi-dimensional model, with a small discrepancy on the absolute values mainly
4 0 8 0 1 2 0 1 6 0 2 0 0 2 4 0 2 8 0
C r a n k A n g l e [ d e g s a B D C ]
- 1 8 0 0 0
- 1 6 0 0 0
- 1 4 0 0 0
- 1 2 0 0 0
- 1 0 0 0 0
- 8 0 0 0
- 6 0 0 0
- 4 0 0 0
- 2 0 0 0
0
2 0 0 0
W
a
l
l

H
e
a
t

T
r
a
n
s
f
e
r

[
W
]
1 2 0 0 r p m
1 - D
Q - D
C F D
4 0 8 0 1 2 0 1 6 0 2 0 0 2 4 0 2 8 0
C r a n k A n g l e [ d e g s a B D C ]
- 1 8 0 0 0
- 1 6 0 0 0
- 1 4 0 0 0
- 1 2 0 0 0
- 1 0 0 0 0
- 8 0 0 0
- 6 0 0 0
- 4 0 0 0
- 2 0 0 0
0
2 0 0 0
W
a
l
l

H
e
a
t

T
r
a
n
s
f
e
r

[
W
]
1 5 0 0 r p m
1 - D
Q - D
C F D
4 0 8 0 1 2 0 1 6 0 2 0 0 2 4 0 2 8 0
C r a n k A n g l e [ d e g s a B D C ]
- 1 8 0 0 0
- 1 6 0 0 0
- 1 4 0 0 0
- 1 2 0 0 0
- 1 0 0 0 0
- 8 0 0 0
- 6 0 0 0
- 4 0 0 0
- 2 0 0 0
0
2 0 0 0
W
a
l
l

H
e
a
t

T
r
a
n
s
f
e
r

[
W
]
2 0 0 0 r p m
1 - D
Q - D
C F D
4 0 8 0 1 2 0 1 6 0 2 0 0 2 4 0 2 8 0
C r a n k A n g l e [ d e g s a B D C ]
- 1 8 0 0 0
- 1 6 0 0 0
- 1 4 0 0 0
- 1 2 0 0 0
- 1 0 0 0 0
- 8 0 0 0
- 6 0 0 0
- 4 0 0 0
- 2 0 0 0
0
2 0 0 0
W
a
l
l

H
e
a
t

T
r
a
n
s
f
e
r

[
W
]
3 0 0 0 r p m
1 - D
Q - D
C F D
Fig. 4. Wall-heat losses calculated using the three numerical models (1-D, Q-D and CFD) for
various engine speeds.
during the expansion phase at all engine speeds examined. However, using the 1-D model a
significant under prediction of the wall heat losses is made during the compression and the first part
of the expansion phase, compared to the other two more sophisticated models. More specifically,
during the compression stroke the in-cylinder temperature rises and the in-cylinder gas velocities
increase, which consequently lead to more intense heat transfer to the cylinder walls. As shown in
Fig. 4 the quasi-dimensional model manages to predict precisely the aforementioned physical
mechanism which affects the heat transfer through the cylinder walls. As the piston approaches the
TDC, the predicted values of the heat losses to the cylinder walls using the quasi-dimensional model,
resemble to the ones predicted using the CFD one, at almost all the engine speeds examined. The only
exception occurs at 3000 rpm engine speed, where a relatively higher difference occurs, but again the
prediction is much more accurate than the one obtained using the 1-D model.
The aforementioned better prediction of the wall heat transfer mechanism obtained using the quasi-
dimensional model compared to the 1-D model, should be attributed to the fact that in the quasi-
dimensional model a fundamental description of the physical process is made. The heat is transferred
from the hot in-cylinder core to the outer cells located next to the cylinder walls through conduction
and convection (taking into account the effect of the calculated 3-D velocity field). In turn, the cells
located at the boundary of the cylinder volume transfer heat to the cylinder walls through convection
according to Annands law [12]. Thus it is taken into account the stratification of the in-cylinder local
temperatures at each time step, which is not possible when using the 1-D model. This is one of the
main attractive features of the proposed quasi-dimensional model. To further extend the comparative
evaluation of the quasi-dimensional model and the CFD one, the local heat fluxes are calculated at
four locations on the cylinder head as shown in Fig. 5.
Fig. 5. Position of the four locations (HT1, HT2, HT3, HT4) where the local heat flux is calculated
Due to the different computational mesh used at the quasi-dimensional and the CFD model, it was not
possible to calculate the local heat flux at exactly the same points, however, the differences where
small. In Table 5, is shown the radial distance of each point from the cylinder axis, when using the
CFD and the quasi-dimensional model.
Table 5. Radial Distance from the cylinder axis of the locations where local heat flux is calculated
Location CFD (mm) Quasi-dimensional (mm)
HT1 0 1.25
HT2 28.80 26.29
HT3 30.33 30.06
HT4 32.92 32.58
In Figure 6 are shown the calculated local heat fluxes using the CFD and the quasi-dimensional model
at the four locations (HT1, HT2, HT3, and HT4) respectively, at 2000 rpm engine speed. As
observed, with the exception of point HT1, the quasi-dimensional model predicts similar heat flux at
the other three points with the CFD model, during the compression stroke. On the expansion phase, it
seems that the quasi-dimensional model, slightly over predicts the local heat flux compared to the
CFD one, which is in accordance to what has already been observed when examining the total wall
heat transfer in Fig.4. On the other hand, at location HT1, significant differences are observed
between the calculated heat fluxes, especially near the TDC. This may be attributed to the fact that
these heat fluxes do not correspond exactly to the same location (as shown in Table 5) and that the
local temperature as being calculated by the CFD model seems to be higher at HT1 point, compared
to the one calculated using the quasi-dimensional model, as it will be shown in Figs. 7a,b and 8a,b.
Moreover, the CFD model predicts high level of turbulence close to the cylinder axis (HT1), which
enhances the heat transfer rate. This is not possible to be captured by the quasi-dimensional model,
since turbulence is not taken into account.
4 0 8 0 1 2 01 6 02 0 02 4 02 8 03 2 0
C r a n k A n g l e [ d e g s a B D C ]
- 1 . 5
- 1 . 4
- 1 . 3
- 1 . 2
- 1 . 1
- 1 . 0
- 0 . 9
- 0 . 8
- 0 . 7
- 0 . 6
- 0 . 5
- 0 . 4
- 0 . 3
- 0 . 2
- 0 . 1
0 . 0
L
o
c
a
l

