Preprints, 8th IFAC International Symposium on

Preprints,
Advances 8th IFAC International
in Automotive Control Symposium on
Preprints, 8th
Preprints, 8th IFAC International
IFAC International Symposium on
Symposium on
Advances
June 19-23,in2016.
Automotive Control
Norrköping, Sweden
Advances
Advances in Automotive Control Available online at www.sciencedirect.com
June 19-23,in2016.
Automotive Control
Norrköping, Sweden
June
June 19-23,
19-23, 2016.
2016. Norrköping,
Norrköping, Sweden
Sweden

ScienceDirect
IFAC-PapersOnLine 49-11 (2016) 353–360
A
A Double-Injection
Double-Injection Control
Control Strategy
Strategy For
For
A Double-Injection
Partially Control
Premixed Strategy
Combustion For
Partially Premixed Combustion
Partially Premixed Combustion
Gabriel ∗
Ingesson ∗ Lianhao Yin ∗∗ Rolf Johansson ∗∗
∗∗
Gabriel Ingesson ∗ Lianhao Yin ∗∗ Rolf Johansson ∗
∗∗ ∗
Gabriel
Gabriel Ingesson ∗Per
Ingesson
Per
Tunestål
Lianhao
Lianhao
Tunestål
Yin
Yin∗∗ ∗∗
∗∗
Rolf
Rolf Johansson
Johansson
Per Tunestål
Per Tunestål ∗∗
∗
∗ Department of Automatic Control,Lund University,Lund,Sweden,
∗ Department of
∗ Department(e-mail: of Automatic Control,Lund University,Lund,Sweden,
Gabriel.Ingesson@control.lth.se,
Department(e-mail: of Automatic
Automatic Control,Lund
Control,Lund University,Lund,Sweden,
University,Lund,Sweden,
Gabriel.Ingesson@control.lth.se,
Rolf.Johansson@control.lth.se)
(e-mail: Gabriel.Ingesson@control.lth.se,
(e-mail: Gabriel.Ingesson@control.lth.se,
∗∗ Rolf.Johansson@control.lth.se)
∗∗ Department of Rolf.Johansson@control.lth.se)
Energy Sciences,Lund University,Lund,Sweden,
Rolf.Johansson@control.lth.se)
∗∗ Department
(e-mail:
Department of Energy Sciences,Lund
Lianhao.Yin@energy.lth.se,
of University,Lund,Sweden,
Per.Tunestal@energy.lth.se)
∗∗
Department
(e-mail: of Energy
Energy Sciences,Lund
Lianhao.Yin@energy.lth.se, Sciences,Lund University,Lund,Sweden,
University,Lund,Sweden,
Per.Tunestal@energy.lth.se)
(e-mail: Lianhao.Yin@energy.lth.se, Per.Tunestal@energy.lth.se)
(e-mail: Lianhao.Yin@energy.lth.se, Per.Tunestal@energy.lth.se)
Abstract: Partially Premixed Combustion has shown to be a promising internal combustion
Abstract:
concept in terms
Abstract: Partially
Partially Premixed
of engine
Premixed Combustion
efficiency and emission
Combustion has
has shown
levels,to
shown be
be aaa promising
however,
to since theinternal
promising combustion
internal combustion
process
combustion
Abstract:
concept in Partially
terms of Premixed
engine Combustion
efficiency and has
emission shown
levels, to be
however, promising
since the internal
combustion combustion
is partly
concept in
concept kinetically
in terms
terms of controlled,
of engine
engine efficiency unacceptably
efficiency and and emission high
emission levels,pressure-rise
levels, however,
however, since rates could
since the occur
the combustion
combustion at process
various
process
process
is
is partly
operating
partly kinetically
points.
kinetically A controlled,
remedy
controlled, to unacceptably
this problem
unacceptably high
is
highthe pressure-rise
use of
pressure-rise a rates
pilot-fuel
rates could occur
injection.
could occur at
This
at various
paper
various
is partly kinetically
operating points. A controlled,
remedy to unacceptably
this problem high
is the pressure-rise
use of rates could
aaexperimental
pilot-fuel occurThis
injection. at various
paper
characterizes
operating
operating double
points.
points. A injection
remedy
A injection
remedy to effects
to this on
this on combustion
problem
problem is the through
use
is thethrough of pilot-fuel
use of aexperimental results
injection.
pilot-fuel injection. andThis
This presents
paper
paper
characterizes
a double
characterizes double double
injection closed-loop
double injection effects
model
injection effects
effects on combustion
predictive
on combustion control
combustion through strategy where
through experimental the results
combustion
experimental results results and and presents
phasing
and presents
presents is
characterizes
a double injection closed-loop model predictive control strategy where the combustion phasing is
controlled
a
a double arbitrarily
injection
double injection as
closed-loop
closed-loopthe pressure
model rise rate
predictive
model predictive is controlled
control
control strategy
strategybelow a
where
where upper
the bound.
combustion
the combustion Experimental
phasing
phasing is is
controlled
results show
controlled arbitrarily
promising
arbitrarily as
as the
the pressure
response
pressure rise
times
rise rate
during
rate is
is controlled below
set-point changes
controlled below aa upper
as wellbound.
upper as load
bound. Experimental
and speed
Experimental
controlled
results showarbitrarily as
promising responsethe pressure rise
times duringrate is controlled below
set-point changes a upper
as well bound.
as load Experimental
and speed
disturbances.
results
results show
show promising
promising response response times times duringduring set-point
set-point changes
changes as as well
well as as load
load and and speed
speed
disturbances.
disturbances.
disturbances.
© 2016, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords: Partially Premixed Combustion, Multiple-Injection Closed-loop Control, Model
Keywords:
Predictive
Keywords: ControlPartially
Partially Premixed
Premixed Combustion, Multiple-Injection Closed-loop
Combustion, Multiple-Injection Closed-loop Control, Model Model
Keywords:
Predictive Partially
Control Premixed Combustion, Multiple-Injection Closed-loop Control, Control, Model
Predictive Control
Predictive Control
1. INTRODUCTION It has however been discovered that in single-injection
1. INTRODUCTION
INTRODUCTION It
PPC,has thehowever
long been ignitiondiscovered
delay gives that in single-injection
1.
1. INTRODUCTION It
It has
PPC,has thehowever
however
long
been
been
ignition
discovered
discovered
delay that rise
that
gives
in
rise
to very high
in single-injection
single-injection
Partially premixed combustion (PPC) is a low tempera- PPC, pressure-rise
PPC, the the long rates,
long ignition due to
ignition delay violent
delay givesgives rise rise to
HCCI-like to
very
very high
to combustion
very high
high
Partially premixed combustion (PPC) is aa low
low tempera- pressure-rise
rates (Manente
pressure-rise rates,
rates, due
et due to
al., to violent
2009).
violent HighHCCI-like
pressure-rise
HCCI-like combustion
rates
combustion
ture combustion
Partially
Partially premixed
premixed concept
combustion
combustion that is controlled
(PPC)
(PPC) is
is a lowboth by
tempera-
tempera- fuel pressure-rise rates, due to violent HCCI-like combustion
ture combustion
injection and chemicalconcept that is
kinetics, is controlled
it can be viewed both by
as by fuel rates
a com- indicate
rates (Manente
(Manente et
for high audible
et al., 2009).
et al., noise levels
al., 2009).
High
Highand
High pressure-rise
could also cause
pressure-rise rates
rates
ture
ture combustion
combustion
injection and
concept
concept
chemical
that
that is
kinetics, it
controlled
controlled
can be viewed
both
both as bya fuel rates
fuel
com- indicate (Manente
mechanical for high
engine audible
damage,
2009).
noise levels
therefore, andpressure-rise
it could
has toalso
be
rates
cause
kept
bination
injection of
and conventional
chemical diesel
kinetics, combustion
it can be with
viewed the
as direct
aa com- indicate for
indicate for engine high audible
high audible noise
noise levels
levels andand could
could also
also cause
cause
injection and chemical kinetics, it can be viewed as com- mechanical damage, therefore, it has to be
bination of conventional
controllability
bination of conventional
of the diesel combustion
combustion
diesel combustion
through with
fuel
with the direct below
injection
the direct mechanical
mechanicalcertain levelsdamage,
engine
engine in order
damage, to ensure
therefore,
therefore, it silentto
it has
has be kept
to and
be safe
kept
kept
bination of conventional
controllability of the diesel combustion
combustion through with
fuel the direct below
injection operation.certain levels
Previous in order
research by to ensure
(Tsurushima silentet and
al., safe
2002)
and homogeneous
controllability of charge
the compression
combustion throughignitionfuel with high below
