вход по аккаунту



код для вставкиСкачать
Int. J. Powertrains, Vol. 6, No. 1, 2017
Natural gas/diesel RCCI CFD simulations using
multi-component fuel surrogates
Andrew G. Hockett*
Department of Mechanical Engineering,
Colorado State University,
Fort Collins, Colorado, USA
Email: [email protected]
*Corresponding author
Greg Hampson
Woodward Inc., Loveland,
Colorado, USA
Email: [email protected]
Anthony J. Marchese
Department of Mechanical Engineering,
Colorado State University,
Fort Collins, Colorado, USA
Email: [email protected]
*Corresponding author
Abstract: Previous attempts to model natural gas/diesel reactivity controlled
compression ignition (RCCI) engines using single fuel component chemical
kinetics have demonstrated difficulties with reproducing the gradual increase in
combustion rate observed experimentally. This study investigates whether
employing a multi-component vaporisation and chemical kinetics model for
diesel fuel can yield closer agreement with experimental combustion rates.
Multi-dimensional CFD simulations are compared against an injection timing
sweep from a GM 1.9 L diesel engine modified with port injected natural gas.
Using the multi-component model for both diesel vaporisation and diesel
chemical kinetics resulted in a closer match with experimental heat release rate
than using single component diesel chemical kinetics. However, the overly fast
combustion rates at ignition could not be completely eliminated. In addition, a
parameter study revealed that the simulation results are strongly sensitive to the
ratio of components in the diesel fuel surrogate, the injected mass, and the
injection velocity.
Keywords: RCCI engine; CFD model; dual fuel; chemical kinetic mechanism;
multi-component surrogate; natural gas; diesel.
Reference to this paper should be made as follows: Hockett, A.G.,
Hampson, G. and Marchese, A.J. (2017) ‘Natural gas/diesel RCCI CFD
simulations using multi-component fuel surrogates’, Int. J. Powertrains,
Vol. 6, No. 1, pp.76–108.
Copyright © 2017 Inderscience Enterprises Ltd.
Natural gas/diesel RCCI CFD simulations
Biographical notes: Andrew G. Hockett is a Practicing Simulation Engineer
for an internal combustion engine research consulting company. He received
his PhD in Mechanical Engineering at Colorado State University in 2015.
Greg Hampson is a senior principal engineer at Woodward, Inc. He received
his PhD in Mechanical Engineering from the University of Wisconsin,
Anthony J. Marchese is a Professor of Mechanical Engineering at Colorado
State University. He received his PhD in Mechanical Engineering from
Princeton University.
Engine manufacturers face increasingly stringent emissions regulations due to concerns
over human health and the environment, specifically with regards to emissions of carbon
monoxide (CO), unburned hydrocarbons (UHC), particulate matter (PM), and oxides of
nitrogen (NOX). Additionally, the US Environmental Protection Agency (EPA) recently
initiated regulations on greenhouse gases (GHG), such as carbon dioxide (CO2), and
raised fuel efficiency standards for light duty vehicles by 2018 and heavy duty vehicles
by 2025 (US Environmental Protection Agency, 2011, 2012). Therefore, engine designers
are facing difficult challenges to meet emissions regulations and increase engine
efficiency. One promising strategy that has received considerable attention over the years
is low temperature combustion, wherein fuel premixed with excess air generates lower
combustion temperatures and less NOX compared to combustion of diesel sprays and
stoichiometric spark ignited (SI) mixtures (Dec, 2009). Additionally, lean premixed
combustion avoids the rich regions within diesel sprays where soot is formed.
Homogeneous charge compression ignition (HCCI) is a low temperature combustion
method whereby compression heating of a lean premixed fuel air mixture in a high
compression ratio engine initiates ignition reactions throughout the chamber nearly
simultaneously. This method generates very fast combustion rates over a short duration
yielding high efficiencies, but also has a narrow operating range and difficulties in
combustion phasing control (Dec, 2009). In recent years some of the obstacles of HCCI
have been overcome by stratifying the charge mixture with two fuels of different
reactivity, known as reactivity controlled compression ignition (RCCI) (Kokjohn et al.,
2009). This strategy uses a premixed low reactivity fuel (e.g., gasoline or natural gas) and
stratification of a high reactivity fuel (e.g., diesel) via early direct injection. By stratifying
the high reactivity fuel, ignition proceeds from areas of high mixture reactivity to areas of
low reactivity, which lengthens the combustion duration, lowers the pressure rise rate and
peak pressure, and allows for high loads to be achieved. Additionally, the load range can
be extended by using a higher percentage of high reactivity fuel at low load and more low
reactivity fuel at high load. Splitter et al. (2011) were able to achieve high efficiency
RCCI operation using gasoline and diesel in a heavy duty single cylinder engine between
4 and 14.5 bar IMEP while meeting EPA emission standards for NOX without after
A.G. Hockett et al.
The use of natural gas for RCCI operation is advantageous for multiple reasons.
Firstly, as shown by Nieman et al. (2012), the difference in reactivity between natural gas
and diesel is wider in comparison to that between gasoline and diesel. This attribute
enables further lengthening of the combustion duration, which consequently results in
lower pressure rise rate, lower peak pressures, and extension to higher engine loads.
Secondly, natural gas can produce less CO2 emissions than gasoline or diesel on an
energy equivalent basis, because it is composed primarily of methane (CH4), which has
higher hydrogen to carbon ratio and a higher heating value. Thirdly, natural gas is
expected to remain less expensive than gasoline or diesel on an energy equivalent basis
for many decades to come (US Energy Information Administration, 2014). Therefore,
utilising natural gas in RCCI combustion allows for a significant range of the operating
map in a diesel engine to achieve higher efficiency with significantly reduced emissions
and lower fuel cost.
A greater understanding of RCCI combustion physics can be obtained through
multi-dimensional modelling, which is necessary to explore emissions formation and
develop optimal control strategies that can overcome the difficulties of combustion
phasing and large pressure rise rate. Previous modelling work with gasoline/diesel RCCI
has been used for optimisation of injection strategies, fuel ratios, and exploring the
effects of EGR (Splitter et al., 2011; Kokjohn et al., 2009; Hanson et al., 2010;
Puduppakkam et al., 2011). Many of these simulations have used n-heptane as a single
component surrogate for diesel fuel, which is based on similarity incetane number and
heating value. However, previous modelling studies on natural gas/diesel RCCI are
limited and have demonstrated additional modelling challenges due to the wider
reactivity difference between diesel and natural gas. Nieman et al. (2012) used reduced
chemical kinetics in a multi-dimensional CFD simulation coupled with a genetic sorting
algorithm to optimise natural gas/diesel RCCI operation over a wide range of speeds and
loads, including 23 bar IMEP. The diesel fuel was modelled using single component fuel
surrogates where n-heptane represented diesel fuel and the natural gas was modelled as
pure methane. However, the study by Neiman et al. did not compare simulated pressure
and heat release rates with experiments and so it is not possible to ascertain the accuracy
of the model when using single component chemistry.
HCCI and RCCI combustion is characterised by both low temperature heat release
(LTHR) and high temperature heat release (HTHR) (Hanson et al., 2010). Doosje et al.
(2014) showed experimentally that the increase in combustion rate for natural gas/diesel
RCCI should be gradual and continuous during HTHR at high natural gas to diesel ratios.
However, the modelling results in the study by Nieman et al. showed apparent heat
release rate (AHRR) profiles with multiple periods of increasing heat release separated by
periods of nearly constant combustion rate during what appears to be HTHR. Dahodwala
et al. (2015) used multi-dimensional CFD to model natural gas diesel RCCI at 6 bar
BMEP to investigate how the mixture changes as the percentage of natural gas is
increased and the causes for high methane and CO emissions. The study by Dahodwala
et al. also used n-heptane to model diesel fuel and methane to model natural gas. Their
comparison between simulation and experiment showed that the CFD model predicted a
noticeably steeper increase in heat release rate at the start of combustion than observed in
experiment. Puduppakkam et al. (2011) showed that for gasoline/diesel RCCI simulations
Natural gas/diesel RCCI CFD simulations
using multi-component gasoline chemistry was able to achieve better agreement with the
AHRR profile in comparison to using a single component. It is therefore of interest to
explore whether a multi-component chemistry and vaporisation model for diesel fuel and
multi-component chemistry model for natural gas can produce a more gradual increase in
combustion rate similar to that observed experimentally.