H
e
a
t

F
l
u
x

[
M
W
/
m
2
]
2 0 0 0 r p m
R
Q - D
= 1 . 2 5 m m
R
C F D
= 0 . 0 0 m m
Q - D
C F D
P o i n t : H T 1

4 0 8 0 1 2 01 6 02 0 02 4 02 8 03 2 0
C r a n k A n g l e [ d e g s a B D C ]
- 1 . 5
- 1 . 4
- 1 . 3
- 1 . 2
- 1 . 1
- 1 . 0
- 0 . 9
- 0 . 8
- 0 . 7
- 0 . 6
- 0 . 5
- 0 . 4
- 0 . 3
- 0 . 2
- 0 . 1
0 . 0
L
o
c
a
l

H
e
a
t

F
l
u
x

[
M
W
/
m
2
]
2 0 0 0 r p m
R
Q - D
= 2 6 . 2 9 m m
R
C F D
= 2 8 . 8 0 m m
Q - D
C F D
P o i n t : H T 2
4 0 8 0 1 2 01 6 02 0 02 4 02 8 03 2 0
C r a n k A n g l e [ d e g s a B D C ]
- 1 . 5
- 1 . 4
- 1 . 3
- 1 . 2
- 1 . 1
- 1 . 0
- 0 . 9
- 0 . 8
- 0 . 7
- 0 . 6
- 0 . 5
- 0 . 4
- 0 . 3
- 0 . 2
- 0 . 1
0 . 0
L
o
c
a
l

H
e
a
t

F
l
u
x

[
M
W
/
m
2
]
2 0 0 0 r p m
R
Q - D
= 3 0 . 0 6 m m
R
C F D
= 3 0 . 3 3 m m
Q - D
C F D
P o i n t : H T 3

4 0 8 0 1 2 01 6 02 0 02 4 02 8 03 2 0
C r a n k A n g l e [ d e g s a B D C ]
- 1 . 5
- 1 . 4
- 1 . 3
- 1 . 2
- 1 . 1
- 1 . 0
- 0 . 9
- 0 . 8
- 0 . 7
- 0 . 6
- 0 . 5
- 0 . 4
- 0 . 3
- 0 . 2
- 0 . 1
0 . 0
L
o
c
a
l

