1
2000-01-2938
Effects of Direct Water Injection on DI Diesel Engine
Combustion
F. Bedford and C. Rutland
Engine Research Center, UW–Madison
P. Dittrich, A. Raab and F. Wirbeleit
DaimlerChrysler Research
Copyright © 2000 Society of Automotive Engineers, Inc.
ABSTRACT
The effects of in-cylinder water injection on a direct
injection (DI) Diesel engine were studied using a
computational fluid dynamics (CFD) program based on
the Kiva-3v code. The spray model is validated against
experimental bomb data with good agreement for vapor
penetration as a function of time. It was found that liquid
penetration increased approximately 35% with 23% of
the fuel volume replaced by water, due mostly to the
increase in latent heat of vaporization.
Engine calculations were compared to experimental
results and showed very good agreement with pressure,
ignition delay and fuel consumption. Trends for
emissions were accurately predicted for both 44% and
86% load conditions. Engine simulations showed that
the vaporization of liquid water as well as a local increase
in specific heat of the gas around the flame resulted in
lower Nitrogen Oxide emissions (NOx) and soot
formation rates. Using stratified fuel-water injection
increases soot at 86% loads due in part to late injection.
Because NOx decreased at all loads, the injection timing
can be advanced to minimize fuel consumption and soot.
INTRODUCTION
Recent fluctuations in fuel prices have underscored some
of the pressures under which the engine industry
operates. In many industries, fuel economy is a primary
concern, yet health risks from high concentrations of
airborne toxins cannot be ignored. Thus, constraints
from legislating bodies combine with market forces to
push engine manufacturers towards creating engines
that simultaneously use less fuel and produce fewer
harmful pollutants. Techniques such as split injection,
high pressure fuel injection (see Han [4] and Reitz [9]),
and stratified fuel-water in cylinder injection (see
Wirbeleit [12] and Takasaki [10]) have been shown
effective at reducing pollutants from DI Diesel engines
while minimizing fuel consumption. This paper describes
a computational study of stratified Direct Water (DW)
injection. Here we use CFD to explore the effects of DW
injection on Diesel engine combustion to gain insight into
the in cylinder processes which make the technique one
of the most promising technologies for NOx reduction.
There are several practical means of inducing water into
the combustion chamber (see Wirbeleit [12]).
Fumigation, emulsions, parallel injection systems and
DW injection are all effective realizations of water
injection technology, each with their own set of
advantages and drawbacks. In this study, it must be
stressed that the fuel and water are separate until just
before the actual injection event – DW injection is not an
emulsion and does not suffer from the same drawbacks.
Fumigation is where liquid water is injected into the
intake manifold upstream of the intake valve. The
fumigation technique has been shown to reduce NOx
emissions in DI Diesel applications but suffers from the
drawback that the liquid water in the combustion
chamber is typically in areas where it is less effective at
reducing emissions. Therefore, fumigation requires
approximately twice the liquid volume for the same
reduction in engine out NOx when compared to DW
injection. Additionally, liquid water present after
combustion can contaminate the oil and increase engine
wear.
Emulsified fuel-water blends can be used as an
alternative fuel and have been shown to reduce NOx and
particulate matter (PM) emissions. However, emulsified
fuel blends tend to lower the combustion temperature
indiscriminately. Lower temperatures too early in
combustion can lead to increased ignition delay and
engine noise. Although there is little or no increase in
engine cost when using emulsified fuels, the engine
2
injection timing must be changed to take advantage of
the new mixture. A more significant drawback to
emulsified fuels is that the percentage of water is
constant and cannot be changed for cold start or other
transient operating conditions. In other words, a
particular blend of fuel and water may be optimal for one
operating condition but degrade performance for other
points in the design envelope.
DW injection has the advantage over fumigation of
having the liquid water close to the flame and away from
the wall. Unlike emulsified fuels, DW injection allows the
fuel-water percentage to be changed for cold start or
different operating ranges. Although the injection system
needs to be modified for DW injection, a single injector
per cylinder is used so that the overall cost would be less
than a parallel injection system. A schematic of a typical
dual feed injector is shown below in figure 1.