injection below certain
certain levels in
levelsresearch order
in orderbyto to ensure
ensure silentsilent and safe
and2002)
safe
controllability
and homogeneous of the combustion
charge compression throughignitionfuel injection
with high operation.
implies that Previous
pressure oscillations (Tsurushima
commonly et al.,
resulting after
thermal
and efficiency
homogeneous and
chargelow NO and
compression
x soot emission
ignition with levels.
high operation.
operation. Previous
Previous research
research by
by (Tsurushima
(Tsurushima et
et al.,
al., 2002)
2002)
and homogeneous
thermal efficiency and charge
and lowcompression
NOx and sootignition
emissionwith high implies
levels. violent that pressurerates oscillations
also arecommonly resulting after
thermal
thermal efficiency
efficiency and low
low NO NOxxisandand soot emission levels. implies combustion
implies that
that pressure
pressurerates oscillations
oscillations able to break
commonly
commonly insulating
resulting
resulting after
after
Low temperature combustion an soot
oftenemission
used name levels.for violent combustion
thermal-boundary
violent combustion layers
rates also
which
also are able
increases
are able to
to break insulating
heat-transfer
break flux
insulating
Low temperature combustion is an often used name for violent combustion
thermal-boundary rates
layers also
which are able
increases to break insulating
heat-transfer flux
combustion
Low
Low temperature
temperatureconcepts where
combustion
combustion the ignition
is an
isignitionoften
an often delayused
usedis prolonged
name for to the cylinder walls.
layers A remedy
name for thermal-boundary layers which increases heat-transfer flux
thermal-boundary which to
increases the problem
heat-transfer of high
combustion concepts where the delay is prolonged to the cylinder walls. A remedy to the problem of flux
high
in direct-injection
combustion concepts
combustion engines
concepts wherewhere the in order
the ignition to
ignition delay enhance
delay is fuel-air
is prolonged
prolonged to pressure-rise
the cylinder rates in PPC
walls. A is to divide
remedy to the the fuel injection
problem of high
in direct-injection engines in order to enhance fuel-air to the cylinder walls. A remedy to the problem of high
premixing.
in An increased
direct-injection engines ignition
in
in orderdelayto togives the fuel
enhance more pressure-rise
fuel-air into rates
several injection in events,
PPC ise.g.,
to divide
by having the fuel injection
in direct-injection
premixing. An engines
increased ignition order
delay enhance
gives the fuelfuel-air
more pressure-rise
pressure-rise
into several
rates
rates in
injection
PPC
PPC is
in events, to
to divide
ise.g., divide
by havingthe an
the fuelearly
fuel
an
pilot
injection
injection
early pilot
time to
premixing.
premixing. penetrate
An
An the
increased
increased gas mixture
ignition
ignition before
delay
delay gives
givesthethe
thecombustion
fuel
fuel more
more injection
into several with less
injectionthan half
events, of the
e.g., fuel
by and
having a main
an injection
early pilot
time to penetrate the gas mixture before the combustion into several
injection with injection
less than events,
half of e.g.,
the by
fuel having
and a an
main early pilot
injection
starts.
time
time to This
to yieldsthe
penetrate
penetrate thelocally
gas leaner mixtures
gas mixture
mixture before
before the theduring
combustion
combustion com- containing
injection theless
with majority
than half ofof the
the fuel amount.
and aa main The pilot
injection
starts. This yields locally leaner mixtures during com- injection with less than half of the fuel and main injection
bustion
starts. This
starts. whichyields
This both locally
yields reducesleaner
locally the formation
leaner mixtures ofduring
mixtures particulate
during com- containing
com- injection
containing
containing
the
sets
the a majority
lean
majority
thea majority mixtureof
of the
of the
fuel
environment
the fuel amount.
amount.
fuel amount. that The pilot
decreases
The
The pilot
pilot
bustion
matter
bustion which
and
which NO both
bothdue reduces
to
reduces lowerthe
the formation
combustion
formation of particulate
temperatures
of particulate injection sets lean mixture environment that decreases
bustion which x
both reduces the formation of particulate the ignition
injection
injection sets delay
a
sets delay lean of the
mixture
a leanofmixture main injection
environment
environment and
that
that therefore
decreases
decreases
matter
(Musculus, and NO x due to lower combustion temperatures the ignition the main injection and therefore
matter
matter and
(Musculus, and 2006).
NO
NOxx due
2006). due to to lower
lower combustion
combustion temperatures
temperatures decreases the ignition
the ignition the delay
combustion
delay therates.
of the
of mainThis
main technique
injection
injection andis also
and used
therefore
therefore
(Musculus, 2006). decreases
in the
conventional combustion
diesel rates.
engines This
both technique
to improve is also
low used
load
(Musculus,
PPC has been 2006).under research for the past decade at decreases the combustion rates. This technique is also used
decreases the combustion rates. This technique is also used
PPC has been under research for the past decade at in
in conventional
performance
conventional diesel
(MacMillan
diesel engines
et
engines al.,both
2009;
both to
to improve
Osuka
improve et low
al.,
low load
1994)
load
Lund
PPC
PPC University
has
has been
been and is aresearch
under
under conceptfor
research where
for the
the apast
combination
past decade
decade of in
at
at conventional
performance diesel engines
(MacMillan et al.,both
2009; toOsuka
improve et low1994)
al., load
Lund
early University and is a concept where a combination of and to decrease
performance
performance emissions
(MacMillan
(MacMillan and
et
et engine
al.,
al., 2009;
2009; noise
Osuka
Osuka levels
et
et (Eriksson
al.,
al., 1994)
1994)
Lund injection
Lund University
University timings,
and is
and is high
a concept
a EGR where
concept ratios a
where aand the usage of
combination
combination of and to decrease emissions and engine noise levels (Eriksson
early and toNielsen, 2014; Kiencke and Nielsen,
noise2000).
early injection
gasoline-like
early injection
injection
timings,
fuels has been
timings,
timings,
high EGR
EGR ratios
highapplied
high EGR
and
and the
to achieve
ratios
ratios and
usage
usage of
a sufficient
the
the usage of and
of and
to decrease
decrease
Nielsen,
emissions
emissions
2014; Kiencke
and
andand
engine
engine noise
Nielsen,
levels (Eriksson
levels
2000).
(Eriksson
gasoline-like
ignition
gasoline-likedelay fuels has
while
fuels has been applied
maintaining
been applied to
low
to achieve
fuel
achieve a sufficient
consumption
a sufficient and
and Nielsen,
Nielsen, 2014;
2014; Kiencke
Kiencke and
and Nielsen,
Nielsen, 2000).
2000).
gasoline-like fuels has maintaining
been applied low to achieve a sufficient Both optical OH chemiluminescence experiments (Tanov
ignition
(Manente
ignition delay
et
delay while
al., 2010).
while maintaining low fuel
fuel consumption
consumption Both optical OH chemiluminescence experiments (Tanov
ignition delay while et
Both al., 2014),
maintaining low fuel consumption Both optical OH chemiluminescence experiments (Tanov
optical OH and computational
chemiluminescence fluid dynamics
experiments CFD
(Tanov
(Manente
(Manente et et al.,
et al., 2010).
al., 2010).
2010). et al., 2014), and computational fluid dynamics CFD
(Manente simulation
et
et al., 2014),
al., 2014), researchand (Solsjö,
computational
and computational 2014) have
fluid been
fluid been dynamics
dynamicsperformed
CFD
CFD
simulation
to better
simulation research
understand
research (Solsjö, 2014)
multiple-injections
(Solsjö, 2014) have
have light-duty
been performed
PPC,
performed
 The authors would like to acknowledge the Competence Center for simulation
to better research (Solsjö,
understand 2014) have light-duty
multiple-injections been performed PPC,
 The authors would like to acknowledge the Competence Center for findings
to bettersuggest
to better understand
understand that multiple-injections
multiple injections light-duty
multiple-injections can be used
light-duty PPC,
PPC, to