H
e
a
t

F
l
u
x

[
M
W
/
m
2
]
2 0 0 0 r p m
R
Q - D
= 3 2 . 5 8 m m
R
C F D
= 3 2 . 9 2 m m
Q - D
C F D
P o i n t : H T 4
Fig. 6. Calculated local heat fluxes at locations HT1, HT2, HT3, HT4 using CFD and Quasi-
dimensional model
However, taking into account that the local heat flux is affected by both the temperature and the flow
field, it seems that the quasi-dimensional model predicts with relative good accuracy the in-cylinder
physical processes (heat and mass transfer).
4.3.3. In-cylinder Spatial Distribution of Temperature
The quasi-dimensional model solves the energy conservation equation at each computational cell,
according to the finite volume method using the velocity field which has been calculated with a
phenomenological velocity sub-model [1]. In this way it becomes feasible to capture how each
parameter (operating or design) affects the physical processes taking place inside the combustion
chamber, without the need to use the more accurate but also much more CPU demanding CFD
models. In Figures 7 a,b are shown the calculated temperature fields by the quasi-dimensional and the
CFD model corresponding to 160 CA degs ABDC, at 2000 rpm engine speed. As observed the
temperature range used in Fig. 7a is slightly shorter than the one used in Fig. 7b, in order to improve
the contrast of the stratification of the local temperature. Comparing these two figures, it can be
concluded that the calculated temperature field using the quasi-dimensional model is similar to the
one obtained using the CFD model. The gas with the higher temperature is restricted in the inner
portion of the cylinder, while near the cylinder and piston walls the gas has lower temperature. This is
due to the heat transfer, from the hot gas to the colder cylinder walls. However, using the CFD model,
as observed in Fig. 7b, the hot core of the cylinder gas is predicted to be slightly nearer to the bottom
of the piston bowl compared to what is predicted using the quasi-dimensional model. Moreover, the
maximum value of the local temperature in Fig. 7b is approx. 15
o
C higher than the one predicted by
the quasi-dimensional model, Fig. 7a.
Fig. 7a. Calculated spatial temperature distribution at 160 CA degs ABDC, at 2000 rpm engine
speed using the quasi-dimensional model
Fig. 7b. Calculated spatial temperature distribution at 160 CA degs ABDC, at 2000 rpm engine
speed using the CFD model
On the other hand, in Figs. 8a,b are shown the calculated temperature fields, during the expansion
phase, at 220 CA degs ABDC, at 2000 rpm engine speed, using the quasi-dimensional and the CFD
model respectively. At the same figures, the velocity vectors at each computational node are also
presented, in order to show a sample of the predicted velocity field using the two models and make a
qualitative comparison. It is noticed that the vectors shown in Fig. 8a,b have a uniform length and
they correspond to the local velocity of the air relative to the computational mesh (which follows the
piston motion). As observed both models predict similar temperature fields.
It is interesting to notice that the shape of the contour surfaces is similar in both figures, with the hot
core of the gas being restricted in the middle of the piston bowl and the temperature drops when
moving towards the cylinder walls. As far as the velocity field is concerned, it is noticed that the
quasi-dimensional model predicts the existence of a vortex in the region upon the cylinder crown and
the flow follows a path from the periphery of the cylinder to the inner of the piston bowl (as sketched
in Fig. 8a).
Fig. 8a. Calculated Temperature and Velocity field (uniform vector length) with indicative flow path
lines at 220 CA degs ABDC, at 2000 rpm engine speed, using the quasi-dimensional model
Fig. 8b. Calculated Temperature and Velocity field (uniform vector length) at 220 CA degs ABDC, at
2000 rpm engine speed, using the CFD model
These results are in accordance to the predictions of the CFD model (Fig. 8b), although the
comparison is only qualitative. Taking into account the simplicity of the phenomenological sub-model
used to calculate the velocity field when using the quasi-dimensional model and comparing it to the
detailed procedure followed by the CFD model, the produced result is considered satisfactory. It
should be mentioned that the alternative procedure followed by most phenomenological models in
order to describe the flow field inside the combustion chamber of an engine with bowl in piston
design, would be to pre-define that the flow would have only a radial component in the region above
the piston crown and an axial component inside the piston bowl. In previous studies, it has been
shown by the authors [19], that the calculated mean mass flow rates through certain reference planes
using the quasi-dimensional and the CFD model are similar, however the predicted velocity field
using the quasi-dimensional model could by no means thought to be as accurate as the one obtained
using the CFD model.
4.4. Comparison of computational time.
