See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/245353880

In-Cylinder Flow Computational Fluid Dynamics Analysis of a Four-Valve

Spark Ignition Engine: Comparison Between Steady and Dynamic Tests

Article in Journal of Engineering for Gas Turbines and Power · May 2010
DOI: 10.1115/1.4000265

CITATIONS READS

8 504

2 authors:

Damian Ramajo Norberto Nigro

Universidad Nacional del Litoral, Santa Fe, Argentina - CONICET National Scientific and Technical Research Council
47 PUBLICATIONS 108 CITATIONS 224 PUBLICATIONS 1,198 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Development of a robust computational solver for Fluidized Bed applications View project

VOF-PLIC solvers - Granular Flow - Mesh dynamics View project

All content following this page was uploaded by Norberto Nigro on 09 January 2016.

The user has requested enhancement of the downloaded file.

 IN-CYLINDER FLOW CFD ANALYSIS OF A 4-VALVE SPARK IGNITION ENGINE. COMPARISON
BETWEEN STEADY AND DYNAMIC TESTS

Damian E. Ramajo and Norberto M. Nigro

International Center for Computational Methods in Engineering (CIMEC)

INTEC-Universidad Nacional del Litoral-CONICET, Güemes 3450
S3000GLN Santa Fe, Argentina, [email protected]

ABSTRACT− Numerical and experimental techniques were applied in order to study the in-cylinder flow field in a
commercial 4-valve per cylinder spark ignition engine. Investigation was aimed at analyzing the generation and
evolution of tumble-vortex structures during the intake and compression strokes, and the capacity of this engine to
promote turbulence enhancement during tumble degradation at the end of the compression stroke. For these purposes,
three different approaches were analyzed. Firstly, steady flow rig tests were experimentally carried out, and then
reproduced by CFD. Once CFD was assessed, cold dynamic simulations of the full engine cycle were performed for
several engine speeds (1500 rpm, 3000 rpm and 4500 rpm). Steady and cold dynamic results were compared in order
to assess the feasibility of the former to quantify the in-cylinder flow. After that, combustion was incorporated by
means of a homogeneous heat source and dynamic boundary conditions were introduced in order to approach real
engine conditions. The combustion model estimates the burning rate as a function of some averaged in-cylinder flow
variables (temperature, pressure, turbulent intensity and piston position). Results were employed to characterize the in-
cylinder flow field of the engine and to establish similarities and differences between the three performed tests that are
currently used to estimate the engine mean-flow characteristics (steady flow rig, cold dynamic and real dynamic
simulation).

KEY WORDS: dynamic engine simulation, steady flow rig tests, CFD, tumble, turbulence

1. INTRODUCTION

In-cylinder charge motion has been steadily gaining importance since the introduction of new technologies like GDI
(gasoline direct injection) or HCCI (homogeneous charge compression ignition). Understanding the behavior of in-
cylinder flow structures is the first step to efficiently control fuel stratification, turbulence generation and heat
losses. These factors play a crucial role on fuel economy, emissions and engine performance.
For years, engine development has been mainly supported by experimental tests like dynamometric and steady flow
rig tests. Latterly, the evolution of optical techniques has allowed sights of the inside of optical access engines.
However, this kind of tests are tedious, expensive and limited to low engine speeds. Moreover, they are restricted to
a small portion of the whole cylinder. On the other hand, steady rig tests are easier, faster and cheaper, but they yield
limited information about engine performance and in-cylinder flow characteristics under real conditions.
Engine research by computational techniques started by means of 0/1-dimensional (0/1-D) simulators. Nowadays,
they are massively employed to study the overall engine gas dynamics, but in-cylinder studies are gaining more
relevance due to the necessity to improve combustion and reduce emissions. For this reason, computational fluid
dynamics (CFD) became a useful tool to design complex engine components like combustion chambers, manifolds
and injectors. During the last years CFD has been significantly improved, especially thanks to High Performance
Computing (HPC), allowing more real engine simulations. Nowadays, efforts are focused on simulating injection of
fuel, combustion and chemical species evolution for emission control. But, on the other hand, experimental
techniques like steady rig and cold dynamic tests (optical) are daily employed. Several papers report CFD
simulations of intake-compression or compression-expansion strokes (Bailly et al., 1999; Weng et al., 1999;
Matarelli, 2000; Hong and Tarng, 2001; Li et al., 2001; Stan et al., 2003; Tan and Reitz, 2006; Ramajo et al., 2007),
but only a few deal with the whole engine cycles (Bianchi et al., 1998; Johan et al., 2001; Fontana et al., 2003; Cao
et al., 2004; Fontana et al., 2006; Mahrous et al., 2007). One possible explanation might be found on the difficulty to
handle the extreme mesh deformations around valves and the topological changes during valves opening and
closing. Moreover, combustion model estimations are far from accurate, so cold in-cylinder simulations are the first
step to understand in-cylinder flow dynamics.
In this work experimental and CFD steady tests were carried out. Subsequently, the in-cylinder flow behaviors
under cold and real dynamic conditions were solved by CFD and results were compared to the steady ones,
concluding about the usefulness of steady and dynamic tests.

2. METHODOLOGY

A commercial 16-valve 4-cylinder spark ignition Fiat Torque engine was employed. Table 1 consigns the main
constructive and operative characteristics of the engine. Crank angles (CA) are given with reference to the top dead
center (TDC) at the beginning of the intake stroke. Steady test conditions are not standardized, so it is important to
describe the methodology to carry on perform the different tests. As regards the performed steady test simulations,
they reproduced the experimental conditions as exactly as possible.
Three kinds of steady tests were carried out: flow-meter, swirl and tumble tests. For all tests, a constant pressure
drop of 6227 Pa (equivalent to 25 inches of water) (SuperFlow SF-600) was applied to the cylinder head to induce
flow. A rotating-honeycomb swirl meter (MS) (AUDIE Technology) was employed to perform the angular
momentum measurements.

Table 1. Constructive and operative data of the engine.