Combustion
 The authors Processes,
would likeKCFP,
to and the Swedish
acknowledge the Energy Agency
Competence Center for
for findings
create
findings suggest
stratified
suggest that
mixtures
that multiple
in
multiple injections
different air-fuel
injections can
can be
ratio
be used
regions
used to
to
The authors
Combustion would like
Processes, to acknowledge
KCFP, and the the Competence
Swedish Energy Center for
Agency for findings suggest mixtures
that multiple injections canratio
be used to
the financial support
Combustion Processes, (project
Processes, KCFP, number
KCFP, and and the 22485-3),
the Swedish Scania
Swedish Energy for supplying
Energy Agency
Agency for for create
where
create stratified
very lean
stratified mixtures in different
created
in by
different air-fuel
pilot
air-fuelinjections
ratio regions
burns
regions
Combustion
the financial support (project number 22485-3), Scania for supplying create stratified mixtures in different air-fuel ratio regions
the
the experimental
financial supportengine and
(project the KFCP
number 22485-3), PPC Control
Scania
Scania for
reference
for supplying
where
more very
slowly leanbut mixtures
assists thecreated
ignition by pilot injections burns
the
the financial
group
the
support
experimental
for consistent
experimental
(project
engine andnumber
feedback
engine and
the the
on
the
KFCP
KFCP
22485-3),
work.PPCThe Control
PPC Control
supplying
authors reference
Ingesson
reference
where
where
more
very
very lean
slowly leanbut
mixtures
mixtures
assists the
created
created
ignition byof
by
of
fuel-richer
pilot
pilot injections
injections
fuel-richer
mixtures
burns
burns
mixtures
the
group
and
experimental
for consistent
Johansson are
engine and
feedback
members of
the
on
the
KFCP
the
LCCC work. PPC Control
The authors
Linnaeus Center
reference
Ingesson
and the created
more
more by
slowly
slowly the but
butmain
assists
assistsinjection.
the
the ignition
ignition of
of fuel-richer
fuel-richer mixtures
mixtures
group
group for consistent
for consistent feedback
feedback on
on the
thethe
work.
work. The authors
The authors Ingesson
CenterIngesson created
created by the main injection.
and
and
Johansson
eLLIIT are Center
Excellence
and Johansson
Johansson are
members
are members
at of
members of
Lund
of the
LCCC
University
the LCCC
Linnaeus
LCCC Linnaeus
Linnaeus Center
Center and
and the
and the
the created by by the
the main
main injection.
injection.
eLLIIT Excellence Center at Lund University
eLLIIT
eLLIIT Excellence
Excellence Center
Center at
at Lund
Lund University
University
Copyright
2405-8963 ©
© 2016
2016, IFAC
IFAC (International 362 Hosting by Elsevier Ltd. All rights reserved.
Federation of Automatic Control)
Copyright © 2016 IFAC 362
Copyright
Peer review© 2016
under IFAC
responsibility
Copyright © 2016 IFAC of International Federation of 362
Automatic
362Control.
10.1016/j.ifacol.2016.08.053
IFAC AAC 2016
354
June 19-23, 2016. Norrköping, Sweden Gabriel Ingesson et al. / IFAC-PapersOnLine 49-11 (2016) 353–360