This study seeks to explore the difficulties and sensitivities associated with
multi-dimensional CFD modelling of natural gas/diesel RCCI. Simulations are compared
against experimental engine data acquired from a GM 1.9 L light duty turbo-charged
diesel engine modified with port injection of natural gas. Comparisons are made between
using single component and multi-component diesel surrogates for vaporisation and
chemical kinetics. To perform these computations, an existing multi-component diesel
reaction mechanism has been combined with a newly reduced mechanism for methane,
ethane, and propane, resulting in a new reduced natural gas/diesel dual fuel mechanism.
A parameter study was also performed to investigate which inputs have the strongest
effect on the AHRR profile. Parameters investigated include the liquid surrogate, the ratio
of diesel components, injected mass, and injection velocity. Finally, the effectiveness of
adding multi-zone chemistry and adaptive mesh refinement are explored.
Reduced chemical kinetic mechanism development
Natural gas and diesel combustion chemistry was modelled using reduced chemical
kinetic mechanisms. Previously, the authors presented a reduced natural gas/diesel
chemical mechanism for conventional dual fuel engines, which used n-heptane for diesel
fuel and methane, ethane, and propane for natural gas (Hockett et al., 2016). This
mechanism, referred to as CSU141, is initially used herein to model RCCI combustion.
The natural gas chemistry in CSU141 was reduced from the detailed methane through
n-pentane mechanism from Healy et al. (2010) at the National University of Ireland
Galway and is referred to in this work as the NUIG mechanism. The n-heptane chemistry
in CSU141 was reduced from the detailed n-heptane mechanism from Curran et al.
(1998, 2002). Recently Pei et al. (2015) presented a reduced chemical kinetic mechanism
for a two component diesel surrogate consisting of n-dodecane and m-xylene, which
consisted of 163 species and 887 reactions, and will be hereafter referred to as the
LLNL163 mechanism. To use the LLNL163 diesel mechanism in the RCCI engine
simulations, the natural gas chemistry from CSU141 was added. The reduction and
combination procedure is shown schematically in Figure 1. The natural gas chemistry
consisted of two reductions of NUIG using the direct relation graph with error
propagation and sensitivity analysis (DRGEPSA). The details of each reduction are listed
in Table 1, which are reproduced here from Hockett et al. (2016). The second reduction
uses a target mixture with higher propane than ethane content, which was necessary to
match the behaviour of the NUIG mechanism for pure propane in air. The species and
reactions from these two reduced natural gas mechanisms were then appended to the
LLNL163 diesel mechanism. Any duplicate reactions between the three reduced
mechanisms were given the rate constant values from the LLNL163 mechanism to ensure
that the kinetics of diesel ignition was preserved. Care was taken to find and add
reactions from the detailed NUIG mechanism that were not in any of the individual
A.G. Hockett et al.
reduced mechanisms, but whose species were all present after combining the reduced
mechanisms. The result was a multi-component mechanism for natural gas/diesel
combustion consisting of 186 species and 1,014 reactions, referred to hereafter as
Figure 1
Schematic diagram illustrating the formulation of the CSU186 multi-component dual
fuel diesel/natural gas mechanism (see online version for colours)
Table 1
Conditions used for reducing the detailed natural gas mechanism using DRGEPSA
Base mechanism
NUIG natural gas
NUIG natural gas
Target species
CH4, C2H6, C3H8
CH4, C2H6, C3H8
Fuel composition
(mole frac.)
XCH4 = 0.85,
XCH4 = 0.93,
XC2H6 = 0.1,
XC2H6 = 0.02,
XC3H8 = 0.05
XC3H8 = 0.05
Temperature (K)
750, 800, 850, 900, 950,
750, 800, 850, 900, 950,
1,000, 1,050, 1,100, 1,150, 1,200
1,000, 1,050, 1,100, 1,150, 1,200
10, 30, 80
10, 30, 80
Equivalence ratio, φ
0.2, 0.4, 0.6, 0.8, 1.0, 1.2, 1.4
0.2, 0.4, 0.6, 0.8, 1.0, 1.2, 1.4
Ignition delay error
Sensitivity analysis
Pressure (bar)
Figure 2 shows comparisons of computed ignition delay period between the CSU186
mechanism and the LLNL163-base mechanism for a diesel surrogate mixture of 70%
n-dodecane and 30% m-xylene at lean and stoichiometric equivalence ratios. The
excellent agreement across the range of temperatures and pressures demonstrates that the
behaviour of the n-dodecane and m-xylene chemistry from LLNL163 has been retained
after addition of the natural gas chemistry. A comparison with the CSU141 n-heptane
mechanism for the 80 bar condition, which is most relevant to engine TDC pressure, is
also shown in Figure 2. The 70% n-dodecane/30% m-xylene mixture has shorter ignition
delay periods compared to n-heptane.
Natural gas/diesel RCCI CFD simulations
Figure 2
Zero-dimensional adiabatic ignition delay calculations comparing the dual fuel
CSU186 to the base LLNL163 mechanism for a diesel surrogate of 70%
n-dodecane/30% m-xylene in air at (a) φ = 0.5 and (b) φ = 1.0 also shown are the
ignition delay periods for n-heptane at 30 and 80 bar using the csu141 mechanism
(see online version for colours)
A similar ignition delay comparison between the reduced CSU186 mechanism and the
detailed NUIG natural gas mechanism is shown in Figure 3 for an equivalence ratio of
0.5 at two different natural gas compositions. One composition consists of 90% methane,
6.7% ethane, and 3.3% propane and the other is a mixture of 70% methane, 20% ethane,
and 10% propane. Close agreement between the two mechanisms across a range of
pressures and temperatures shows that the reduced mechanism can replicate the
behaviour of the detailed mechanism at conditions indicative of engine top dead centre
(TDC) conditions for different natural gas compositions.
Figure 3
Zero-dimensional adiabatic ignition delay calculations comparing the reduced dual fuel
CSU186 mechanism to the detailed NUIG-base mechanism for two different natural
gas compositions of methane/ethane/propane in air, (a) 90%CH4/6.7%C2H6/3.3%C3H8
in air, φ = 0.5 (b) 70%CH4/20%C2H6/10%C3H8 in air, φ = 0.5 (see online version
for colours)
A.G. Hockett et al.
Experimental procedure
Engine experiments were conducted using a four cylinder light-duty General Motors
(GM) 1.9 L common rail, turbo-charged diesel engine. The specifications of the engine
are listed in Table 2, the diesel injector specifications are listed in Table 3, and the
measured gas composition is listed in Table 4. For the experiments described herein, the
engine was modified to allow multi-port injection of natural gas. This modification was
accomplished by building a natural gas rail with Woodward natural gas injectors attached
to the rail and connecting tubing from each injector to the intake manifold runners, such
that the gas was injected perpendicular to the air flow direction.
Table 2
Engine specifications
Cylinder arrangement
4 Inline
Displacement volume
1.9 L
82 mm
90.4 mm
Connecting rod length
161 mm
Compression ratio
Injection system
Max injection pressure
Valves per cylinder
Bosch common rail
1,600 bar
Rated power
110 kW (147.5 hp) @ 4,000 RPM
Rated torque
315 Nm (232 ft-lbf) @ 2,000 RPM
Table 3
Diesel injector specifications
Bosch CRIP 2-MI
Nozzle orifice diameter
Included spray angle
Flow specification
Discharge coefficient
Table 4
141 micron
440 mm3/30 sec
Measured natural gas composition
Mole fraction
The engine was controlled by a Woodward engine control unit. The diesel fuel
consumption for the engine was measured using Micro Motion CMF025 Coriolis mass
flow meters on the supply and return of the fuel tank, which have an accuracy of ±0.1%
of the measured rate. Natural gas consumption was measured using a Micro Motion
Natural gas/diesel RCCI CFD simulations
CMF010 Coriolis mass flow meter, which has an accuracy of ±0.35% of the measured
rate. The air mass flow rate was measured with an automotive hot wire anemometer,
which has an accuracy of ±3%. In-cylinder pressure transducers (Kistler 6058A) were
installed in the glow plug hole of each cylinder and pressure data were acquired using a
National Instruments data acquisition system. Crank angle data were acquired with a
resolution of 0.25 crank angle degrees. Pressure traces were averaged over 100 cycles for
each cylinder and used to calculate the AHRRs for each cylinder, which assumed a
constant ratio of specific heats of 1.3. A three-point moving average was applied to
smooth out small amplitude noise in the AHRR profiles.
Table 5
Experimental operating conditions.