Cylinder Valves characteristics
Fuel Gasoline Intake Exhaust
Number of cylinders 4 Diameter mm 30.4 29.9
Piston displacement 1582cm3 Max. lift mm 9 at 8.5 at
Bore 87 mm at CA 102.5º 618.5º
Opening
Stroke 68 mm 255º 253º
period (CA)
Compression ratio 10.5:1 OVA (CA) 25º 48º
Connecting rod length 118 mm CVR (CA) 50º 25º

2.1. Experimental procedure. Steady tests

As regards flow discharge-coefficients both the intake and the exhaust systems were studied. Flow entered the
cylinder through the intake system, and left the cylinder through the exhaust. In order to reduce flow turbulence and
wave effects, a 100 mm cylinder was placed between the cylinder head and the flowbench. Additionally, a nozzle
and a diffuser were placed at the beginning and at the end of the intake and exhaust ducts, respectively.
For swirl tests the swirl meter was inserted between the cylinder and the flowbench. As regards tumble tests, they
were performed through a L-shape tumble bench (figure 1) of our own design and based on CFD analysis (Ramajo,
2008).

Possible swirl meter locations

Bridle to mount
the cylinder head
MS MS MS
Blind cap in P2 in P3 in P4

φ 87

155 25 95 95

Figure 1. Sketch of the tumble bench employed.

The employed tumble bench has some interesting characteristics; firstly, the lateral duct has a diameter larger
than the cylinder bore, thus reducing tumble loss. Moreover, tumble loss is fairly linear along the lateral duct.
Secondly, the fact that the tumble bench has only one lateral duct produces more intense angular momentum.
Thirdly, the tumble loss at the joint between the cylinder and the lateral duct is negligible, so tumble below the
cylinder head can be estimated by linear extrapolation with two measurements along the lateral duct (P2 and P4, for
example). Finally, in general L-shape benches are asymmetric with respect to the cylinder mean plane, so one of the
intake valves produces more tumble than the other (Grimaldi et al., 2004). This can be observed by measuring the
tumble generated while one of both valves is kept closed. However, no differences were experimentally observed
for this tumble bench. That is, both valves produced the same tumble regardless of which one was closed. That
might be explained by the significant distance that exists from the lateral duct to the bridle (30 mm), and the fact
that the joint between the cylinder and the lateral duct is symmetric. From experimental results it was found that this
tumble bench introduces a pressure drop, reducing the mass flow rate at maximum valve lift in 7%.

2.2. Numerical procedure

2.2.1. Computational model

Due to the narrow dimensions of the engine ports, in order to obtain the engine geometry it was necessary to
make positive replicas from the intake and exhaust ports (low-contraction silicone) and the combustion chamber
(high-rigid polyester resin). Afterwards, replicas were measured by means of a 3-dimensional rotary laser scanner
(Roland LPX-250), obtaining around 550.000 points over the surfaces of replicas. Pathlines were drawn from
selected points (GID 7.2) and then exported to ANSYS-ICEM 10.0 to generate the overall surface by adding small
patches (figure 2).
 Inletpor Exhaus

Figure 2. Computational model.

The meshes were generated in ANSYS-ICEM 10.0. The optimum element size – 1.1 mm (Ramajo, 2008) – was
obtained through a mesh convergence study. Around 2x10 6 tetrahedral elements were required to get an accurate
description of the layout of the steady rig tests.

2.2.2. Steady test simulations

Results were obtained by using the incompressible and isothermal Navier-Stokes equations. Although air is
clearly a compressible fluid, due to the relatively low flow velocities close to the valves (maximum Mach number
around 0.2) the compressible effects on the air are less than 4%, therefore incompressible formulation is appropriate
in this context. Turbulence was modeled through a standard k-ε model. Simulations were unsteady, so a suitable
time step was obtained through a time-convergence study, this being 5x10-5 seconds (Ramajo, 2008). At least 1000
time steps were solved for each valve lift. Results were obtained as a time average of 10 solutions in time along the
simulated period.
Steady test simulations reproduced the experimental conditions. Cylinder head was subjected to a pressure drop
that was progressively increased from 0 to 6227 Pa (25 inches of water) in order to avoid overflow problems.
A boundary condition of total pressure was fixed at the inlet, while an opening condition was employed for the
outlet (this allows the flow to leave or to enter the domain according to pressure conditions). A non slip boundary
condition was fixed at walls, and a roughness of 0.2 mm was applied at the intake and exhaust duct walls.
Models were solved with distributed calculus in a Beowulf cluster with 20 single-core processors; around 24
hours were necessary to obtain each valve lift result.

2.2.3. Dynamic test simulations

For dynamic tests simulation the compressible Navier-Stokes equations were solved, including the thermal
equation for both cold and real dynamic simulations. Cold dynamic simulations were performed under atmospheric
pressure and room temperature conditions at the inlet and outlet boundaries and room temperature was also set at the
engine walls. Although these engine test conditions are far from the real ones, experimental tests under these
conditions are usually carried on in order to visualize the in-cylinder flow patterns and measure the vortex dynamics,
so cold versus real dynamic simulations can be useful to extract conclusions about the suitability of the firsts.
Tackling the dynamic simulation of the whole engine cycle involves at least two main difficulties: the mesh
deformation and the topological changes of the geometry. Mesh deformation was handled by a Laplacian algorithm
that at each time step computes the displacement of the inner nodes due to the imposed motions at some boundary
walls. Moreover, that was combined with remeshing after several time steps, thus preserving mesh quality.
Seventeen remeshes were needed to complete the whole engine cycle. The mesh deformation algorithm of ANSYS-
CFX 10.0 did not prove to be robust enough, so better algorithms, like the one introduced by Lopez et al. (2008), are
being considered in order to reduce the remeshing steps from 17 to only 4.
Simulations start at the top dead center (TDC) when all valves are opened and the intake and exhaust systems are
connected to the cylinder (domain ICE at figure 3). Then, the exhaust valves close and the exhaust duct is
disconnected to the cylinder. After that, the exhaust system is solved separately. Six different IC domains are were
needed to represent the intake process (from 0 to 208 CA). Once the intake valves are completely closed (at 208
CA), three isolated domains are solved (I, C and E). At 492 CA degrees the exhaust valves are opened, connecting
the exhaust duct to the cylinder (domain CE), so initial conditions have to be fed from the isolated domains C and E.
For this purpose, a User External Fortran Routine (UEFR) was incorporated to ANSYS-CFX. This routine uses a
directed-search algorithm to find the cell of the C domain or E domain where each one of the nodes of the
assembled CE domain is located. Then, a finite element interpolation is employed to find initial conditions for
temperature, pressure and turbulence (Ramajo, 2008). The assembled methodology is also applied when the intake
valves are opened.
The extreme deformation of the mesh around the valve seats is a really hard problem for the mesh deformation
algorithm. So, the minimum valve lift was limited to 0.6 mm. That is, the domains were connected or disconnected
when the valve lifts reached a lift threshold of 0.6 mm. Although this reduces the valve opening period, the mass
flow rate at low valve lifts can be neglected.
Even though the simulation of several isolated domains is laborious, it allows the use of an increased time step to
simulate the intake and exhaust systems due to the low gas velocity inside ducts, thus reducing the cost of
simulations without introducing convergence problems.
The real valve displacements were computed through the well known crank-slider mechanism expression
(Heywood, 1988) and then introduced to simulations through an UEFR.
The time step at 1500 rpm was 5x10-5 seconds (the same as for steady tests), while it was reduced to 2.5x10 -5
seconds at 3000 rpm and 1.66x10-5 seconds at 4500 rpm in order to keep the Courant number almost constant.
 Domain IC
Domain ICE