With the increased amount of fuel injection events, the the real-time system using Drivven drivers. The fuel used
amount of calibration work for optimized engine perfor- in the experiments was a mix of 80 volume % gasoline and
mance for different loads and speeds grows exponentially, 20 volume % N-heptane.
(Meyer, 2011). Therefore it is of course very appealing to
find automatic fuel-injection controllers that automatically The NOx and HC emission levels were measured with a
Horiba measurement system (MEXA-9100E) while soot
find fuel-injection timings and fuel distribution among the
levels were measured with an AVL micro soot sensor
multiple injections. Previous work on pilot-injection com-
measurement unit.
bustion control in low-temperature combustion concepts
were presented by (Ott et al., 2013; Eichmeier et al., 2012; The heat-release analysis and controller computations
Ekholm et al., 2008; Kokjohn et al., 2009). were run in MathScript RT Module nodes inside a timed-
loop block on the real-time target. The timed loop was
This paper presents experimental PPC results from a
triggered by the FPGA once the in-cylinder pressure was
Scania multi-cylinder engine that reflects engine output
sampled. All computations were done using floating point
characteristics w.r.t. double-injection parameters. The ex-
arithmetic.
perimental results are then used for design of a model pre-
dictive controller that is able to track combustion phasing
3. SIGNAL PROCESSING AND DEFINITIONS
while guaranteeing an upper bound on the maximum pres-
sure rise rate. Controller implementation and experimental
performance during engine load and speed variations are 3.1 Heat Release Analysis
also presented.
In order to extract combustion information from the mea-
sured in-cylinder pressure p, heat-release analysis was per-
2. EXPERIMENTAL SETUP
formed every cycle for each cylinder. In order to account
for the pressure-sensor offset, the p was pegged w.r.t. to
2.1 Engine Specifications the measured inlet manifold pressure at inlet valve closing
(IVC), high-frequency noise on p was attenuated using a
The experimental engine was a Scania D13 heavy-duty zero-phase digital filter.
diesel engine with engine specifications displayed in Table
2.1. The engine speed was controlled with a 355 kW AC The heat-release rate dQc /dθ was estimated by treating
motor that worked both as an engine motor and brake. the combustion chamber content as a single-zone open
The engine was boosted with a fixed geometry turbine system. Applying the first law of thermodynamics and the
compressor combination. ideal gas law yields
dQc γ dV 1 dp dQht
Table 1. Engine Specifications = p + V + , (1)
dθ γ − 1 dθ γ − 1 dθ dθ
Total Displaced volume 12.74 dm3 where V is the cylinder volume and θ the crank angle.
Stroke 160 mm The ratio of specific heats γ = cp /cv depends on the
Bore 130 mm cylinder gas composition and temperature, in this work γ
Connecting Rod 255 mm
Compression ratio 18:1
was estimated assuming that p and V satisfy the adiabatic
Number of Valves 4 relation
pV γ = C (2)
γ and C could then be estimated from p and V using linear
2.2 Engine control and measurement system regression methods. The convective heat-transfer rate from
the combustion chamber gases to the combustion chamber
The entire engine control system was programmed with wall dQht /dθ was modeled according to (Woshni, 1967)
LabVIEW which is a graphical programming environment where the in-cylinder temperature T was estimated using
developed by National Instruments. The real-time system the measured intake manifold temperature TIM at IVC
consisted of a NI PXIe-8135 embedded controller (2.3 GHz (TIM = TIVC ) which gives
quad-core processor), NI PXI-7854/7854 R (Multifunc- pV
T = TIVC . (3)
tion reconfigurable I/O (RIO) with Virtex 5-LX110/LX30 pIVC VIVC
FPGA). The FPGA was considered as a configurable hard- With the accumulated heat release Q obtained from Eq.
ware that worked as a flexible AO / DIO, it was also used (1), the combustion timing θ50 and the combustion start
for AD acquisition. θ10 was computed using
The in-cylinder pressure was measured with water-cooled Q(θx )
Kistler 7061B pressure sensors and was sampled with the x = 100 , x = {10, 50}. (4)
Leine-Linde crank angle encoder pulse every 0.2 crank maxθ Q(θ)
angle degree. Inlet manifold and exhaust pressures were
measured with Keller PAA-23S absolute pressure sensors. 3.2 The Maximum Pressure Rise Rate dpmax
Inlet manifold and exhaust temperatures were measured
with K-type thermocouples. The maximum pressure rise rate dpmax is in this work
defined as
The fuel injection system was a production Xtra high Pres- dp
dpmax = max . (5)
sure Injection (XPI) common-rail injection system for the θ dθ
Scania D13 engines. The common-rail pressure, injection Due to the high cycle-to-cycle variation of this signal it has
timings, durations and valve positions were controlled from to be filtered in some way in order to be used as a feedback

363
 IFAC AAC 2016
June 19-23, 2016. Norrköping, Sweden Gabriel Ingesson et al. / IFAC-PapersOnLine 49-11 (2016) 353–360 355

Injector Current Signal Table 2. The investigated operating points.

2
The notion x : y : z means that the corre-
sponding parameter was swept from x to z in
1.5 steps of y, every combination of rp and dSOI
were tested.
Injector Current [a.u]

m
θSOI
p
1 θSOI pIMEPg [bar] 10
Nspeed [rpm] 1200
λ 1.8
0.5
rEGR [rpm] 25
θ50 [CAD] 10
0 p m
rp 0 : 0.075 : 0.3
τ θ τth θDOI
th DOI dSOI [CAD] 0 : 12.5 : 50