Diesel volume per cylinder (μL/st)
4.6 ± 0.2
Natural gas per cylinder (mg)
Natural gas energy share (%)
Engine speed (RPM)
Diesel rail pressure (bar)
Injection duration (ms)
Commanded start of injection (°ATDC)
Boost pressure (kPa abs)
Manifold temperature (°C)
Air mass per cylinder (mg)
Net IMEP for cylinder 2 (bar)
NOx (ppm)
The experimental operating conditions are listed in Table 5. Only cylinder 2 was operated
in RCCI mode, while the other cylinders operated using conventional dual fuel
combustion, whereby the injection is near TDC. This was done to avoid the inherent
controls issues with maintaining steady state with RCCI. An injection timing sweep at
8.8 bar net IMEP for cylinder 2 at 2,000 rpm using a single diesel injection was
performed while keeping both fuel masses constant. The injected diesel volume listed in
Table 5 is from the injector characterisation experiments described in the computational
method section. The mass from the injector characterisation is preferred as a model input
because it is specific to each injector and avoids the inherent inaccuracies from averaging
the total engine consumption over all four cylinders. Also the injector characterisation is
a direct measurement of fuel flow as opposed to fuel consumption which measures a
supply and return flow, which are both much higher flow rates compared to the
consumption rate. Therefore, the uncertainty in mass per cylinder from the injector
characterisation experiments is much lower than the uncertainty from the consumption
rate. The injector characterisation used an AVL PLU131 mass flow meter, which has a
precision of 0.1%, to directly measure the mass per stroke of the injector needle. The
density of the fuel is then found from a correlation with injected fuel temperature and
used with the mass measurement to calculate a volume per stroke. This volume can then
be used with the engine’s fuel temperature to calculate density and then estimate the mass
per injection. Fuel temperature was measured in the rate tube near the injector. The actual
mass estimates used are described in the computational section. The mass measured from
the injector characterisation was within uncertainty of the average diesel mass per
A.G. Hockett et al.
cylinder calculated from the engine fuel consumption. Small fluctuations in the rate tube
pressure caused the mass flow measurement to fluctuate and resulted in an overall
injected volume uncertainty of ±0.2 μL/st. With the uncertainty in the injected fuel
temperature on the engine, the uncertainty in injected mass is estimated to be ±0.5 mg/st.
Computational method
Computational modelling of the in-cylinder physics was performed using
CONVERGE™ 2.2.0. A 1/7th sector of the cylinder geometry was used for all
simulations in this work, wherein each sector included one of the injector’s seven nozzles
as depicted in Figure 4. All engine sector simulations presented herein began at intake
valve closure (IVC) and were run until 20 crank angle degrees ATDC. A preliminary
GT-POWER simulation incorporating the experimental AHRR profile was performed to
obtain the initial conditions at IVC, such as the mass of exhaust residual, wall
temperatures, and the initial pressure and temperature. This 1-D simulation tool is useful
for matching the experimental compression pressure when using homogenous initial
conditions and constant uniform wall temperatures. The initial velocity field was obtained
from a gas exchange simulation using the full cylinder geometry with intake and exhaust
ports and valve motions. This provided a mapping file that was used to initialise the
velocity and turbulence fields in the sector geometry. The wall temperatures in the sector
simulation were adjusted slightly to better match the start of combustion. The
CONVERGE™ code uses a fixed and uniform Cartesian grid and a cut cell method at the
boundaries. The simulations used a maximum cell width of 1 mm. Adaptive mesh
refinement down to 0.5 mm was used during the compression stroke up until start of
injection (SOI). Fixed embedding with 0.25 mm cells was used throughout the entire
domain from SOI to the end of the simulation. Second order central differencing was
used for the surface fluxes and a fully implicit time marching method was used with a
variable time step.
For the single component diesel surrogate, the gas phase chemistry was modelled
using n-heptane and the CSU141 mechanism. For the multi-component surrogate, diesel
fuel was represented with a blend of n-dodecane and m-xylene as required by the
CSU186 mechanism. For all simulations, natural gas was represented with a methane,
ethane, and propane mixture, whereby all higher carbon number species were represented
with propane. The assumption of a well stirred reactor (WSR) in each cell was used along
with detailed chemical kinetics and a Reynolds averaged Navier-Stokes (RANS)
turbulence model. The RANS turbulence model used was the renormalisation group
(RNG) κ-ε model (Han and Reitz, 1995) and the turbulent heat transfer model used the
law of the wall model from Amsden (1997). While the WSR assumption does not
account for many of sub-grid turbulence-chemistry interactions (TCI), Pomraning et al.
(2014) found that the WSR assumption with a RANS turbulence model is able to achieve
grid convergence for premixed and non-premixed combustion at cell sizes of 0.25 mm
and smaller and that the converged solution agreed well with experimental measurements
for non-premixed jet flames and premixed propagating flames. Pomraning et al.
concluded that while the TCI commutation error between cells averaged temperature and
cell averaged reaction rate cannot be eliminated, in a well-resolved RANS simulation the
magnitude of this error must not be as significant as other sources of uncertainty.
However, this study did not investigate the TCI error for RANS simulations of
Natural gas/diesel RCCI CFD simulations
volumetric premixed auto-ignition characteristic of HCCI and RCCI engines. Resolving
the turbulent eddies may be necessary to replicate the smooth increase in combustion rate
for these types of engines, but that requires a large eddy simulation (LES), which carries
a very large increase in computational time. This work attempts to work with RANS
types models to reduce computational cost and therefore the 0.25mm recommendation of
Pomraning et al. is used.
Figure 4
Computational sector geometry used for engine simulations (see online version
for colours)
The spray model considered in this study was that of the Lagrangian drop Eulerian fluid
type. The dynamics of the spray droplets were modelled using a statistical approach
where parcels are used to represent a group of identical droplets. Droplet breakup was
modelled using the modified KH-RT droplet breakup model, as described in the
CONVERGE™ theory manual (Convergent Science, 2014), which does not use a defined
breakup length. The initial liquid surface disturbances responsible for the primary KH
breakup are modelled using the KH-ACT method of Som and Aggarwal (2010), which
includes effects of aerodynamics, cavitations, and turbulence. Droplet/wall interactions
were modelled using the CONVERGE™ hybrid film model, which uses both particle and
film height quantities to model wall impingement (Convergent Science, 2014). Collision
and coalescence were modelled using the no time counter collision model of Schmidt and
Rutland (2000). Collision outcomes included reflexive and stretching separation as
described by Post and Abraham (2002). The Frossling correlation (Amsden et al., 1989)
was used to model droplet vaporisation. The droplet surface temperature used in the
Frossling correlation is found from an energy balance that assumes a homogeneous
droplet temperature. For multi-component liquid vaporisation, the composition of the
vapor at the droplet surface is found via Raoult’s law and the vapour pressure used in the
Frossling correlation is found through Dalton’s law of partial pressures. Each liquid
component vaporised into its gas phase species. Other properties for multi-component
liquids are mass weighted averages.
The spray model was validated against liquid and vapour penetration measurements
from experiments done by the Engine Combustion Network (ECN) (2012), which
consisted of a vaporising, but non-reacting spray injected into a constant volume chamber
at typical TDC temperature and pressure conditions. Spray simulations were performed
A.G. Hockett et al.
for the 500 bar injection pressure spray A condition, which is an n-dodecane spray, and
the 1,500 bar injection pressure spray H condition, which is an n-heptane spray.
Comparisons of liquid and vapour penetration between the simulations and the
experiments demonstrated close agreement within the first millisecond (Hockett, 2015).
These findings are consistent with those of Senecal et al. (2012), who employed many of
the same spray sub-models and found grid convergent behaviour at a cell size of
0.25 mm.
To model the diesel injection, CONVERGE™ requires a rate of injection profile or
rate shape, injection duration, a discharge coefficient, and a total injected mass to
determine how much mass to inject per time step. Rate shape profiles, applied current
traces, time averaged mass flow, and fuel temperature were collected for each injector
across sweeps of electronic pulse width at fixed rail pressure using a Bosch style rate of
injection tube apparatus (Hockett, 2015). This sweep was repeated for rail pressures from
400 to 1,200 bar. These measurements also provided the dwell time between applied
injector current and start of fuel flow, which determined the SOI timing in the
simulations. The injector characterisation data also provided a means to determine the
volume per injection for a specific electronic pulse width at a specific rail pressure. These
injection experiments were conducted using an industry standard calibration fluid for
diesel injection equipment, which was provided with a correlation for density with
temperature to determine the volume flow rate from the mass flow measurement. For the
experimental SOI timing sweep described in Table 5, an electronic pulse width of 400 μs
was used for all diesel injections and the diesel rail pressure was held at 700 bar. At this
condition, the injector characterisation predicts that the volume of diesel fuel per
injection is 4.6 ± 0.2 microlitres per stroke (μL/st).
