International journal of automotive technology, tập 12, số 1, 2011

KEY WORDS : Spark ignition SI engine, Quasi-dimensional combustion model, Variable intake system, Intake charge motion control, Calibration NOMENCLATURE A f : flame front area B : cylin

International Journal of Automotive Technology, Vol 12, No 1, pp 1−9 (2011)

DOI 10.1007/s12239−011−0001−4

Copyright © 2011 KSAE 1229−9138/2011/056−01

1

IMPROVING THE PREDICTIVENESS OF THE QUASI-D COMBUSTION MODEL FOR SPARK IGNITION ENGINES WITH FLEXIBLE INTAKE

SYSTEMS

T.-K LEE and Z S FILIPI*

Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI 48105, USA

(Received 24 March 2009; Revised 10 September 2009)

development A particularly attractive tradeoff between speed and fidelity is achieved with a co-simulation approach thatmarries a commercial gas dynamic code WAVETM with an in-house quasi-dimensional combustion model Gas dynamics arecritical for predicting the effect of wave action in intake and exhaust systems, while the quasi-D turbulent flame entrainmentmodel provides sensitivity to variations of composition and turbulence in the cylinder This paper proposes a calibrationprocedure for such a tool that maximizes its range of validity and therefore achieves a fully predictive combustion model forthe analysis of a high degree of freedom (HDOF) engines Inclusion of a charge motion control device in the intake runnerpresented a particular challenge, since anything altering the flow upstream of the intake valve remains “invisible” to the zero-

D turbulence model applied to the cylinder control volume The solution is based on the use of turbulence multiplier and

scheduling of its value Consequently, proposed calibration procedure considers two scalar variables (dissipation constant Cβ

and turbulence multiplier C M), and the refinements of flame front area maps to capture details of the spark-plug design, i.e.the actual distance between the spark and the surface of the cylinder head The procedure is demonstrated using an SI enginesystem with dual-independent cam phasing and charge motion control valves (CMCV) in the intake runner A limited number

of iterations led to convergence, thanks to a small number of adjustable constants After calibrating constants at the referenceoperating point, the predictions are validated for a range of engine speeds, loads and residual fractions

KEY WORDS : Spark ignition (SI) engine, Quasi-dimensional combustion model, Variable intake system, Intake charge

motion control, Calibration

NOMENCLATURE

A f : flame front area

B : cylinder bore diameter

Cβ : adjustable constant of the quasi-D combustion

model

C M : adjustable constant of the quasi-D combustion

model

K : mean flow kinetic energy

k : turbulent kinetic energy

P : production rate of turbulent kinetic energy

m b : mass of burned products

The gasoline spark ignition (SI) engine dominates the light

vehicle markets in US and many other regions due to its

inherent high power density, low cost, effective exhaustaftertreatment, and smooth operation The continuedsuccess hinges upon continuous improvements over time

As shown by Heywood (2009), the main performanceattributes have been improved at the rate of 2% per year Tocompete with turbocharged common rail direct injectiondiesel engines, gasoline engine developers have adopted aslew of new technologies, many of which pertain to theflexible devices for improving engine breathing Thisexpands the operating range and allows unprecedentedopportunities for optimizing the engine system, but alsoincreases the complexity as well On the business side,cost-reduction requires further shortening of thedevelopment cycle Achieving design objectives withinsevere cost constraints critically depends on effective use

of predictive simulation tools Simulations allow earlyexplorations, optimization of design, and full characteriza-tion of the engine system for controls work In order toaccomplish these goals, the simulation must be sensitive tothe variations of process variables resulting from the action

of devices under consideration This provides the impetusfor the work presented here

As an alternative to costly engine experiments, computer

*Corresponding author e-mail: filipi@umich.edu

2 T.-K LEE and Z S FILIPI

simulations have been widely used to predict engine

per-formance characteristics Simulations range in complexity

from highly detailed three dimensional computational fluid

dynamics (CFD) models (Choi et al., 2005; Haworth,

2005; Gosman, 1994) to simplified mean value engine

models (Cook and Powell, 1998; Hendricks and Sorenson,

1990) Every class of models has its place, and the

selection depends on the simulation goals Development of

engine control strategy requires a large number of

simulations that cover all possible engine operating

conditions Therefore, fast computations are highly desired,

but with sufficient accuracy of predictions Clearly, a need

for high computational speed eliminates CFD codes, while

simplifications in the mean value models limit their

predictiveness An optimum compromise can be found

with a one-dimensional (one-D) gas dynamic simulation

coupled to a thermodynamic cycle simulation, as long as

the latter includes models capable of capturing the physics

of all relevant phenomena

As an example, application of a flexible valve actuation

system will lead to variations of the flow parameters,

including turbulence, and the residual fraction in a very

wide range Hence, the use of a semi-empirical model such

as the Wiebe function (Wiebe, 1956; Katsumata, 2007) will

not suffice as the combustion model needs to be able to

predict the effect of turbulence and residual fraction on burn

rates A promising approach is the application of a two-zone

quasi-dimensional (quasi-D) combustion model It includes

the effect of turbulence on the rate of flame entrainment,

and the effect of laminar flame speed on the burn-up of

entrained mixture The concept was first pro-posed in the

previously published literature (Tabaczynski et al., 1977,

1980), but until recently the quasi-D simulations were

viewed as relatively computationally intensive tools

intended primarily for engine development work Advances

of the computer technology and refinements of the models

open the doors to a wider use of quasi-D tools for

simu-lation-based engine control development A much more

predictive tool can replace the mean-value model and allow

explorations in a much wider range of operating

condi-tions The objectives of our work are to maximize the

fide-lity of such tools through a co-simulation approach

marry-ing a commercial gas dynamic code and an in-house

com-bustion model, and to subsequently propose a systematic

methodology for calibrating model constants based on the

limited set of experimental data In doing so, we address a

particular challenge and capture the effect of a charge

motion control device mounted in the intake runner,

upstream of the cylinder The challenge stems from the fact

that a zero-dimensional turbulence model can normally be

applied only to the cylinder control volume, so anything

altering the flow upstream of the intake valve remains

“invisible” After characterizing the sensitivity of the flow

and combustion predictions to model constants, we

propose a solution based on the use of a turbulence

multiplier and scheduling of its value

The co-simulation approach combines strengths of thecommercial code Ricardo WAVETM in gas dynamics model-ing and strengths of the in-house quasi-dimensional SparkIgnition Simulation (SIS) in combustion modeling WAVETMhas been widely used for engine performance predictions (e.g

Kim et al., 2005) We use it to model gas dynamics in the

intake and exhaust systems from the air filter to the tailpipe.SIS is a research code written in FORTRAN language thathas been refined over time and used routinely at theUniversity of Michigan for a variety of simulation studies

(Filipi and Assanis, 2000; Wu et al., 2005, 2006) The

combustion sub-model in the code is based on the turbulent

flame entrainment model proposed by Tabaczynski et al.

(1977, 1980) and further refined by Poulos and Heywood(1983) Figure 1 illustrates our vision for a co-simulationapproach The experiments are required only in the develop-ment stage for the calibration of model constants and thevalidation of predictions However, it is important to notethat the real engine does not necessarily have to compriseall technologies under consideration Once the code hasbeen validated within a given range of operating variables,

it will be suitable for studies of many configurations ducing similar changes of in-cylinder conditions The development of the calibration methodology and thepredictiveness of the co-simulation tool are demonstratedusing an engine with a dual-independent variable valvetiming (di-VVT) system and charge motion control valves(CMCVs) The CMCV is an air flow restriction devicelocated upstream of the intake valves It generates turbu-lence in the flow entering the combustion chamber in order

pro-to produce faster burning rates – see Figure 2 The CMCV

is simple and inexpensive to use, but developing controlstrategies requires full characterization of its impact oncombustion Therefore, creating a fast and accurate “virtualengine” is critical for efficient engine control design.This paper is organized as follows First, the predictivephysics-based simulation is created based on the co-simu-lation approach Then, we propose a calibration procedure

to improve the prediction accuracy of the quasi-D lation The calibration procedure considers the dissipation

simu-constant Cβ, and the multiplier C in the turbulence model

Figure 1 Illustration of the procedure for building a fastand predictive simulation tool using a co-simulationapproach

IMPROVING THE PREDICTIVENESS OF THE QUASI-D COMBUSTION MODEL FOR SPARK IGNITION 3

In addition, it discusses the need for an in-depth look at the

early flame growth in the combustion chamber, and

intro-duces adjustments of the flame maps based on the actual

distance from the spark to the combustion chamber wall

Finally, the accuracy of predictions is demonstrated through

the comparison with experimental data obtained with the

engine equipped with the CMCV

2 ENGINE CONFIGURATION

The engine used as the platform for simulation

develop-ment, calibration and validation is a Chrysler

dual-overhead camshaft 2.4 liter inline four (I4) cylinder spark

ignition (SI) engine with the di-VVT device and the

CMCV Two intake valves and two exhaust valves are used

per cylinder and are actuated by the dual overhead

camshaft The CMCV is introduced upstream of the

combustion chamber in the intake runner to generate high

turbulence for fast combustion and reduced combustion

variability at low loads The relevant engine parameters are

summarized in Table 1

3 SIMULATION TOOL

The high-fidelity simulation consists of a one-D gas dynamicssimulation model, a quasi-D combustion model, and anintegration module To achieve the combustion predictive-ness over all possible engine operating conditions, thequasi-D combustion model is integrated into the one-D gasdynamics simulation Engine states related to gas exchangeprocess, such as mass flow rate, gas velocity, temperatureand composition through intake and exhaust valves, arepredicted by the one-D simulation Engine responsesrelated to the combustion process are predicted by thequasi-D combustion simulation

3.1 Integration of One-D and Quasi-D Models to Create aVersatile Engine System Simulation Tool

A top-level program written in the C++ language is duced to realize the co-simulation approach by integratingthe WAVETM–based gas dynamic model with the quasi-Dcombustion model The program was originally developed

intro-by Wu et al (2005) and refined for this study The

integ-ration procedure is illustrated in Figure 3 The integinteg-rationprogram calls the one-D simulation with an initial guess ofthe burning rate profile to calculate the gas flows throughthe intake and exhaust valves Next, the program transfersgas flow predictions to the in-cylinder quasi-D simulation,which calculates the burning rate profile, in-cylinder pre-ssure, engine output and emissions Then, the predictedburning rate profile is passed back to the one-D simulationfor the next iteration The convergence is established based

on the error tolerances for indicated mean effective ssure (IMEP), residual fraction, and volumetric efficiency.3.2 One-D Gas Dynamics Simulation Model

pre-The one-D gas dynamics model is created using the mercial software Ricardo WAVETM It includes all air flowpaths from the air box to the intake valve as well as fromthe exhaust valve to the tail pipe Figure 4 shows the gasdynamics simulation model of the entire engine system.First, the cylinder block is modeled Each cylinder has twointake and exhaust valves and ports Gas flow paths areconnected to the cylinder head via the intake and exhaustrunners Air flow coefficients through the valves are found

com-by using experimental data provided com-by Chrysler LLC, and

Figure 2 Schematic of the engine configuration with

dual-independent cam phasing (di-VVT) and the charge motion

Max intake valve lift 8.25 mm

Max exhaust valve lift 6.52 mm

Default intake valve timing

Closes/Opens/Centerline 51

o ABDC/1o BTDC/

115o ATDCDefault exhaust valve timing

Closes/Opens/Centerline 9

o ATDC/51o BBDC/111o BTDCDefault valve overlap 9o@0.5 mm lift

Intake cam-phasing range 15o Crank angle

Exhaust cam-phasing range 15o Crank angle

Figure 3 Integration of a one-D gas dynamics simulationand quasi-D combustion model

4 T.-K LEE and Z S FILIPI

these values are critical for correct estimation of air mass

flow rate into the cylinders Then, the piping and manifolds

are modeled by using duct and junction components Using

exact three-dimensional CAD data and two-dimensional

drawings provided by Chrysler LLC guarantees the

accuracy of gas exchange predictions

The throttle valve is emulated by an orifice The

maxi-mum orifice diameter at the wide open throttle (WOT) is

restricted to the maximum intake air path diameter at the

throttle body For part load conditions, an equivalent orifice

diameter is determined to achieve the air mass flow rate

corresponding to a given throttle position Finally, the

integ-ration program establishes an interface between the WAVETM

and the external combustion model at the valve seat

3.3 Quasi-D Spark-ignition Combustion Model

The quasi-D model is based on mass and energy

conserva-tion and phenomenological models for mean flow,

turbu-lence, combustion, and heat transfer in the cylinder

(Tabaczynski et al., 1977, 1980; Poulos and Heywood,

1983) The quasi-D model includes detailed physics and

has previously been validated for a range of applications

(Poulos et al., 1983; Filipi, 1994), hence it has the

cap-ability to extrapolate once calibrated at several important

engine operating conditions

Flame is assumed to propagate spherically from an

igni-tion point The main governing equaigni-tions are given here to

facilitate further discussions, while the details of the model

can be found in the already referenced papers

The rate of mass entrainment is

where m e is the mass entrained, t is time, ρu is density of

unburned charge, A f is the flame front area, u' is turbulent

intensity, and S L is laminar flame speed The flame area

term takes into account the effect of combustion chamber

geometry, and turbulence intensity captures the effect of

charge motion, while the laminar flame speed ensures

sensitivity to residual fraction and air-to-fuel ratio

The rate of burning is

where m b is the mass of burned products, λ is the Taylormicroscale, and τ=λ/S L Clearly, everything affecting thelaminar flame speed will have significant influence on therate of burn-up

The combustion model is complemented by a dimensional turbulence model, since turbulence intensityplays a major role in the prediction of the flame entrain-ment, and Taylor microscale is essential for determiningthe rate of burn-up in the reaction zone The model calcu-lates crank-angle resolved global turbulence throughout thewhole cycle based on the energy cascade concept shown inFigure 5 The equations for the zero-dimensional energycascade are as follows

where and are mass flow rates into and out of the

cylinder respectively v i is the gas flow velocity into thecylinder, and ε is the dissipation rate of turbulent kinetic

energy per unit mass by assuming turbulence is isotropic P

is the production rate of turbulent kinetic energy and it iscalculated assuming analogy to the turbulence production

over flat plates K is the mean kinetic energy and k is the

turbulent kinetic energy defined as:

- = 12

- m· i v i2− P − m·e

m

-dk dt

Figure 4 One-dimensional gas dynamics simulation model

built with the Ricardo WAVETM™ software

Figure 5 Turbulent energy cascade model for estimatingmean and turbulent flow parameters