−0.5

−50 −45 −40 −35 −30 −25 −20 −15 −10 −5 0 5

dp [bar/CAD], 10 bar p
max IMEPg
θ [CAD ATDC] 50
21

45 20
x x
Fig. 1. Definitions of θSOI , θDOI , and τth . 19
40
variable. It was therefore assumed that the measured 18

dpm
max was corrupted with Gaussian measurement noise

[CAD]
35 17

dpm 2
max = dpmax + e, e ∼ N (0, σ ). (6) 16
30

SOI
15
A Kalman filter was then used to recover dpmax , this is

d
presented more in detail in Section 5.2. 25 14

13
20
3.3 The Fuel Injection Signal 12

11
15
The start of injection of the main and pilot injection 0 0.05 0.1 0.15 0.2 0.25 0.3
m p
θSOI , θSOI is here defined as the crank-angle timing of rp
the current pulse sent to the injector. The fuel-injection
m p
duration of the injections θDOI , θDOI is defined as the Fig. 2. Maximum pressure-rise rate contour plot versus
difference width in milliseconds between the current pulse injection timing difference and pilot ratio.
sent to the injector and a threshold value τth that is the
minimum injection duration for which fuel flows into the thermal efficiency, ηth , NOx , HC and soot emission levels
m
cylinder, here it was assumed that τth = 0.25 ms, see Fig. were measured in steady state. The gains from θSOI and
m p m p p
1. Instead of studying θSOI , θSOI , θDOI and θDOI explicitly, θSOI to θ50 were also investigated.
the pilot ratio rp The results of these experiments and its implications for
θp controller design are presented in the following sections.
rp = p DOI m , (7)
θDOI + θDOI
the injection separation dSOI 4.1 The Maximum Pressure Rise Rate dpmax
m p
dSOI = θSOI − θSOI , (8) In Fig. 2 the measured dpmax is presented as level curves
m
the main injection timing θSOI and the total injection of a linear interpolations between the operation points in
tot
duration θDOI were studied Table 2. It is clear that rp can be used to control dpmax
tot
θSOI m
= θDOI p
+ θDOI . (9) since dpmax decreases with rp , however, the controllability
decreases as dSOI is increased. A possible explanation for
This change of variables was done since at a given
tot m the observed trends is that the air-fuel mixture that is set
operating point, θSOI and θSOI would be determined by
up by the pilot injection decreases the ignition delay of the
the desired load and combustion phasing, rp and dSOI are
main-injection fuel charge, making the combustion of this
then free parameters whose influence on engine output
charge less violent. This effect then becomes stronger for
parameters are discussed in the preceding section.
larger pilot-fuel amounts and weakens as the pilot charge
is more dilute as dSOI increases. The ignition delay of the
4. DOUBLE-INJECTION EXPERIMENTS
main injection τ m is presented in Fig. 3, where τ m was
computed accordingly
In order to investigate the engine-output properties w.r.t. m
different double-injection configurations, rp and dSOI were τm = θ10 − θSOI [ms]. (11)
m
varied while keeping θ50 and the indicated gross mean It can be seen that τ and dpmax correlate in Figs. 2 and
effective pressure pIMEPg , 3.
 EV O
1 4.2 Thermodynamic Efficiency
pIMEPg = pdV (10)
Vd IVC
constant at the operating points (OP) presented in Table The thermodynamic efficiency ηth is here defined as the
2. At every operating point, 1000 cycles of data were ratio between the indicated gross mean effective pressure
sampled. At each operating point dpmax , the indicated pIMEPg and the fuel mean effective pressure pFuelMEP

364
IFAC AAC 2016
356
June 19-23, 2016. Norrköping, Sweden Gabriel Ingesson et al. / IFAC-PapersOnLine 49-11 (2016) 353–360

τm [ms], 10 bar pIMEPg HC [ppm], 10 bar p

IMEPg
50 50
1.9
450
45 45
1.8

400
40 1.7 40

1.6

dSOI [CAD]
350
[CAD]

35 35

1.5
30 30 300
SOI

1.4
d

25 25
1.3 250

20 1.2 20
200

15 1.1 15

0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 0.1 0.15 0.2 0.25 0.3
r r
p p

Fig. 3. Main-injection ignition delay contour plot versus Fig. 5. HC emission-level contour plot versus injection
injection timing difference and pilot ratio. timing difference and pilot ratio at 10 bar pIMEPg .
η , 10 bar p
th IMEPg NO [ppm], 10 bar p
50 x IMEPg
0.488 50
850
45 0.486
45

40 0.484 800
40
[CAD]

35 0.482
dSOI [CAD]
35
750
30 0.48
SOI

30
d

0.478
25
25 700
0.476
20
20
0.474
650
15
15
0 0.05 0.1 0.15 0.2 0.25 0.3
rp 0 0.05 0.1 0.15 0.2 0.25 0.3
rp

Fig. 4. Thermodynamic efficiency contour plot versus

injection timing difference and pilot ratio.
pIMEPg
ηth = , (12)
pFuelMEP
where
QLHV mf
pFuelMEP = , (13)
Vd
mf is the injected fuel amount and QLHV the lower
heating value of the fuel. Note that this definition of ηth
includes the combustion efficiency. The result presented in
Fig. 4 shows that ηth has a slight increase when having a
pilot injection close to the main injection. Another visible
trend is that ηth drops when dSOI increases and that this
effect is stronger when rp is higher. This indicates that it
could be more efficient to have a pilot injection but that
this effect is reversed when the pilot is advanced.
Fig. 6. NOx emission-level contour plot versus injection
timing difference and pilot ratio at 10 bar pIMEPg .
by the HC emission-level increase could also explain the
efficiency drop seen in Fig. 4.
4.4 NOx
The measured NOx emission levels are presented in Fig.
6 where it can be seen that NOx mainly depends on
rp and decreases with an increased rp . A hypothetical
explanation for this is that the increase in τm gives
more violent combustion with increased temperatures with
leaner charge mixtures at the start of combustion and that
this gives an increased NOx formation.

4.3 HC
The measured HC emission levels are presented in Fig. 5.
The HC emission levels increase steeply as dSOI and rp are
simultaneously increased, the explanation for this could
be that the pilot fuel is injected into the crevice regions
outside of the combustion chamber and does not burn
completely, it could also be that the pilot-fuel mixture
becomes too lean to burn due to the increased mixing
period. The decrease in combustion efficiency indicated
4.5 Soot
In Fig. 7 it is shown that the soot levels increase with
rp and more so when dSOI is small, the reason for this
is probably that the ignition delays for the fuel injections
are minimizes at these operating points which makes the
charge burns more rich and this increases soot formation.
Similar NOx and soot emission trends were presented by
(Manente et al., 2010a, 2009).