Domain C

Domain CE

Domain E
Domain I

Figure 3. Different kinds of domains employed to simulate the whole engine cycle.

Depending on the mesh sizes, between 8 and 20 cluster processors were employed to solve the dynamic
simulations. The simulation of one engine cycle at 1500 rpm demanded about 40 hours, this value increasing to 80
hours at 3000 rpm and to 120 hours at 4500 rpm.

2.2.4. Combustion model

For homogeneous charge spark ignition engines (HCSI), combustion takes place in a very thin zone (a few
millimeters) where flame propagation is fairly laminar. Turbulence distorts, stretches and corrugates the thin
combustion zone, increasing the flame front surface area and enhancing the mixing of reactants. Combustion
phenomenon can be classified under the turbulent premixed combustion flamelet regime. This regime has the
advantage of allowing the flow behavior and the combustion phenomenon to be separately solved, only introducing
the combustion effect by means of a heat source in the energy balance equation (Williams, 1985; Tan and Reitz,
2006).
In this work, a phenomenological simple combustion model was implemented in order to roughly quantify the
effect of combustion (change on temperature, pressure and gas properties) over the in-cylinder flow behavior. As
expected, combustion increases gas temperature and pressure affecting mainly gas viscosity, thermal conductivity
and heat capacity. But the most evident effect is that the in-cylinder pressure increment that affects the in-cylinder
mass flow rate during the intake and exhaust processes, modifying the in-cylinder flow structures. The combustion
model was incorporated through an UEFR. This takes into account the main in-cylinder flow variables (average
pressure and temperature and the turbulence quantities k and ε), applying a simple fractal model to estimate the
turbulent burning velocity (St):

D3 − 2
St At  le 
= ≈  , (1)
Sl Al  li 
being Sl the laminar burning velocity, At and Al the turbulent and laminar flame front surface areas, and le and li the
outer and inner turbulent spatial cutoff scales (Gouldin, 1987). Finally, D3 is a fractal dimension. The scales le and li
are related to the biggest and the smallest turbulent structures of the flow. le and li were assumed as the integral scale
and the Kolmogorov one, respectively. The fractal dimension D3 was calculated through a correlation proposed by
North and Santavicca (1990):

u´ Sl
D3 = C1 + C2 . (2)
u´+ Sl u´+ Sl

In equation (2), u´ is the turbulent intensity and C1 and C2 are model constants (C1 = 2.35 and C2 = 2.0). Returning
to equation (1), the well-known correlation proposed by Meghalchi and Keck (1982) was employed to estimate the
laminar burning velocity (Sl):

α β
T   p
S l = S l 0  u    (1 − 1.5Yres ) , (3)
 T0   p0 

being Sl0 a reference laminar burning velocity related with the fuel and the stoichiometric ratio of the mixture, Tu the
temperature of the unburned gas and Yres the residual gas fraction. Sl0, α and β are obtained from:

S l 0 = B1 + B2 ( ϕ~ − ϕ 0 )
2

α = 2.18 − 0.8( ϕ~ − 1) . (4)

β = −0.16 + 0.22( ϕ~ − 1)

Due to the combustion model not being a true multi-zone model, there is no real interface separating burned and
unburned gases. So, an algebraical estimation of the laminar flame front surface area (Al) must be performed,
assuming that an imaginary flame is spherically growing from the spark plug (St is isotropic). Then, the estimation
of Al must be performed in order to know how much fuel is being burned at each time step due to the flame front
• •
advance. That allows the introduction of a homogeneous heat source xb in Navier-Stokes equations. xb has the

following form:

• m fuel E fuel
xb = S t . Al .Q fuel = S t . Al . , (5)
Vcyl

being mfuel the mass of fuel at the ignition time, Efuel the specific heat of the fuel and Vcyl the current volume of the
combustion chamber (Vcyl = f(t))
Fluid properties notoriously change with temperature variations, so some suitable correlations were introduced to
estimate the constant pressure heat capacity (Cp), the thermal conductivity (λ) and the dynamic viscosity (µ).
As it was previously mentioned, cold dynamic simulations were performed under atmospheric conditions at inlet
and outlet. However, for real dynamic simulations constant boundary conditions are far from realistic because
combustion induces significant temperature and pressure waves inside the ducts. Moreover, the computational
model considers only a fraction of the overall engine, so the influence of the other cylinders has to be properly
represented. For these two reasons, the overall engine was modeled by 0/1-D engine software of our own
development, (http://www.fceia.unr.edu.ar/fceia1/mecanica/Automotores/laboratorio_de_automotores_index.htm,
López and Nigro, 2009). The 0/1-D engine simulator allows to handle the overall layout of pluri cylinder engines,
modeling ducts by one-dimensional mass, momentum and energy Euler equations, adding appropriate terms for
taking into account heat transfer and frictional losses, and solving by finite element method. As for cylinders,
plenums and tanks they are approximated by zero-dimensional mass and energy balances. Finally the in-cylinder
heat release (combustion) is modeled by the Wiebe function (Heywood, 1988). 0/1-D Simulator parameters were
tuned by fitting the power and torque curves of the engine along the overall engine speed range (1000 to 6500 rpm),
using typical SI engine parameters for the Wiebe equation (a=2.0, m=5.0, ∆φ=80 CA and φ0=340 CA). Once a good
agreement was reached, dynamic boundary conditions (pressure, temperature and turbulence) were extracted from
the 0/1-D results and incorporated to the 3D computational model by means of an UEFR. 3D simulations were
carried on and the combustion durations estimated from the CFD combustion model were feed on the 0/1-D
simulator to improve the estimation of the dynamic boundary conditions for the 3D simulations.

3. RESULTS AND DISCUSSION

This section shows numerical and experimental steady tests results. Then, the results concerning dynamic test
simulation are presented focusing on describing the main flow characteristics and averaged flow quantities. Finally,
steady and dynamic results are compared so as to evaluate the suitability of the former and the in-cylinder flow
behavior.

3.1. Numerical and experimental steady rig tests

3.1.1. Flowmeter tests

Figure 4 shows numerical and experimental results of flow discharge coefficients for the intake (called test 1) and
exhaust (called test 2) systems. Note that an acceptable concordance is obtained between CFD and experimental
results. Differences are bigger at the mean valve-lift range, the average relative errors being closer to 6.5%, and the
maximum relative errors around 10%. Both the discharge (CD) and flow (Cf) coefficients are displayed because the
former has more sensitivity at low valve lifts (CD is calculated based on the current valve curtain area), while the
second (Cf) better reflects the behavior of the overall system at high valve lifts (Cf is calculated based on the constant
cross section area of the valve seat) (Xu, 2001).
 Experimental (Cf)
Experimental (CD)
0.8
Numeric (Cf)
0.7 Numeric (CD)
0.6

Flow coefficients
0.5
0.4
0.3 Test 1
0.2
Intake system
0.1
0.0
1 2 3 4 5 6 7 8 9 10 11
Intake valve lift (mm)

Experimental (Cf)
0.8 Experimental (CD)
Numeric (Cf)
0.7
Numeric (CD)
0.6
Flow coefficients

0.5
0.4
0.3
Test 2
0.2
Exhaust system
0.1
0.0
1 2 3 4 5 6 7 8 9 10 11
Exhaust valve lift (mm)

Figure 4. Discharge coefficients (CD and Cf). Upper: intake system. Bottom: exhaust system.

A comparison of the flow coefficients (Cf) of both systems showed that the exhaust one has a better efficiency at
low valve lifts (below 4.5 mm), while the intake system is 17% more efficient at maximum valve lift. For this
engine, the maximum valve lifts are 9 mm and 8.5 mm for intake and exhaust systems respectively (Table 1).
Regarding the flow efficiency, these valve timings seem to be close to the optimum. The gain in mass flow rate is
only 2.5% when the intake valve lift is increased from 9.09 mm to 10.6 mm, while for the exhaust system the
improvement on Cf is less than 1%.

3.1.2. Swirl tests

Typical 4-valve cylinder engines produce negligible swirl, so 2 timing strategies were employed to explore the
potentiality of this cylinder head to produce swirl. Firstly, one intake valve was kept completely closed while the
other was opened (called test 3). Secondly, both valves were opened but one of them was opened 1.51 mm more
than the other (called test 4). In order to compare numerical and experimental results, the measured angular honey
comb speed from the swirl meter was converted to volume-average angular momentum (kg/ms) applying the rigid
body assumption for the in-cylinder flow:

R
Bo 4
Ms = ∫ ρ .V .r.dV = ∫ 2.π .r.l.ρ .r.ω.r.dr = π .l.ρ .ω
vol
t
0
32
, (6)

being Bo the cylinder bore, ρ the fluid density and l the cylinder height (100 mm) (Ramajo 2008). Figure 5 shows
the swirl and mass flow rate results of both timing strategies.
 0.5
Test 3 (experimental)

Swirl momentum (kg/ms)

Test 3 (numeric)
0.4
Test 4 (experimental)
Test 4 (numeric)
0.3

0.2

0.1

0
0 1 2 3 4 5 6 7 8 9 10 11
Right-valve lift (mm)

Test 3 (experimental)
0.1 Test 3 (numeric)
Test 4 (experimental)
Mass flow rate (kg/s)

0.08 Test 4 (numeric)

0.06

0.04

0.02

0
0 1 2 3 4 5 6 7 8 9 10 11
Right-valve lift (mm)

Figure 5. Numerical and experimental results from swirl tests. Upper: swirl momentum. Bottom: mass flow rate.

Correlation between numerical and experimental data is good enough. However, differences on mass flow rate
increase for the second timing strategy (test 4). A look at the upper part of figure 5 shows that swirl generation is
fairly linear when only one valve is opened (test 3), even though at high valve lifts the mass flow rate keeps almost
constant. One possible explanation for this could be found in the flow redistribution around the valve curtains while
the valve lift is increased. Small valve lift differences between the intake valves (test 4) induce little swirl at medium
valve lifts, but once the valve lift reaches 7.57 mm (starting the asymptotic mass flow rate zone) the mass flow rate
through both valves becomes almost the same and swirl generation falls down. Grimaldi et al. (2005) obtained
similar results in experimental tests and Mahrous et al. (2007) found that valve-lift differences greater than 100%
are required to induce significant swirl by means of dynamic tests (CFD).
Figure 6 shows numerical results of the mass flow distribution around one valve curtain at 10.6 mm. Note that, if
only one valve is opened (test 3), flow turns to the 2-3 direction, clearly influenced by the corresponding intake
duct. On the other hand, if both valves are equally opened (test 1) the flow turns to the middle zone between valves.
Similar conclusions have also been reported in other papers (Mattarelli, 2000; Grimaldi et al., 2004).