IMPROVING THE PREDICTIVENESS OF THE QUASI-D COMBUSTION MODEL FOR SPARK IGNITION 5

the minimum vessel dimension:

where V is the instantaneous volume of the combustion

chamber, and B is the cylinder bore diameter Cβ is an

adjustable constant that tunes the production and

dissipa-tion rate of turbulent kinetic energy during the compression

and expansion processes After ignition, the conservation

of mass and angular momentum of individual eddies leads

to the following expressions,

where subscript “0” denotes conditions at the time of

ignition Multiplier C M is a tunable parameter useful for any

situation involving additional devices for generating

turbu-lence

3.4 Implementation of the CMCV in the High-Fidelity

Simulation

Implementing the CMCV into the high-fidelity simulation

requires the prediction of its impact on air mass flow rate

into the cylinder and turbulence intensity The air mass

flow rate can be easily predicted by adding an orifice at the

CMCV position in the one-D gas dynamics model in order

to emulate the pressure drop across the CMCV Since the

one-D code provides only the mean flow parameters and

the calculation of the energy cascade begins with the flow

velocity through the intake valve, there is no sensitivity of

the in-cylinder calculations to the turbulence-enhancing

devices mounted upstream Another way must be found to

simulate the effect of turbulence generation in the intake

runner A promising solution is to use the multiplier C M in

equation (11) of the quasi-D combustion model and tune it

until burn rates with the CMCV blocked match the

mea-sured burn rates Meanwhile, the overall behavior of the

in-cylinder turbulence model depends on the values chosen

for the dissipation constant Cβ

4 CALIBRATION PROCEDURE OF A QUASI-D

COMBUSTION MODEL

The ultimate goal of model calibration is to select the

smallest number of constants that will be evaluated over a

relevant range of operation This is achieved by

investigat-ing governinvestigat-ing equations of the quasi-D model Flame front

area maps in equation (1) have a very direct impact on

predictions of flame entrainment The dissipation constant

Cβ in equation (8) influences predictions of turbulence

intensity used in (1) throughout compression, while the

multiplier C M in equation (11) allows adjusting the

turbu-lence intensity level after ignition to simulate the impact of

the CMCV

The mass fraction burned profile is highly influenced by

the flame front area maps The maps need to be prepared in

advance using a dedicated code for calculating the action between the spherical front and the combustionchamber walls While this is a purely geometric calculationand there is no possibility for adjustments, one aspect of themap generation process deserves special attention Theproximity of combustion chamber walls to the spark, i.e.gap between electrodes, determines the flame kernelgrowth In many cases the predictions of the flamedevelopment stage are crucial for the overall accuracy ofthe calculated mass fraction burned, and yet this detail caneasily be overlooked Thus, we include assessments of theignition delay predictions based on the spark location intothe overall calibration procedure

inter-4.1 Overall Calibration ProcedureThe overall calibration procedure is illustrated in Figure 6.The flame front area maps are generated from 3-D CADdata of the combustion chamber geometry by consideringthe interaction between a spherical front growing outwardsfrom the spark and the combustion chamber walls Aseparate map is generated for every piston position Thefirst iteration is carried out using the best available infor-mation about the spark electrode length, but small adjust-ments are made in case there are obvious deficiencies inpredictions of the early part of the mass fraction burnedprofile Details of the flame frontal area calculation arepresented in the next sub-section

Next, the multiplier C M is calibrated to account for adevice such as the CMCV that manipulates the turbulentintensity upstream of the combustion chamber Then, the

parameter Cβ is calibrated to emulate the realistic energy

cascade process It is worth noting that adjustments of Cβallow capturing the global effect of 3-D flow patterns onturbulence in the context of the zero-dimensional model

6 T.-K LEE and Z S FILIPI

In general, a single value for C M and Cβ, respectively,

may not be sufficient to cover the whole operating range,

but a small set of values will still provide a robust

simu-lation tool If the engine includes a device that significantly

alters the flow in the intake system, as is the case with the

CMCV, the scheduling of the constant accounts for the

state of the device The iterations continue until the

satis-factory match between the predicted and experimental

mass fraction burned is achieved, and the overall procedure

is then repeated for several selected operating conditions

4.2 Flame Front Area Maps and Their Effect on

Combus-tion

The flame front area is critical for the accuracy of

combus-tion prediccombus-tions, such as the mass fraccombus-tion burned profiles

The mass fraction burned profile is a function of crank

angle, and has a typical S-shaped curve It consists of the

flame-development angle (∆θd) and the rapid-burning

angle (∆θb) The flame-development angle (the 0~10%

mass burned) is the crank angle interval between the spark

discharge and the time when a small but significant fraction

of the cylinder mass has burned The flame-development

stage is influenced by mixture composition and charge

motion in the vicinity of the spark plug Initially, the flame

develops freely around the point of ignition, as shown in

Figure 7(a) When the flame touches the surface of cylinder

head, the interaction between the flame front area and the

combustion chamber walls becomes a factor as well – see

Figure 7(b) Hence, the exact location of the spark can be

very influential for the growth of the flame kernel, e.g

longer electrodes will allow more space for the spherical

flame kernel and lead to a shorter ∆θd The rapid-burning

angle (the 10~90% burn duration) characterizes the main

stage of combustion During this stage, shown in Figures

7(c) and 7(d), the details of combustion chamber geometry,

including the shape of the piston top, become dominant.The complexity of combustion chamber geometry poses

a special challenge In our case, the combustion chamber is

a pent-roof shape and the piston top is raised up to maintaincompression ratio The 3-D CAD geometry is converted toadequate 3-D mesh data for calculating the flame front areamaps using a finite element pre-processing tool Re-mesh-ing procedure generates coarse mesh shown in Figure 8and enables fast calculations of geometric interactions The

Figure 7 Illustration of flame front area propagation with

respect to the crank angle

Figure 8 Pre-processed and simplified combustionchamber 3-D geometry using finite element pre-processortools

Figure 9 Comparison of flame front area maps: (a) with aninaccurate spark plug position; (b) with the accurate sparkplug position

IMPROVING THE PREDICTIVENESS OF THE QUASI-D COMBUSTION MODEL FOR SPARK IGNITION 7

accuracy is confirmed by verifying the clearance volume

and compression ratio

Next, we assess the sensitivity of the flame area

calcu-lations to the location of the spark Figures 9(a) and 9(b)

illustrate the flame front area development for two spark

locations Each plot contains a set of lines, each calculated

for a different piston position The first line from the

bottom corresponds to the piston located at the top dead

center (TDC), and the top line corresponds to 120 degCA

position The slope of the flame front area line at the very

beginning largely influences the flame-development angle

The peak and the slope observed for larger flame radius

influence the rapid burning stage

When the spark is located near the wall, only 1 mm from

the back surface of the head, the flame touches the wall

early and a significant portion of the front is cut out This

leads to a mild slope of the flame area line with respect to

flame radius, and a relatively flat appearance of the profiles

shown in Figure 9(a) When the distance between the spark

and the wall is increased to 5 mm, the flame area profiles

become much sharper thus leading to larger flame front

size for a given radius − see Figure 9(b) The flame front

area maps are obviously highly sensitive to the spark plug

position

Different flame front area maps are expected to produce

significant variations of mass fraction burned profiles.Figure 10 compares burning rate and mass fraction burnedprofiles predicted using flame area maps shown in Figures9(a) and 9(b) Indeed, burn rates predicted for case 1 (i.e.flame area maps shown in Figure 9(a)), are very differentfrom those obtained for case 2 (i.e flame area maps shown

in Figure 9(a)) Case 1 produces an asymmetric burn rateprofile with a retarded peak, as shown in Figure 10(a) Thisleads to a reduced slope of the mass fraction burned duringthe main stage of combustion – see Figure 10(b) In addi-tion, Case 1 demonstrates slower burning during the flamedevelopment stage Experiments confirm that Case 2 cap-tures the flame front evolution during the cycle much moreaccurately In summary, flame area calculations deservespecial attention, and in case there is any uncertainty aboutthe details of the geometric interaction close to the spark-plug electrodes, the experimentally measured burn ratescan indirectly verify the accuracy of flame area maps thatare subsequently being used as input the quasi-Dsimulation

4.3 Sensitivity to C M The parameter C M in equation (11) is introduced as a multi-plier for adjusting the turbulent intensity when additionaldevices are mounted upstream of the combustion chamber

to increase the turbulent intensity Figure 11 shows the

influence of the C M on the mass fraction burned profiles.Multiplier values larger than unity imply enhanced turbu-lent intensity due to a device such as the CMCV Thissignificantly increases the slope of the mass fraction burn-

ed curves In other words, combustion predictions are very

sensitive to the multiplier C M and its value will indicate thesuccess in enhancing turbulence with the CMCV

4.4 Sensitivity to Cβ

The dissipation constant Cβ in equation (9) influences thezero dimensional energy cascade by varying the rate ofmean kinetic energy dissipation and turbulence production

Larger Cβ implies faster conversion of the mean kineticenergy into turbulent kinetic energy The effect on turbu-

Figure 11 Influence of the C M on the mass fraction burnedprofiles

Figure 10 Influence of different flame front area maps: (a)

normalized burning rate profiles; (b) mass fraction burned

profiles

8 T.-K LEE and Z S FILIPI

lence during combustion is somewhat non-intuitive

Greater turbulence production leads to high values of u' at

the beginning of intake process, but this is accompanied by

relatively faster dissipation of mean kinetic energy

Conse-quently, the mean kinetic energy drops to lower levels by

the end of intake and beginning of compression, and higher

turbulence intensity values cannot be sustained Close to

the TDC, when it matters for combustion, the turbulence

intensity is lower for higher values of Cβ, and combustion

speed is reduced as well (see Figure 12) Calibrating Cβ

based on burn-rates enhances the versatility of the quasi-D

combustion model by indirect compensation of in-cylinder

flow patterns

5 CALIBRATION RESULTS

The proposed calibration procedure is validated for the

di-VVT engine with the CMCV using experimental results

obtained in the University of Michigan Automotive

Labo-ratory The proposed procedure completes calibration with

a small number of iterations due to only three calibration

parameters and the sequential approach

First, the flame front area maps are generated from the

3-D CA3-D geometry using the methodology introduced in the

section 4.2 Then, the adjustable constants C M and Cβ are

separately calibrated for the CMCV blocked and

unblock-ed cases When the CMCV is blockunblock-ed, C M value is swept

from a unit value to larger value to account for the

increas-ed turbulent intensity upstream of the combustion chamber

Then, the Cβ value is fine tuned in order to reproduce an

experimentally measured combustion profile When the

CMCV is unblocked, C M value is set to a unit value

because there is no increase in turbulence upstream of the

valve, and Cβ is adjusted in the range 1 to 2 until the

experimental combustion profile is reproduced

Validation of the predictions at the engine speed of 2000

rpm and the break mean effective pressure (BMEP) of 2

bar is shown in Figure 13 Mass fraction burned profiles

change significantly between the two CMCV positions, but

in both cases the agreement between predicted and

experi-mental curves is excellent Similar agreement is observedfor all low to medium engine speeds, for load ranging fromidle to WOT, and residual fractions ranging from 0% to34% Therefore, after calibrating the constants for theCMCV blocked and unblocked cases at a reference point,the quasi-D combustion model can be used over the entirerange of engine operating points relevant for fuel economystudies Prediction errors may be larger under some ex-treme conditions, but the combustion sensitivity related tomain control variables, such as throttle input, EGR, enginespeed, and variable valve actuation, is preserved in thewhole range Thus, the co-simulation approach coupled to

a systematic calibration procedure yields a truly predictivetool for HDOF engine optimization and control develop-ment

6 CONCLUSION

This work proposes a systematic calibration procedure forthe predictive SI engine simulation tool that maximizes itsrange of validity The simulation is based on the co-simu-lation approach marrying a commercial gas dynamic codeWAVETM and an in-house quasi-dimensional combustionmodel The latter is based on the turbulent flame entrain-ment concept and it is chosen because of its ability tocapture the effects of key process variables on combustion

In particular, the model is sensitive to the changes of bustion chamber shape, engine speed, manifold absolutepressure, air-to-fuel ratio, residual fraction, and turbulencelevel in the cylinder

com-A particular challenge arises with the introduction of acharge motion control device in the intake runner, upstream

of the cylinder The zero-dimensional turbulence modelfollows the energy cascade that starts with mean kineticenergy generation in the intake gas jet, and it is insensitive

to the phenomena occurring upstream of the valve In order

to mitigate this problem, a multiplier C M is introduced inthe equation that tracks the turbulence intensity evolution

Figure 13 Comparison of predictions and experimentalresults for the mass fraction burned at the CMCVunblocked and blocked cases; engine speed of 2000 rpmand BMEP of 2 bar

Figure 12 Influence of the Cβ on the mass fraction burned

profiles

IMPROVING THE PREDICTIVENESS OF THE QUASI-D COMBUSTION MODEL FOR SPARK IGNITION 9

during combustion The dissipation constant Cβ, which

affects the rate of mean kinetic energy dissipation and

turbulence production, is considered next

Sensitivity analysis emphasizes a need for in-depth look

at the flame area maps and their impact on early flame

growth Flame area maps depend on the combustion

chamber shape and the spark location A detail that plays a

big role is the distance of the spark from the cylinder head

surface dictated by the electrode length Greater distance

leads to delayed contact of a spherical flame front and the

wall, increased flame areas, and faster burn rates In case

there is any uncertainty, the experimentally measured burn

rates should be used to indirectly verify the accuracy of

flame area maps

In summary, calibration of only two constants pertaining

to the in-cylinder model and possible adjustments of the

flame area maps are sufficient to provide a predictive SI

engine simulation based on a gas dynamics model and a

quasi-D combustion model A sequence of steps begins with

the assessment of flame area maps before moving on the

adjustments of the turbulence multiplier and the dissipation

constant In case the engine is equipped with a device for

altering charge motion in the intake runner, calibration

needs to be repeated for different settings of the device

The procedure is demonstrated using an SI engine

system with dual-independent cam phasing and charge

motion control valves in the intake runner A limited

number of iterations led to convergence, thanks to a small

number of adjustable constants After calibrating constants

at the refer-ence operating point, the predictions were

validated for a range of engine speeds, loads and residual

fractions The results indicate that the co-simulation

approach combined with a systematic calibration procedure

yields a predictive and robust tool for HDOF engine

optimization and control development

Robert Prucka for providing the experimental results, Chrysler

LLC for financial support, and Roger Vick, Denise Kramer and

Greg Ohl for providing engine geometry data

REFERENCES

Choi, H., Lim, J., Min, K and Lee, D (2005) Simulation

of knock with different piston shapes in a heavy-duty

LPG Engine Int J Automotive Technology 6, 2, 133−

139

Cook, J A and Powell, B K (1988) Modeling of an

internal combustion engine for control analysis IEEE

Control Systems Magazine 8, 4, 20−26

Filipi, Z S and Assanis, D N (1991) Quasi-dimensional

computer simulation of the turbocharged spark-ignition

engine and its use for 2- and 4-valve engine matching

studies SAE Paper No 910075.

Filipi, Z S (1994) Investigation of variable valve area

strategies for a turbocharged SI-engine Proc IMechE

1994-6, 5th Int Conf Turbocharging and Turbochargers,

93–102

Filipi, Z S and Assanis, D N (2000) The effect of to-bore ratio on combustion, heat transfer and performance

stroke-of a homogeneous-charge spark-ignited engine stroke-of given

displacement Int J Engine Research 1, 2, 191−208.