Fuel Feed
Low Pressure
Water Feed
Water Loading Operation
Fuel
Injection
Check Valve
Fuel Injection Operation
Fuel
Water
a) b)
Figure 1: Operation of a typical Fuel-Water Injection
system.
The key to the DW injection system is the dual feed
injection nozzle with the corresponding water supply
system. Note that the water supply system does not
need to support high pressures like the fuel injection
system. The water loading event (Figure 1a) shows the
water being pushed first through a one-way valve, then
through the hollow passage in the injector body and
eventually into the annular sac region surrounding the
central pintle valve. The fuel displaced by the water is
returned to the fuel injection pump.
At the start of the injection event, the pressure in the fuel
line is increased which closes the one-way valve so that
no fuel contaminates the water line. The fuel-water
mixture is forced out of the injector by the high pressure
fuel. The amount of mixing inside the injector is not well
known and can vary with the design of the injector body.
Typically, a larger volume of the secondary chamber in
the injector allows more mixing of fuel and water to occur
before the injection event and a smaller chamber (with
less fuel in front of the water slug) allows less mixing. As
will be seen in the engine calculation section, having the
water towards the front of the injection causes significant
ignition delay.
NUMERICAL MODELING
A modified version of the KIVA-3v (see Amsden [1]) CFD
program, originally developed at Los Alamos National
Laboratory, was used for this study. KIVA solves the
conservation equations for unsteady, compressible,
turbulent reacting flows on finite volume grids. There
have been numerous additions and modifications of
many submodels at the ERC and DaimlerChrysler
Research that have been validated for engine
combustion simulations. The additional submodels have
been described in the literature and so are only briefly
mentioned below (see Rutland [8] and Dittrich [3] for
details). The multidimensional simulations for Diesel
engine combustion have been extensively verified with
modern optical experimental methods in high
temperature combustion bombs as well as in transparent
single cylinder engines (see Schwarz [11]).
The standard k-ε model was used to account for
turbulence, and the spray breakup is computed with the
"wave" breakup model of Reitz [7] which has been
modified to account for the effects of drop distortion on
the drag coefficient of the drops (see Han [4]). For this
study, the new parcel diameter is calculated using an
SMR conserving breakup concept (from Patterson [6]).
Ignition delay is modeled with a Wolfer type equation for
the ignition delay which takes into account the local
equivalence ratio, pressure and temperature (see Dittrich
[3]).
Using a Wolfer type equation for ignition delay in the
presence of water seems to capture the effects of
temperature and pressure adequately. As a check,
constant pressure oxidations of n-heptane with varying
amounts of initial water vapor were calculated using a
detailed chemical kinetic mechanism of 550 species and
2450 reactions (see Curran et.al. [2]) using the CONP
program from Chemkin II. At an equivalence ratio of 1,
86 bar initial pressure and 900 K initial temperature, a
140% water/fuel vapor mass ratio increased the ignition
delay less than half a millisecond, or less than half a
crank angle at the engine speed considered. Higher
equivalence ratios were even less sensitive to additional
water vapor. Thus, the effect of water vapor was not
explicitly included in the ignition correlation. Since other
operating points could show more significant changes in
ignition delay in the presence of large amounts of water
vapor, the correlation could be modified for future
studies.
3
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
Water
Fuel
Normalized Velocity
Normalized Injection Duration
Figure 2: Distribution of water in the fuel spray for a wide distribution
function (σ
w
= 5.00 and x
c
= 0.5).
Combustion is modeled with the seven species
characteristic time model that is explained in detail in
Patterson [6] and Han [4]. Emissions of nitrogen oxides
(NOx) are modeled with the extended Zel’dovich
mechanism and soot is modeled with the Hiroyasu
model. Detailed descriptions of the implementation of
the current models are available in papers by Han [4] and
Kazakov [5].
Including the effects of a second liquid species required
extensive modification to the existing subroutines. The
two liquids are always separate and no mixing is allowed
in during the collision process, however the species do
interact in the gas phase. Treating the liquids separately
allows modeling of parallel injection systems as well as
typical single nozzle (multiple hole) injection systems.