A significant parameter taken into account in the present investigation is the computation time
required by each model to complete the engine closed cycle simulation. As mentioned earlier, the
limitations and advantages of the 1-D models are well known to the research community, however,
there are applications where the significantly lower computational time required by the quasi-
dimensional model makes it preferable compared to the more accurate CFD model. This does not
mean that the quasi-dimensional model can directly substitute the CFD model, but sometimes (i.e.
engine cycle parametric studies) it can be proved to be a fair compromise.
In Table 6 are given the actual simulation times required by each model examined to fully complete
the engine closed cycle simulation, at all engine speeds. It should be mentioned that for the CFD
model the computational time decreases as the engine speed increases. All the simulation runs have
been accomplished on a PC with a CPU running at 2.2 GHz.
Table 6. Simulation times of the 1-D, Q-D and CFD model at various engine speeds
Engine speed 1-D (sec) Q-D (sec) CFD (sec)
1200 rpm ~1 360 72x10
3
1500 rpm ~1 360 58x10
3
2000 rpm ~1 360 52x10
3
3000 rpm ~1 360 43x10
3
As was expected, the CFD model requires much more computational time than the other two models
(it is 120 to 200 times more time consuming than the quasi-dimensional model), but it is the most
detailed one, proper for fundamental research. The 1-D model is very fast, it is ideal for quick
parametric investigations, however it has certain limitations on the extent of its usage. Finally, the
quasi-dimensional model is 360 times more time consuming than the 1-D model, but in a generally
acceptable time interval it manages to predict reliably the thermodynamic properties of the gas on a
cycle basis and also provides a rough estimation of the in-cylinder temperature and velocity field.
5. Conclusions
In the present study, three engine simulation models (1-D, Q-D and CFD) have been applied to
simulate the closed part of the engine cycle of a HSDI Diesel engine, running under motoring
conditions at various engine speeds. The single zone model calculates with reasonable accuracy the
in-cylinder pressure, while it rather underestimates the mean gas temperature near the top dead centre
(TDC), compared to the other two models (quasi-dimensional and CFD one) at all engine speeds
examined. Among the positive features of this type of model is definitely the very low computational
time, which is approximately equal to 1 sec for the simulation of the engine closed cycle, when
running on a personal computer. On the other hand the computational time required by the quasi-
dimensional model is approximately 6 minutes to simulate the closed part of the engine cycle,
whereas the CFD model needs up to 20 hours. Additionally, the quasi-dimensional model provides
information for the local in-cylinder temperature distribution and describes qualitatively correctly
how the cylinder geometrical design affects the in-cylinder flow field and spatial temperature
distribution, as revealed by comparing its results with the pertinent obtained by the more accurate and
time consuming CFD model.
Moreover, comparing the quasi-dimensional with the 1-D model, it calculates the cylinder pressure
trace closer to the corresponding measured one, while the calculated mean cylinder temperature is
closer to the one obtained using the CFD model, at all cases examined. Concerning the heat transfer
process, the calculated heat transferred through the cylinder walls during the compression phase is
almost identical between the quasi-dimensional and the CFD model at all cases examined (with the
exception of the case of 3000 rpm engine speed, where a small deviation is observed). Thus, the
timing and the peak value of the heat transferred is accurately predicted. During the expansion phase,
a discrepancy is observed between the calculated heat transferred through the cylinder walls using the
CFD and the other two models, with the quasi-dimensional giving much closer predictions to the ones
obtained using the CFD model, than the 1-D model.
Based on the previous analysis, it can be concluded that the usage of the quasi-dimensional model
might be beneficial as an intermediate tool to appropriately analyze, on an engine cycle basis, the
effect of combustion chamber geometry on the physical processes taking place inside the cylinder. In
these applications, the 1-D model fails, since it does not take into account the design of combustion
chamber, although it seems capable of predicting the effect of engine operating parameters (engine
speed in the present investigation) on engine performance (cylinder pressure trace). On the other
hand, for a detailed investigation of how each design parameter of the combustion chamber affects the
in-cylinder flow-field, and consequently all the relevant physical processes, the CFD model is the
most appropriate. On a real case, where a parametric investigation is to be conducted, it could be