Figure 5
Comparison of fuel liquid density with fuel temperature for the diesel liquid surrogates
investigated (see online version for colours)
In this study different diesel surrogate liquid properties and different injected liquid
temperatures were used to determine the density and therefore the mass of diesel to use as
an input to the model. Figure 5 shows a plot of density as a function of temperature for
the different diesel surrogates used. The CONVERGE™ liquids property library has a
built-in single component surrogate for diesel fuel called DIESEL2. However, the
DIESEL2 density does not vary with temperature, unlike the other properties. To remedy
this issue, the DIESEL2 surrogate was modified by adding the density correlation used
for the calibration fluid in the rate of injection experiments, which is shown in Figure 5.
For the multi-component diesel surrogate, the liquid density was represented by a mass
Natural gas/diesel RCCI CFD simulations
weighted average between n-dodecane and m-xylene. The initial two-component diesel
surrogate investigated consisted of 30% m-xylene and 70% n-dodecane and its density is
shown in Figure 5.
Figure 6
Comparison of experimental rate of injection profiles for two different injection
durations (see online version for colours)
Note: The rate shapes have been normalised by the steady state injection rate.
Table 6
Calculated injected mass and velocity for different diesel surrogates investigated
temp (K)
ModifiedDIESEL2 liquid,
n-heptane chemistry
70% n-dodecane/30%
70% n-dodecane/30%
75% n-dodecane/25%
65% n-dodecane/35%
70% n-dodecane/30%
Diesel surrogate
The rate shape for the experimental condition investigated is compared with the steady
state injection rate in Figure 6. The rate shape for this condition shows an injection
duration of 3.5 ± 0.5 crank angle degrees. Dual fuel combustion has been shown to have
higher injector tip temperatures than conventional diesel operation at high substitution
due to lower diesel flow rates providing less cooling. Königsson and Stalhammar (2012)
measured injector tip temperatures as high as 600 K in a dual fuel engine operating at
80% substitution and 13 bar BMEP. This high tip temperature can cause the first injected
fuel to have higher temperature than that in the rail. For short injection durations such as
that in Figure 6, this can imply that the average fuel temperature used as an input should
A.G. Hockett et al.
be closer to the tip temperature (Leuthel et al., 2008). While the temperatures are lower in
RCCI, the short injection duration used and high substitution is reason to use a higher
fuel temperature than the measured fuel return temperature as a model input. The initial
fuel temperature used in the simulations was 500 K. The injected mass and peak injection
velocities for the various liquid surrogates, fuel temperatures, and injected volumes
investigated in the simulations are shown in Table 6. The steady state injection velocity
from the rate shape in Figure 6 was 213 m/s.
The experimental pressure and AHRR plots for the injection timing sweep are shown in
Figure 7. The six crank angle degree advancements resulted in a four degree retard in
CA10 timing, a six degree retard in CA50 timing, and a 7.5 degree increase in CA10-90
combustion duration. This behaviour is in agreement with other experimental studies on
RCCI combustion (Splitter et al., 2011; Hanson et al., 2010). The peak LTHR and HTHR
decreases as combustion phasing retards.
Figure 7
(a) Experimental pressure and AHRRs for the SOI timing sweep (b) AHRRs during
LTHR for the same SOI sweep (see online version for colours)
5.1 Single component diesel model
Computational modelling was first performed for the –46° ATDC diesel injection timing
case. The simulation results using the modified DIESEL2 liquid properties and n-heptane
chemistry from CSU141 to represent diesel fuel are compared against experimental
pressure and AHRR in Figure 8. The phasing of LTHR is slightly later in comparison to
experiment, but the start of the HTHR is in close agreement. However, the CA10-90
combustion duration is longer than the experimental and the CA50 has later phasing.
Additionally, there is a significant difference in the rate of change in the AHRR profile at
the start of HTHR with the simulation predicting a larger and faster rise in AHRR. After
the initial fast combustion rate at main ignition the simulation shows a period of nearly
constant heat release rate between –8° and –6°ATDC. This is followed by a second rise
in heat release rate to a maximum that is less than the experimental peak and occurs at a
more retarded phasing, which causes the lower peak pressure. The simulated AHRR
Natural gas/diesel RCCI CFD simulations
profile therefore shows two distinct events where the heat release rate is increasing, while
the experimental shows a single continuous increase in combustion rate. For the single
component n-heptane diesel chemistry surrogate, the faster rise in AHRR at main ignition
could not be eliminated with further tuning of such parameters as injected mass, natural
gas mass, exhaust residual mass, wall temperatures, turbulent kinetic energy, or swirl
Figure 8
Comparison between experimental and simulated pressure and AHRR for the
SOI = –46° ATDC case with the single component n-heptane chemistry used to model
diesel chemical kinetics (see online version for colours)
5.2 Multi-component diesel surrogate vaporisation and chemical kinetic model
The initial fast rise in AHRR at ignition could be a consequence of modelling the diesel
fuel as a single component liquid, both in terms of vaporisation and the gas phase
chemical kinetics. To separate these two competing effects, the simulation was
initially performed using a two-component liquid surrogate of 70% n-dodecane/30%
m-xylene, while the gas phase kinetics were modelled with n-heptane. The simulation
was then repeated using both the two-component liquid surrogate and the CSU186
two-component chemical mechanism. Figure 9 is a comparison of simulated pressure and
AHRR between the single component n-heptane diesel surrogate model from Figure 8,
the two-component liquid surrogate/single component gas phase surrogate, and the
two-component liquid/two-component gas phase surrogate. For all three simulations the
natural gas mixture is the same.
Figure 9
A.G. Hockett et al.
Experimental and simulated pressure and AHRR for SOI = –46° ATDC (see online
version for colours)
Notes: Simulation results include single component n-heptane diesel surrogate model (red
lines), the two-component liquid surrogate/single component gas phase kinetics
model (green lines), and the two-component liquid/two-component gas phase
kinetics model (blue lines).
A comparison of the red and green curves in Figure 9 shows that adding a
two-component liquid surrogate changed the start of combustion phasing by one crank
angle degree and consequently lowered the peak pressure. Other than the small change in
start of combustion phasing, the AHRR profiles are very similar in shape and magnitude.
The LTHR is still under predicted and has later phasing than the experiment. Therefore,
the two-component vaporisation model did not have a dramatic effect on the AHRR. In
contrast, the addition of both the two-component liquid surrogate and the two-component
diesel gas phase surrogate, as shown by the blue curve, results in a LTHR profile that is
in much closer agreement with experiment and also results in a peak heat release rate that
is closer to the experimental peak AHRR, which leads to a closer match in peak pressure.
The start of combustion phasing also has closer agreement with experiment. The initial
fast rise in combustion rate at ignition is still present. However, the transition to the
second period of increasing heat release rate occurs sooner and increases at a faster rate,
which is closer to the experimental result. The fact that the fast increase in combustion
rate at ignition could not be reduced with the two-component liquid surrogate/
two-component gas phase surrogate for diesel fuel suggests that it is caused by other
effects not accounted for in this model, such as turbulence chemistry interactions and
error in the convection and diffusion rates of the injected fuel affecting the air/fuel ratio
In-cylinder images of local equivalence ratio and temperature from the simulations in
Figure 9 are compared in Figure 10. The top row of images shows a cross-section plane
coloured by equivalence ratio just before the start of LTHR at –20° ATDC for each of the
three diesel surrogate models. The middle and bottom rows show equivalence ratio and
temperature, respectively, for each of the three diesel surrogate models at –12° ATDC,
which is just prior to HTHR. For all three diesel surrogate cases, a vapour plume of
injected fuel travels towards the liner wall it rotates counter-clockwise through the
periodic boundaries due to the swirl motion. A volume of rich mixture is apparent at the
leading edge of the injected fuel. The presence of injected fuel in the squish region prior
to ignition agrees with optical observations of RCCI made by Kokjohn et al. (2012).