When modeling one multiple hole injector, the fuel and
water are injected from the same spatial location and the
velocity profile is calculated using the sum of the fuel and
water masses. The water mass is added to the fuel
mass and the velocity profile accordingly scaled.
The water distribution in the spray is biased by the
following Gaussian function
(
)
wc
xxAxp
σ
2
)(exp)( −= , (1)
where p(x) is the probability of finding a water parcel in
the fuel spray. The normalization constant A is chosen
so that the fuel/water mass ratio is correct, the center of
the Gaussian is denoted x
c
and the width is given by σ
w
.
If the distribution function causes the spray mass to go
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
Water
Fuel
Fuel
Normalized Velocity
Normalized Injection Duration
Figure 3: Distribution of water in the fuel spray for a narrow distribution
function (σ
w
= 0.05 and x
c
= 0.5).
too high, the function is clipped and rescaled to ensure
mass conservation. By clipping the top of the Gaussian
distribution function, the water in the spray can
completely displace the fuel in that portion of the
injection. A small value for σ
w
implies a narrow, high
distribution with sharp stratification of fuel and water,
while a large value for σ
w
implies a uniform mixing of the
fuel and water inside the injector.
An example of a wide distribution is shown in Figure 2 for
parameter values of σ
w
= 5.00 and x
c
= 0.5 with 45% of
the fuel mass was input as water. A velocity distribution
for a narrow distribution function (σ
w
= 0.05) is shown in
Figure 3 for the same 45% fuel/water ratio.
For the engine cases, the fuel consumption values were
calculated in Kiva by integrating p dV from intake valve
closing to exhaust valve opening to obtain an IMEP value
and dividing the result by the fuel injected in the cycle.
SPRAY MODEL VALIDATION
Experiments for Diesel fuel/water injection were
performed with a prototype Bosch dual feed injector,
where the stratification of the fuel and the water can be
controlled. In this study, the volumetric flow rate was
kept constant. In each case, 136
mm
3
of liquid was
injected into a quiescent chamber filled with Nitrogen at
830 K and 41 bar. The spray was illuminated from below
by a laser sheet and vapor penetration was obtained by
fluorescence. More detail on the experimental setup can
be found in Renner [9].
4
0.0 0.5 1.0 1.5 2.0
0
2
4
6
8
10
12
14
Penetration (cm)
Time (ms)
Vapor Data
Liquid Kiva
Vapor Kiva
Figure 4: Comparison of liquid and vapor penetration for the fuel only
spray as functions of time with experimental data.
For the spray considered, the nozzle diameter is 345 µm
and the injection duration is 1.7 ms. Peak injection
pressures varied from 600 to 800 bar depending on the
composition of the spray and injection duration. Because
the magnitude of the mixing inside of the injector was not
known, the spray was considered as uniformly mixed
(σ
w
=5.0).
In the experimental study, the volume injected was held
constant and the composition varied by feeding water
into a secondary annular chamber inside the fuel injector.
For the water case, 77% (105 mm
3
) of the injected
volume was fuel and the remaining 23% (31 mm
3
) was
water. The mesh spacing was chosen to be close to that
used in typical engine calculations.
The total number of parcels injected was 500, which was
sufficient for accuracy in this two-dimensional simulation.
The number of parcels in the course of the calculation
varied due to differences in breakup and evaporation
rates for the different sprays. Typically, the cases where
water was injected had a higher number of parcels,
despite an identical number of total injected parcels. This
computational study used a representative Diesel fuel,
C
14
H
30
, and H
2
O as the two liquids. The fuel species,
C
14
H
30
was chosen because it has physical properties that
are close to an average Diesel fuel. For these
calculations, a model constant proportional to the
breakup time was changed to characterize the injector for
the fuel only spray. This same constant was used for
both the water and fuel injectors for subsequent
calculations.
The liquid penetration values were defined in the
computation as the minimum distance along the injection
direction that includes 95% of the liquid mass.