365
 IFAC AAC 2016
June 19-23, 2016. Norrköping, Sweden Gabriel Ingesson et al. / IFAC-PapersOnLine 49-11 (2016) 353–360 357

Soot [mg/m3], 10 bar pIMEPg dθ50 / dθpSOI, 10 bar pIMEPg

50 50 0.08
130

45 120 45 0.07

110
40 40 0.06
100
dSOI [CAD]

[CAD]
35 90 35
0.05
80
30 30

SOI
70 0.04

d
25 60 25
0.03
50
20 20
40 0.02

15 30 15
0.01
0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 0.1 0.15 0.2 0.25 0.3
r rp
p

p
Fig. 7. Soot emission-level contour plot versus injection Fig. 9. The estimated gain from θSOI contour plot versus
timing difference and pilot ratio. injection timing difference and pilot ratio at 10 bar
pIMEPg .
4.7 Summary
dθ / dθm , 10 bar p
50 SOI IMEPg
50 1.14
The findings in the preceding experimental results are
45
1.12 summarized accordingly
1.1
• rp can be used to control dpmax , the controllability
40
1.08 decreases with dSOI .
1.06 • ηth increases slightly with rp , the opposite holds with
dSOI [CAD]

35

1.04 large dSOI , this is probably linked with increased HC-

30
1.02
emission levels.
• rp is a trade-off between NOx and soot emission levels.
25 1 m
• θ50 is controlled by θSOI .
0.98
20
0.96 5. CONTROLLER DESIGN
15
0.94
0 0.05 0.1 0.15 0.2 0.25 0.3 The goal of the controller design is to find a controller that
r
p simultaneously keeps θ50 at a predefined value while dpmax
is maintained below an upper bound.
m
Fig. 8. The estimated gain from θSOI contour plot versus
injection timing difference and pilot ratio at 10 bar The experimental results above show that dpmax can be
m
pIMEPg . controlled with rp while θ50 is controlled with θSOI . To
maintain high controllability of dpmax , dSOI should be
kept low, in this work dSOI was therefore chosen constant,
dSOI = 20. In order to maximize the ignition delay τ m
4.6 Combustion-Phasing Controllability which is a key parameter in PPC it was also decided to
try to keep rp low while fulfilling the constraint on dpmax .
The controllability of the combustion phasing θ50 was tot
The engine load was then controlled with θDOI , in a
m p
investigated by varying θSOI and θSOI for the operating separate closed loop.
points presented in Table 2.
In order to build a state-space model and apply model-
For each operating point and rp and dSOI combination, based controller design, the state vector x and input vector
m,p
θSOI was varied individually as square waves with ampli- u are introduced accordingly
tudes of 1 CAD and periods of 25 cycles during 500 cycles. xk = [θ50,k , dpmax,k , rp,k ]T , (14)
m,p
The gains from θSOI to θ50 were estimated from these m
uk = [∆θSOI,k , ∆rp,k ]T ,
experiments and are presented in Figs. 8 and 9.
where k denotes the cycle index and ∆ is the forward-shift
m
It is clear from the results that it is θSOI that controls θ50 operator, rp is here introduced as a state in order to keep
but that the controllability somewhat decreases with rp , track of its absolute value.
p
while the gain from θSOI is insignificant. A hypothetical
A simple linear cycle-to-cycle state-space model could then
explanation is that the stratification from the main injec-
be formulated accordingly
tion that ignites the charge but that the pilot injection
charge affects the combustion timing more when rp is
increased, the same trend was shown by (Hagesawa and xk+1 = Axk + Buk + vk ,
(15)
Yanigihara, 2003; Manente et al., 2009). yk = Cxk + ek .

366
IFAC AAC 2016
358
June 19-23, 2016. Norrköping, Sweden Gabriel Ingesson et al. / IFAC-PapersOnLine 49-11 (2016) 353–360
During the experimental campaign it was observed that
the static relation between inputs u and states x explained
most of the system behaviour and that there were none or
very little system dynamics, therefore, the assumption of
a static relation between u and x was made, this gave
A = I3x3 , and B which contains the gains from u to x
 
B11 B12
B= B21 B22 . (16)
0 1
From the experimental results in Figs. 8 and 9 it was found
that B11 = 1. The gain B12 = −3 was found by studying
Fig. 3. When advancing θ50 , τ m increases and θ50 is moved
closed to TDC which gives more violent combustion and
an increased dpmax , it was found that B21 was quite
dependent on the operating point as well as θ50 , however, a
constant value of −0.5 was chosen. The dpmax experiments
presented in Fig. 2 gave B22 = −25. The observation
matrix C = I3x3 since both states could be estimated
directly using the pressure sensor information. In order
to incorporate model uncertainty and measurement noise
in the model, zero-mean white-noise processes ωk and ek
were introduced with variance matrices Qv and Qe
   
Qv = E vkT vk , Qe = E eTk ek . (17)
Since the system (15) to be controlled has multiple in-
puts and outputs with interaction effects and constraints,
the model predictive control (MPC) framework was cho-
sen as a solution to the controller-design problem. With
active constraints, an MPC design obtains non-linear
properties that are not obtainable with standard linear-
controller designs such as LQR or PID. The principle
of MPC is briefly presented in the following section.
5.1 Model Predictive Control
MPC is a finite-horizon optimal-control principle where
T This section only considers the first two state equations of
the input sequence U = [uk , . . . , uk+Hc−1 ] from sample
k to Hc is computed iteratively at every new sample k
by solving an optimization problem, the first input u∗ (k)
of the optimal sequence U ∗ is actuated to the system
each sample, (Maciejowski, 2002). MPC, receding-horizon
control or dynamic programming as it has also been called,
has previously been applied to a wide range of problems
including chemical-process control (Maciejowski, 2002),
supply-chain management (Cho et al., 2003), finance
(Dawid, 2005) and internal-combustion engines (Widd et
al., 2009; Lewander et al., 2008).
For the control problem studied in this article the opti-
mization problem was defined as
minimize Z T QZ + U T RU + γE T E
U
subject to
X = AX + BU
xk−1 = x0
lUb ≤U ≤ ubU
X ≤ ubX + E
E ≥0
 T
where X = xk , . . . , xk+Hp −1 is the state vector over the
prediction horizon Hp ,
 T
Z = rk − xk , . . . , rk+Hp −1 − xk+Hp −1 (19)
is the set-point error. The variable
 T
E = k , . . . , k+Hp −1 (20)
is a cost-variable that penalizes violation of the state-
inequality constraint, Q, R and γ are cost weights while lUb ,
ubU and ubX are input and state boundaries. The matrices
A and B are built from the state-space model (15).
The optimization problem (18) is a quadratic program
(QP) and was solved in LabVIEW using the built-in QP
active-set solver. In order to speed up the computational
time, the solver was allowed to use the previous solution
and active set as initial guess. Early termination was also
used in order to allow the solver to finish faster, these are
well known methods for speeding up MPC execution times
(Wang and Boyd, 2010). Average computational times for
solving (18) was 1-5 ms for one cylinder. The computations
were done every two engine revolutions after the pressure
traces of the latest cycle for each cylinder was sampled.
The weighting matrices in the cost function in Eq. (18)
where manually tuned in order to obtain adequate closed-
loop response times subject to overshoot and actuator
cycle-to-cycle variation, the weighting matrices used in the
results section are given by
   