Flow entering
through the
valve positive 3 4
5
curtain 2 6

1 7
Flow entering 12 8
through the 11 10 9
valve negative
curtain
 Figure 6. Mass flow rate distribution around one valve curtain when only one valve is opened (dashed line) and
when both valves are equally opened (solid line).

The honey comb of the swirl meter introduces negligible pressure drop. A comparison of the results obtained with
and without the device showed that the mass flow rate loss at maximum valve lift was less than 0.4% (Ramajo
2008).

3.1.3. Tumble tests

Figure 7 shows numerical and experimental results obtained at 2 locations (P 2 and P4 in figure 1) along the lateral
duct of the tumble bench. Test 5 and Test 6 refer to P 2 and P4 locations, respectively. A good agreement at low valve
lifts at both locations is found. Moreover, the tumble fall at medium valve lifts is also well predicted by CFD.
However, at the location P2 (Test 5) tumble is significantly overestimated, being the maximum relative error closer
to 61% (at 7.57 mm valve lift) and the average relative error 25%. As regards position P 4 (Test 6), the maximum
relative error reaches 32% (also at 7.57 mm valve lift), and the average one is 32%. Note that even though
discrepancies between numerical and experimental results are noticeable similar numerical investigations had
yielded errors closer to 100% (Grimaldi et al., 2003; 2004). Simulations showed that the tumble falling effect also
occurs inside the cylinder, but curiously it is weaker than the measured in the lateral duct of the tumble bench. The
results at P2 and P4 positions show a similar behavior indicating that momentum loss is fairly linear along the lateral
duct of the tumble bench.
Note that tumble generation is almost linearly proportional to valve lift, except at medium valve-lift range, where
the tumble falling effect takes place.

Test 5 (experimental)
1 Test 5 (numeric)
Tumble momentum (kg/ms)

Test 6 (experimental)
0.8 Test 6 (numeric)
Cylinder (numeric)
0.6

0.4

0.2

0
0 1.51 3.03 4.54 6.06 7.57 9.09 10.6

Intake valve lift (mm)

Figure 7. Numerical and experimental results from tumble tests for P2 and P4 measurement positions.

Figure 8 displays numerical results from four strategies to produce angular momentum (swirl or tumble) using
different timing configurations.
 0.9 full tumble
0.8 full swirl

0.7 tumble valve retard = 1.51mm

swirl valve retard = 1.51mm

Momentum (kg/ms)
0.6
0.5
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8 9 10 11
Intake valve lift (mm)

0.12 full tumble

full swirl
0.1 tumble valve retard = 1.51mm
Mass flow rate (kg/s)

swirl valve retard = 1.51mm

0.08

0.06

0.04

0.02

0
0 1 2 3 4 5 6 7 8 9 10 11
Intake valve lift (mm)

Figure 8. Numerical results for four strategies to produce angular momentum (swirl or tumble). Upper: angular
momentum. Bottom: mass flow rate.

The full tumble strategy (valves equally opened) seems to be the most effective option to induce momentum. On
the other hand, a small lift difference (of 1.51 mm) smooths tumble production at mean valve lifts, but also produces
a significant reduction on mass flow rate at low and medium valve lifts. If only one valve is opened (full swirl), the
maximum momentum is less than a half that for the full tumble strategy. Also, the mass flow rate reduces more than
50%. This comparative analysis allows us to estimate the intensity of the flow structures (tumble and swirl) that
might be induced using variable valve lift (VVL). As shown in several papers, steady tests are the first step to design
valve timing strategies to control the in-cylinder flow motion (Wilson et al., 1993; Geiger et al., 1999; Li et al.,
2001; Grimaldi et al., 2004; 2005).
Any valve opening configuration that improves tumble or swirl will probably negatively affect the volumetric
efficiency of the engine. On the other hand, the effect of introducing adjustable tumble deflectors inside the intake
port of this engine has been numerically evaluated (Ramajo, 2008), enhancing tumble by 18%, but reducing the
mass flow rate by 10%. In regards of swirl, the use of a swirl deflector produced swirl values closer to the
previously reported for test 3, but enhancing the mass flow rate by 46% with respect to test 3.

3.2. Numerical dynamic tests

3.2.1 Cold dynamic simulations

Three engine speed simulations were performed to cover almost the whole engine speed range (1500 rpm, 3000
rpm and 4500 rpm). Initial conditions were atmospheric over a quiescent flow,
but steady results were quickly obtained after the fourth cycle. Of course, this fast convergence probably will be
slower for multi-cylinder engines, due to the combination of pressure waves generated by each cylinder inside the
intake and exhaust systems.

102 CA 142 CA

200 CA
256 CA

Figure 9. Evolution of flow during the intake and compression strokes on a plane cutting one intake valve.

Figure 9 shows the evolution of the velocity pattern on a cutting plane at the intake valve, during the intake and
compression strokes. Note that vortex dynamics is very complex. Structures are constantly created and destroyed. In
figure 9, the main structures have been identified using circles for clockwise rotation sense vortexes (positive
tumble) and squares for counter clockwise ones (negative tumble). Moreover, structures are classified according to
its size (big with solid; small with hollow).
Figure 10 shows the averaged tumble momentum MT (m2/s) along with the turbulent kinetic energy k (m2/s2) for
the last simulated engine cycle. MT is negative until the first 50 CA. Then, tumble quickly increases, reaching the
maximum at 150 CA. Subsequently, tumble evolution linearly decreases until 280 CA, and it quickly decreases at
the end of the compression stroke. After that, during the expansion stroke, the tumble is negligible, but negative
tumble is produced between 630 and 720 CA due to the flow evacuation when piston rises.
Figure 10. Tumble momentum (Upper) and turbulent kinetic energy (bottom) for three engine speeds.