Gosman, A D (1994) CFD modeling of flow and

com-bustion for IC engines Proc 2nd European Computational

Fluid Dynamics Conf 5, 1, 133−143

Haworth, D C (2005) A review of turbulent combustion

modeling for multidimensional in-cylinder CFD SAE

Paper No 2005-01-0993.

Heisler, H (1995) Advanced Engine Technology Society

of Automotive Engineers, Inc

Hendricks, E and Sorenson, S C (1990) Mean value

modeling of spark ignition engines SAE Paper No 900616 Heywood, J B (1998) Internal Combustion Engine Fund-

amentals McGrawHill

Heywood, J B (2009) Trends in performance characteristics

of modern automobile SI and diesel engines SAE Paper

Kim, C.-S and Oh, B (2005) Performance simulation of

an in-line 6 engine considering VVT strategy using wave

SAE Paper No 2005-01-0036

Poulos, S G and Heywood, J B (1983) The effect ofchamber geometry on spark-ignition engine combustion

SAE Paper No 830334.

Tabaczynski, R J., Ferguson, C R and Radhakrishnan, K.(1977) A turbulent entrainment model for spark-ignition

engine combustion SAE Paper No 770647.

Tabaczynski, R J., Trinker, F H and Shannon, B A (1980).Further refinement and validation of a turbulent flame

propagation model for spark-ignition engines Combustion

and Flame 39, 2, 111−121

Taylor, C F (1985) The Internal-combustion Engine in

Theory and Practice Volume I: Thermodynamics, Fluid Flow, Performance The M.I.T Press.

Wiebe, I I (1956) The combustion speed in internal bustion piston engines; Collected works of piston engineresearch, Laboratory of Engines, Academy of Sciences,USSR, Moscow 1956 (Translated to English by M Kiisa,KTH 1993)

com-Wu, B., Prucka, R G., Filipi, Z S., Kramer, D M and Ohl,

G L (2005) Cam-phasing optimization using artificialneural networks as surrogate models – maximizing torque

output SAE Trans J Engines, SAE Paper No

2005-01-3757

Wu, B., Prucka, R G., Filipi, Z S., Kramer, D M and Ohl,

G L (2006) Cam-phasing optimization using artificialneural networks as surrogate models – Fuel consumption

and NOx emissions SAE Trans J Engines, SAE Paper

No 2006-01-1512.

International Journal of Automotive Technology, Vol 12, No 1, pp 11−20 (2011)

DOI 10.1007/s12239−011−0002−3

Copyright © 2011 KSAE 1229−9138/2011/056−02

11

DEVELOPMENT OF AN IDLE SPEED ENGINE MODEL USING IN-CYLINDER PRESSURE DATA AND AN IDLE SPEED CONTROLLER FOR A SMALL CAPACITY PORT FUEL INJECTED SI ENGINE

P V MANIVANNAN*, M SINGAPERUMAL and A RAMESH

Department of Mechanical Engineering, Indian Institute of Technology, Chennai 600036, India

(Received 26 August 2009; Revised 21 June 2010)

ABSTRACT−An idle speed engine model has been proposed and applied for the development of an idle speed controller for

a 125 cc two wheeler spark ignition engine The procedure uses the measured Indicated Mean Effective Pressure (IMEP) atdifferent speeds at a constant fuel rate and throttle position obtained by varying the spark timing At idling conditions, IMEPcorresponds to the friction mean effective pressure A retardation test was conducted to determine the moment of inertia of theengine Using these data, a model for simulating the idle speed fluctuations, when there are unknown torque disturbances, wasdeveloped This model was successfully applied to the development of a closed loop idle speed controller based on sparktiming The controller was then implemented on a dSPACE Micro Autobox on the actual engine The Proportional DerivativeIntegral (PID) controller parameters obtained from the model were found to match fairly well with the experimental values,indicating the usefulness of the developed idle speed model Finally, the optimized idle speed control algorithm was embedded

in and successfully demonstrated with an in-house built, low cost engine management system (EMS) specifically designed fortwo-wheeler applications

KEY WORDS : In-cylinder pressure data, Model for engine control, Idle speed control, PID control, Engine management

system

1 INTRODUCTION

In large Asian metropolitan cities where two-wheelers are

widely used, most of the city drive cycle will involve the

idle mode of operation Hence, it is necessary to develop an

optimal idle speed controller that can effectively reduce

fuel consumption and in turn considerably reduce

emissions of green house gases like CO2 and other toxic

gasses like CO and HC In a spark ignition engine, the idle

speed changes as the engine warms up due to heating of the

lubricating oil and consequent variations in friction In

addition, because of random disturbances in the air fuel

ratio, spark system, and other parameters, the idle speed

always fluctuates, leading to vibrations and rough running

of the engine Even switching on auxiliary equipment, like

the headlamp and horn, can load a small engine and change

the idle speed Thus, it is necessary to continuously control

the idle speed Such control can enable the setting of lower

idle speeds with consequent savings in fuel consumption

and reduction in emissions

Several methods, such as varying the amount of mixture

admitted using an idle air control valve (Lee, 2001;

Thornhill and Thompson, 1999), changing the spark timing

(Srail et al., 2002), and a combination of air quantity and

spark time (Manzie and Watson, 2003; Osburn andFranchek, 2006), have been suggested to control idle speed.The spark control method can be easily implemented on asmall engine without additional cost Here the spark timing

is set at a level that is slightly less advanced than theoptimum Thus, advancing the spark timing will lead to anincrease in the indicated torque and idle speed Retardingthe spark timing will lower the idle speed

Idle speed control is a regulatory problem in which theprimary role of the controller is to maintain a constant idlespeed in the presence of external torque disturbances (i.e.,from auxiliary loads) Further, the controller must be able

to reject internal torque disturbances generated by theengine due to unstable combustion at low throttle openings

A wide range of linear control techniques, from the proven

Proportional Integral (PI) controller (Thornhill et al., 2000;

Hrovat and Sun, 1997) to complex and computationallydemanding techniques such as optimal control theory basedcontrollers (Joo and Chun, 1997), the H loop shaping method(Ford and Glover, 2001), model predictive control (Manzie

and Watson, 2003), the Pole placement technique (Hsieh et

al., 2007), the linear loop shaping control technique (Osburn

and Franchek, 2006), and the linear quadratic regulatormethod (Nagashima and Levine, 2006), have been applied

to reduce idle speed fluctuations Idle speed control becomes

a non-linear control problem when control of the air fuel

*Corresponding author e-mail: pvm@iitm.ac.in

12 P V MANIVANNAN, M SINGAPERUMAL and A RAMESH

ratio is included along with spark time control In such

cases, non-linear control techniques like fuzzy logic

(Thornhill et al., 2000), adaptive fuzzy logic (Thornhill and

Thompson, 1999), the sliding mode method (Srail et al.,

2002), genetic algorithms (Kim and Park, 2007), the

non-linear autoregressive exogenous (NARX) model (De Nicolao

et al., 1999), and on-line adaptive Proportional Integral

Derivative (PID) tuning and the Continuous Action

Rein-forcement Learning Automata (CARLA) algorithm (Howell

and Best, 2000), have been used to achieve superior

per-formance in terms of improved controller robustness and

reduction in speed fluctuations, fuel consumption, and noise

vibration and harshness

The development of engine controllers requires numerous

experiments, which involves considerable time and cost

Initial development using simulators, which use fast and

robust engine models, can significantly reduce the number

of needed experiments and trials Engine models normally

use thermodynamic and engine dynamics sub-models to

describe engine processes As the combustion models are

very complex and require high computational power for

execution, they are seldom used in real-time control

ap-plications (i.e., embedding them in Engine Management

Systems, or EMS) Moskwa and Hedrick (1987) developed

a mean value model for control applications Other widely

used models are those by Cook and Powell (1988),

Andersson et al (1999), Chaing et al (2007) and Wu et al.

(2007) These models consist of sub-modules that describe

throttle dynamics, manifold filling dynamics, fuel injector

dynamics, engine (or) combustion models, and crankshaft

dynamics models, as represented graphically in Figure 1

In the above model, the mass flow rate of air through the

throttle opening is given by:

In the above equation, the function C d(αth), which

com-putes the discharge rate, is a complex function and is

generally determined through experimental data The term

A(αth) represents the effective throttle area, which is

pro-portional to the throttle angle (αth) Similarly, the function

ϕ(p r), which limits the flow at low intake pressures, is

dependent on the pressure ratio (p r) and the ratio of specific

heats (γ)

(2)Equation (2) represents the mass flow rate of the airentering the cylinder, while the volumetric efficiency of theengine ηvol can be determined with empirical data Theother parameters that influence the mass of air entering thecylinder are the engine speed (N), the manifold pressure(Pman), the engine displacement volume (Vd), the manifoldtemperature (Tman), and the universal gas constant (R)

In the port fuel injection system, a portion of the injectedfuel (χfp) is deposited on the manifold wall as a fuel puddle,which leads to wall wetting phenomena The remainingfuel that enters the cylinder ( ) as a fuel-air mixture,along with the fuel evaporated from the fuel puddle withtime constant (τfp), is given as:

In an internal combustion engine, the torque generated bycombustion of the fuel air mixture is mainly dependent onthe air fuel ratio and the spark timing The air fuel ratio isgenerally represented as a normalized air fuel ratio (λ) thatcan be computed using the following equation:

DEVELOPMENT OF AN IDLE SPEED ENGINE MODEL USING IN-CYLINDER PRESSURE DATA 13

spark occurrence (θ) before top dead center (° bTDC)

Finally, the net torque available from the engine that will

be used to drive the engine crankshaft is given by:

The angular motion dynamics are represented by:

In the angular motion equation (7), the terms ‘I’ and dω/dt

represent the engine’s moment of inertia and the angular

acceleration, respectively The speed dependent frictional

torque TF is a complex function to model; hence, it is

modeled with experimental data TL represents the load

torque

Experimental data are often collected from several

engines and used to make the model robust and fast by

avoiding complex, time-consuming algorithms Hence, a

simpler robust model will be beneficial, particularly for the

development of cost-effective controllers for two-wheeler

applications

2 PRESENT WORK

The present work is an attempt toward the development of

a simple method that can be used to develop idle speed

controllers with limited experimental data Although the

developed model is engine specific, the procedure can be

extended to other small-capacity engines Experimental

data obtained from a single cylinder scooter engine, whose

details are given in Table 1, have been used to formulate

the model Cylinder pressure data were used to obtain the

indicated mean effective pressure (IMEP) at different spark

timings under idle conditions at a fixed throttle position

and fuel injection pulse width (constant overall equivalence

ratio) These data were also used to obtain a model for

friction, as described later A retardation test was

performed to determine the moment of inertia of the

engine From the observed speed fluctuations at idle

conditions, the disturbance to the engine from an unknown

external torque variation was computed A model was

formulated by integrating these and was then used to

determine the PID constants for speed control based on

spark timing The previously computed disturbance torque

was used in the model to create a disturbance The most

suitable set of PID constants obtained by running the model

were used on the actual engine's idle speed controller,

which was based on dSPACE The system was able to

decrease idle speed fluctuations considerably In addition,

any desired idle speed could be set from the dSPACE

controller-based electronic control unit (ECU) developed

in this work, which was developed in Matlab/Simulink

The obtained set of PID constants was then used and tested

on an in-house built, low cost engine management system

(EMS) specifically designed for two-wheeler applications

The details of the experiments and model and results are

discussed in subsequent sections

3 EXPERIMENTAL SETUP

In this work, a commercially available four stroke, 125 cc,single-cylinder scooter engine was used for all of theexperi-ments The details of the engine are given in Table

1 This engine was modified for the Port Fuel Injection(PFI) mode of operation by replacing the carburetor with aspecially made throttle body assembly The intakemanifold was equipped with a Throttle Position Sensor(TPS) and Manifold Absolute Pressure (MAP) sensor Thethrottle body and fuel injector adapter assembly werefabricated in-house

The engine was coupled to an eddy current meter The experimental test rig was extensivelyinstrumented with various sensors to acquire engineparameters such as load, torque, speed, air flow, and inletand exhaust temperature The mass flow rate of the fuelwas obtained using suitable instrumentation HC and COexhaust emissions were measured using a NDIR type gasanalyzer (manufactured by HORIBA) The measurementswere done on dry exhaust gases A dSPACE MicroAutoBox interfaced with an IBM PC compatibleworkstation was used for the data acquisition and real-time

Figure 2 Line sketch of the experimental setup

14 P V MANIVANNAN, M SINGAPERUMAL and A RAMESH

engine control The signals from the crank angle and cam

position reference sensors were conditioned using specially

developed circuitry to generate square pulses that were fed

into the dSPACE system The signal from the cam position

sensor was used to locate the crank position with respect to

the cycle Additional power amplifier circuits were

developed for driving the fuel injector and the electronic

ignition unit A line sketch diagram of the complete

experimental setup is shown in Figure 2

As part of this work, a complete Engine Management

System (EMS), based on a Philips P89C51RD2 8-bit

micro-controller, was designed and developed Figure 3 shows the

functional block diagram of the EMS As shown in Figure

2, the test engine can be controlled either by dSPACE or by

the prototype EMS system A real-time operating system

(RTOS) and other control software modules (PID algorithm)

were developed using assembly & C programming

langu-ages The final compiled code was embedded into the memory

of the micro controller

4 DEVELOPMENT OF THE IDLE SPEED

MODEL

Experiments were initially conducted under idle

conditions with the air fuel ratio set at a slightly rich

value (≈14.3) to avoid misfiring and stalling of the

engine The spark timing was then set at different

values, which enabled the engine to run at different

speeds with no external load The pressure crank angle

variations at different spark advance angles (whichresulted in different idle speeds) were recorded using ahigh-speed data acquisition system The pressure crankangle data obtained at different idling speeds isgraphically shown in Figure 4

First, the IMEP was computed from the average cylinderpressure data based on 100 cycles The variation of averageindicated torque (TI) at every spark advance was thenobtained from the IMEP data Figure 5 shows the variation

of indicated torque (TI) with spark advance

The curve fit to this data is the indicated torque modelequation, which is given below

T I = (0.00005*SA2)+(0.0011*SA)+1.1265 (8)Here, TI is the indicated torque in N.m., and SA refers tospark advance in degrees before TDC (° bTDC) Thisequation was used in the model to obtain the indicatedtorque Thus, combustion was not explicitly modeled inthis method The indicated torque increases with speedbecause it has to balance the frictional torque at idle condi-tions In the range tested, the indicated torque increasedwith an increase in spark advance because the sparktimings are always more retarded than the best condition.Only under these circumstances can the spark timing beused to control idle speed variations

The next step was to determine engine friction at variousaverage speeds at idling conditions At idling conditions,

no useful work is done, and hence, the measured IMEP isthe frictional mean effective pressure (FMEP) Figure 6indicates the frictional torque at different idle speedsobtained by varying the spark timing and the final frictionalmodel equation derived from the curve fit, as given below

