Prediction of In-Cylinder Flow in DI Diesel Engine with

RNG k -ε Model
A S Gowd, Non-member
Prof V Ganesan, Member
Prof K Rajagopal, Member
Prof K H Reddy, Member
In a four stroke CI engine the fluid motion inside the engine cylinder plays an important role in proper
combustion and emission characteristics of the engines. The KIVA code generally is used for such
predictions, which solves three-dimensional unsteady turbulent flow equations and capable of predicting
the behaviour of fluid flow in four-stroke DI diesel engine. In this work, the RNG k-ε turbulent model,
which has the capability of predicting the fluid flow in four stroke DI diesel engines in a much better way
was used in place of standard k-ε model. A hemi-spherical bowl-in-piston was considered to carryout the
analysis. Global variables like pressure, turbulent kinetic energy during compression and expansion
stroke and pollution formation, namely, CO, HC were predicted with both the turbulence models. The
theoretical predictions were compared with the available experimental results in the literature and from
the comparisons it was seen that the predictive capabilities of RNG k-ε turbulent model is better than the
standard k-ε model. The results obtained with RNG k-ε turbulence model were found to be closer to the
experimental results.
Keywords : KIVA code; DI diesel engine; RNG k -ε model; Hemi-spherical bowl

INTRODUCTION equations which are associated with this averaging process

In a four-stroke CI engine the fluid motion inside the engine
cylinder influences the combustion, performance and
emissions of the engine. The fluid motion inside the engine
cylinder is inherently unsteady, turbulent and three-
dimensional. Insight in understanding engine flows comes
from both the experimental measurements and numerical
computations. The gas motion is unsteady during the
intake as well as the compression process and in cylinder
air motion three-dimensional and turbulent. Obtaining
fluid flow patterns by means of experimental techniques
is time consuming and requires large resources. With the
advent of high-speed computers, fluid dynamic based
models are attracting the attention of the researchers in
understanding the in-cylinder fluid dynamics.
In this work, an effort was made to replace the standard
k -ε model in KIVA-based code with RNG k -ε model and
their relative performance was compared. The widely
accepted computational approach for flow, spray and
combustion problems in technical applications is the
solution of the Reynolds/Favre averaged conservation
A S Gowd and Prof K H Reddy are with the Department
of Mechanical Engineering, JNTU College of
Engineering, Anantapur 515 002, Andhra Pradesh; Prof
V Ganesan is with the Department of Mechanical
Engineering, IIT
IIT,, Chennai; and Prof K Rajagopal is with
the JNT University
University,, Hyderabad, Andhra Pradesh.
This paper was received on September 18, 2006. Written discussion
on the paper will be entertained till December 31, 2007.
of the Reynolds fluctuation terms. The Reynolds
fluctuation terms were considered to be responsible for
the turbulence-increased diffusion of the transport
properties, such as, momentum and heat. Therefore, the
introduction of the respective turbulence transport
coefficients, vt and Kt , constitute the set of desired closure
relations. A widely used turbulence model based on this
eddy viscosity approach was the linear, two equation k -ε
model, where the turbulent kinetic energy and the
turbulent dissipation rate (ε) were described by
corresponding transport equations 1 . One of the
shortcomings of their model was in rapidly varying
transient flows where the equilibrium turbulence
hypothesis may be violated. In fact, due to piston
movement, the fuel injection and the combustion process
in DI diesel engines are transient in nature.
Improvements of the linear k -ε model, based on the theory
of re-normalization groups (RNG), have been achieved2
and successfully introduced to spray combustion
simulations3. However, uncertainties in the dissipation
transport equation still remain and have been investigated
in the recent study4.
THEORETICAL MODEL
The gas flow within the cylinder is extremely complex
and three-dimensional. They are unsteady and turbulent
in nature due to the reciprocating piston movement. In
the present work, a time marching three-dimensional,

Vol 88, October 2007 45

finite difference program were used to steady the fluid The total fluid density ρ satisfies the equation
flow. This program solves three-dimensional differential
equations of conservation of mass, momentum, energy and
species concentrations. To get the solution for a continuum (6)
problem, such as, flow field inside the engine cylinder, the
continuum was represented by a finite number of The momentum equation for the fluid mixture was given by
discretization. This means to divide the region of interest
into a number of small cells. These cells form a mesh,
which serves as a framework for constructing finite (7)
volume approximations to the governing partial differential
equations. The time variable was similarly discretized into
a sequence of small time intervals called time steps and a where P is the fluid pressure; σ, the viscous stress tensor
transient solution was marched out in time. and Fs is the momentum per unit volume per unit time
transferred from the spray droplets to the fluid.
Governing Equations
The stress tensor was given by
The governing equations were written in vector notations.
The unit vectors in the x , y and z directions were denoted
by i, j and k, respectively. The position vector x is denoted by
where µ and λ are first and second viscosity coefficients; I,
(1)
the unit dyadic and super script; and T denotes the
transpose. The details of the internal energy equations,
The vector operator ∇ was given by, state relations, spray dynamics and chemical reactions
were given by Amsden and others.

