Modeling Spark Ignition and
Flame Propagation for More
Efficient Engine Designs
November 6, 2013
REACTION DESIGN
www.reactiondesign.com
+1 858-550-1920
Reaction Design 1
Abstract
Today’s engine manufacturers face many challenges in the design of low‐emissions, high‐performance
engines that can run on a wide variety of fuels. New design strategies for emissions control make fuel and
pollutant chemistry critical factors in determining overall engine performance. Combustion computational
fluid dynamics (CFD) simulation has the potential to help create better engines without spending millions of
dollars in physical prototypes or hundreds of hours building elaborate software models. However, to be a
trusted part of the engine design workflow and aid in the production of cleaner burning, higher‐
performance, and more‐efficient engines, these simulations must be able to account for real fuel effects
while employing an appropriate level of kinetics detail.
The Challenges of Spark Ignition Modeling
To efficiently design spark‐ignited engines with high fuel economy and low pollutant emissions, effective
combustion simulation is a must. Dynamic virtual prototyping allows engine designers to e
xploit predicted
dependencies between inputs and outcomes to control combustion. Effective simulations guide design
decisions that improve performance and emissions, while
reducing development costs and schedules
.
However, to ensure that engine simulations can accurately predict real‐life engine behavior requires both
advanced computational algorithms to describe the physics and thermodynamic mechanisms of
combustion, and also a detailed representation of the chemical makeup of fuels. Accurately predicting fuel
effects, ignition dependencies and emissions requires more detailed chemistry inputs than can be handled
by conventional CFD approaches. Many designers don’t believe they can incorporate these predictive
mechanisms into their simulations without significantly increasing compute expense and Time‐to‐Solution.
One of the foremost challenges of combustion engine modeling involves the accurate representation of
flame propagation within the combustion chamber. A spark of energy ignites the highly compressed fuel,
and combustion spreads across the chamber, led by an extremely thin flame front that compresses the
unburned gas as it propagates into the fuel‐air mixture (Figure 1). The pressurized fuel‐air mixture can also
auto‐ignite, either before or after the spark event that is meant to trigger combustion. This auto‐ignition
results in engine knocking, which, aside from being annoying to drivers, can lead to significant wear and
tear on engine parts. Auto‐ignition also reduces engine performance and can increase pollutant emissions.
The phenomenon known as “mega‐knock” or “super‐knock”, where auto‐ignition occurs prior to the spark
event, is particularly harmful to the engine.
Reaction Design 2
Figure 1: Representation of flame front and pollutants in the combustion chamber.
1
Accurately modeling auto‐ignition within the combustion chamber is very difficult. The pressure within the
combustion chamber and the speed of the flame front are just two of many properties that must be
represented accurately. The initially controlled spark starts out as an ignition kernel, which may be
modeled as a collection of “particles” carrying the spark energy. The kernel then quickly transitions into a
thin flame surface that moves through the turbulent flowfield, leaving burned gas in its wake.
Flame thickness is measured in the tens of microns, and the higher the pressure in the chamber, the
thinner the flame. As the flame encounters turbulence within the chamber, the flame front wrinkles and
becomes corrugated, which affects the flame propagation speed. Yet accurately modeling the location and
structure of the flame front as it expands is extremely important for predicting combustion heat release
rate, fuel‐burn efficiency, auto‐ignition and the resultant impacts on knocking and emissions.
Flame Propagation Modeling
Building simulations that accurately predict the dynamics and outcomes of flame propagation within a
spark‐ignited engine is clearly very challenging, and a number of modeling approaches have been
developed to simulate these behaviors.
Since the scale of the flame front thickness is significantly smaller than the computational mesh, even with
severe grid refinement, a method for tracking the flame‐front location is needed that is independent of the
mesh resolution. CFD simulations that rely solely on the mesh to resolve the flame location will yield results
that are highly mesh‐dependent or that require an inordinate number of cells to resolve the flame topology
sufficiently. Use of such severe mesh refinements to resolve flame location and topology requires very
small cell sizes to chart the infinitesimally thin and discontinuous flame front. Meshes with excessively large
numbers of tiny computational cells will get bogged down by the proportionally small time steps needed to
maintain simulation stability with the small cells. In practice then, there must be a compromise on mesh
resolution and accuracy, either around the flame or elsewhere, in order to keep compute times reasonable.
Reaction Design 3
This lack of reliability due to mesh‐resolution dependency is further compounded by the lack of detail in
the way the combustion kinetics is typically represented in the cells. Conventional CFD solutions are limited
in the amount of chemistry and kinetics detail they can incorporate within a simulation, particularly when
dealing with complex meshes and large numbers of cells. At the same time, mesh‐dependent methods can
only be successful for predicting flame location when they incorporate complete and accurate kinetics over
all ranges of temperature, fuel‐air equivalence ratio, pressure, and dilution for the cell‐by‐cell kinetics
calculations. However, this too will greatly increase compute time and solution complexity. The use of
highly reduced kinetics will deliver a more reasonable compute time, but at the price of accuracy.
Compromises on the kinetics to achieve practical turn‐around times directly undermine the notion of a
purely kinetics‐driven outcome.
Complete reliance on adaptive mesh refinements to drive resolution may also result in missing key events
that occur spontaneously outside of the adapted region, such as knocking related to auto‐ignition of the
end gas, or emissions production. A mesh that is too coarse in other regions will smear out the ability to
accurately predict auto‐ignition in those regions.