(9)The moment of inertia of the engine was obtained by aretardation test The engine was run at a speed of 4500 rpm,

T F= 5 10( × 7×N2)− 0.0014 N( × )+2.085

Figure 3 Block diagram of the in-house built Engine

Management System (EMS)

Figure 4 P-Theta diagram at different idling speeds

Figure 5 Variation of indicated torque variation withrespect to spark angle at idling

DEVELOPMENT OF AN IDLE SPEED ENGINE MODEL USING IN-CYLINDER PRESSURE DATA 15

and then the ignition was cut off at no load The decrease in

speed was recorded as a function of time

Finally, the following equation was used to determine

the moment of inertia of the engine

(10)

In this equation, TI is the indicated torque, and TF refers

to friction torque, which is obtained based on equation (9)

The disturbance torque (TD) is taken to be zero during the

retardation test, i.e., when combustion does not take place

The term ‘I’ refers to the moment of inertia of the rotating

parts of the engine, while ω is the angular velocity When

the spark ignition is cut, TI= 0 From these results, the

moment of inertia of the engine was computed to be

0.00145 kg·m2

Subsequently, the idle speed was recorded at the same

throttle position and fuel injection pulse width for a given

time From these data, the disturbance torque TD was

obtained as a function of time using equations (8), (9), and(10) Figure 7 indicates the raw idle speed data recordedalong with the disturbance torque computed from this data.This disturbance torque was then used to evaluate thedeveloped PID controller

A program was written in Matlab to determine the actualidle speed fluctuations at any spark ignition timing, based

on the above-mentioned equations for various quantities,which were in turn based on equation (10) The disturbancetorque calculated earlier was given as the disturbing input

as a function of time This program was used as a Simulinkblock Closed loop control of the spark timing was imple-mented using a PID control block in the Matlab program.The PID controller can be mathematically described as:

where u(t) is the input signal to the plante(t) is the error signal, defined as e(t) = r(t)− y(t)r(t) is the reference signal

y(t) is the plant output signal

In equation (11), the terms Kp, Ki, and Kd represent the PIDcontroller’s parameters of proportional, integral, and deri-vative gain The proportional gain (Kp) acts on the instant-aneous error value e(t), and increasing this value will re-duce the settling time, i.e., the system reaches the set pointquickly The integral gain (Ki) acts on the accumulatederror and reduces the steady error, i.e., the differencebetween the set point and the actual value The derivativegain factor (Kd) is effective only when there is a rate ofchange of error and helps in damping the systemoscillations due to disturbances

In this work, the initial PID parameters (Kp, Ki, and Kd)were computed using the Process Reaction Curve (PRC)method After warming up the engine, a step input in terms

of spark time was applied to the engine in the open loop

Figure 7 Raw idle speed fluctuation and disturbance

torque variation Figure 8 Engine open loop response for a step change ofspark timing

16 P V MANIVANNAN, M SINGAPERUMAL and A RAMESH

mode and the PRC; the engine speed was recorded for a

sufficiently long period for the engine speed to settle at a

new value The PRC of the engine speed change to a step

change in the spark angle is shown in Figure 8

From the PRC data, the process parameters of time delay

(L=0.06) and time constant (τ= 1.31) were extracted by

drawing a tangent at the inflection point of the PRC, as

shown in Figure 9 The process gain (P=25.84) is also

marked in the same figure

With the process parameters P, L, and τ, the constant K

was computed as shown below:

Finally, using the Ziegler-Nichols open-loop tuning

method, the approximate PID controller gains (Kp, Ki, and

Kd) were computed with following equations:

Proportional gain (Kp) = 1.2 × K (13)

Integral gain (Ki ) = 2 × L (14)

The calculated PID gain values are Kp= 1.01, Ki= 0.12,

and Kd= 0.03 These PID gain values were used as initial

values for all of the simulation and experimental

investi-gations of idle speed control The spark timing was

com-puted with the PID controller and then applied on the

sub-sequent cycle During the simulation, the PID parameters

were varied to obtain the best set and were fine tuned later

through engine experiments using actual controllers

5 RESULTS AND DISCUSSION

5.1 Simulation Results

As mentioned earlier, the disturbance torque indicated in

Figure 7 was fed into the model to predict the speed

varia-tions as a function of time This model was linked to the

PID control module Based on the PID values set in the

controller, the spark timing needed to control the speed

fluctuations was calculated for the next cycle This

com-puted spark timing was then applied on the subsequentcycle The best set of PID parameters was evaluated fromthe coefficient of variation in speed and the ISE (integral-squared-error) produced by each PID set The abovecontrollers were tuned for the minimal ISE The ISE ismathematically defined as:

where y(t) and ysp are the output and the desirable output ofthe process model, respectively The ISE performance cri-terion is widely used to tune controllers because its minimi-zation is related to the minimization of the error magnitude,i.e., the peak value (in this case, the idle speed fluctuationover the set point)

The influence of proportional (Kp), integral (Ki), andderivative (Kd) gains was evaluated We find that the fluctu-ations in the engine speed can be controlled by the properselection of the three gain constants With the right combi-nation, the fluctuations can be as low as 3 to 5 rpm Thesystem was stable even when there was a change in ex-ternal loads, such as switching on a headlamp This methodcan be adapted to any engine and used for tuning the P, I,and D constants easily

During the simulations with the developed model, theproportional gain was varied from an initial value of 1 to amaximum value of 6 As expected, the idle speed fluctua-tions decreased for increased Kp values, and the systembecame unstable when the Kp gain is was set to 5 For thiscase, we find from Figure 10 that a KP value of about 4yields the best stability under idle conditions

In the classical closed loop PID controller, it is typicallymandatory to include the Derivative control (D-control) tosuppress unwanted system oscillations that arise due toexternal disturbances Another advantage of adding the D-control is the ability to increase the proportional gain (Kp),

Figure 9 Approximated Process Reaction Curve (PRC)

with process parameters

Figure 10 Idle speed model response for the variations inproportional gain (Kp)

DEVELOPMENT OF AN IDLE SPEED ENGINE MODEL USING IN-CYLINDER PRESSURE DATA 17

which helps in reducing the system settling time From

Figure 11, it can be noted that the Proportional Derivative

controller's (PD controller’s) performance is optimized

when the proportional gain Kp and the derivative gain Kd

are set to values of 4.5 and 0.15, respectively, for this idle

speed control problem

The Integral control (I-control) is not effective when the

idle speed is controlled only with spark timing because the

control signal (spark time) is determined based on the

instantaneous speed variation (cyclic) of the engine, which

is random in nature Hence, the final idle speed controller

structure can be a Proportional plus Derivative (PD)

cont-roller

Table 2 shows that the ISE value is minimal for a PDcontroller with gain values of Kp=4.5 and Kd=0.15.5.2 Experimental Results with the dSPACE ControllerThe validation of the idle speed controller was done using adSPACE Micro Autobox hardware system The controlalgorithm implemented with Matlab/Simulink software isshown in Figure 12 The Graphical User Interface (GUI)and data acquisition part of the idle control system wasimplemented with dSPACE ControlDesk software.During the open loop experimental investigation, theengine was operated at a fixed air fuel ratio of ≈14:35 and

Figure 11 Idle speed model response for the variations in

derivative gain (Kd)

Figure 12 Idle speed controller with Matlab/Simulink

Table 2 Integral-squared-error (ISE) performance index criterion

18 P V MANIVANNAN, M SINGAPERUMAL and A RAMESH

with constant spark ignition timing (10° BTDC) By

adjust-ing the throttle openadjust-ing, the engine’s idle speed was set at

the recommended value (1700 rpm) Figure 13 shows large

speed fluctuations around the set idle speed value of 1700

rpm in the open loop condition (controller is turned OFF),

mainly due to unstable combustion and external

disturbances

The engine idle speed response obtained with the best

proportional controller gain value (Kp=5) is also shown in

Figure 13 In the experiments, it was found that a

propor-tional controller gain (Kp) value of 5 gives the best results,

whereas a Kp value of 4 resulted in the best performance

during the simulation This minor variation of Kp can be

attributed to small air fuel ratio variations induced due to

idle speed fluctuations (even though in idling mode, the

throttle position and fuel injection pulse width were kept at

a constant value) These air fuel ratio fluctuations were not

taken into account in the model

Another set of experiments was conducted to study the

performance of the Proportional Derivative (PD)

controller In these experiments, the proportional gain (Kp)

value was set at 5, and the derivative controller gain (Kd)

was set to 0.15, resulting in reduced idle speed fluctuations,

as shown in Figure 14

5.3 Experimental Results with the In-house Built Engine

Management System (EMS)

The optimal controller parameters obtained with the dSPACE

controller were embedded into the target idle speed troller (the in-house built Engine Management System).Table 3 summarizes the effectiveness of the controller inthe modified Port Fuel Injected (PFI) engine with respect tothe original engine, which used the carburetor as the fuelmetering device

con-With the controller, the spark advance time is controlled

on a cycle-by-cycle basis, leading to better combustionand, hence, reductions in idle speed fluctuations, fuel con-sumption, and CO emissions We were also able to operatethe engine with a leaner mixture (AFR 14.3) and withlower cyclic speed fluctuations compared to the carburetor(AFR=13.9) The HC emissions with the injection systemusing the developed idle speed controller are, however,slightly higher than with the carburetor At idling condi-tions, the amount of exhaust gas trapped in the combustionchamber will be significant, which is one of the reasons forthe richer stoichiometric mixtures that are used Becausethe mixture is leaner with the injection system, the HClevels could be higher In addition, the spark timing withthe present idle speed controller is not minimal advance forbest torque MBT time, as we need a torque margin forcontrol of idle speed When the spark time is set at otherthan the MBT timing, higher levels of HC are generated inthe combustion chamber Variations in spark timing thatresult during idle speed control will also change the ex-haust temperature and influence the post oxidation of HC

in the tail pipe These factors could contribute to the smallincrease in the HC emissions that was observed

Figure 14 Idle speed response of the engine with

Propor-tional Derivative (PD) controller

Table 3 Idle speed performance under different control schemes

Fuel

metering Type of idle speedcontrol Avg speed(rpm) Std Div ofspeed Fuel time(secs) (%)CO (ppm)HC ratioA/F

Fuel injection Closed loop (Kp=5, Kd=0.15) 1702 4.73 155.6 1.48 491 14.29

Figure 15 Engine response to load disturbance at an idlecondition

DEVELOPMENT OF AN IDLE SPEED ENGINE MODEL USING IN-CYLINDER PRESSURE DATA 19

The primary role of the closed-loop idle speed controller

is to reject disturbances and reduce the idle speed

fluctua-tions Another important benefit is that the controller

allows the engine to operate at lower idle speeds and helps

in leaving and entering the idle to cruise mode smoothly

Hence, it is necessary to evaluate the developed

controller’s disturbance rejection capability by applying

sudden load changes, i.e., by suddenly demanding a higher

torque Figure 15 shows the engine’s response to a step

load (electrical load of a 40 watt headlamp) disturbance at

idle conditions The engine speed dips to a lower value of

1575 rpm from the recommended 1700 rpm when the

torque developed by the engine is not sufficient, which can

result in engine stalling

Figure 16 shows that the speed fluctuations are

minimi-zed and that the speed is maintained around the set value of

1700 rpm, when the closed loop idle speed controller (PD

controller) is in operation When the electrical load

(head-lamp) is switched on, the engine speed drops only by 50

rpm to 1650 rpm However, the controller brings the engine

idle speed back to 1700 rpm within 40 cycles (1.5 secs).Because the closed loop controller was implemented indSPACE, the model for idle speed control can be used totune the PID parameters of the ECU

The closed loop idle speed controller reduces the NoiseVibration Harshness (NVH) of the vehicle by suppressingengine speed fluctuations Apart from this, with theclosedloop control it is possible to reduce the idle speed setpoint without stalling the engine Lowering the idle speedhelps in reducing fuel consumption and CO2 emissions,especially in the city drive cycle Figure 17 shows theengine responses for the originally recommended idlespeed of 1700 rpm and a lower idle speed condition of

1600 rpm

6 CONCLUSION

Based on this work, the following conclusions are drawn:The simple idle speed model developed in this work wasfound to be effective in determining a set of PID controlparameters that are similar to the best values obtained usingexperiments

Both the model and experimental results showed that a

PD controller is effective in controlling idle speed on acycle-by-cycle basis The developed model, along with a

PD controller having gain values of Kp=4.5 and Kd=0.15,shows optimal performance in simulations, whereas thebest experimentally obtained constants are similar at Kp=5and Kd=0.15 These results indicate the effectiveness of thepresent model for use in controller development

When the idle speed control system was implemented on

a modified Port Fuel Injected (PFI) engine, the idle speedfluctuations (Std deviation) decreased from 35.1 to 4.7,and CO emissions decreased by 60% There was areduction in fuel consumption of 11.2%

Though the developed model is engine-specific, the cedure can be adopted to simulate idle speed fluctuationseasily for use in the development of controllers of anyengine

pro-REFERENCES

Andersson, P., Eriksson, L and Nielsen, L (1999) ing and architecture examples of model based engine

Model-control 2nd Conf Computer Science and Systems

Engi-neering in Linköping Sweden.

Chaing, W.-P., Zhu L and Patankar, R (2007) Meanvalue engine modeling and validation for a 4-stroke

single cylinder gasoline engine Trends in Applied

Sciences Research 2, 2, 124−131

Cook, J A and Powell, B K (1988) Modeling of an

inter-nal combustion engine for control ainter-nalysis IEEE Control

Systems Magazine, 20−26

De Nicolao, G., Rossi, C., Scattolini, R and Suffritti, M.(1999) Identification and idle speed control of internal

combustion engines Control Engineering Practice, 7,

Figure 16 Engine response to a step load with idle speed

controller

Figure 17 Engine response for different idle set points

(with and without controller)

20 P V MANIVANNAN, M SINGAPERUMAL and A RAMESH

1061−1069

Ford, R and Glover, K (2001) Spark ignition engine idle

speed control using a novel framework and enabling

control of the tradeoff between fuel consumption and

load rejection capability Vehicle System Dynamics 36,

Howell, M N and Best, M C (2000) On-line PID tuning

for engine idle-speed control using continuous action

reinforcement learning automata Control Engineering

Practice, 8, 147−154

Hrovat, D and Sun, J (1997) Models and control

metho-dologies for IC engine idle speed control design Control

Eng Practice 5, 8, 1093−1100

Hsieh, F.-C., Chen, B C and Wu, Y Y (2007) Adaptive

idle speed control for spark-ignition engines SAE Paper

No 2007-01-1197.

Joo, S H and Chun, K M (1997) Idle speed modeling

and optimal control of a spark-ignition engine KSME Int.

J 11, 1, 88−95

Kim, D and Park, J (2007) Application of adaptive

con-trol to the fluctuation of engine speed at idle Elsevier:

Information Sciences, 177, 3341–3355.