(2) Boundary Conditions

Temperature and velocity boundary conditions were

and the fluid velocity vector u was given by required to calculate the flow field computational domain.
In this work, the law of the wall boundary condition was
(3) considered to resolve the boundary layer near the solid
walls while calculating velocities as shown below.
where t was the time. Here and . (i) When the wall was flat the components of tangential
velocity calculated.
The program solved the equations of motion for fluid along
with the equations for spray droplets and chemical kinetics. (ii) When the wall was having curvature in one direction,
the component of tangential velocity in the direction of
Equations of Motion for the Fluid Motion curvature was set to zero and the other component in
the direction of no curvature was set equal to the next
The total mass density excluding the mass of the spray closest corner (but not on the wall).
droplets was given by
(iii) When the wall was curved in both principal directions
(the case of hemi-spherical bowl) both components of
(4) tangential velocity were set to zero.

The normal velocity was always set equal to the velocity

The continuity equation for species m was given by
of the corresponding wall.

THE RNG k - ε MODEL

(5)
The computations presented in this study have been
performed with KIVA code. This code was equipped with
where D is the species diffusivity (assumed to be same for standard k -ε model. The code was modified by
incorporating RNG k -ε turbulence model in place of
all species); , the rate of change of ; , the rate
standard k -ε model.
of change of due to spray evaporation ( not considered
The RNG k -ε turbulence model was derived from the
in the present study); , the Kronecker delta; and l is instantaneous Navier-Stokes equations, using a
the species of which the spray droplets were composed. mathematical technique called ‘renormalization group’
46 IE(I) Journal–MC
(RNG) methods. The sensitivity of the combustion results Table 1 Some important engine details (input values)
depends on the appropriate turbulence models used. The
KIVA code, which was used, has a generalized standard Bore, mm : 87.5
turbulence model. The standard k -ε turbulence model is Stroke, mm : 110
given below Connecting rod length, mm : 232
Piston bowl configuration : Hemi-spherical bowl
Engine speed, rpm : 1500
(12)
Initial swirl ratio :2
Initial change temperature, K : 363
Fuel : Normal decane
Density of fuel, kg/m3 : 771.9
Critical temperature of fuel droplets, K : 619
Injector : Single hole
(13)
Type of spray : Solid cone
Tilt of injector, degree : 27
where Mean injection velocity, m/s : 80
Inflow temperature of fuel droplets, K : 350
Total mass of fuel injected, g/cycle : 0.02591
Total number of spray droplets : 2000
Start of injection (bTDC), degree : 28
where k is the turbulent kinetic energy; ε, the turbulent End of injection (bTDC), degree : 4.5
kinetic energy dissipation rate; D , the diffusion term and
the conventional turbulence model constants are Cε1 =1.45, Mesh Generation
Cε2 = 1.92, Cε3 = 1, σk = 1, σε = 0.7 and Cµ = 0.09 in
An automatic mesh generator was provided in the
subroutine SETUP that creates a two-dimensional or
.
three-dimensional cylindrical grid. The code can generate
mesh for a wide variety of piston shapes for CI engine
combustion chambers. The mesh generator requires only
The standard k -ε model was improved uses
a simple set of tabular information of the input data. Given
renormalization group theory (RNG) and introduces a new
the dimensions of the piston geometry, a half cross-sectional
term R in the ε equation. This modified k -ε model is called
view of it is drawn on a graph paper. Based on this the
RNG k -ε turbulence model. The modification to the
grid points were defined along the piston boundary, based
ε equation was given below
on the level of resolution available, starting at the bowl
axis and ending at cylinder wall. For these mesh points
corresponding to the logical (i and k ) and physical co-
(14) ordinates (r and z ) were to be set (z = 0 being the lowest
z co-ordinate). As per this, the mesh generator first assigns
the vertices those have been set and then assigns the
remaining vertices, those within the fluid region at
average positions of their neighbours. The generated mesh
where , where and S = was a two-dimensional one. Figure 1 shows the top and
front views of a mesh generated for a diesel engine
(2 s ij s ij ) 0.5 . combustion chamber with hemi-spherical bowl in the
piston.
Here, sij is the mean strain rate; VT , the shear stress
tensor and η0 and β are constants5. Cε1 and Cε2 are the
constants used in the standard k -ε turbulence model.