Natural gas/diesel RCCI CFD simulations
Figure 10 Comparisons of planar cross-sections of local equivalence ratio and temperature for the
three diesel surrogates investigated (see online version for colours)
Notes: The top row of images shows equivalence ratio prior to LTHR. The middle and
bottom rows show equivalence ratio and temperature, respectively, prior to
At –20° ATDC, the single component model has the largest volume of rich mixture,
which is located closer to the liner showing greater vapour penetration. The larger
volume of rich mixture can be explained by referring to Table 6, where it is seen that, for
liquid fuel temperature of 500 K, the mass of injected fuel is highest for the modified
DIESEL2 liquid surrogate due to its higher density. The greater injected fuel mass creates
more LTHR as indicated by the higher gas temperature in the bottom row of images of
Figure 10. Comparing the two-component liquid surrogate cases shows that the fuel
distributions are nearly identical, but the case with the 70% n-dodecane gas phase
kinetics has a larger region of high temperature than the case with n-heptane gas phase
kinetics. This result is due to n-dodecane being more reactive than the n-heptane as
indicated by the ignition delay comparison in Figure 2, which results in more energy
released during LTHR. Therefore, the higher reactivity of n-dodecane in the
two-component gas phase surrogate results in closer agreement with the experimental
AHRR profile during LTHR in terms of phasing and magnitude.
A.G. Hockett et al.
Figure 11 (a) Planar cross-section of equivalence ratio distribution with a line showing where
data was extracted. Variation in (b) equivalence ratio, (c) temperature, and (d) heat
release rate are presented at different crank angles as a function of distance along the
white line with the origin at the white dot in the lower right hand side of the plane (see
online version for colours)
Figure 11(a) shows a line through the rich region of injected fuel for the two-component
liquid and gas phase diesel surrogate model. Figure 11(b) and Figure 11(c) plot the
equivalence ratio and temperature across this line prior to ignition at –12° ATDC and just
after ignition at –10° ATDC. The white dot marks the origin for distance in the plots. The
peak in temperature (1,150 K) along this line at –12° ATDC is a result of the LTHR as
explained in Figure 10. The temperature of the natural gas far from any injected diesel
fuel at this time is approximately 900 K due to compression. Because the natural gas is
assumed to be homogeneous, the gradient in equivalence ratio is a consequence of the
gradient in diesel fuel concentration and therefore mixture reactivity. The results suggest
that just prior to ignition, large spatial gradients exist in the diesel fuel concentration (and
therefore equivalence ratio) and large homogeneous regions of diesel fuel concentration
are not observed. Therefore, the overly fast combustion rate at ignition is not likely due to
the model’s inability to generate a gradient in gas mixture reactivity. Figure 11(d) shows
profiles of local heat release rate along the same line from Figure 11(a) at different crank
angles during ignition. During this period, the gas is slowly travelling to greater distances
Natural gas/diesel RCCI CFD simulations
along the line due to the counter-clockwise swirl motion. Heat release begins at the
richest location as seen by the profile at –11.5° ATDC. Then at –11° ATDC, the heat
production grows to a nearly constant value from 7 to 9 mm. The heat release rate near
7 mm stays constant at about 50 W/mm3 between –11.5° and –11° ATDC, but then
increases at –10.5°. Between –10.5° and –10° ATDC, the gas between 8 and 10 mm
releases all of its energy from combustion reactions thereby demonstrating a nearly
simultaneous volumetric ignition. The profile at –10° ATDC also shows two thin reaction
zones, which subsequently propagate into the unburned gas. It is the fast combustion
between –10.5° and –10° ATDC that is responsible for the sudden rise in global AHRR at
ignition seen in Figure 9.
The flat heat release rate profile at –11° ATDC and the nearly simultaneous ignition
between 8 and 10 mm in Figure 11(d) are unexpected behaviours because there is a
gradient in temperature and equivalence ratio across this region that should create a more
sequential ignition process. The equivalence ratio and temperature profiles in
Figure 11(b) and Figure 11(c) show that the gas located near the highest concentration of
injected fuel is between 900 and 1100 K and has an equivalence ratio between 0.9 and
1.1. Referring to the ignition delay plots in Figure 2 shows that the ignition delay periods
decrease substantially at temperatures greater than 1,000 K and for near stoichiometric
mixtures at this pressure (55 bar) the ignition delay periods become close to one crank
angle at 2,000 rpm (0.08ms). Moreover, the difference in ignition delay between the
stoichiometric mixture and the φ = 0.5 mixture at these temperatures and pressures is a
fraction of a crank angle. While the ignition delay calculations are not truly representative
of the time temperature history of the gas in the engine, the ignition delay calculations
demonstrate that having a volume of the mixture near stoichiometric and at temperatures
greater than 1,000 K can create very similar chemical kinetic reaction rates that are fast
enough relative to one crank angle degree to behave like a simultaneous ignition. Because
such a simultaneous ignition was not observed in experiment, it is possible that the model
could be over predicting the maximum equivalence ratio at the time of ignition. Optical
RCCI engine experiments by Kokjohn et al. (2012) did not observe any fuel rich regions
in the squish region for a similar injection timing of –50° ATDC and instead measured a
maximum equivalence ratio of 0.5. It is unclear if the mixture differences between the
optical experiments of Kokjohn et al. and this model are due to differences in engine
geometry and operating conditions or a consequence of shortcomings in the RNG κ-ε
turbulence model to accurately capture turbulent mixing rates in the squish region.
Figure 12 shows the progression of a heat release rate iso-surface of 30 W/mm3,
which represents the boundary between the burned and unburned regions of the gas. The
period of nearly constant heat release rate after ignition occurs from –8° to –6° ATDC.
The image at –8°ATDC in Figure 12 shows the burned region has consumed most of the
region where the injected fuel was concentrated. The image at –5° ATDC shows that
burned region has expanded into the bowl region and therefore increased in surface area
significantly. A larger surface area over which combustion reactions occur generates
higher global AHRR, which is confirmed by the acceleration in AHRR in Figure 9
beginning around –5° ATDC. Therefore, the period of nearly constant AHRR is caused
by a lack of injected fuel available to burn and continues until the burned region expands
into the piston bowl. This suggests that the model could be over predicting the vapour
penetration into the squish region.
A.G. Hockett et al.
Figure 12 Time progression of heat release rate iso-surfaces for different diesel surrogates
(see online version for colours)
The two-component chemistry model had a higher peak AHRR than the other two
models which was in closer agreement with the experimental peak AHRR. The
explanation for the faster heat release can be explained in Figure 13. The image at –12°
ATDC depicts the distribution of the injected fuel prior to HTHR and clearly shows
injected fuel present in bowl region due to the tail end of the vaporising spray. The local
heat release rate and equivalence ratio are shown at –1° ATDC, which is when the
differences in global AHRR are apparent in Figure 9. The model with the two-component
diesel chemical kinetics surrogate clearly shows ignition reactions occurring in the gas
between the burned regions whereas the two models using n-heptane for diesel chemical
kinetics does not. The equivalence ratio image shows that three diesel surrogate models
have essentially equal concentration of injected fuel in the bottom of the bowl at this
time. This additional volumetric heat release causes faster propagation of the burned
region. Therefore, the faster acceleration in AHRR and higher peak AHRR observed for
the two-component gas phase surrogate is due to the more reactive chemical kinetics of
n-dodecane in comparison to n-heptane causing volumetric heat release rate throughout
the bottom of the piston bowl.
Natural gas/diesel RCCI CFD simulations
Figure 13 Cross-sectional planes showing the distribution of injected fuel at –12° ATDC and
local heat release rate and equivalence ratio in the bottom of the piston bowl at –1°
ATDC for the different diesel surrogates (see online version for colours)
5.3 Ratio of diesel fuel components
The effect of changing the ratio of n-dodecane to m-xylene in the diesel surrogate is
investigated in Figure 14. The 70% n-dodecane/30% m-xylene simulation from Figure 9
was used as base case. A comparison was made between the base case and a 65%
n-dodecane/35% m-xylene diesel surrogate and a 60% n-dodecane/30% m-xylene
surrogate. The volume of injected fuel was kept constant and the injected mass varied by
less than 1% as detailed in Table 6. As more n-dodecane is added to the diesel surrogate
the start of combustion advances and there is greater energy released during the fast
combustion rate at ignition. This leads to a larger difference in pressure rise rate at start
of combustion. The experimental peak AHRR, CA50 timing, peak pressure, and
combustion duration are most closely matched by the 75% n-dodecane/25% m-xylene
case. In addition, 75% n-docecane surrogate provides the closest prediction to the
experimental AHRR between –5° and 5° ATDC, which represents the majority of the
combustion duration. Furthermore, the peak AHRR during LTHR is more closely
matched by the 75% n-dodecane case even though the phasing of the LTHR peak is about
1 crank angle degree more advanced than the experimental peak. The 65% n-dodecane/
A.G. Hockett et al.