Lee, Y C (2001) Design of controllers for the stable idle

speed in the internal combustion engines Int J Korean

Society of Precision Engineering 2, 4, 54−60

Manzie, C and Watson, H C (2003) A novel approach to

disturbance rejection in idle speed control towards reduced

idle fuel consumption IMechE Part D: J Automobile

Engineering, 217, 677−690

Moskwa, J J and J Karl Hedrick, J K (1987) tive engine modeling for real time control application

Automo-Proc American Control Conf., 341−346

Nagashima, M and Levine, W S (2006) Development of

an engine idle speed and emission controller Proc 2006

American Control Conf., Minneapolis, Minnesota, USA.

Osburn, A W and Franchek, M A (2006) Reducing engineidle speed deviations using the internal model principle

J Dynamic Systems, Measurement and Control, 128,

869−877

Srail, M S., Sindano, H., Gough, N E and Cole, A C (2002).Sliding mode idle speed ignition control strategies for

automotive engines Electronic Proc 15th Int Symp

Mathe-matical Theory of Networks and Systems, University of

Notre Dame, USA

Thornhill, M and Thompson, S (1999) Adaptive fuzzy

logic control of engine idle speed IMechE, Part I, 213,

145−155

Thornhill, M., Thompson, S and Sindano, H (2000) A

comparison of idle speed control schemes Control

International Journal of Automotive Technology , Vol 12, No 1, pp 21−28 (2011)

DOI 10.1007/s12239−011−0003−2 Copyright © 2011 KSAE1229−9138/2011/056−03

21

SOOT AND TEMPERATURE DISTRIBUTION IN A DIESEL

DIFFUSION FLAME: 3-D CFD SIMULATION AND MEASUREMENT WITH LASER DIAGNOSTICS

Y HAN, W PARK and K MIN*

School of Mechanical and Aerospace Engineering, Seoul National University, Seoul 135-080, Korea

(Received 29 December 2008; Revised 6 July 2010)

ABSTRACT−In this study, a 3-D CFD simulation and laser diagnostics were developed to understand the characteristics of soot generation in a diesel diffusion flame The recently developed RANS (Reynolds-averaged Navier-Stokes equations) hybrid combustion model (Extended Coherent Flame Model - 3 Zones, ECFM-3Z model) was used This industrial, state-of- the-art model of the diffusion flame is commonly used in diesel combustion models as well as for propagating (premixed) flame combustion The simulation results were validated with measurements from a constant volume combustion chamber The experiment revealed that soot accumulated in the chamber where the temperature decreased Where the temperature increased rapidly, only a little soot accumulated The temperature and soot distribution were independently examined using both the two-color method and a 3-D CFD simulation for a turbulent diesel diffusion flame.

KEY WORDS :Constant volume vessel, Two-color Method, KL factor, ECFM-3Z, Soot, Diffusion flame

NOMENCLATURE

Ta1, Ta2: temperature at the 550 nm and 750 nm wavelength

1 INTRODUCTION

Current diesel engine research is focused on reducing exhaust

emissions and creating a reasonable fuel economy due to

increased environmental concerns, strict government

regulat-ions on exhaust emission standards and the increased prices of

petroleum-based fuels In anticipation of upcoming

regula-tions, the automotive industry is working to reduce vehicle

NOx, and PM) by developing high-efficiency engines.

As a result, the High-Speed Direct Injection (HSDI)

engine is gaining recognition and prominence for its

ultra-low emissions and ultra-high fuel efficiency However, the

HSDI diesel engine releases more pollutants into the

atmosphere than a gasoline engine This release is large

enough that consumers and environmental groups have

requested stronger regulations on the amount of legal

emissions In response, governments are introducing

stricter environmental regulations.

With the new regulations, every vehicle company is devoting significant resources to develop a low-emission automobile Unlike the gasoline engine, post-treatment technology cannot be applied to a diesel engine, which produces NOx and soot, because of its lean combustion conditions Consequently, to design the optimum combustion chamber shape, research must be done on high pressure and temperature conditions in a combustion chamber.

Therefore, many researchers are interested in developing

a process to measure soot measurements within a

two-color method in a diesel engine using the visible wavelength area to measure the temperature and soot

temperature under high temperature spray conditions In

(2004) measured the soot integral ratio and temperature in

a laminar flow diffusion flame and then compared the measurements to a steady-state diffusion flame where the soot integral ratio and temperature were measured using laser-based techniques.

However, diesel engine combustion conditions are complex due to the turbulent nature of flow; thus, the accuracy of the temperature and soot measurements is

2008, Vattulainen et al , 2000, Han et al , 2009) was used to measure the temperature and the soot concentration factor

α π D λ≡ ⁄

* Corresponding author e-mail: kdmin@snu.ac.kr

22 Y HAN, W PARK and K MIN

in diesel engines Two wavelengths of flames in a diesel

combustion chamber were analyzed.

Theoretical soot generation research has progressed.

However, to predict soot particle generation, the

configura-tion of the diffusion flame must be precisely modeled In

several established works, only the diffusion flame was

studied in a diesel engine; however, a real diesel engine has

both a diffusion flame and a rich premixed flame, as shown

in Figure 1B (Chomiak and Karlsson, 1996) The premixed

flame area must be included because most of the soot

particles collect in this area.

Therefore, many combustion models have been

developed to model diesel combustion characteristics In

particular, the ECFM model, which was developed by

The result was the ECFM-3Z model, which can be applied

to every combustion pattern (Colin and Benkenida, 2004).

In this study, the characteristics of diesel diffusion

flames were examined by measuring soot accumulation

and temperature in a visualized constant-volume chamber.

The soot and temperature distributions were experimentally

verified using the two-color method In addition, by

comparing two theoretical methods, the Eddy Break-Up

model, which considers only the diffusion flame, and the

ECFM-3Z combustion model, which considers both the

premixed flame and the diffusion flame, the characteristics

of the flame temperature and the soot generation

mechanism were found The accuracy of the exhaust gases

for combustion conditions was estimated by comparing the

experimental and simulation results.

2 EXPERIMENTAL SET-UP AND METHODS

2.1 Measurement of the Turbulent Diffusion Flame using a

Constant-volume Chamber

The two-color method (Zhao and Broughton, 1998, Reitz

and Hampson, 1998) detects the radiation of soot particles

by calculating the value of flame emissivity at two different

wavelengths to determine the flame temperature According to Wien’s law, blackbody emissivity at short wavelengths is defined in Equation (1) and can be used to calculate the black body emissivity over the visible wavelength range when temperatures are less than 3,000 K (Matsui et al , 1979).

(1) Where λ is the wavelength, T is absolute temperature, c1

c2 and are Frank constants, c1= 3.742×108w ·µm4/ m2, c2= 1.439×108µm4· K andελis the short wavelength emissivity

of the flame A turbulent diesel diffusion flame was created

in a visualized constant-volume chamber, as shown in Figure 2 The figure also shows the premixed combustion equipment injector for making sprays, the fuel supply system, that data acquisition system (which consists of a pressure sensor and an R-type thermocouple), and a high- speed camera to capture the spray and combustion phenomena High temperatures and pressures were attained

by igniting the gases (C2H2, O2, and N2) with a spark plug, which was insulated and kept at atmospheric temperature and pressure prior to injection.

Diesel fuel was sprayed through the one-hole injector when the atmospheric pressure and temperature inside the constant-volume chamber reached 19 bar and 1,200 K The diameter of the injector nozzle was 0.295 mm with an injection pressure of 1,000 bar Twenty milligrams of fuel were injected through the nozzle.

In addition, to observe combustion phenomena within the chamber, high-speed color and black-and-white cameras were positioned beside the bottom and side windows The cameras had 550-nm and 750-nm narrow band pass filters, and saturation of the 750-nm filtered image was prevented

by using a 2.62 ND (neutral density) filter (Lyn, 1957) Figure 3 shows the black-and-white images for when the largest and smallest flames occurred (1.667 msec and 5.667 msec, respectively) using the 550- and 750-nm narrow band pass filters.

I λ T( , ) ελc1λ5 c2

λT -–

SOOT AND TEMPERATURE DISTRIBUTION IN A DIESEL DIFFUSION FLAME 23

3 SIMULATION MODEL AND METHODS

3.1 Simulation Method

Two different combustion models, the Eddy Break-Up

model and the ECFM-3Z (Extended Coherent Flame

Model-3 Zones), were used to determine temperature and

soot distribution in the diesel flame A summary of the

spray sub-model parameters is shown in Table 1.

3.1.1 Combustion model

1) Eddy breakup model

The Eddy Break-Up (EBU) model, which was

developed by Magnussen and Hjertager (1981), is one of

the earliest models of turbulent chemical reactions The

EBU model was originally developed for combustion

applications and is based on the following assumptions: the

reaction is single step, irreversible, involves fuel (F), an

inert species, and the reaction has such a small time scale

that the rate-controlling mechanism is turbulent mixing

(Kim, 2003).

2) ECFM-3Z model

The ECFM-3Z model (Campbell and Gosman, 2008)

was developed by the Groupement Scientifique Moteur

(GSM) consortium along with their partners IFP, Renault, and PSA Peugeot-Citroen Figure 4 shows the conceptual sub-grid view of the ECFM-3Z combustion model The

following: the combustion sub-models (auto-ignition, premixed propagating flame, and diffusion flame) are combined by accounting for the local sub-grid state of the gases (i.e., their composition and temperature) by applying

a simple form of double conditioning This conditioning is done by dividing the mixing state into the following three zones: the unmixed fuel zone (labeled F), the mixed zone containing fuel, air and EGR (labeled M), and the unmixed air + EGR zone (labeled A) In addition, the reaction states

of the gases correspond to either the unburned (labeled u)

or burnt gas (labeled b) mixture In mathematical terms, the mixing space was defined by a three-point Dirac delta probability distribution function (PDF) For the reaction state, a reaction progress variable (c = 1) tracked the increase of burnt gases relative to the unburned gases Hence, the progress variable 'c' assumed a double Dirac (unburned or burnt) PDF description This double conditioning (for the mixing and reaction states) was applied to the cell mean values, which was solved by the transport equations in the CFD simulation The composition and temperature conditioned (sub-grid) values were used in each of the component combustion reaction rate models This approach accounted for the turbulence- chemistry interaction due to microscale fuel and temperature stratification.

3.1.2 Emission model for soot formation

additional transport equation that accounts for the soot mass fraction The modeling of the soot/flow-field interaction was based on a flamelet approach Source terms for the soot volume fraction were taken from a flamelet

Figure 3 Photograph of a raw image acquired by 550 and

750 nm narrow band-pass filters (Han et al., 2008).

Table 1 Sub-models used in spray model.

Atomization Reitz-Diwaker (Reitz and Diwakar, 1986)

Figure 4 Conceptual sub-grid view of the ECFM-3Z combustion model.

24 Y HAN, W PARK and K MIN

library using a presumed probability density function and

were integrated over the fraction space of the mixture To

save computer storage and CPU time, the flamelet library

of sources was constructed using a multi-parameter fitting

procedure, which resulted in simple algebraic equations

and a proper set of parameters.

The transport equation for soot mass fraction is given by:

(2) where Ys is the soot mass fraction, and ùõ is the source term

for the soot volume fraction.

3.2 Simulation Model and Set-up

The shape of the constant volume vessel is shown in Figure

5 For convenience, a 3-D cylinder mesh was created with

the same volume as the actual constant volume vessel.

cell size that was put near the flame extent position The

mesh contains 261,000 cells.

Unlike the experimental conditions, there was no

pre-combustion process for the auto-ignition condition in the

simulation Instead, the gas composition, temperature, and

pressure after pre-combustion were used for the initial

conditions The initial temperature and pressure of the

constant volume vessel were 1,200 K and 19 bar,

respectively The software Star CD version 3.26 was used

for the calculations.

4 EXPERIMENTAL AND THEORETICAL RESULTS

4.1 Temperature and Soot Measurement Results for a Diffusion Flame in a Visualized Constant-volume Chamber

Figure 7 shows the injected diesel fuel auto-igniting in the combustion chamber at a 1,000-bar injection pressure with

of 3,000 fps.

The fuel, which was injected from the upper side, ignited and traveled perpendicularly down to the bottom The largest diffusion occurred approximately between 1.333 and 1.667 msec During this phase, the amount of soot from the flame was measured using the two-color method.

⎛ ⎞ ρ+ Sωv

=

Figure 5 3-D CAD of a constant volume vessel.

Figure 6 Refined mesh for a constant-volume chamber.

Figure 7 Flame visualization of a constant volume

Figure 8 Soot and temperature distributions acquired by the flame image at 1.667 msec (Han et al., 2008).

SOOT AND TEMPERATURE DISTRIBUTION IN A DIESEL DIFFUSION FLAME 25

A short time later, at approximately 5.667 msec, the soot

measurements were taken again using the two-color

method The temperature distribution was also measured.

Figure 8 shows the results from the two-color method for

the qualitative soot amount and the temperature

distribution of the largest flame The soot distribution was

maximized in the region with the greatest flame generation.

This result suggests that the injected fuel did not fully

oxidize in the area that contained the most injected diesel

fuel In addition, some of the highest soot distributions

were seen around fuel droplets clustered together at the

nozzle tip area Note that the maximum temperature was

approximately 2,300 K and occurred at the bottom of the

constant-volume chamber.

and temperature distributions opposite from their statistical

representations in (a) and (c) As seen in the figure, the soot

created within the diffusion flame was reduced by rapid

oxidation under the high temperature flame.

Figure 9 shows the soot and temperature distribution 5.667 msec after the injection Due to the nozzle characteristics, agglomerated fuel droplets were injected

towards the end of the injection process, which caused a relatively large amount of soot to accumulate in a different area because of the oxidation process of the diffusion flame The maximum temperature was approximately 1,500 K, which was approximately 800 K below the overall maximum temperature In addition, the temperature distribution at this last stage (6.667 msec) was comparatively uniform.

Figure 10 shows the temperature distributions found by the two-color method with the same time sequence as Figure 7 The temperature decreased after the middle phase

of the process and decreased further during the latter part of the combustion process In addition, the maximum temperature occurred in the middle of the flame development.

Figure 11 shows the soot distributions for the same time sequence as Figure 7 From Figure 8, it is known that

a large amount of soot was created as a result of the initial flame generation when the flame developed The timing of the soot generation, which was different from the results for temperature generation, was during the latter part of maximum temperature generation The reason for this result could be that during the injection of fuel, the flame was generated and oxidized quickly Then, the air-fuel ratio changed rapidly during the later phase of flame generation Additionally, the maximum soot generation occurred shortly after the maximum temperature generation 4.2 Simulation Results

4.2.1 Comparison of combustion model The simulation results from the ECFM-3Z model were compared with the Eddy Break-Up model for the constant- volume vessel condition.

Figure 12 compares the pressure and the largest flame temperature of the vessel during combustion for the two

Figure 9 Soot and temperature distributions acquired by

the flame image at 5.667 msec (Han et al., 2008).

Figure 10 Results of flame temperature for a constant

Figure 11 Results of soot distribution for a constant

26 Y HAN, W PARK and K MIN

different models The pressure curve in Figure 12 shows

the pressure, which decreased initially and then increased.