Turbulent kinetic energy (k) is mainly produced due to two mechanisms; firstly, frictional efforts, flow
detachment around valve vortexes and large strain gradients during the intake and exhaust strokes. Secondly, tumble
degradation at the end of the compression stroke. The first mechanism explains the peak of k around 170 CA, and
the second causes the significant increment of k around 360 CA (top dead center), this being the maximum k of the
whole engine cycle. k keeps almost null during the expansion stroke and the first stages of the exhaust stroke.
Finally, little turbulence is produced during the gas evacuation.
The generation of coherent macro-vortex structures guarantees the conservation of a large amount of kinetic
energy of the entering flow until the end of the compression stroke. Then, the destruction of the macro vortexes
transfers the kinetic energy from the mean flow to smaller turbulent structures (micro eddies), causing the peak of
turbulence close to the spark ignition time. That increases the burning velocity (S t) and improves the overall
combustion process (Jeon et al., 1998; Aleiferis et al, 2004; Li et al., 2001; Mariani and Cavalletti, 2004).
Table 2 highlights the effect of the engine speed over MT and k. For low engine speeds (increment from 1500 to
3000 rpm), MT and k increase more than the engine speed. MT increases between 150% and 300% while k grows
over 350%. However, when the engine speed is incremented from 3000 rpm to 4500 rpm, M T is scarcely
incremented and at 360 CA the increment is negligible. On the other hand, k increases similarly to the engine speed.
So, the increment of k at high engine speed seems to be caused by strain and squish than by tumble degradation.

Table 2. Increments of tumble momentum (MT) and turbulent kinetic energy (k) by increasing engine speed.
 Percentage increments (%)
Engine speed
increments Maximum at 342 CA at 360 CA
MT k MT k MT k
1500 to 3000 153 387 157 350 300 365
3000 to 4500 38 111 39 119 5 100
1500 to 4500 250 929 257 889 320 830

3.2.2 Steady vs. cold dynamic simulations

It is interesting to compare the mass-flow distribution around one valve curtain for steady (tumble test 5) and
dynamic cases at different valve lifts. Figure 11 shows the mass flow distribution (percentage) at 6 mm, 8 mm and 9
mm during valve opening and closing. In the same graphics, distributions corresponding to steady tests are also
included. Naturally, for steady tests the opening and closing events are the same. In the polar diagrams, the flow
crossing the upper segment from 1 to 7 produces positive tumble momentum, so this segment is called positive
curtain of the valve. The bottom segment from 1 to 7 is called the negative curtain of the valve.

Valve opening Valve closing

Valve closing – lift: 6 mm

Valve opening – lift: 6 mm

6 mm 6 mm

Valve closing – lift: 8 mm

8 mm 8 mm

Maximum valve lift: 9 mm

9 mm Static
Dynamic 1500 rpm
Dynamic 3000 rpm
Dynamic 4500 rpm

Figure 11. Mass flow distribution around one valve curtain for several valve lifts and steady and cold dynamic
simulation.
 Note that for the three engine speeds flow distributions for dynamic tests are fairly similar during opening and
also for the maximum valve lift. However, they completely change during the closing phase. Steady distributions are
roughly similar to dynamic ones when valves are opening, but the differences become significant once the
maximum valve lift is reached and start the closing phase. The orientation of the flow at the middle zone between
the intake valves is more evident for the dynamic cases. At maximum valve lift the entering flow through the
negative curtain of the valve diminishes with respect to the steady case.
The significant change on flow distribution at high valve lifts for dynamic tests enhances tumble generation with
respect to the steady estimations. Moreover, the mass flow rate entering through the negative curtain is reduced even
more during valve closing, improving tumble production.

1
Static case
Dynamic 1500 rpm
0.8 Dynamic 3000 rpm
Dynamic 4500 rpm
Tumble momentum (m^2/s)

0.6

0.4

0.2

0
0

0 1 2 3 4 5 6 7 8 9 8 7 6 5 4 3 2 1 0
-0.2
Intake valve lift (mm)

Static case
0.13
Dynamic 1500 rpm

0.11 Dynamic 3000 rpm

Dynamic 4500 rpm
0.09
Mass flow rate (kg/s)

0.07

0.05

0.03

0.01

-0.01 0 30 60 90 120 150 180 210

Crank angle (º)
-0.03 1 2 3 4 5 6 7 8 9 8 7 6 5 4 3 2 1

Figure 12. Upper: tumble momentum as a function of valve lift for steady and dynamic tests. Bottom: mass flow rate
along the intake process.

Figure 12 in its upper part shows the tumble generation while valves are opened as a function of the valve lift.
Note that positive tumble appears after 7 mm valve lift in the opening phase, even though the mass flow rate before
this valve lift is quite significant as shown in figure 12 at the bottom. The negative tumble at the beginning of the
cycle might be explained by the flow leaving the cylinder through the exhaust valves at the end of the previous
cycle, and due to the significant fraction of flow entering through the negative curtain of the intake valves at valve
lifts below 7 mm. The fast production of positive tumble after 7 mm can be due to the change on flow distribution
around the valve curtains. Curiously, for the three engine speeds analyzed, around 70% of the maximum tumble is
reached at the maximum valve lift and the maximum tumble is reached at 8 mm valve lift (while valves are closing).
Thereafter, tumble decreases almost linearly during valve closing.
As regards the mass flow rates, no conclusions can be drawn about the relationship between the steady and cold
dynamic behavior of the engine because it is difficult to do both experiments at the same operation conditions.
While steady tests are performed under constant pressure drops with a flow completely developed, in cold dynamic
tests the cylinder pressure continuously changes depending on piston and valves motion with a significant influence
to the flow field.
Flow discharge coefficients obtained through steady tests are quite independent of the pressure drop (Xu, 2001;
Grimaldi et al., 2003). In order to compare steady vs. dynamic results it is valid to scale the steady curve of mass
flow rate in such a way that it matches the dynamic one at maximum valve lift (9 mm) for each engine speed. Then,
it is possible to see the overestimation in mass flow rate predicted by steady tests with respect to the dynamic ones,
especially during valves closing. As previously observed in figure 11, differences between steady and dynamic
results are bigger during valve closing than during valve opening.