preferable to use the quasi-dimensional model, to determine a limited number of cases with higher
interest, and then focus on these cases, by applying the CFD code, to produce more detailed and
accurate results.
In a future work it is interesting to compare the quasi-dimensional model with a multi-zone
phenomenological model and a CFD one, applied to simulate a diesel engine running under firing
conditions. In this situation it will be possible to explore the beneficial characteristics of the quasi-
dimensional model, on the description of the air-fuel mixing mechanism compared to the
conventional multi-zone model, and determine to which extent its usage would be preferable.
Acknowledgments
The authors would like to thank A.M. Dimaratos for his significant contribution in the experimental
facilities of the present study. Also, G.M. Kosmadakis wishes to thank the Greek State Scholarships
Foundation for granting him a Ph.D. research scholarship.
Abbreviations
ABDC after bottom dead center
ATDC after top dead center
BBDC before bottom dead center
BTDC before top dead center
o
CA degrees of crank angle
CFD computational fluid dynamics
CPU central processing unit
IVC inlet valve closure
PISO pressure implicit splitting of operators
rpm revolutions per minute
SIMPLE Semi-Implicit Method for Pressure Linked Equations
TDC top dead center
References
[1] Hountalas DT, Pariotis EG. A simplified model for the spatial distribution of temperature in a
motored DI diesel engine. SAE Paper no. 2001-01-1235; 2001.
[2] Pariotis EG, Hountalas DT. A new quasi-three dimensional combustion model for prediction of
DI diesel engines performance and pollutant emissions. SAE Paper no. 2003-01-1060; 2003.
[3] Pariotis EG, Hountalas DT. Validation of a newly developed quasi-dimensional combustion
model application on a heavy duty DI diesel engine. SAE Paper no. 2004-01-0923; 2004.
[4] Rakopoulos CD, Kosmadakis GM, Pariotis EG. Critical evaluation of current heat transfer
models used in CFD in-cylinder engine simulations and establishment of a comprehensive wall-
function formulation. Appl Energy 2010;87(5):161230.
[5] Rakopoulos CD, Kosmadakis GM, Demuynck J, De Paepe M, Verhelst S. A combined
experimental and numerical study of thermal processes, performance and nitric oxide emissions
in a hydrogen-fueled spark-ignition engine. Int J Hydrogen Energy 2011;36(8):516380.
[6] Rakopoulos CD, Kosmadakis GM, Dimaratos AM, Pariotis EG. Investigating the effect of
crevice flow on internal combustion engines using a new simple crevice model implemented in
a CFD code. Appl Energy 2011;88(1):11126.
[7] Rakopoulos CD, Kosmadakis GM, Pariotis EG. Evaluation of a new computational fluid
dynamics model for internal combustion engines using hydrogen under motoring conditions.
Energy 2009;34(12):215866.
[8] Rakopoulos CD, Kosmadakis GM, Pariotis EG. Evaluation of a combustion model for the
simulation of hydrogen spark-ignition engines using a CFD code. Int J Hydrogen Energy
2010;35(22):1254560.
[9] Rakopoulos CD, Kosmadakis GM, Pariotis EG. Simulation of a motored internal combustion
engine using an improved CFD code. In: Proceedings of the 23
rd
International Conference on
Efficiency, Cost, Optimization, Simulation and Environmental Impact of Energy Systems
(ECOS 2010), 2010 June 1417, Lausanne, Switzerland, Paper No. 295.
[10]Borgnakke C, Arpaci VS, Tabaczynski VS. A model for the instantaneous heat transfer and
turbulence in a spark ignition engine. SAE Paper no. 800287; 1980.
[11]Nefischer A, Hallmannsegger M, Wimmer A, Pirker G. Application of a flow field based heat
transfer model to hydrogen internal combustion engines. SAE Int J Engines 2009;2(1):125163.
[12]Annand WJD. Heat transfer in the cylinders of reciprocating internal combustion engines. Proc
Inst Mech Engrs 1963;177:97390.
[13]Rakopoulos CD, Antonopoulos KA, Rakopoulos DC, Hountalas DT. Multi-zone modeling of
combustion and emissions formation in DI diesel engine operating on ethanol-diesel fuel blends.
Energy Convers Manage 2008;49(4):625-43.
[14]Demirdzic I, Gosman AD, Issa RI, Peric M. A calculation procedure for turbulent flow in
complex geometries. Computer Fluids 1987;15(3):25173.
[15]Han Z, Reitz RD. Turbulence modeling of internal combustion engines using RNG k- models.
Combust Sci Technol 1995;106(4):26795.
[16]Jennings MJ, Morel T. Observations on the application of the k- model to internal combustion
engine flows. Combust Sci Technol 1988;58(1):17793.
[17]Ferziger JH, Peric M. Computational methods for fluid dynamics. Berlin, Germany: Springer;
2002.
[18]McBride BJ, Zehe MJ, Gordon S. NASA Glenn coefficients for calculating thermodynamic
properties of individual species. NASA/TP2002-211556; September 2002 Available at:
<http://gltrs.grc.nasa.gov/reports/2002/TP-2002-211556.pdf> [accessed 04.02.2011].
[19]Rakopoulos CD, Kosmadakis GM, Pariotis EG. Investigation of piston bowl geometry and speed
effects in a motored HSDI diesel engine using a CFD against a quasi-dimensional model.
Energy Convers Manage 2010;51(3):47084.

S-ar putea să vă placă și