The starting point for the pressure increase was possible the

point when the combustion started The Eddy Break-Up

model predicted the start of combustion earlier than the

ECFM-3Z model The pressure and temperature curves of

the Eddy Break-Up model were also higher amplitude than

the curves for the ECFM-3Z model.

Figure 13 shows the temperature distribution of the

vessel when the flame was the largest The flame and

temperature distributions of the two models appeared

markedly because the ECFM-3Z model accounted for the

diffusion part as well as the premixed part of the diesel

flame The ECFM-3Z model simulated real diesel flame

phenomenon more closely than the Eddy Break-Up model.

The ECFM-3Z model accurately predicted the onset of combustion, while the Eddy Break-Up model predicted an early start of combustion This result indicated that the auto-ignition model based on tabulated chemistry in the ECFM-3Z model was more reliable than the simple chemical mechanism used in the Eddy Break-Up model 4.2.2 Comparison of the experimental results

After comparing the two combustion models, the 3Z model provided a better description of diesel combustion In this subsection, the results of the simulation were compared with the experimental results.

ECFM-Figure 14 shows the temperature and the soot formation

of the largest flame during the combustion period in the constant-volume vessel After fuel injection, the flame began to form and temperature increased rapidly Soot formation also increased rapidly, although the soot subsequently met the hot air and was oxidized.

Figure 12 Pressure and temperature evolution of a

constant-volume vessel using the different combustion

models.

Figure 13 Temperature distribution of the largest flame

using different combustion models.

Figure 14 Temperature and soot evolution of a constant volume vessel.

Figure 15 Temperature distribution during combustion.

SOOT AND TEMPERATURE DISTRIBUTION IN A DIESEL DIFFUSION FLAME 27

As predicted by the ECFM-3Z model, Figure 15 shows

the temperature distribution in the middle of the vessel

during combustion with the same time sequence as Figure

7 The white circles indicate the visible window of the

vessel The largest flame, which occurs when the

maximum intensity of the fame was reached, occurred

1.667 msec after injection, after which, the flame

temperature decreased The maximum flame temperature

of the simulation was 2,600 K, while the maximum

temperature of the experiment was 2,300 K This

discrepancy occurred because the points of view for the

flame image were different The experimental results

showed the outside of the flame, and the simulation results

showed the center of the flame.

Figure 16 shows the soot distribution in the vessel during

combustion at the middle section of the vessel using the

ECFM-3Z model The time sequence for Figure 16 is the

same as Figure 7 The white circles indicate the visible

window of the vessel After fuel injection, the soot began to

form in the rich air-fuel mixture region inside the flame

with the same profile shape as the flame temperature in

Figure 15 Subsequently, the soot oxidized At the end of

the combustion, most of the soot that formed at the center

of the flames was oxidized, while the rest resided on the

flame surface The soot formation and oxidation process

was not well matched the experimental results Thus, the

soot model requires additional improvement.

5 CONCLUSIONS

In this study, the soot and temperature characteristics of a diesel

diffusion flame were examined through measurements taken

from a visualized constant-volume chamber The simulation

and the experiments yielded the following results:

(1) The largest flame temperature occurred at the front of

the flame, and the maximum amount of soot was

generated at the rear of the largest flame

(2) For the constant volume vessel case, the largest flame temperature was 2,300 K during the early combustion period, and maximum soot generation occurred after the maximum temperature was reached

be lower under high-temperature conditions and higher under low-temperature conditions.

(4) The experimental results were compared with two different combustion models for diesel flame: the ECFM-3Z model and the Eddy Break-Up model The ECFM-3Z model described diesel combustion better than the Eddy Break-Up model.

(5) The ECFM-3Z model results for temperature and soot distribution in a constant-volume vessel during combustion were compared with the experimental results The diesel fuel auto-ignition process matched the model well, but the soot model needs improvement.

Technology59, 6, 593−609.

Colin, O and Pires da Cruz, A and Jay, S (2005) Detailed chemistry based auto-ignition model including low temperature phenomena applied to 3D engine calculations Proc Combustion Institute

Han, Y T., Kim, K B and Lee, K H (2008) The investigation of soot and temperature distributions in a visualized direct injection diesel engine using laser diagnostics Meas Sci Tech 19, 11, 1−11.

Han, Y T., Lee, K H and Min, K D (2008) A Study on the measurement of temperature and soot in a constant- volume chamber and a visualized diesel engine using the two-color method J Mech Sci Tech , 22, 1537−1543 Han, Y T., Lee, K H and Min, K D (2009) A study on the measurement of temperature and soot in a constant- volume chamber and a visualized diesel engine using the

Consideration of Cavitation and Spray Impingement

Ph D Dissertation Seoul National University Korea Lyn, Y T (1957) Diesel combustion study by infrared emission spectroscopy J Inst Petrol , 43, 25.

Figure 16 Soot distribution during combustion.

28 Y HAN, W PARK and K MIN

Magnussen, B F and Hjertager, B W (1981) On the

structure of turbulence and a generalised eddy

dissipation concept for chemical reaction in turbulent

Matsui, Y., Kaminoto, T and Matruoka, S (1979) A study

on the flame temperature and soot concentration in a D.I.

790491.

Nam, Y W and Lee, W (2004) The measurement of soot

Symp

Regimes of Liquid Jet Ph D Dissertation Princeton

University USA.

Reitz, R D and Diwakar, R (1986) Effect of drop

Reitz, R D and Hampson, G J (1998) Two-color imaging

of in-cylinder soot concentration and temperature in a

heavy-duty D.I diesel engine with comparison to multidimensional modeling for single and split injection.

SAE Paper No 980524.

STAR-CD London.

Vattulainen, J., Nummela, V., Hernberg, R and Kytola, J (2000) A system for quantitative imaging diagnostics and its application to pyrometric in-cylinder flame-

Sci Tech , 11, 103−119.

Yamaguchi, I and Nakahira, T., Komori, M and Kobayashi, S (1990) An image analysis of high speed combustion photographs for D.I diesel engine with high

Zhao, H and Broughton, F (1998) Optical diagnostics for soot and temperature measurement in diesel engines.

Prog Energy Combust Sci , 24, 221−225.

International Journal of Automotive Technology, Vol 12, No 1, pp 29−38 (2011)

DOI 10.1007/s12239−011−0004−1

Copyright © 2011 KSAE 1229−9138/2011/056−04

29

EMISSION ANALYSIS OF A COMPRESSED NATURAL GAS

DIRECT-INJECTION ENGINE WITH A HOMOGENOUS MIXTURE

S ABDULLAH*, W H KURNIAWAN, M KHAMAS and Y ALI

Centre for Automotive Research, Faculty of Engineering & Built Environment,

National University of Malaysia, UKM Bangi 43600, Malaysia

(Received 16 March 2009; Revised 27 January 2010)

ABSTRACT−In an era in which environmental pollution and depletion of world oil reserves are of major concern, emissionsproduced by automotive vehicles need to be controlled and reduced An ideal solution is to switch to a cleaner fuel such asnatural gas, which generates cleaner emissions In addition, control over the in-cylinder air-fuel mixture can be best achievedthrough a direct-injection mechanism, which can further improve combustion efficiency This need for cleaner automobilesprovides the motivation for this paper’s examination of the use of computational fluid dynamic (CFD) simulations to analyzethe concentrations of the exhaust gases produced by a compressed natural gas engine with a direct-fuel-injection system Inthis work, a compressed natural gas direct-injection engine has been designed and developed through a numerical simulationusing computational fluid dynamics (CFD) to provide an insight into complex in-cylinder behavior The emissions analyzed

in this study were carbon monoxide (CO), nitric oxide (NO) and carbon dioxide (CO2), i.e the main pollutants produced bynatural gas combustion Based on a stoichiometric mixture, the concentrations of CO and NO were computed using thedissociation of carbon dioxide and the extended Zeldovich mechanism CO2 was calculated using a mass balance of the speciesinvolved in the combustion process The simulation results were then compared with the experimental data generated by asingle-cylinder research engine test rig A good agreement was obtained with the experimental data for the engine speedsconsidered for all emissions concentrations

KEY WORDS : Computational fluid dynamics, Compressed natural gas, Direct injection, Exhaust gases, Emissions,

Homogeneous mixture

NOMENCLATURE

ρ : density

µ : dynamic viscosity

c v : heat capacity at constant volume

k : thermal conductivity coefficient

Since the 19th century, gasoline and diesel based internal

combustion engines (ICEs) have been used to power

auto-motive vehicles and have achieved high levels of success in

terms of performance and features These engines have

been increasingly optimized for the best performance with

reduced exhaust emissions

In an era in which concerns over environmental

pollu-tion and the gradual deplepollu-tion of world oil reserves are

becoming major issues In light of the continuing use ofpetroleum-based fuels, the development of new enginesbased on available clean or renewable energy sources areideal solutions to address these critical issues However,hydrogen-based technologies, which include fuel cells andhydrogen ICEs, are still under intensive research and areanticipated to become available in approximately 20 years.While waiting for such technologies to become feasible,clean alternative fuels such as natural gas are available fordirect consumption by existing ICEs with some minormodifications Therefore, the usage of ICEs can be extend-

ed by switching to this fuel for commercial and domesticapplications

It is well-understood that configuring conventional ICEs

to improve efficiency while reducing exhaust emissions isdifficult Combustion strategies used to improve engineefficiency will also increase harmful emissions, namelycarbon monoxide (CO) and nitrogen oxides (NOx) On theother hand, reducing NOx emissions in certain ICEs mayincrease the levels of hydrocarbon output and particulatematters Conversely, applying some approaches to reducinghydrocarbons may increase NOx emissions (Turns, 1999).Having been a major source of pollution in the form ofspark ignition or diesel engines, ICEs’ environmental

*Corresponding author e-mail: shahrir@ukm.my

30 S ABDULLAH, W H KURNIAWAN, M KHAMAS and Y ALI

impact from their exhaust emissions and the inevitable

depletion of crude oil reserves result in a need to create a

clean and efficient engine (Soylu, 2005) Due to their

advantages in controlling fuel economy, direct-injection

(DI) systems appear to be a key to the successful

application of the spark ignition engine in the commercial

passenger car market However, mechanisms that can lead

to a reduction in exhaust emissions should be planned

properly to meet the requirements of regulations such as

Euro IV− after all, a common DI engine produces

relatively high levels of CO and NOx emissions (Belardini

and Bertoli, 1999)

For natural gas engines, research ranging from

numeri-cal analysis to experimental studies has been carried out by

previous researchers on issues related to performance and

emission Zhang and Frankel (1998) performed a

multi-dimensional numerical simulation to optimize the

perfor-mance of a fuel-lean burn with a homogeneous-charge,

natural gas, spark-ignition IC engine by examining the

effects of swirl, combustion chamber geometry and spark

plug location Chen and Milovanovic (2002) analyzed the

effects of exhaust gas recirculation (EGR) on a

homogene-ous charged compression ignition (HCCI) engine fuelled

with natural gas Selamet et al (2004) studied the unsteady

motion of chemical species including exhaust emissions in

the intake and exhaust ducts of a spark ignition engine

using a finite-difference-based simulation code A

multi-dimensional modeling of the formation of NO in a

direct-injection natural gas engine (modified from a diesel engine

with an auto-ignition system) was performed by Agarwal

and Assanis (2000) The combustion process and fluid flow

in a compression ignition natural gas engine with a

separat-ed chamber was analyzseparat-ed by Zheng et al (2005) by

coupling commercial CFD software with detailed chemical

kinetics

In efforts to achieve better emission quality for gasoline

engines, Duclos et al (1999) performed numerical studies

on DI spark ignition gasoline engines for stratified loads

and found good agreement between the computational and

experimental results for pressure trace, NO and CO For

diesel engines, CFD modeling of non-premixed

combus-tion in DI engines was performed by Barths et al (2000)

using the eddy dissipation concept for the combustion

model followed by a direct calculation of NO and soot

formation based on the simulation results Zellat et al.

(2005) carried out advanced modeling of DI diesel engines

to analyze the formation of NO and soot emissions and

then minimized the emissions using a multi-objective

optimization code to find the engine configuration

For experimental work on a natural gas engine, Shiga

et al.(2002) determined the characteristics of combustion

and emission of a CNG direct-injection combustion engine

by using a rapid compression machine (RCM) with a

compression ratio of 10 and a disc-shaped combustion

chamber The burned gases analyzed were methane (CH4),

NO and NO An analysis of fuel injection timing in

relation to ignition timing for a natural gas, direct-injection

mechanism was carried out by Huang et al (2003) using

RCM, where the exhaust emissions considered were burned CH4, CO, NOx, and CO2 Zeng et al (2006) studied

un-combustion characteristics of a direct-injection natural gasengine under various fuel injection timings The resultsshowed that injection timing can significantly influenceengine performance, combustion and emissions (CO, NO,HC) Cho and He (2008) performed a combustion andemission analysis on a spark ignited, port injection, naturalgas engine and found that lean burn could significantlyreduce NOx emissions but resulted in high cyclic variations

In terms of performance (torque and power output), thenatural gas engine was still constrained by its lowercalorific value compared to gasoline and diesel engines,and a pressure boost was recommended (Choa and Heb,

2007) A similar finding was reported by Abianeh et al.

(2009) using a bi-fuel engine for natural gas and gasolinefuels The study included influences on wall temperature,performance and emissions Another limitation of a naturalgas engine is that better emissions quality can be achieved

by operating the engine in a stoichiometric conditionbecause lean operation will increase the level of NOx Inaddition, lean operation is also possible by blending natural

gas with hydrogen (Wang et al., 2008).

Based on a review of the subject, it is obvious that anynumerical simulation used to predict emissions must beverified with an experiment For a compressed natural gasengine with spark ignition and direct fuel injection systems,however, experimental studies on emissions are still scarceand verification with numerical simulation is required toprovide good understanding of the overall process There-fore, this paper presents an in-depth investigation of theemissions produced by the combustion process of a com-pressed natural gas, direct-injection engine (CNGDI) Theobjective of this work is to analyze the emission gasesformed as a result of the combustion process in the engine.Finally, the DI system has been used as a method to add thefuel directly into the combustion chamber at a preciseamount according to engine load and speed during theintake stroke This mechanism was selected instead of theport-injection system because it can provide more controlover the in-cylinder mixture profile before combustion, andthus better fuel economy and engine performance Thenumerical simulation and the corresponding experimentwere performed for CO, NO and CO2 for engine speeds of1000~3000 rpm

2 ENGINE GEOMETRY AND CONDITIONS

In this work, the single-cylinder model of a combustionchamber was taken from a 1.6-liter four-stroke CNGDIengine with two intakes and exhausts valves as shown inFigure 1(a) The chamber was equipped with a pistoncrown designed specifically to generate a homogeneousmixture and a compression ratio of 14:1 for the CNGDI

EMISSION ANALYSIS OF A COMPRESSED NATURAL GAS DIRECT-INJECTION ENGINE 31

operation The section view of the combustion chamber

geometry is illustrated in Figure 1(b), which shows the

locations of the intake and exhaust ports, the fuel injector

and spark plug, and the intake and exhaust valves For this

engine, the spark plug position was maintained on the

central axis of the combustion chamber while the location

of the fuel injector was shifted slightly aside of the spark

plug Due to the combustibility issue of natural gas at a

lower concentration, the nozzle end of the injector was kept

within 5 mm of the spark plug tip By doing so, the overall

DI system could still be categorized as a central injection

system (Abdullah et al., 2008).