3.2.3 Real vs. cold dynamic simulations

Dynamic boundary conditions extracted from 0/1-dimensional simulation were achieved and a phenomenological
combustion model was implemented in order to perform more real dynamic simulations. Results for the same 3
engine speeds (1500 rpm, 3000 rpm y 4500 rpm) were obtained, these highlighting the differences with respect to
cold dynamic tests.
As for the mass flow rate, the main differences between cold and real hot simulations are found at the beginning
and at the end of the cycles. Under real engine conditions, flow motion is mainly governed by pressure differences,
but for cold dynamic tests flow is pushed or sucked by the piston motion (figure 13). Note that the intake process is
fairly similar for cold and real simulations, but significant differences appear when exhaust valves are opened. The
maximum mass flow rate for cold dynamic tests takes place close to the maximum exhaust-valve lift, while for real
dynamic tests it immediately occurs once the valves open.
Comparing figures 10 and 14, the evolution of tumble is quite similar for cold and real simulations, although the
maximum tumble increases for real tests (20% at 1500 rpm, 18% at 3000 rpm, and only 4% at 4500 rpm). An
oscillatory behavior is found at the end of the cycles for real dynamic tests, possibly due to the variations on mass
flow rate caused by pressure waves at the intake and exhaust systems.
 Cold simulation

PMI

PMI

Hot simulation

PMI

PMI

Figure 13. Mass flow rate for cold (upper) and real (bottom) dynamic simulations.

Turbulent kinetic energy (k) for real dynamic simulations is greater than for cold ones, and differences become
more significant as the engine speed is increased. The first peak during the intake stroke gains importance during
real dynamic simulations and for 4500 rpm it becomes the highest, indicating that tumble degradation at the end of
the compression stroke produces less turbulence than the strain efforts during the intake stroke.

Table 3. Increments of MT and k (percentage) as a result of introducing the combustion process.

Percentage increments (%)
Engine speed Maximum at 342 CA at 360 CA
MT k MT k MT k
1500 22 37 71 33 190 26
3000 18 16 39 17 40 19
4500 4 64 0 4 5 3

Table 3 consigns the increments of MT and k (percentage) after the introduction of the combustion process. Note
that the effect of combustion over tumble diminishes with engine speed increment, but tumble is always bigger than
that for the cold tests. Similar conclusions can be reached by analyzing the turbulent kinetic energy.
Figure 14. Tumble momentum MT (upper) and turbulent kinetic energy k (bottom) for three engine speeds.

3.2.4 Combustion behavior

Figure 15 on the upper part of the graphic displays the evolution of the main characteristic variables with the
implemented combustion model (at 1500 rpm), while on the bottom part the turbulent burning velocity (St) for the
three engine speeds is drawn. The flame front radius is the time-integrate turbulent burning velocity. Note that the
flame front area changes when the flame front reaches the piston crown surface (around 355 CA) and the cylinder
surface (around 405 CA). Although the implemented combustion model is a rough simplification of the complex
combustion phenomenon, it is still adequate to estimate the influence of turbulence due to tumble degradation. As
shown in figure 15, St is directly related with turbulence enhancement, but St quickly diminishes after the TDC.
Moreover, the increment on St due to turbulence is not enough to compensate the increment in engine speed and
combustion process is extended during the expansion stroke.
Figure 15. Combustion variables for 1500 rpm (upper) and burning velocity (bottom) for three engine speeds.

4. CONCLUSION

In this work numerical and experimental tests were performed over a commercial engine. Some mean-flow
characteristics of this engine like mass flow discharge coefficients and tumble and swirl production were
investigated. About steady tests, the following conclusions were reached:
• CFD proved to be adequate to capture the tumble falling effect at medium valve lifts, making it possible to relate
this effect with the vortex dynamics below the intake valves
• considering exclusively the generation of swirl the greater the differences between valves lifts the greater the swirl
production, not taking into account the loss of volumetric efficiency produced
• there is a cooperative effect between the intake valves, improving significantly the discharge coefficients when
both of them are opened. when both valves are opened there is a cooperative effect between them that improves the
flow discharge coefficients.
Cold dynamic simulation results allow us to find the following conclusions: Cold dynamic simulations led us to
the following conclusions:
• the steady state engine conditions are reached after no more than 4 cycles. convergence to a steady operation is
reached after four cycles. Probably, more cycles are necessary for multi-cylinder engines
• positive tumble dominates the whole cycle. It starts, starting to grow after 50 CA, and reaching a peak around 144
CA. Then a A fast vortex degradation takes place from between 280 and to 360 CA, enhancing turbulence. After
630 CA negative tumble is induced by gas evacuation
• the abrupt increment of tumble nearby the maximum valve lift is mainly a consequence of a the change in the mass
flow rate distribution around the valves curtains
• k has two peaks; the first one during the intake process, and the other one due to tumble degradation near to the
TDC at the end of the compression stroke
• MT and k are proportional to the engine speed, but their increments are more significant from 1500 rpm to 3000
rpm than from 3000 rpm to 4500 rpm.
Regarding the combustion process the following conclusions seem appropriate:
• differences between cold and real dynamic simulations are highlighted by drawing the in-cylinder mass flow rates.
For cold tests, flow is governed by piston kinematics, while for real tests pressure differences drive flow dynamics
• MT is slightly affected by the combustion process. On the other hand, k is substantially increased during the intake
stroke. Moreover, at 4500 rpm the maximum k is not due to tumble degradation but to frictional efforts during the
intake process.

ACKNOWLEDGEMENT− Authors want to thanks to CONICET and ANPCyT (grants PICT Lambda 12-14573/2003,
PME 209/2003). Also, they are gratefully to Fiat Argentina for the donation of the engine analyzed in this work.