0.0 0.5 1.0 1.5 2.0
0
2
4
6
8
10
12
14
Penetration (cm)
Time (ms)
Vapor Data
Liquid Kiva
Vapor Kiva
Figure 5: Comparison of liquid and vapor penetration for the fuel water
spray as functions of time with experimental data.
The vapor penetration was defined as the point along the
injection direction nearest the tip where 5% of the
maximum vapor concentration occurs. In all simulations,
the liquid and vapor penetrations split from each other,
the liquid penetration reaching a plateau of between 4
and 6 cm while the vapor continued to penetrate.
Figures 4 and 5 show a comparison of the experimental
vapor penetration to the calculated vapor penetration and
also show the calculated liquid penetration, all as
functions of time. For these calculations, the calculated
vapor penetration is very close to the experimentally
measured values. For this injector, which produces
relatively large drops, the vapor and liquid penetrations
were very similar initially, but diverged from each other
when the jet penetrated about 4 cm from the nozzle for
the fuel only spray.
The presence of water in the fuel spray increases both
the liquid and vapor penetration, however the liquid
penetration is changed more significantly as shown in
Figure 5. As can be seen from the comparison of
calculated spray tip penetrations in Figures 4 and 5, the
water lengthens the liquid penetration at 150 ms by
approximately 35% when 23% of the injected liquid is
water.
Note that the vapor penetration is not changed as much
as the liquid penetration when water is added to the
spray, in fact most of the difference can be attributed to
the additional momentum from the increased density of
the water vapor.
5
0.0000 0.0005 0.0010 0.0015
0
2
4
6
8
Penetration (cm)
Time (s)
Fuel + Water
Surface Tension
Latent Heat
Fuel Only
Figure 6: Liquid Penetration vs. Time - Sensitivity to Physical
Properties
To gage the relative importance of the physical
parameters of the second injected liquid (e.g. water), the
physical properties of the injected water were selectively
changed to equal those of the fuel. The liquid penetration
as a function of time of the modified spray is shown in
figure 6.
When the surface tension of the water is set equal to the
fuel, the Weber number of the liquid water increases
which shortens the breakup time of the water spray. The
effect of the water fades more rapidly when the water
breaks up sooner, and approximately 15% of the original
liquid penetration distance is recovered at 1.5 ms.
The effect of higher liquid latent heat is that more energy
is required to vaporize the spray. The liquid persists for a
longer time and travels further into the domain.
Additionally, the vaporization rate of the overall spray per
unit volume decreases due to the lower temperature from
the water vaporization. Thus, one would expect a high
sensitivity of liquid penetration to latent heat effects.
Figure 6 shows the effect of changing the latent heat of
vaporization of the water equal to that of the fuel. When
the latent heats of the two liquids are set equal, nearly
85% of the original liquid spray penetration is recovered.
As a check on the model, when the physical properties of
the second liquid are set equal to the first liquid, the
original penetration is recovered.
Figure 7: Computational mesh at CA = 0.0.
If a highly stratified distribution is used the effects of the
water are isolated in that section of the injection. If the
water is biased towards the front of the spray, both liquid
and vapor fuel penetrate further because the gas is
already moving when the fuel is injected. When the
water slug is biased towards the end of the spray, fuel
towards the end of the injection penetrates further.
ENGINE RESULTS
The combustion models have been validated by Dittrich
[3] and Schwarz [11] in a number of studies of
DaimlerChrysler heavy-duty diesel engines.
Specifications and operating conditions are given in
Table 1. The computational mesh for the engine
simulations is shown in Figure 7. At TDC, there are 25
cells in the radial direction, 14 cells in the vertical
direction (perpendicular to the piston) and 20 cells in the
azimuthal direction. The injector is a centrally located, 8-
hole common rail type which is modeled as a 45 degree
sector with the parcels initially entering the domain at the
vertex of the sector mesh.
Table 1: Engine Parameters
Parameter Value (units)
Bore 13.00 (cm)
Stroke 15.00 (cm)
Connecting Rod Length 27.3 (cm)
Displacement 2.00 (Liters)
Compression Ratio 17.25
Engine Speed 1080 (rev/min)
Start of Injection -2 ATDC