By using this geometry, the CFD simulation was carried

out to predict emissions produced by the combustion

process for engine speeds of 1000~3000 rpm with 500 rpm

increments The specifications of the CNGDI engine under

consideration are summarized in Table 1 For the purpose

of verifying the numerical results, experiments were

per-formed on a single-cylinder research engine (SCRE) test

rig, and the results were then compared with the CFD

simulation for the selected exhaust gases through a

mea-surement with a gas analyzer

The conditions for the internal combustion process ofthe engine were controlled by three types of timing para-meters: the start of injection (SOI), the end of injection(EOI) and the spark ignition (SI), which was managed by aprogrammable electronic control unit (ECU) These com-bustion parameters played an important role in achievingthe optimal engine performance and minimal exhaustemissions Initially, the parameters of SOI, EOI and SIwere set for a stoichiometric combustion so as to achievecomplete combustion and thus, better engine performance.Further adjustments to the parameters were made to reducethe exhaust emissions After performing a comparisonbetween the experimental and simulated results, an enginemapping database was established for the ECU operation.The engine setting for the SOI, EOI and SI timings usedduring the experiment and for the CFD simulation is given

Figure 1 Schematic diagram of the CNGDI engine

Table 1 Specifications of the CNGDI engine

Intake valve open (° c) 12° before TDC

Intake valve close (° CA) 48° after BDC

Exhaust valve open (° CA) 45° before BDC

Exhaust valve close (° CA) 10° after TDC

Maximum intake valve (mm) 8.1

Maximum exhaust valve (mm) 7.5

Table 2 Engine timings used for the CFD simulation andthe SCRE experiment at 20 bar injection pressure

Speed (rpm) (oSOI timing before TDC) (oEOI timing before TDC) (obefore TDC)SI timing

32 S ABDULLAH, W H KURNIAWAN, M KHAMAS and Y ALI

hand, at bottom dead center (BDC), the number of cells

and vertices exceeded 163,110 cells and 48,550 vertices

(see Figure 3(b)) During the mesh generation, a

hexa-hedral cell was preferred due to its accuracy and stability

when performing the CFD simulation with a moving mesh

and boundaries For the numerical solution, STAR-CD

software was employed, coupled with the user-defined

subroutines for controlling the gaseous fuel injection event

and the moving mesh event (which allowed the valves and

piston to move according to the crank angle) The

simu-lation covered the full four-stroke cycle, which was

measured as degrees of crank angle (°CA) The simulation

started just before the intake valve opened during the intake

stroke and continued until the residual gases exited through

the exhaust port during the exhaust stroke

3.2 Governing Equations for the CFD Simulation

The governing equations of mass, momentum and energy

conservation were based on the continuity, Navier-Stokes

and energy equations for an ideal gas and are respectively

given as follows:

(1)

(2)

(3)Initially, the computational domain was assumed to be

occupied with stationary fresh air in the form of an ideal

gas, where its temperature and pressure were assumed to be

homogeneous at the standard atmospheric conditions Thepressure and temperature at the intake manifolds were set

at 103 kPa and 305 K, respectively, while at the exhaustport, the pressure was kept constant at the atmosphericcondition and the temperature was anticipated to reach ashigh as 802 K A constant pressure condition was used atboth the intake and exhaust ports so that dynamic behavior

in the ports was determined through the mass and tum balance of the solution The walls for the intake andexhaust ports as well as the lateral walls for the valves wereconsidered to be adiabatic, while constant temperatureconditions were specified separately at the cylinder head,the cylinder wall and the piston crown, which defined theinternal structure of the combustion chamber based on thetypical values observed during the experiments

momen-The turbulence model used in this work for the CFD

simulation was the k-ε model for a high Reynolds numberwhich was found to be adequate for reciprocating engines(El-Tahry, 1983) The initial values for pressure andtemperature for each engine operating condition wereobtained from the experiment on the SCRE test rig Theinitial turbulence intensity was set at 5% of the mean flowwhich was found to be suitable for in-cylinder turbulentflow The integral length scale was estimated at 0.4%(Launder and Spalding, 1972) The initial value of the

turbulent kinetic energy k was assumed to be spatially

uniform and was set equal to 3% of the kinetic energy ofthe mean piston speed

As mentioned above, the simulation started at 348º CAafter TDC and finished at 855º CA after TDC following thecompletion of all four strokes (intake, compression, powerand exhaust strokes) During the compression stroke, thefuel injection event specified by the SOI and EOI timingswas invoked, followed by an ignition event defined by the

SI timing just before the end of the compression stroke Inaddition, the initial pressure and temperature within theengine cylinder were also defined to provide a more re-gulated initial condition, which assisted in the convergence

of the solution at an early stage of the simulation Theinitial pressure was set at 100 kPa and the initial temper-ature was set at 302 K The time step used for the simu-lation had units of degrees of crank angles and the incre-ment value was set as 0.1° CA This value was consideredsmall and was used to avoid the formation of the negativedensities that occurred during the simulation as a result ofhaving a local Courant number exceeding the limit of 2.0 atcertain parts of the computational domain, (causing themesh to distort when the intake and exhaust valves openedand closed)

To maintain numerical stability during the solver tion, the temporal discretization was set as implicit with anunder-relaxation parameter of 0.1 For greater accuracy ofthe simulation, the second-order differencing scheme ofMARS (monotone advection and reconstruction scheme)was employed for solving the momentum, energy and tur-bulence equations This was coupled with the arbitrary

+∇ k∇T⋅( )+λ(∇ u⋅ )2+∇u⋅µ[∇u ∇u+( )T]

Figure 3 Computational mesh of the CNGDI engine at

TDC and BDC positions

EMISSION ANALYSIS OF A COMPRESSED NATURAL GAS DIRECT-INJECTION ENGINE 33

Lagrangian-Eulerian (ALE) subroutine for controlling grid

movement associated with moving pistons and valves The

overall system of the resulting algebraic equations was

solv-ed using the popular PISO algorithm for unsteady flows

3.3 Combustion and Emissions Modeling

The combustion modeling utilized to perform the CFD

analysis for this CNG/DI engine was the standard eddy

break-up (EBU) model developed by Magnussen (1981)

for an upremixed or diffusion reaction, which consisted of

three global reactions The main reason for using this

scheme was that the natural gas fuel in the direct-injection

system was merely segregated from air as the oxidizer, so

the rate of energy release was primarily limited by the

mixing process within the engine cylinder In the case of

the unpremix reaction, there was no fundamental flame

speed (as in the case of premixed flames) and the flames

were not one-dimensional, i.e the chemical kinetics played

a secondary role in the behavior of the diffusion of flames

The EBU model was originally developed for combustion

reactions and is based on the following assumptions (Borman

and Ragland, 1998):

a The reaction is a single-step irreversible reaction

involv-ing fuel, an oxidant and products, plus possible

back-ground inert species;

b The reaction time scale is so small that the

rate-controlling mechanism is turbulent micromixing

The standard EBU model involves three global reactions

for an unpremix or diffusion reaction for CH4 (the primary

compound of natural gas) and is described as follows:

(4)

To determine the concentration of the exhaust emissions,

the concept of carbon dioxide dissociation and three

ex-tended Zeldovich mechanisms were employed for the

prediction of CO and NO concentrations, respectively CO

is usually generated in ICEs when they are operated in a

fuel-rich conditions When there is inadequate oxygen to

convert all carbon to CO2, some portion of the fuel is not

burned, producing CO Typically, the exhaust of a spark

ignition engine will contain about 0.2% to 5% CO As for

NO, it is generated concurrently with the combustion

pro-cess due to the reaction between oxygen and nitrogen

atoms The formation of NO is dependent on in-cylinder

temperature: Its formation is relatively low during engine

start-up and begins to increase when the high-temperature

burned gases are left behind by the flame front The three

chemical reactions that form the extended Zeldovich

reaction known as thermal NO can be written as follows:

(5)

In general, CO2 is not considered as a pollutant ever, it is a major greenhouse gas with a potentially signi-ficant impact on the global warming of the earth if itscomposition in the environment exceeds a certainthreshold In addition, for the combustion of natural gasand any other hydrocarbon fuels, CO2 is one of the majorcombustion products besides water vapor Therefore, inthis work, the exhaust gases of CO2 were also considered asone of the pollutants generated by the combustion process

How-A sufficient amount of oxygen and a specific portion offuel could react in a full stoichiometric condition and form

CO2 without any CO or NO A more complete combustionproduced the CO2 and all chemical energy contained in thefuel was converted into thermal energy and kinetic energyduring the combustion process The CO2 emissionmodeling was completed using mass equilibria of thechemical species The calculation of concentrations insidethe CFD code was performed by considering the chemicalreaction rate of each species as stipulated in Equation (4).3.4 Experiment on the Single-cylinder Research Engine(SCRE) Test Rig

For the purpose of verifying the CFD simulation, a bustion experiment was carried out on the SCRE test rigusing an eddy current dynamometer as shown in Figure 4.During the experiment, the fuel injection and ignition tim-ings were adjusted via a programmable ECU kit installedinside the SCRE This ECU kit was operated through thesoftware interface to yield the best optimal configurationfor engine performance, namely torque and power Toobserve and monitor the boundary conditions at the intakeand exhaust for the CFD analysis, thermocouples wereinstalled as close as possible to the cylinder head so that thetemperature of the intake and exhaust ports could bedetermined and the values could to be used inside the CFDsimulation All other engine boundary conditions werefixed The concentrations of exhaust gases were measuredusing the in-situ gas analyzer The gas analyzer was located

com-at a distance of 3.0 meters from the exhaust port and was

34 S ABDULLAH, W H KURNIAWAN, M KHAMAS and Y ALI

connected through a pipe These measured emissions levels

were compared with the concentrations of CO, NO and

CO2 predicted by the CFD simulation

4 RESULTS AND DISCUSSIONS

In this section, the results of the CFD simulation of exhaust

emissions for various engine speeds are presented

How-ever, due to similarity in CFD results for many engine

speeds, only the results for 2000 rpm are presented in the

form of color contours For other speeds, the results have

been converted into key engine parameters presented in the

form of graphs and histograms All the contours of the

simulation results are depicted and displayed according to

the degrees of crank angle By showing the emissions in

colored contours, localities where the formation of high

concentrations of CO, NO and CO2 can be identified within

the combustion chamber and analyzed further

4.1 Engine PerformanceFor a speed of 2000 rpm, the power unleashed by com-bustion can be best represented by the pressure contours asdepicted in Figure 5 Using pressure contours that employ-

ed the appropriate volume integral technique, the averagein-cylinder pressure was calculated and plotted againsttime (along with the experimental data for verification) asgiven in Figure 6 This figure shows a very good agreementbetween the CFD simulation and the experimental dataobtained from the SCRE test rig Additionally, the samevalues were re-plotted against degrees of crank angle to

give a simulated p-V curve, as illustrated in Figure 7 The area of the p-V loop yields the indicated power for this

single-cylinder engine

4.2 CO ConcentrationFigure 8 presents the concentration of CO found inside thecombustion chamber for several crank angles during andafter combustion A high concentration of CO was locatednext to the fuel injector and close to the piston bowl wherethere was a rich mixture of fuel This led to incompletecombustion around the area The variation of CO along the

Figure 5 In-cylinder pressure distribution at 2000 rpm

Figure 6 Simulated and measured in-cylinder pressure at

2000 rpm

Figure 7 p-V diagram for the indicated power at 2000 rpm.

Figure 8 Distribution of CO at 2000 rpm

EMISSION ANALYSIS OF A COMPRESSED NATURAL GAS DIRECT-INJECTION ENGINE 35

crank angle degrees is shown in Figure 9

As can be seen, the CO emissions were initially formed

at a crank angle of 703° at the time of ignition, and they

increased until the crank angle reached 731° The

maximum concentration of CO was 0.74% and exited at a

crank angle of 731° After that, the CO level decreased and

stabilized at 0.44% until the beginning of the exhaust

stroke Although the concentration of CO seemed to be

stable after reaching the maximum value (due to a heated

environment), a further oxidation process from CO to CO2

occurred when the flue gases exited the combustion

chamber through the exhaust pipe Hence, the simulated

value of CO emissions was higher than the measured level

because the data was measured at a distance of 3.0 meters

from the exhaust port

4.3 NO Formation

Figure 10 depicts the formation of NO gases inside the

combustion chamber for several degrees of crank angle An

area containing NO formation was found in the region

around the spark plug This was the result of a high local

temperature produced during ignition In addition to the

temperature, the amount of NO generated also depended on

pressure, air-fuel ratio, combustion time inside the cylinder,

and the locality within the combustion chamber The NO

gases were initially produced at a crank angle of 708°,

which is the crank angle degree just after ignition The

level of NO increased to 1350 ppm around a crank angle of

740° (as shown in Figure 11) and then decreased until itstabilized around 600 ppm before the exhaust valveopened However, the simulated value of NO emissionswas higher than the measured levels since the datameasured by the gas analyzer was obtained at a distance of3.0 meters from the exhaust port

4.4 CO2 DistributionFigure 12 illustrates the distribution of CO2 (one of theprimary combustion products) as a function of crank angledegrees Here, CO2 was formed in the upper part of thepiston surface, and its concentration increased further dur-ing and even after combustion until the beginning of theexhaust stroke However, the pattern for CO2 gases was nothomogeneous due to the asymmetrical geometry of thepiston crown and cylinder head The CO2 tended to beconcentrated in the localities where the spark plug waslocated and the combustion process began to take place.Figure 13 shows the variation of CO2 along the degree ofcrank angle It appeared that the initial CO2 was firstproduced at a crank angle of 703° due to the combustionprocess The CO2 level continued to increase until themaximum value of 6.89% was reached before the exhaustvalve opened

Figure 9 CO distribution versus crank angle at 2000 rpm

Figure 10 Distribution of NO at 2000 rpm

Figure 11 NO distribution versus crank angle at 2000 rpm

Figure 12 Distribution of CO at 2000 rpm

36 S ABDULLAH, W H KURNIAWAN, M KHAMAS and Y ALI

4.5 Emissions Analysis for Various Engine Speeds (1000~

3000 rpm)