100 0 0
6000 0
Q= 0 0.01 0 , R = , γ = 1, (21)
0 8000
0 0 25
the horizons were chosen according to Hp = 16 and
Hc = 8. Note that the tracking cost of dpmax was kept
very low, this is because the controller objective is to keep
dpmax below the upper bound rather keeping it close to
zero.
5.2 Kalman Filter
Eq. (15), the system-matrices presented here are therefore
truncated accordingly in the following equations. In order
to estimate x from y, u and the model (15), a stationary
Kalman filter was used. The Kalman filter estimate x̂ is
updated recursively according to
x̂k+1 = Ax̂k + Buk + K(yk − Cxk ), (22)
where the Kalman gain is given by
K = AP C T (Qv + CP C T )−1 (23)
and P is the solution to the Riccati equation
P = AP AT + Qv − AP C T (Qe + CP C T )−1 CP AT . (24)
The covariance matrices
 used
  
25 0 10 0
Qe = , Qv = (25)
0 800 0 50
were chosen to get sufficient attenuation of measurement
noise, a more detailed description of the Kalman filter can
be found in Kay (1993).
(18)
6. RESULTS
With the controller design presented above, the controller
performance was tested experimentally during set-point
367
 IFAC AAC 2016
June 19-23, 2016. Norrköping, Sweden Gabriel Ingesson et al. / IFAC-PapersOnLine 49-11 (2016) 353–360 359

14 15 12 14

dpmax [bar/CAD]

dpmax [bar/CAD]
12 10 12
θ50 [CAD]

θ50 [CAD]
10 10 8 10
8 5 6 8
6 4 6
4 0 2 4
400 500 600 700 800 900 400 500 600 700 800 900 200 400 600 800 200 400 600 800
cycle cycle cycle cycle
−6 0.4 −10 0.4
[CAD]

[CAD]
−8
−10 0.2 −12 0.2

rp

rp
SOI

SOI
−12
θ

θ
−14 0 −14 0
400 500 600 700 800 900 400 500 600 700 800 900 200 400 600 800 200 400 600 800
cycle cycle cycle cycle
1.04 7.4
1.02 7.2 1.2 8
pIMEPg

pIMEPg
1 7
DOI

DOI
θtot

θtot
0.98 6.8 1 6
0.96 6.6
0.94 6.4 0.8 4
400 500 600 700 800 900 400 500 600 700 800 900 200 400 600 800 200 400 600 800
cycle cycle cycle cycle

Fig. 10. Input and output data during a sequence of θ50 set- Fig. 12. Input and output data during pIMEPg set-point
point changes. In the upper diagrams x̂ is presented changes.
in black together with y which is presented in grey.
20

dpmax [bar/CAD]
10

θ50 [CAD]
12 20
10
dpmax [bar/CAD]

8
θ50 [CAD]

10
10
8 6 0
500 600 700 800 900 500 600 700 800 900
cycle cycle
6 0 −8 0.4
100 200 300 400 500 600 100 200 300 400 500 600
cycle cycle
[CAD]

−10
−9 0.4
−12 0.2

rp
[CAD]

SOI

−10 −14
θ

0.2
rp

−16 0
SOI

500 600 700 800 900 500 600 700 800 900
−11 cycle cycle
θ

0 1600
100 200 300 400 500 600 100 200 300 400 500 600
cycle cycle
Nspeed [rpm]

1.15 7.5
1400
1.1 7
pIMEPg
DOI
θtot

1.05 6.5 1200

500 600 700 800 900
cycle
1 6
100 200 300 400 500 600 100 200 300 400 500 600
cycle cycle
Fig. 13. Input and output data during Nspeed set-point
Fig. 11. Input and output data during step changes in the changes.
dpmax upper bound.
System response during pIMEPg set-point changes are pre-
sented in Fig. 12, the response time for pIMEPg is around 20
changes in θ50 and pIMEPg , during changes in the dpmax tot
cycles, and as θDOI m
varies, θSOI is adjusted in order keep
upper bound and during changes in Nspeed . θ50 constant, rp is only doing minor adjustments in order
The load, pIMEPg , was changed by varying θDOItot
while to keep dpmax below the upper limit, the variation in rp
tot
keeping the common-rail pressure constant at 800 bar. indicates that dpmax is not so sensitive to changes in θDOI
During the load changes, pIMEPg and the common-rail at this operating point. Some input signal oscillations can
pressure level were controlled using manually tuned PI be observed around cycles 200 and 500.
controllers and pre-calibrated feedforward signals. The In Fig 13, system response to Nspeed transients are pre-
Nspeed changes were performed by changing the used m
sented, here rp have to increase and θSOI advance in order
engine-brake motor-speed set point. to fulfil the constraint and set-point level when Nspeed
Input and output data for one cylinder are presented increases. It can also be seen that the dpmax noise level
during a sequence of θ50 set-point changes in Fig. 10. In decreases with speed.
the upper diagrams x̂ is presented in black together with y
which is presented in grey. As θ50 is advanced, rp is forced 7. CONCLUSIONS
to increase in order to fulfil the specified dpmax constraint
of 8 bar/CAD, when θ50 is retarded, rp is decreased due
The results indicate that the designed controller was
to its absolute-value cost in order to maximize τ m . The
successful in maintaining an upper bound for dpmax using
response time for θ50 is in the range of 3 cycles while the
response time of the decrease in rp is around 20 cycles. rp while keeping θ50 at a predefined value, both in steady
state and during load and speed transients. There are some
In Fig. 11, input and output data are presented during significant cycle-to-cycle variation in the system inputs in
step changes in the dpmax upper bound where the response steady state for some of the experiments, see Figs. 13,
time of dpmax during a negative upper-bound step change 12, this variation could be decreased by increasing the R
is around 2 cycles while the response time of dpmax during matrix weights in Eq. (18) and by increasing Qe in Eq.
an positive upper-bound step change is around 10 cycles. (15).