In the present work, the original KIVA code was modified

by replacing the standard turbulence model with RNG
k -ε turbulence model.

COMPUT
COMPUTAATIONS
(a) front view (b) top view
Some important engine details considered for the analysis
were given in Table 1. Figure 1 Computing mesh

Vol 88, October 2007 47

RESULTS AND DISCUSSIONS
RESULTS
The computed histories of certain global parameters like
average cylinder pressure and pollutant emissions were
compared with the available experimental results5. The
global parameters were predicted with both standard k -ε
turbulent model and RNG k -ε turbulent model and
comparisons were drawn between them.
Comparison between Experimental and
Predicted Pressure Histories
Figure 2 shows the variation of pressure (experimental
and predicted) with variations in crank angle. It is clear
from the figure that the predicted results were closely
following the experimental results. The major deviations
of predicted pressure from the experimental value were
noticed near TDC. From the deviations two important
aspects can be observed; (i) the predicted peak pressures
higher than that of experimentally obtained pressures.
The peak pressures with the standard k -ε turbulence
model, RNG k -ε turbulence model and experimental
values were 6.7 MPa, 6.3 MPa and 6.1 MPa, respectively,
and (ii) the computed peak pressure was occurring a little
earlier than the experimental case.
One reason for lower experimental pressures may be due
to the greater ignition delay during combustion. The
increase in ignition delay causes the combustion to continue
even after the piston crosses TDC towards expansion
stroke, resulting in lower peak pressures.
The other reasons for the difference in peak pressures may
be due to the retaining of residual gasses in the clearance
volume during exhaust stroke. The residual gasses at
higher temperature reduce the fresh charge entry during
8 0.16
RNG k -ε
6 Standard k -ε
Experimental
Pressure, MPa
4 0.08
2 0.04
0 0
–80 –60 0 40 80
Crank angle (aTDC), degree
Figure 2 Variation of in-cylinder pressure with crank
Variation
angle
48
suction, as it destroys some vacuum by expansion. Blow-
by and crevice flow also effect the in-cylinder pressures.
Both these features were not incorporated in this model.
The occurrence of predicted peak pressures even before
TDC indicates that majority of the fuel was consumed in
combustion before the piston reaches TDC. This supports
the argument that the ignition delay was more in
experimental case. Another important observation that can
be made was that RNG k -ε turbulence model predicted
the pressure variations closer to the experimental
variations.
Variation of Turbulent Kinetic Energy with
Turbulent
Crank Angle
Turbulent flows are important in diesel engine combustion
chambers. The turbulence of the flow ensures rapid mixing
of air and fuel because, the local fluctuations in the
turbulent flow field results in molecular diffusion. The
turbulent flows are always dissipative. The deformation
work done on the fluid due to viscous shear stresses
increases is internal energy at the expense of its turbulent
kinetic energy (TKE). Figure 3 shows the variation of TKE
with crank angle. A noticeable increase in the TKE was
observed at 7° bTDC indicating the initiation of
combustion. From this point, within a very short variation
of crank angle the TKE is increasing very sharply. The
maximum value of TKE was noticed near TDC. The
sudden rise in the TKE can be attributed to the energy
release in the combustion chamber due to combustion.
During the expansion stroke the TKE was showing a
decreasing trend. Very sharp fall in TKE was noticed at
the beginning of expansion and a gradual fall there after.
The peak TKE predicted with RNG k -ε model was 0.133 J
and that predicted with standard k -ε model was 0.128 J.
0.12
Turbulent kinetic energy, J
Standard
RNG k -ε
120 –100 –50 0 50 100
Crank angle (aTDC), degree
Figure 3 Variation of turbulent kinetic energy with crank
Variation
angle
IE(I) Journal–MC
Variation of Emission Formation
The predicted HC and CO variations were compared with
measured values.
UBHC Emissions
Unburnt hydrocarbon (UBHC) emissions with exhaust
gasses indicate the combustion efficiency and uniformity
in fuel-air mixing. The presence of UBHC in exhaust may
be because of the following reasons.
(i) Too lean fuel-air mixture can not support the