In this section, the average values for the emissions

concentrations at the beginning of the exhaust stroke, as

simulated by the CFD, are compared with the measured

data obtained from the gas analyzer connected to the SCRE

test rig The latter reported more diluted values because the

concentrations were measured 3.0 meters from the SCRE

combustion chamber In fact, the CFD computation was

finished before the exhaust valve opened Both analyses

were performed using the injection and ignition timings

listed in Table 2 because these timings had led to optimal

torque and power during the SCRE experiment

In general, all of the data measured inside the

combustion chamber were less than the corresponding

simulated values for a number of reasons For CO and CO2,

the oxidation reaction for the conversion of CO to CO2

continued inside the exhaust pipe of the SCRE test rig

Figures 14 and 15 show a comparison of the emission

levels between the simulated value in the cylinder and the

data measured by the gas analyzer

From these diagrams, the overall trend of CO emissions

for various engine speeds was quite inconsistent according

to the mass of fuel injected into the combustion chamber

per cycle However, the balance between CO and CO2 can

clearly be seen, where a relatively low CO concentrationresulted in a high concentration of CO2 at 1000 rpm, andvice versa At a low engine speed, the injected fuel insidethe cylinder was initially (relatively) smaller, resulting inlower CO levels Then, the CO values increased as thespeed increased due to the addition of more fuel At themid-range speeds, 1500~2500 rpm, CO seemed to bestabilized, and not much had been converted to CO2 Themeasured CO2 level can be said to have come from thecombustion of the fuel itself At the same time, the levels of

CO in the combustion chamber and in the exhaust tail piperepresented the completeness of the combustion As for ahomogeneous mixture, the primary objective of the optimi-zation process was to maximize engine performance.Hence, the occurrence of incomplete combustion must bereduced From Figure 14, it can be seen that there was anagreement between the CO emissions levels calculated bythe CFD simulation and the gas analyzer measurement atthe exhaust tail pipe

For the CO2 concentration, the combustion productgiven in Figure 15 showed a slightly lower level at a lowerspeed due to the low equivalence ratio between fuel and air

At 2000 rpm, the CO2 concentration of the CFD simulationincreased due to the large quantity of CO2 produced fromthe air-fuel mixture At a medium speed around 2500 rpm,the CO2 concentration decreased slightly due to the reducedamount of intake air even though the quantity of fuelincreased because of the injection timing used Similar tothe CO concentration, there was some agreement betweenthe CO2 levels measured by the CFD simulation and thegas analyzer measurement at the exhaust tail pipe at certainengine speeds

On the other hand, the concentration of NO wasrelatively small at a lower speed because of the lowercombustion temperature in the engine cylinder Figure 16shows the comparison between the CFD simulation and thevalue measured by the gas analyzer at the exhaust exit Thesame trends were observed for engine speeds of1000~2000 rpm At mid-range speeds, the NO levelincreased The highest levels of NO occurred at speeds of

Figure 13 CO2 distribution versus crank angle at 2000

rpm

Figure 14 Calculated and experimental values of CO

emission levels Figure 15 Calculated and experimental values of COemission levels 2

EMISSION ANALYSIS OF A COMPRESSED NATURAL GAS DIRECT-INJECTION ENGINE 37

2500~3000 rpm due to higher temperatures during the

combustion process and a large amount of air within the

combustion chamber The discrepancies with the measured

gas analyzer data occurred at higher engine speeds where

the measured values were much lower than those predicted

by the CFD simulation This interesting finding will be

covered in future works because compliance to some NOx

levels is a requirement for most vehicle standards, such as

those in EuroIV However, the agreement at the lower

engine speeds (1000~2000 rpm) is visible in Figure 16

Finally, the CFD numerical works presented in this

paper proved that it is possible to predict in-cylinder

combustion behavior which can then compared with the

experimental data obtained from a test rig By doing so, the

phenomena that occur inside the combustion chamber can

be well-understood In terms of flame propagation and its

variation with respect to the crank angle, natural gas

possesses a resistance to knocking due to its high rating

octane number (107), its higher flash point for auto-ignition

and its good mixture profile, supplied by the central DI

system Consequently, the ignition timing (15~20°CA

before TDC for this work) can be set at a wider range of

crank angle without knocking

5 CONCLUSION

In this work, the CO, NO and CO2 emissions

concent-rations occurring in a CNGDI engine running at speeds of

1000~3000 rpm were calculated numerically by

perform-ing a CFD simulation from the intake stroke until the

beginning of exhaust stroke At certain engine speeds, the

results showed good agreement between the CO, NO and

CO2 concentration values calculated from the CFD results

(to represent in-cylinder behaviors) and the experimental

data measured by the gas analyzer at the exhaust tail pipe

However, for higher engine speeds, the level of NO inside

the combustion chamber anticipated by the CFD results

was higher than the measured data These interesting

phen-omena will be studied in future works Nevertheless, the

engine performance still lagged behind that of the gasoline

engine, especially for high-speed and high-load operations,

due to natural gas’s limited energy content

REFERENCES

Abdullah, S., Kurniawan, W H and Shamsudeen, A.(2008) Numerical analysis of the combustion process in

a com-pressed natural gas direct injection engine J.

Applied Fluid Mechanics 1, 2, 65−86

Abianeh, O S., Mirsalim, M and Ommi, F (2009)

Com-bustion development of a bi-fuel engine Int J.

new technology light duty diesel engines Oil & Gas

Science and Technology - Rev IFP, 54, 251−257

Borman, G L and Ragland, K W (1998) Combustion

Engineering McGraw-Hill New York.

Chen, R and Milovanovic, N (2002) A computationalstudy into the effect of exhaust gas recycling on homo-geneous charge compression ignition combustion in

internal combustion engines fuelled with methane Int J.

Thermal Science, 41, 805−813

Cho, H M and He, B Q (2008) Combustion andemission characteristics of a lean burn natural gas

engine Int J Automotive Technology 9, 4, 415−422.

Choa, H M and Heb, B Q (2007) Spark ignition natural

gas engines - A review Energy Conversion &

Manage-ment 48, 2, 608−618

Duclos, J M., Zolver, M and Baritaud, T (1999) 3D

modeling of combustion for DI-SI engines Oil & Gas

Science and Technology - Rev IFP 54, 2, 259−264

El-Tahry, S H (1983) k-ε equation for compressible

reciprocating engine flows AIAA J Energy 7, 4, 345−

353

Huang, Z., Shiga, S., Ueda, T., Nakamura, H., Ishima, T.,Obokata, T., Tsue, M and Kono, M (2003) Effect offuel injection timing relative to ignition timing on the

natural gas direct-injection combustion J Engineering

for Gas Turbines and Power, 125, 783−790

Launder, B E and Spalding, D B (1972) Lectures in

Mathematical Models of Turbulence Academic Press

Inc London

Magnussen, B F (1981) On the structure of turbulenceand a generalised eddy dissipation concept for chemical

reaction in turbulent flow Proc Nineteenth AIAA

Aero-space Meeting St Louis, Missouri, USA.

Selamet, E E., Selamet, A and Novak, J M (2004).Predicting chemical species in spark-ignition engines.Figure 16 Calculated and experimental values of NO

emission levels

38 S ABDULLAH, W H KURNIAWAN, M KHAMAS and Y ALI

Energy, 29, 449−465

Shiga, S., Ozone, S., Machacon, H T C., Karasawa, T.,

Nakamura, H., Ueda, T., Jingu, N., Huang, Z., Tsue, M

and Kono, M (2002) A study of the combustion and

emission characteristics of compressed-natural-gas

direct-injection stratified combustion using a

rapid-compression-machine Combustion and Flame, 129, 1−

10

Soylu, S (2005) Examination of combustion characteristics

and phasing strategies of a natural gas HCCI engine

Energy and Conversion Management, 46, 101−119

Turns, S R (1999) An Introduction to Combustion -

Con-cepts and Applications 2nd Edn McGraw-Hill Boston.

Wang, J., Huang, Z., Miaoa, H., Wanga, X and Jianga, D

(2008) Characteristics of direct injection combustion

fuelled by natural gas–hydrogen mixtures using a

constant volume vessel Int J Hydrogen Energy 33, 7,

1947−1956

Zellat, M., Duranti, S., Liang, Y J., Kralj, C., Schmidt, G.and Duclos, J M (2005) Towards a universal com-bustion model in STAR-CD for IC engines: from GDI toHCCI and application to DI Diesel combustion optimi-

zation Proc 14th Int Multidimensional Engine User’s

Meeting, SAE Cong., Detroit, Michigan, USA.

Zeng, K., Huang, Z H., Liu, B., Liu, L X., Jiang, D M.,Ren, Y and Wang, J H (2006) Combustioncharacteristics of a direct-injection natural gas engine

under various fuel injection timings Applied Thermal

Engineering, 26, 806−813

Zhang, D H and Frankel, S H (1998) A numerical study

of natural gas combustion in a lean burn engine Fuel 77,

International Journal of Automotive Technology, Vol 12, No 1, pp 39−44 (2011)

DOI 10.1007/s12239−011−0005−0

Copyright © 2011 KSAE 1229−9138/2011/056−05

39

AUTOMOBILE DEFROSTING SYSTEM ANALYSIS THROUGH

A FULL-SCALE MODEL

S J KANG, M F KADER, Y D JUN and K B LEE*

Department of Mechanical Engineering, Kongju National University, Chungnam 330-717, Korea

(Received 18 June 2009; Revised 18 April 2010)

ABSTRACT−Adequate visibility through the automobile windshield is of paramount practical significance, most often atvery low temperatures when ice tends to form on the windshield screen But the numerical simulation of the defrost process

is a challenging task because phase change is involved In this study numerical solution was computed by a finite volumecomputational fluid dynamics (CFD) program and experimental investigations were performed to validate the numericalresults It was found that the airflow produced by the defrost nozzle is highly nonuniform in nature and does not cover thewhole windshield area The nonuniformity also severely affected the heating temperature pattern on the windshield Thewindshield temperature reached a maximum in the vicinity of the defroster nozzle in the lower part of the windshield andranged from 9~31°C over a period of 30 min, which caused the frost to melt on the windshield The melting time was under

10 minutes, which satisfied the NHTSA standard The numerical predictions were in close agreement with the experimentalresults Thus, CFD can be a very useful design tool for an automobile HVAC system

KEY WORDS : CFD, Automobile HVAC, Windshield, Defrosting

1 INTRODUCTION

During the winter season, at very low temperatures, ice

usually forms on the windshield of an automobile Defrost

analysis is essential to improve the capacity of the

wind-shield defrost system to melt ice completely from the outer

screen surface and to eliminate the mist formed on the

inner surface within an expected time period The advent of

unstructured grid technology and improved physical

modeling capabilities in areas such as phase change and

radiation have contributed to the increased use of CFD in

automotive applications, especially in the field of

automobile heating, ventilating, and air-conditioning

(HVAC) systems Earlier investigators searched for ways

to improve the design of windshield defroster/demister

systems They recognized the problem and applied recent

advances in experimental diagnostics techniques and

computational fluid dynamics (CFD) to study the air flow

Dugand and Vitali (1990) carried out an experimental

investigation where a thermographic technique was used to

detect thermal fields on emitting surfaces The authors

proposed a specific combination of hardware/software for

the processing of the obtained images and recommended

various ways of improving windshield/defrosting systems

Carignano and Pippione (1990) used a computer assisted

thermographic technique to optimize the perfor mance of

windscreen defrosting for an industrial vehicle system Lee

et al (1994) utilized a CFD program, namely ICEM-CFD,

to simulate the mechanism of windshield de-icing Thecomplete vehicle configuration was transformed fromCAD, and the mesh was created and assembled using amulti-domain approach The authors demonstrated thecapability of the developed module in simulating coldroom de-icing tests to supplement the experimental work

Brewster et al (1997) used the CFD program (STAR-CD)

to simulate the mechanism of ice building on the shield in three-dimensional form The authors used a non-linear enthalpy-temperature relationship to simulate theice/water layer Melting contours were predicted every 5minutes, and the authors reported close agreement betweenthe numerical simulations and cold-room test data for theice coverage contours Abdul Nour (1998) conducted asimilar study, also using the STAR-CD program Heexamin ed the windshield flow fields and vehicle defrostersystem under various operating conditions Thecomparison between hot-wire velocity measurements andthe numerical predictions showed close agreement forvarious defroster and windshield flows Aroussi andHassan (2003) compared the performances of the sidewindow defrosting mechanism of several current vehicle

wind-models An additional study by Aroussi et al (2003)

concentrated on simulating the turbulent fluid flow over,along with heat transfer through, a model of a vehiclewindshield defrosting and demisting system Furthermore,

Park et al (2006) simulated the flow and temperature field

on the interior of an automobile cabin when the hot air is

*Corresponding author e-mail: kumbae@kongju.ac.kr

40 S J KANG, M F KADER, Y D JUN and K B LEE

discharged from the defrost nozzle to melt the frost on the

windshield glass Kader et al (2009) used both numerical

and experimental methods to study temperature

distribu-tion characteristics of an automobile interior when the

HVAC system is operated through defrost mode and

Instrument Pane (IP) mode

From the above review, it is clear that numerical

simu-lation of defrosting is a challenging task Though some

achievement has been made in understanding the defrost

analysis, there is still a need for further scrutiny In the

present study, the fluid flow pattern and temperature

distri-bution on the windshield inner surface and outer surface

are investigated using CFD to determine the capability of

the method Thermography and K-type probes were used to

determine the elevated temperature on the windshield and

some particular positions in Figure 2(b), respectively

2 EXPERIMENTAL SETUP

The experiment was performed on an SM3 2006 model

vehicle of Samsung Automobile Company with a 1,500 cc

diesel engine, presented in Figure 1 The automobile was

instrumented with sensors (K-type probes) to measure the

temperature on the inner and outer surfaces of the

wind-shield as shown in Figure 2 The fully opened defroster

nozzles were used to supply the flow at an average velocity

of 13 m/s The experimental period was 30 min Because

the data were automatically recorded to a PC every 5

seconds, a data acquisition system integrated with a PC

was employed to control this complicated task The

am-bient temperature was −6.9°C Before the experiment was

started, ice was allowed to form naturally on the

wind-shield

Thermography was used to determine the temperature

contours developed on the windshield due to the flow from

the defroster grillers Thermography has the advantage of

providing an instantaneous map of the object surface

temperature or velocity rather than point by point

measure-ments in space The experimental setup associated with this

technique is shown in Figure 3 All the trial runs were

carried out at ambient room temperature under defroster

conditions in which air was discharged into the windshield

through the defroster nozzle located on the dashboard The

system consisted of a thermal image camera, which

re-corded the thermal evolution of the windshield, togetherwith a PC to capture, analyze and process the imagesobtained The lens was perpendicular to the plane of thewindshield The thermal image camera was positioned on atripod at a distance of about 3 m in front of the windshield.After turning on the blower of the HVAC system andsetting up the thermographic equipment, thermal mapswere recorded at 30-second intervals

3 NUMERICAL INVESTIGATION

The numerical code used in this study was the finitevolume CFD program Scryu Tetra (SC/T) version 7 (Anon,2007) The software has three main components, namelythe pre-processor, solver and post-processor andabbreviated as SC/T-pre, SC/T-solver and SC/T-post,respectively The three dimensional geometry of the modelwas imported to the pre-processor of the Scryu TetraFigure 1 Automobile used in experiment

Figure 2 Measurement locations

Figure 3 Experimental setup for thermograph