368
IFAC AAC 2016
360
June 19-23, 2016. Norrköping, Sweden Gabriel Ingesson et al. / IFAC-PapersOnLine 49-11 (2016) 353–360
The pressure-rise rate dpmax is in this paper treated as a
noisy signal whose mean value is to be controlled below an
upper bound, the validity of this treatment could of course
be questioned since dpmax levels above the upper bound
will occur even if the mean dpmax level is kept below. A
more sophisticated design would be to choose the upper
bound based on the statistical distribution of dpmax and
in this way control the frequency or probability of the
dpmax boundary violations, similar to what was presented
by (Jones and Frey, 2015).
It is possible that comparable performance could be ob-
tained with a simpler controller structure for instance by
formulating the dpmax boundary problem as a set-point
problem as presented in (Ott et al., 2013), this would how-
ever demand more ad-hoc logic, and the framework would
not be as general and easily expandable if more states and
inputs were to be added to the controller framework.
REFERENCES
Cho, E., Thoney, K., Hodgson, T., and King, R. (2003).
Rolling horizon scheduling of multi-factory supply
chains. In Simulation Conference, 2003. Proceedings of
the 2003 Winter, volume 2, 1409–1416 vol.2.
Dawid, H. (2005). Long horizon versus short horizon plan-
ning in dynamic optimization problems with incomplete
information. Economic Theory, 25(3), 575–597.
Eichmeier, J., Wagner, U., and Spicher, U. (2012). Con-
trolling gasoline low temperature combustion by diesel
micro pilot injection. Journal of Engineering for Gas
Turbines and Power, 134(7), 072802.
Ekholm, K., Karlsson, M., Tunestl, P., Johansson, R.,
Johansson, B., and Strandh, P. (2008). Ethanol-diesel
fumigation in a multi-cylinder engine. SAE Int. J. Fuels
Lubr., 1, 26–36.
Eriksson, L. and Nielsen, L. (2014). Modeling and control
of engines and drivelines. John Wiley & Sons, Chich-
ester, West Sussex, United Kingdom.
Hasegawa, R. and Yanagihara, H. (2003). HCCI combus-
tion in DI diesel engine. 2003-01-0745, SAE Technical
Paper.
Jones, J.C.P. and Frey, J. (2015). Threshold optimization
and performance evaluation of a classical knock con-
troller. SAE Int. J. Engines, 8, 1021–1028.
Kay, S.M. (1993). Fundamentals of statistical signal
processing, Estimation theory, volume 1. Prentice Hall,
Upper Saddle River N.J.
Kiencke, U. and Nielsen, L. (2000). Automotive control
systems. for engine, driveline, and vehicle.
Kokjohn, S.L., Hanson, R.M., Splitter, D.A., and Re-
itz, R.D. (2009). Experiments and modeling of dual-
fuel HCCI and PCCI combustion using in-cylinder fuel
blending. 2009-01-2647, SAE Technical Paper.
Lewander, M., Johansson, B., Tunestål, P., Keeler, N.,
Milovanovic, N., and Bergstrand, P. (2008). Closed loop
control of a partially premixed combustion engine using
model predictive control strategies. In Proceedings of
AVEC, volume 8. Kobe, Japan.
Maciejowski, J.M. (2002). Predictive control: with con-
straints. Pearson education, Essex, England.
MacMillan, D., La Rocca, A., Shayler, P., Morris, T.,
Murphy, M., and Pegg, I. (2009). Investigating the
effects of multiple injections on stability at cold idle
369
for DI diesel engines. SAE International Journal of
Engines. 2009-01-0612.
Manente, V., Johansson, B., and Tunestal, P. (2009). Par-
tially premixed combustion at high load using gasoline
and ethanol, a comparison with diesel. 2009-01-0944,
SAE Technical Paper.
Manente, V., Johansson, B., and Tunestål, P. (2010a).
Characterization of partially premixed combustion with
ethanol: EGR sweeps, low and maximum loads. Jour-
nal of Engineering for Gas Turbines and Power-
Transactions of the ASME, 132(8).
Manente, V., Zander, C.G., Johansson, B., Tunestal, P.,
and Cannella, W. (2010b). An advanced internal com-
bustion engine concept for low emissions and high ef-
ficiency from idle to max load using gasoline partially
premixed combustion. 2010-01-2198.
Meyer, J. (2011). Calibration reduction in internal com-
bustion engine fueling control: modeling, estimation and
stability robustness. (Electronic Thesis or Disserta-
tion). Retrieved from https://etd.ohiolink.edu/, The
Ohio State University.
Musculus, M.P. (2006). Multiple simultaneous optical
diagnostic imaging of early-injection low-temperature
combustion in a heavy-duty diesel engine. 2006-01-0079,
SAE Technical Paper.
Osuka, I., Nishimura, M., Tanaka, Y., and Miyaki, M.
(1994). Benefits of new fuel injection system technology
on cold startability of diesel engines-improvement of
cold startability and white smoke reduction by means
of multi injection with common rail fuel system ECD-
U2. 940586, SAE Technical Paper.
Ott, T., Zurbriggen, F., Onder, C.H., and Guzzella, L.
(2013). Cylinder individual feedback control of combus-
tion in a dual fuel engine. In Advances in Automotive
Control, volume 7, 600–605. Tokyo, Japan.
Solsjö, R. (2014). Large Eddy Simulation of Turbulent
Combustion in PPC and Diesel Engines. Ph.D. thesis,
Lund University, Lund, Sweden. Department of Energy
Sciences.
Tanov, S., Collin, R., Johansson, B., and Tuner, M. (2014).
Combustion stratification with partially premixed com-
bustion, PPC, using NVO and split injection in a LD-
diesel engine. SAE International Journal of Engines,
7(2014-01-2677), 1911–1919.
Tsurushima, T., Kunishima, E., Asaumi, Y., Aoyagi, Y.,
and Enomoto, Y. (2002). The effect of knock on heat loss
in homogeneous charge compression ignition engines.
2002-01-0108, SAE Technical Paper.
Wang, Y. and Boyd, S. (2010). Fast model predictive
control using online optimization. Control Systems
Technology, IEEE Transactions on, 18(2), 267–278.
Widd, A., Ekholm, K., Tunestal, P., and Johansson, R.
(2009). Experimental evaluation of predictive combus-
tion phasing control in an hcci engine using fast thermal
management and VVA. In Control Applications, (CCA)
Intelligent Control, (ISIC), 2009 IEEE, 334–339. Saint
Petersburg.
Woshni, G. (1967). A universally applicable equation for
the instantaneous heat transfer coefficient in the internal
combustion engine. 670931, SAE Technical Paper.