REFERENCES

Aleiferis P.G., Taylor A.M.K.P., Ishii K. and Urata Y. (2004). The nature of early flame development in a lean-burn
stratified-charge spark-ignition engine. Combustion and Flame, 136, 283-302.
Bailly O., Buchou C., Floch A. and Sainsaulieu L. (1999). Simulation of the Intake and Compression Strokes of a
Motored 4-Valve SI Engine with a Finite Element Code, Oil & Gas Science and Technology, 54, 2, 161-168.
Bianchi G., Cantore G, Mattarelli E., Guerrini G. and Papetti F. (1998). The influence of Stroke-to-Bore Ration and
Combustion Chamber Design on Formula One Engines Performance, SAE Paper No. 980126.
Cao L., Zhao H. and Jiang X. (2004). Effects of Intake Valve Timing on Premixed Gasoline Engine with CAI
Combustion, SAE Paper No. 2004-01-2953.
Fontana G., Galloni E., Jannelli E. and Palmaccio R. (2003). Influence if the Intake System Design on a Small
Spark-Ignition Engine Performance. A Theoretical Analysis, SAE Paper No. 2003-01-3134.
Fontana G., Galloni E., Palmaccio R. and Torella E. (2006). The influence of Variable Timing on the Combustion
Process of a Small Spark-Ignition Engine, SAE Paper No. 2006-01-0445.
Geiger J., Grigo M., Lang O., Wolters P. and Hupperich P. (1999). Direct Injection Gasoline Engines – Combustion
and Design, SAE Paper No. 199-01-0170.
Gouldin F.C. (1987). An Application of Fractals to Modelling Premixed Turbulent Flames, Combustion and Flame,
vol. 68, pp. 249-266.
Grimaldi C., Battistone M. and Mariani F. (2005). Experimental and Numerical Analysis of Charge Motion
Characteristics Depending on Intake Valves Actuation Strategies, SAE Paper No. 2005-01-0242.
Grimaldi C., Battistone M. and Uccellani M. (2004). Dependence of Flow Characteristics of a High Performance
S.I. Engine Intake System on Test Pressure and Tumble Generation Conditions- Part 1: Experimental Analysis,
SAE Paper No. 2004-01-1530.
Grimaldi C., Battistone M. and Postrioti L. (2003). Flow Characterization of a High Performance S.I. Engine Intake
System- Part 1: Experimental Analysis, SAE Paper No. 2003-01-0623.
Heywood J.B. (1988). Internal Combustion Engines Fundamentals, Mc Graw Hill. U.S.A.
Hong C. and Tarng S. (2001). In-Cylinder Tumble Flow Field Measurements and Predictions, J. of Engineering for
Gas Turbines and Power, 123, 139-145.
Jeon C., Chang Y., Cho K. B. and Kang K. Y. (1998). Effects of Intake Ports on In-Cylinder Flow and Lean
Combustion in a 4-Valve Engine, SAE Paper No. 981048.
Johan Z., Moraes C.M., Buell J.C. and Ferencz R.M. (2001). In-cylinder cold flow simulation using a finite element
method, Computer methods in applied mechanics and engineering, 190, 3069-3080.
Kim M.J., Lee H. S., Kim W. T. (2006). Correlation Study of the Measured Tumble Ratios Using Three Different
Methods: Steady Flow Rig; 2-Dimensional PIV, and 3-Dimensional PTV Water Flow Rig, Int. J. of Automotive
Technology, 7, 4, 441-448.
Mahrous A., Wysznski M., Xu H. and Tsolakis A. (2007). A CFD Investigation into the Effects of Intake Valves
Events on Airflow Characteristics in a Motored 4-Valve Engine Cylinder with Negative Valve Overlapping, 07-
NAPLES-74.
Mariani F. and Cavalletti M. (2004). Dependence of Flow Characteristics of a High Performance S.I. Engine Intake
System on Test Pressure and Tumble Generation Conditions – Part 2: Numerical Analysis, SAE Paper No.
2004-01-1531.
Mattarelli E., Comparison between two Combustion Chambers for a Motorcycle Racing Engine, SAE paper
2000-01-1894, 2000.
North G. and Santavicca D. (1990). The Fractal Nature of Premixed Turbulent Flames, Combustion Science and
Technology, 72, 215-232.
Li L., Tao J., Wang Y., Su Y. and Xiao M. (2001). Effects of Intake Valve Closing Timing on Gasoline Engine
Performance and Emissions, SAE Paper No. 2001-01-3564.
López E., Nigro N. and Storti M. (2009). Validation of a 0D/1D Computational Code for the Design of Several Kind
of Internal Combustion Engines, in press LAAR.
López E., Nigro N. and Storti M. (2008). Simultaneous untangling and smoothing of moving grids, Int. J. for
Numerical Methods in Engineering, 76, 994-1019.
Ramajo D. (2008). Simulación computacional de los procesos fluido-dinámicos en el interior de motores de
combustión interna, PhD thesis, Universidad Nacional del Litoral, Argentina.
http://www.cimec.org.ar/ojs/index.php/cimec-repo/article/view/1384
Ramajo D., Zanotti A. and Nigro N. (2007). Assessment of a zero-dimensional model of tumble in four-valve high
performance engine, Int. J. of Numerical Methods for Heat and Fluid Flow, 17, 8, 770-787.
Stan C., Stanciu A., Troeger R., Martorano L., Tarantino C., Antonelli M. and Lensi R. (2003). Direct Injection
Concept as a Support of Engine Down-Sizing, SAE Paper No. 2003-01-0541.
Xu H. (2001). Some Critical Technical Issues on the Steady Flow Testing of Cylinder Heads, SAE Paper No.
2001-01-1308.
Tan Z., Reitz R.D. (2006). An ignition and combustion model based on the level-set method for spark ignition
 engine multidimensional modeling, Combustion and Flame, 145, 1-15.
Weng V., Gindele J., Töpfer G., Spicher U., Latsch R. and Kuhnert D. (1999). Investigation on the Bowl-
Prechamber-Ignition (BPI) Concept in a Direct Injection Gasoline Engine at Part Load, SAE Paper No.
1999-01-3658.
Williams F. (1985). Turbulent Combustion, SIAM, Philadelphia, U.S.A.
Wilson N., Watkins A. and Dopson C. (1993). Asymmetric Valve Strategies and Their Effect on Combustion, SAE
Paper No. 930821.

View publication stats