35% m-xylene has the most gradual increase in combustion rate at ignition, which tends
to agree with the experimental rise in AHRR. However, the 65% n-dodecane case has a
more retarded start of combustion and CA50 phasing and consequently has lower peak
AHRR and peak pressure. For the purpose of matching general engine performance
metrics, such as CA50, peak pressure, peak AHRR, and combustion duration, the 75%
n-dodecane/25% m-xylene case predicts the closest match with experiment. The results
from Figure 14 demonstrate that the combustion phasing and consequently the engine
performance predicted by this RCCI model is sensitive to a 5% change in the ratio of
diesel surrogate components. This presents a challenging obstacle for RCCI modelling as
the ideal ratio of n-dodecane to m-xylene to use as a surrogate could vary from one diesel
fuel sample to another.
Figure 14 Comparison of pressure and AHRR as the ratio of n-dodecane to m-xylene in the
diesel surrogate is changed (see online version for colours)
The mass fraction distribution of the two diesel fuel components for the n-dodecane
percentage sweep are shown in Figure 15 at –20° ATDC, which is prior to start of LTHR,
and at –14°ATDC, which is after LTHR but before HTHR. The image at –20° ATDC
clearly shows that the location of the richest region is nearly equivalent in all three cases
with the peak concentration of n-dodecane scaling with the percentage of n-dodecane
used for liquid diesel surrogate. Examination of the equivalence ratio at –20° ATDC
shows equivalent distribution between all three surrogate mixtures, thereby indicating
that differences in combustion behaviour are due to mixture reactivity and not air/fuel
ratio. The images at –14° ATDC shows that as the n-dodecane percentage in the diesel
surrogate increases the volume of n-dodecane consumed in the richest region during
Natural gas/diesel RCCI CFD simulations
LTHR increases and so too does the peak temperature. This agrees with the greater peak
AHRR during LTHR seen in Figure 14. The higher peak temperature at this location then
leads to the more advanced phasing of the HTHR ignition. Furthermore, the model with
75% n-dodecane has a larger volume of gas with a temperature greater than 1,000 K,
which creates a larger volume of nearly simultaneous auto-ignition and results in the
larger rise in AHRR at ignition seen in Figure 14. The images in Figure 15 demonstrates
how the reactivity of the diesel surrogate employed strongly affects the energy released
during LTHR, which in turn determines the start of combustion phasing for HTHR and
consequently the engine performance predicted.
Figure 15 Cross-section planes showing the distribution of injected fuel and temperature
showing where LTHR has occurred as the ratio of n-dodecane to m-xylene is changed
(see online version for colours)
A.G. Hockett et al.
The simulation results demonstrated that as percentage of n-dodecane in diesel surrogate
increased, there was a faster acceleration in AHRR that resulted in a higher peak AHRR.
This trend can be explained in Figure 16, which depicts a cross-sectional plane of
temperature at –5° ATDC when the global AHRR begins accelerating. The unburned gas
temperature is higher for a higher percentage of n-dodecane because the start of
combustion phasing is more advanced and therefore a greater portion of the combustion
occurs during compression. The greater temperatures in the piston bowl cause earlier
volumetric heat release throughout the bottom of the piston bowl as seen in the image at –
2° ATDC. With earlier onset of heat release throughout the piston bowl, compression
heating helps keep the temperatures high, yielding faster reaction rates and faster
propagation of the burned region into the piston bowl. Therefore, differences in the
acceleration of AHRR and resulting peak AHRR between the three simulations are due to
differences in combustion phasing, which is a result of the differences in reactivity
between the three diesel surrogates.
Figure 16 Distributions of temperature and local heat release rate as the ratio of n-dodecane to
m-xylene is changed (see online version for colours)
5.3 Sensitivity to injected diesel mass
The injector characterisation experiments allow for the injected mass to be known within
±0.5 mg thus allowing for a range of values to use as a model input. Figure 17 explores
the sensitivity of the injected mass by comparing the base case of 4.6 μL per injection to
a case with 5.4 6 μL per injection. Using the density at 500 K, these injected volumes
correspond to 2.78 mg and 3.3 mg per injection respectively, which spans the range of
uncertainty in injected mass. Both of these cases used the two-component diesel
surrogate with 70% n-dodecane and 30% m-xylene. Because the injection duration was
Natural gas/diesel RCCI CFD simulations
kept constant, the 3.3 mg/st case has a higher peak injection velocity of 229 m/s
compared to the 194 m/s for the 2.78 mg/st case. The greater diesel injection has greater
LTHR, which leads to earlier start of HTHR and greater energy released at ignition. Later
the AHRR profiles for the two simulations become nearly equal to each other from –8° to
–1° ATDC. In general, a small change of 0.5 mg, which is within the uncertainty of the
measurement, has advanced the start of combustion by 1.5 crank angle degrees and
increased the peak pressure by 13 bar, thereby demonstrating that this RCCI model is
sensitive to the injected mass.
Figure 17 Comparisons of pressure and AHRR as injected mass is increased and injection
duration is held constant (see online version for colours)
The fuel and temperature distribution between LTHR and HTHR is shown in Figure 18 at
–14° ATDC. The fuel in the 3.3 mg/st case has penetrated farther due to the greater
injection velocity. The further penetration leads to the highest concentration of fuel being
near the liner wall in the squish region. Once the rich volume of the plume reaches the
wall there is further enrichening as upstream fuel continues to travel towards the same
location. The larger volume of rich mixture leads to more LTHR at this location. Also the
injected fuel is distributed over a larger volume. The iso-surface of local heat release rate
at TDC shows that more of the squish region has been consumed and the surface area of
the burned region is larger. Therefore, the higher peak AHRR for the 3.3 mg/st case is
due to faster consumption of the natural gas in the squish region.
A.G. Hockett et al.
Figure 18 In-cylinder images for the two different injected masses (see online version
for colours)
5.4 Sensitivity to injection velocity
The model has been shown to be sensitive to the injected mass, but as injected mass was
changed the injection velocity also changed because the injection duration was held
constant. Therefore, it is of interest to explore how sensitive the model might be to
injection velocity when the injection mass is held constant. In order to achieve higher
injection velocity the discharge coefficient in the base case was lowered from 0.86 to 0.7,
which changed the peak injection velocity from 194 m/s to 228 m/s. Attempts to increase
the discharge coefficient were not able to appreciably increase the injection velocity.
Therefore, the injection duration was increased from 3.5 crank angle degrees to 4.2 crank
angle degrees, which lowered the injection velocity from 194 m/s to 160 m/s. This
change in injection duration is only slightly greater than the reported uncertainty range of
±0.5 crank angle degrees. A comparison of the pressure and AHRR between the
simulations is presented in Figure 19. The combustion rates during LTHR are essentially
equal between the cases as is the main combustion timing. However, the period of nearly
constant combustion rate after ignition is very different between the three simulations,
with the higher injection velocity having a longer period of constant AHRR. The
differences in the period of constant AHRR also causes the CA50 timing to vary between
the three simulations. The slowest injection velocity case appears to have the closest
agreement with the gradual increase in AHRR and therefore the closest agreement in
pressure rise rate. The results of Figure 19 demonstrate that the model is sensitive to
injection velocity.
Natural gas/diesel RCCI CFD simulations
Figure 19 Comparisons between pressure and AHRR as injection velocity is changed and
injection mass is held constant (see online version for colours)
Note: The base case uses the 70% n-dodecane/30% m-xylene diesel surrogate.
In order to explore the cause for the difference in the predicted CA50 timings, a
comparison of heat release rate iso-surfaces of 30 W/mm3 are presented in Figure 20. The
image at –11° ATDC shows the ignition location which is located where the richest
mixture of injected fuel exists. As the injection velocity is increased the vapour
penetration into the squish region increases. The ignition location for the slowest
injection velocity case does not occur in the squish region and instead occurs near the
bowl rim. This leads to a dramatic difference in how the rest of the burned region grows.
The image at –5°ATDC shows that the highest injection velocity case has a burned
region that is beginning to expand into the piston bowl, the base case has expanded
further into the bowl, and the slowest injection velocity case has grown significantly into
the bowl, but has not yet burned into the squish region. These differences in surface area
of the burned region correspond to the differences in global AHRR seen in Figure 19.
Examination of Figure 20 at 5°ATDC, when the AHRR is decreasing for all three cases,
shows that the highest injection velocity case burned the furthest into the squish region.