propagation of the flame resulting in incomplete
combustion.
(ii) Too rich fuel-air mixture also can not support the
flame propagation for the lack of oxygen, resulting in
incomplete combustion.
Figure 4 gives the mass of UBHC present in the cylinder
at different crank angle positions of the piston during
expansion stroke. The unburnt fuel mass includes both
the liquid and vapour fuel. The experimentally measured
HC emission was marked on Y-axis. The measured value
is 0.0022 g. From the figure it can be noticed that the fuel
mass is gradually reducing from TDC (crank angle 0°).
The declining fuel mass indicates that the combustion was
proceeding even during the exhaust stroke. HC mass
remains constant during the later part of expansion stroke.
This mass was considered to be exhausted during the
exhaust stroke. The unburned hydrocarbon emission
predicted with standard k -ε model is 0.003 g, whereas, it
is 0.00255 g with RNG k -ε model. The UBHC predicted
with RNG k -ε model was close to the measured value
(0.0022 g).
2.00E–2
1.60E–2
HC mass, g/cycle
1.20E–2 Standard
RNG k -ε
8.00E–3
4.00E–3
Measured value
0.00E+0
0 20 40 60
Crank angle (aTDC), degree
Figure 4 Variation of HC emission with crank angle
Variation
Vol 88, October 2007
4
3
Standard k -ε
CO mass, g/cycle
RNG k -ε
2
1 Experimental
value
0
–20 0 20 40 60 80
Crank angle (aTDC), degree
Figure 5 Variation of CO emission with crank angle
Variation
CO Emissions
Figure 5 shows that the variations of CO formation
increase rapidly from the start of combustion (7° bTDC)
to a little after TDC. Ideally the CO formation should be
low during this period as the availability of oxygen is high.
But due to the heterogeneous nature of diesel combustion,
those pockets in which the fuel-air mixture is
stoichiometric burn completely whereas, at those regions
where rich fuel-air mixture is present the combustion will
be incomplete resulting in CO formation. The CO thus
formed will be oxidized in the later part of the expansion
stroke. This justifies the declining trend of CO in the
expansion stroke.
It can also be observed from the figure that there was a
poor agreement between the experimental and predicted
CO emissions. The experimental value was 0.0011 g.
Whereas the CO predicted with both the turbulent models
were far below this. The main reason for this disagreement
was the dissociation of CO2 into CO late in the expansion
stroke.
CONCLUSION
The four -stroke DI diesel engine with hemi-spherical bowl
in piston was considered for analysis. The global
parameters, such as, pressure, turbulent kinetic energy
and emissions were predicted using KIVA CFD code with
two turbulence models, namely, standard k -ε and RNG
k -ε turbulence models. The parameters were compared
with the experimental values. The predicted in-cylinder
peak pressures were 6.7 MPa and 6.3 MPa with the
standard k -ε and RNG k -ε turbulence models,
80 respectively. The experimental peak pressure was only
6.1 MPa. The CO emission predicted with both the
turbulence models was around 0.000318 g/cycle, whereas,
the experimental value was 0.0011 g/cycle. The unburned
49
hydrocarbons emission predicted with standard k -ε model
was 0.003 g, whereas, it was 0.00255 g with RNG k -ε
model. The UBHC predicted with RNG k -ε model was
close to the measured value (0.0022 g). Except for the CO
prediction, the rest of the parameters computed were
reasonably close to the experimental values. Between the
two turbulence models used in the computations, the RNG
k -ε turbulence model predictions were very close to the
measured values.
REFERENCES
1. F H Harlow and P I Nakayama. ‘Transport of Turbulence
Energy Decay Rate’. Combust Flame , vol 30, 1968, pp 45-319.
50
2. V Yakhot and L M Smith. ‘The Renormalization Group, the
ε-expansion and Derivation of Turbulence Models’. J Sci Comp ,
vol 7, 1992, pp 35-61.
3. Z Y Han and R D Reitz. ‘Turbulence Modelling of Internal
Combustion Engines using RNG k -ε Models’. Combust Sci and
Tech, vol 106, 1995, pp 267-295.
4. G M Bianchi, V Michelassi and R D Reitz. ‘Modelling the
Isotropic Turbulence Dissipation in Engine Flows by using the
Linear k -ε Model. Proceedings of ICE Division of ASME , October
1999.
5. K Rajagopal. ‘Three-Dimensional Modelling of In-Cylinder
Flows, Fuel-Air Mixing and Combustion in a Four Stroke
Compression Ignition Engine’. Ph D Thesis, IIT Madras, 1991.
IE(I) Journal–MC