Simulating Flame Propagation Quickly and Accurately
The FORTÉ CFD solution from Reaction Design offers engine designers a state‐of‐the‐art approach to flame‐
front tracking, while including highly detailed reaction kinetics throughout the simulation. FORTÉ carefully
charts the growth of the ignition kernel as it expands from a tight ball of flame “particles” into a flame front
that moves across the combustion chamber, accounting for both chemistry and turbulence effects
explicitly. FORTÉ’s ability to model combustion with accurate chemical kinetics speeds simulations that lead
to better designs in less time.
The software is built to handle complex algorithms and fuel models without a
Time‐to‐Solution penalty, and without reliance on costly and excessive mesh refinement.
For spark‐ignited engines, the initial spark energy is released into a microscopic volume of gas. That volume
is initially much smaller than the mesh cell size needed to resolve the RANS equations, which describe
turbulent flow over time. The initial spark explosion must be modeled, since it would be intractable to
include the full plasma dynamics of the spark event in the system of equations solved on the computational
mesh. A best practice is to capture the initial spark dynamics using an analytical approach, such as the
Discrete Particle Ignition Kernel (DPIK) model. Spark ignition is very different from compression‐induced
auto‐ignition. A pure chemical‐kinetics approach cannot capture the dynamics of the initial spark‐kernel
shape. In FORTÉ, the DPIK model employs a Lagrangian approach to track the initial energy release
accurately in time and space, until the kernel particles begin to intersect with computational cell edges.
When the initial flame‐kernel particles start to disperse across cell boundaries, FORT
É
transitions to a
different computational method to track the location of the flame surface as it expands across the
chamber. Once the flame structure is big enough, a surface can be defined by connecting the points of the
particle locations and then the particles are no longer needed. A front‐tracking algorithm called the G‐
equation takes over (Figure 2). Rather than relying on the forced approach of severe mesh refinement,
FORTÉ uses the G‐equation to determine the flame location accurately without reliance on the mesh used
in the solution of the fluid dynamic equations.
Reaction Design 4
Figure 2: The G‐equation helps track the flame front without mesh refinement.
2
The G‐equation, mathematically known as the level‐set method, tracks the location of the flame front,
independent of mesh resolution, with a highly efficient numerical technique. Front‐tracking algorithms for
resolving sharp gradients in flow systems are well established in computational mathematics. For example,
techniques such as the level‐set method are essential for supersonic systems that resolve shock‐wave
locations.
As the flame expands, the flame front may ripple and buckle based on the stratification of the fuel/air
mixture and turbulence conditions in the cylinder. These complex flame motion phenomena can be
included when the flame front is tracked accurately. These phenomena would be neglected if only the
kinetics within each cell were considered in determining flame location.
FORT
É
accounts for local conditions at each flame‐front location by accessing pre‐established look‐up
tables that provide laminar flame‐speed values dynamically at each location of the flame front. Turbulent
flame speeds are derived from the fundamental laminar flame speed and from the local turbulence
parameters. In this way the flame propagation at each time step and at each point along the flame surface
is determined by the fundamental chemical kinetics relevant to those conditions as well as the turbulence
conditions.
This provides a fundamental link to fuel effects, as well as local impacts of dilution, pressure, temperature
and equivalence ratios. Laminar flame speed is measurable and detailed kinetics calculations will predict
values that agree with data under fundamental flame conditions. It accounts for the effects of molecular
transport as well as detailed reactions between the fuel and air.
Turbulent flame speed cannot be pre‐determined in the same way as laminar flame speed, because of the
dependence on local and dynamic turbulence parameters and the fact that turbulence scales can vary over
several orders of magnitude within the same simulation. However, there are fundamental correlations that
Reaction Design 5
allow on‐the‐fly determination of turbulent flame‐speed values using the kinetics‐derived laminar flame
speed and the local turbulent kinetic energy and turbulent length scale. Such correlations have been tested
and verified experimentally.
Using this computational approach to modeling flame‐front propagation, FORT
É
is able to solve the
chemistry in all the cells, even as the flame transits the combustion chamber, tracking whether each cell
contains the flame front at any given time, the unburned and burned‐gas composition in each cell, whether
the compressed gas will ignite (Figure 3), and the local production rates of NO
x
, CO and soot (Figure 1).
Instead of concentrating mesh around the flame front to try to resolve the flame itself, mesh resolution can
be applied in non‐flame‐containing cells to more accurately predict emissions, as well as pre‐ignition and
knocking phenomena.
Figure 3: FORTÉ tracks the progress of the flame front across the combustion chamber.
3
More Accurate Modeling for Better Engine Design
FORTÉ provides realistic 3‐D modeling of fuel effects in internal combustion engines. FORTÉ uses proven
mathematical techniques and algorithms coupled with direct use of detailed kinetics to simulate spark‐
ignited engine combustion and
predict the effects of operating conditions and fuel variations on engine
performance.
FORT
É
provides robust and fast simulations of spark‐ignited engines, built on well‐established
computational techniques rather than compute‐heavy approaches that rely on extreme mesh refinements.
DPIK and G‐equation flame‐front tracking techniques provide efficient and accurate resolutions of flame
propagation, which help ensure that engine designers have accurate simulations for predicting flame
location and quenching, combustion duration, auto‐ignition and knocking, and the formations of pollutants.
Rapid design of cleaner burning, higher‐performance, and more‐efficient engines is the result.
Reaction Design 6
References
1. C.V. Naik, K.V. Puduppakkam, L. Liang and E. Meeks, SAE Technical Paper Series, in preparation
2014.
2. L. Liang, R. D. Reitz, C. O. Iyer, and J. Yi, SAE Technical Paper Series SAE2007‐01‐0165.
3. Ibid