At this time, all three cases have burned a similar volume of gas in the bowl region.
Therefore, the highest injection velocity case has the largest surface area and
consequently a higher global AHRR. The differences in burn behaviour between the
highest and lowest injection velocity cases are significant, because the initial growth in
the burned region occurs in the squish region for the highest injection velocity case and in
the bowl region for the lowest injection velocity case. Given that the lowest injection
velocity case has closer agreement with the experimental rise in AHRR, it is likely that
A.G. Hockett et al.
injection duration of 4.2 degrees might be more accurate than the 3.5degree measurement
taken from the rate shape. This is possible as the uncertainty range allows for injection
duration as long as 4.0 degrees. The results of Figure 20 show that the location of ignition
is an important factor that it is sensitive to the injection velocity. For the short injection
duration used in this experiment, the value for injection duration needs to be known with
greater accuracy than ±0.5 crank angle degrees in order to predict the injection velocity
Figure 20 Comparison between heat release rate iso-surfaces of 30 W/mm3 at different injection
velocities (see online version for colours)
5.5 Use of multi-zone chemistry and adaptive mesh refinement
The simulations presented thus far have not included any speed up techniques and so the
computational time has been expensive. One such speed up technique is multi-zone
chemistry whereby cells with similar temperature and composition are solved for
simultaneously by the chemistry solver and the solution is remapped to the cells. This
procedure reduces the number of calls to the chemistry solver, which can often be the
most computationally expensive part of each time step. However, because RCCI is a
chemical kinetics dominated combustion regime, the simulation results can be sensitive
to how the multi-zone chemistry method chooses cells to be similar in temperature and
composition. Another speed-up technique is the use of adaptive mesh refinement (AMR),
Natural gas/diesel RCCI CFD simulations
which only refines the grid where gradients in temperature, velocity, and/or species mass
fraction are above a specified tolerance. Thus a larger-base cell width can be used
throughout large portions of the domain and the total cell count is reduced. To test these
speed-up techniques the same base case from Sections 5.4 and 5.5 was repeated first with
multi-zone chemistry, then with AMR, and finally with both multi-zone and AMR.
Figure 21 Comparison between simulations with and without multi-zone chemistry and adaptive
mesh refinement (see online version for colours)
Table 7
A multi-zone chemistry strategy for RCCI simulations
Multi-zone bin: temperature (K)
Multi-zone bin: progress φ
Multi-zone bin: n-C12H26 mass fraction
Multi-zone bin: m-xylene mass fraction
Multi-zone bin: HO2mass fraction
Multi-zone bin: OH mass fraction
Table 8
An adaptive mesh refinement strategy for RCCI simulations
AMR dimension
Sub-grid error tolerance
velocity (m/s)
temperature (K)
n-dodecane mass fraction
Methane mass fraction
m-xylene mass fraction
Figure 21 shows the base simulation compared to a simulation using the multi-zone
chemistry strategy in Table 7 and the AMR strategy in Table 8. The simulations using
only multi-zone and only AMR are not presented for conciseness, but the agreement was
similar that in Figure 21. The agreement between the two simulations in Figure 22 is
excellent and the computational time was reduced from 47.2 hours to 16.4 hours when
using 64 cores. The AMR strategy reduced the maximum cell count from 986,290 cells to
313,868 cells. The agreement in Figure 21 was obtained after multiple iterations using
different multi-zone bin sizes and bin dimensions and using different AMR refinement
tolerances. It was found that species-based AMR was necessary to obtain agreement in
the start of combustion phasing. Also including more species in the multi-zone strategy
A.G. Hockett et al.
did not always yield more accurate results. Therefore the model can be sensitive to the
multi-zone and AMR strategy employed. Future work is still needed to verify that the
speedup strategies presented will perform well at other diesel to natural gas fuel ratios
and different overall equivalence ratios.
Figure 22 Comparison between experimental data and simulation prediction for the SOI = –52°
ATDC case using the 75% n-dodecane/25% m-xylene diesel surrogate (see online
version for colours)
Simulation Results for SOI = –52° ATDC
The 75% n-dodecane/25% m-xylene surrogate in Figure 14 had the closest agreement
with the experimental start of combustion, peak pressure, CA50, and combustion
duration. For this reason the 75% n-dodecane surrogate was used to model the more
advanced injection timing case of SOI = –52° ATDC, which is compared with
experimental pressure and AHRR in Figure 22. A slight variation in initial temperature
and pressure was applied to the inputs files to reflect the slight change in trapped air mass
as described in Table 5. The piston and head temperatures were reduced by 30 K because
this more retarded combustion phasing had lower load and therefore lower peak gas
temperatures. The lower gas temperatures also result in less heat transfer to the injector
tip and so the injected fuel temperature was lowered to 400 K. Figure 22 shows that the
phasing and magnitude of LTHR are well captured by the model. The start of combustion
is close to the experimental, but again the simulation shows a much more rapid increase
in combustion rate at ignition than seen in experiment. The pressure rise rate and the
combustion duration agree well with experiment. The peak heat release is lower than
Natural gas/diesel RCCI CFD simulations
experiment leading to a 4 bar lower peak pressure and lower pressure during expansion.
Overall, the agreement with experimental pressure and AHRR shows that the model is
able to capture the changes in pressure and AHRR due to injection timing change.
This study investigated multi-dimensional CFD modelling of natural gas/diesel RCCI by
incorporating multi-component fuel surrogates for both diesel and natural gas. This
necessitated the formulation of a new chemical kinetic mechanism for natural gas and
diesel dual fuel combustion, which represents diesel fuel as a blend of n-dodecane and
m-xylene and natural gas as a mixture of methane, ethane, and propane. The n-dodecane
and m-xylene chemistry was taken from an existing reduced mechanism from Pei et al.
(2015) and the natural gas chemistry was reduced from the detailed mechanism from
Healy et al. (2010). Experimental RCCI data from a SOI timing sweep on a light duty
diesel engine was modelled using different combinations of the multi-component and
single component vaporisation and chemical kinetics models for diesel fuel. Model
sensitivity studies were then performed on the ratio of diesel surrogate components, the
injected mass, and the injection velocity. Finally, a multi-zone chemistry and adaptive
mesh refinement strategy was investigated to assess the accuracy of adding these speed
up techniques. These comparison studies resulted in the following conclusions.
Introducing a two-component diesel surrogate did not eliminate the large sudden
increases in combustion rate at ignition. In-cylinder images from the simulations
showed a volume of concentrated diesel fuel vapour in the squish region that
demonstrated a nearly simultaneous ignition throughout. While, there was a gradient
in equivalence ratio and therefore mixture reactivity throughout this volume, the
equivalence ratios were close to stoichiometric causing fast reaction rates with
differences in ignition time across the volume of less than one crank angle.
Therefore, the model employed in this work could be under predicting the turbulent
mixing rates of the injected fuel vapour and consequently over predicting the peak
concentration of high reactivity fuel.
Applying the two-component liquid vaporisation and two-component gas phase
chemical kinetics diesel surrogate resulted in closer agreement with the experimental
peak AHRR, combustion duration, and phasing of LTHR compared to the single
component vaporisation and single component gas phase chemical kinetics model.
Including multi-component chemical kinetics had a more appreciably change in the
predicted AHRR than including multi-component vaporisation.
The simulations showed a period of nearly constant combustion rate after ignition.
This period becomes longer as the location of ignition begins further into the squish
region because the combustion zone must propagate a further distance to reach the
piston bowl, where the combustion surface area is able to expand greatly.
Volumetric heat release occurs in the bottom of the piston bowl when using the
two-component diesel chemistry model. This was not seen when using the single
component surrogate due to the lower reactivity of n-heptane compared to
n-dodecane. This volumetric heat release was responsible for the closer agreement
with the experimental peak AHRR.
A.G. Hockett et al.
The RCCI model was shown to be sensitive to the ratio of diesel surrogate
components. A 5% change in the ratio of diesel components resulted in a two crank
angle degree advancement in start of combustion timing. This advancement in start
of combustion increased caused faster volumetric heat release in the piston bowl
resulting in higher peak AHRR. The 75% n-dodecane mixture yielded the closest
agreement with start of combustion phasing, peak AHRR, combustion duration, and
peak pressure for both injection timings.
The RCCI model was shown to be sensitive to injected mass. A 0.5 mg change in
injected mass resulted in a 1.5 crank angle degree advancement in ignition timing
and an increase in peak pressure of 13 bar. Given that the injected mass measurement
has an uncertainty of ±0.5 mg/st, greater accuracy in the diesel mass measurement is
needed at the condition investigated.
The RCCI model was shown to be sensitive to injection velocity, because it strongly
effects the location of ignition. A lower injection velocity obtained using a longer
injection duration of 0.75 crank angles degrees resulted in the initial growth of the
burned zone occurring in the piston bowl region rather than the squish region, which
yielded closer agreement with the experimental rise in AHRR. Given that the
injection duration has an uncertainty of ±0.5 crank angle degrees, the accuracy of the
injection duration measurement needs improvement at the condition investigated.
A multi-zone chemistry and AMR strategy is presented which yielded very small
error in the pressure and AHRR and had a computational time savings of 65%. This
required dimensions of chemical species in both the multi-zone and AMR strategies.
The accuracy of this strategy needs further validation at other diesel to natural gas
ratios, different loads, and different air/fuel ratios.
Future work should investigate the validity of the spray model to predict vapour
penetration with long mixing times when using short duration pulse injections, because
the transient behaviour of the injector has a large impact on the distribution of the high
reactivity fuel. This will require experimental spray data from high speed imaging and
laser diagnostics of short pilot injections with long mixing times. In particular, this data
would test whether the RNG κ-ε turbulence model employed is accurately representing
the vapour penetration and mixing rates that govern the ignition behaviour. In addition,
modelling of RCCI with higher diesel to natural gas ratios should be investigated to
determine if the observed high sensitivities to injected mass and injection duration are
only present when the injected mass is small.
Amsden, A. (1997) KIVA 3-V: A Block Structured KIVA Program for Engines with Vertical or
Canted Valves, Los Alamos National Laboratory Report No. LA-13313-MS.
Amsden, A., O-Rourke, P.J. and Butler, T.D. (1989) KIVA-II: A Computer Program for Chemically
Reactive Flows with Sprays, Los Alamos National Laboratory, Report No. LA-11560-MS.
Convergent Science Inc (2014) CONVERGE 2.2.0 Theory Manual, Madison, WI.
Curran, H., Gaffuri, P., Pitz, W. and Westbrook, C. (1998) ‘A comprehensive modeling study of nheptane oxidation’, Combustion and Flame, Vol. 114, Nos. 1–2, pp.149–177.
Curran, H., Gaffuri, P., Pitz, W. and Westbrook, C. (2002) ‘A comprehensive modeling study of
iso-octane oxidation’, Combustion and Flame, Vol. 129, No. 3, pp.253–280.
Natural gas/diesel RCCI CFD simulations
Dahodwala, M., Joshni, S., Koehler, E., Franke, M. and Tomazic, D. (2015) Experimental and
Computational Analysis of Diesel-Natural Gas RCCI Combustion in Heavy-Duty Engines,
SAE Technical Paper 2015-01-0849.
Dec, J. (2009) ‘Advanced compression-ignition engines – understanding the in-cylinder processes’,
Proceedings of the Combustion Institute, Vol. 32, No. 2, pp.2727–2742.
Doosje, E., Willems, F. and Baert, R. (2014) Experimental Demonstration of RCCI in Heavy-Duty
Engines using Diesel and Natural Gas, SAE Technical Paper 2014-01-1318.
Engine Combustion Network (2012) Diesel Spray Combustion, Sandia National Laboratories
[online] (accessed 2012–2015).
Han, Z. and Reitz, R. (1995) ‘Turbulence modeling of internal combustion engines using RNG k-e
models’, Combustion Science and Technology, Vol. 106, Nos. 4–6, pp.267–295.
Hanson, R., Kokjohn, S., Splitter, D. and Reitz, R. (2010) An Experimental Investigation of Fuel
Reactivity Controlled PCCI Combustion in a Heavy-Duty Engine, SAE Technical Paper 201001-0864.
Healy, D., Kalitan, D., Aul, C., Petersen, E., Bourque, G. and Curran, H. (2010) ‘Oxidation of
C1-C5 alkane quinternary natural gas mixtures at high pressures’, Energy and Fuels, Vol. 24,
No. 3, pp.1521–1528.
Hockett, A. (2015) A Computational and Experimental Study on Combustion Processes in Natural
Gas/Diesel Dual Fuel Engines, PhD dissertation, Colorado State University, Fort Collins, CO.
Hockett, A., Hampson, G. and Marchese, A. (2016) ‘Development and validation of a reduced
chemical kinetic mechanism for computational fluid dynamics simulations of natural
gas/diesel dual-fuel engines’, Energy and Fuels, Vol. 30, No. 3, pp.2414–2427.
Kokjohn, S., Hanson, R., Splitter, D. and Reitz, R. (2009) Experiments and Modeling of Dual-Fuel
HCCI and PCCI Combustion Using In-Cylinder Fuel Blending, SAE Technical Paper 200901-2647.
Kokjohn, S., Reitz, R., Splitter, D. and Musculus, M. (2012) Investigation of Fuel Reactivity
Stratification for Controlling PCI Heat-Release Rates Using High-Speed Chemiluminescence
Imaging and Fuel Tracer Fluorescence, SAE Technical Paper 2012-01-0375.
Königsson, F. and Stalhammar, P. (2012) Controlling the Injector Tip Temperature in a Diesel
Dual Fuel Engine, SAE Technical Paper 2012-01-0826.
Leuthel, R., Pfitzner, M. and Frobenius, M. (2008) Numerical Study of Thermal-Fluid-Interaction
in a Diesel Fuel Injector, SAE Technical Paper 2008-01-2760.
Nieman, D., Dempsey, A. and Reitz, R. (2012) Heavy-Duty RCCI Operation Using Natural Gas
and Diesel, SAE Technical Paper 2012-01-0379.
Pei, Y., Mehl, M., Liu, W., Lu, T., Pitz, W. and Som, S. (2015) ‘A multicomponent blend as a
diesel fuel surrogate for compression ignition engine applications’, ASME Journal of
Engineering for Gas Turbines and Power, Vol. 137, No. 11, pp.11502–11511, GTP-15-1057.
Pomraning, E., Richards, K. and Senecal, P. (2014) Modeling Turbulent Combustion Using a RANS
Model, Detailed Chemistry, and Adaptive Mesh Refinement, SAE Technical Paper 2014-011116.
Post, S. and Abraham, J. (2002) ‘Modeling the outcome of drop-drop collisions in diesel sprays’,
Int. J. Multiphase Flow, Vol. 28, No. 6, pp.997–1019.
Puduppakkam, K., Liang, L., Naik, C., Meeks, E., Kokjohn, S. and Reitz, R. (2011) Use of Detailed
Kinetics and Advanced Chemistry-Solution Techniques in CFD to Investigate Dual-Fuel
Engine Concepts, SAE Technical Paper 2011-01-0895.
Schmidt, D. and Rutland, C. (2000) ‘A new droplet collision algorithm’, J. Comp. Phys., Vol. 164,
No. 1, pp.62–80.
Senecal, P., Pomraning, E., Richards, K. and Som, S. (2012) ‘Grid-convergent spray models for
internal combustion engine CFD simulations’, Proc. ASME Internal Combustion Engine
Division Fall Technical Conference, ICEF2012-92043.
A.G. Hockett et al.
Som, S. and Aggarwal, S. (2010) ‘Effect of primary breakup modeling on spray and combustion
characteristics of compression ignition engines’, Combustion and Flame, Vol. 157, No. 6,
Splitter, D., Hanson, R., Kokjohn, S. and Reitz, R. (2011) Reactivity Controlled Compression
Ignition (RCCI) Heavy-Duty Engine Operation at Mid- and High-Loads with Conventional
and Alternative Fuels, SAE Technical Paper 2011-01-0363.
US Energy Information Administration (2014) Annual Energy Outlook 2014 with Projections to
2040, DOE/EIA-0383(2014).
US Environmental Protection Agency (2011) EPA and NHTSA Adopt First-Ever Program to
Reduce Greenhouse Gas Emissions and Improve Fuel Efficiency of Medium- and Heavy-Duty
Vehicles, Regulatory Announcement, EPA-420-F-11-031.
US Environmental Protection Agency (2012) EPA and NHTSA Set Standards to Reduce
Greenhouse Gases and Improve Fuel Economy for Model Years 2017-2025 Cars and Light
Trucks, Regulatory Announcement, EPA-420-F-12-051.
Без категории
Размер файла
4 858 Кб
ijpt, 082915, 2017
Пожаловаться на содержимое документа