International Journal of Automotive Technology, Vol. 17, No. 1, pp.

13−34 (2016) Copyright © 2016 KSAE/ 088−02

DOI 10.1007/s12239−016−0002−4 pISSN 1229−9138/ eISSN 1976−3832

MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE

AIRPATH AND COMBUSTION SYSTEMS

B. UNVER1), Y. KOYUNCUOGLU1), M. GOKASAN2)* and S. BOGOSYAN2, 3)

1)
Department of Powertrain Calibration & Control, Ford Otomotiv Sanayi A.S., Sancaktepe Research and Development
Cener, Akplnar Mah., Hasan Basri Cd., No. 2, 34885 Sancaktepe, I·stanbul, Turkey
2)
Department of Control & Automation Engineering Electrical-Electronics Engineering Faculty,
Istanbul Technical University, Maslak Istanbul 34469, Turkey
3)
Department of Electrical& Computer Engineering, College of Engineering and Mines University of Alaska
Fairbanks, Fairbanks AK 99775, USA

(Received 8 May 2014; Revised 2 December 2014; Accepted 16 February 2015)

ABSTRACT−The ultimate aim of this study is the development of an engine modeling approach that would facilitate the
design of model-based control techniques for diesel engines. This will allow for the development of more generalized,
modular control strategies for different engine types and sizes as opposed to the commonly practiced map-based engine
control strategies that depend on maps and feedforward control and require lengthy modifications every time a change is
made. Also, most engine modeling studies focus on either airpath or combustion systems, treating these models and their
validation individually, and not as an integrated system as is actually the case. To address the need for more realistic models
suitable for model-based control design, this study develops a combined airpath and combustion model for the engine, using
analytical models wherever possible and derives a model with appropriate control inputs and outputs that could be used in a
control scheme. The inclusion of the actuator dynamics of the Exhaust gas recirculation (EGR), variable geometry turbine
(VGT), and Throttle (THR) valves in the airpath model and the consideration of nonlinearities in the combustion model allow
for the development of a more thorough engine model, as well as the validation of subsystems and the whole integrated engine
model using a complete World Harmonized Transient Cycle (WHTC). This test cycle finds limited use due to its challenging
transients, and yet, is the demanded test cycle for emission regulations nowadays. These are unique aspects of this modeling
study, the results of which indicate that the developed engine model could be used in control design and hardware-in-the-loop
simulation (HILS) based engine control prototyping.

KEY WORDS : Control based diesel engine model, Airpath model, Combustion model, Parameter estimation, Model
validation, EGR valve, VGT valve, THR valve, WHTC

NOMENCLATURE qHV : fuel heating value

qin : specific heat energy
OFR : stoichiometric ratio between oxygen and fuel R : ideal gas constant
Amax : maximum effective area of EGR valve rc : cylinder compression ratio
Apipe : effective pipe area causing temperature drop Rt : blade radius
cp : air constant pressure heat capacity T : temperature
cv : constant volume air heat capacity u : model inputs as valve position, injection timings
H : pressure drop coefficients & quantities
htot : total heat transfer coefficient uinj_qty : total injected fuel quantity
J : inertia V : volume
K : ratio of heat capacities at constant volume and W : mass flow rate
pressure xcv : consumed fuel ratio during constant volume com-
kJt : turbocharger friction coefficient bustion
ncyl : number of engine cylinders XO : oxygen concentration
ne : engine speed xp : pressure ratio between after and before combus-
P : mechanical power tion gases
p : pressure xr : cylinder remaining gas ratio
xv : volume ratio of the after and before combustion
*Corresponding author. e-mail: [email protected] gases

13
14 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

η : efficiency In the literature, there are numerous studies related to

λO : oxygen fuel ratio in the cylinders airpath and combustion modeling. The most detailed
Π : pressure ratio models (like GT Power, Wave and AMESim) for airpath
ω : rotational speed modeling are based on design geometry. Detailed (1D) GT
ψ : energy transfer coefficient Power models are mainly used (Wu et al., 2011) during
Ф : volumetric flow coefficient engine design phase and run much slower than real time.
To increase simulation speed and obtain real time capable
SUBSCRIPTS model, detailed models are converted to mean value
egr (EGR) : exhaust gas recirculation valve models and trained with test data that are obtained from
thr (THR) : throttle valve detailed models. Conversion from detailed models to real-
opt : optimum time models is time consuming, and model errors inevitab-
eg : exhaust gas ly increase after conversion operation. Converted models
T : turbine are imported to MATLAB/Simulink as a black box s-
C : compressor function. Only after this operation, these models can be run
em : exhaust manifold at hardware in the loop systems (with additional hardware
im : intake manifold and software requirements, e.g. Advanced Support Pack-
ic : intercooler age for commonly used dSPACE based HIL systems) to
cool : coolant design model-based controllers.
egrc : EGR cooler Similar to GT Power models, detailed (1D) AMESim
e : engine models (Ahmed, 2013) can be utilized to determine engine
amb : ambient behavior provided the system parameters and initial condi-
d : displacement tions are known accurately. While AMESim’s physical
ei : engine cylinder in models can also be used for controller design, a known
eo : engine cylinder out limitation of the system is the lost accuracy when used in
f : fuel combination with Matlab/Simulink, which is currently the
m : mechanical most commonly used software in controller design. AMESim
max : maximum has two interfaces with Simulink (Bonnet, 2007), however
tm : thermodynamic the co-simulation interface faces state transfer issues when
vol : volumetric working with Simulink, and the regular interface gives rise
crit : critical to reduced accuracy. Hence, currently, AMESim poses
rail : rail some problems for Simulink based Hardware-in-the-Loop
soi : start of injection (HIL) simulation implementations, as is the case with our
vgt : variable geometry turbine engine control research that will use the dSpace’s HIL
inj : injection system. AMESim models are also known to need some
modifications to increase simulation speed in running real
1. INTRODUCTION time HIL systems (Patil et al., 2012).
In this study, we aim to develop an engine model that has
The ubiquitous use of diesel engines in a variety of trans- an analytical form for the most part, could be used in the
portation and energy systems is mostly due to their low fuel design of model-based controllers and would be repro-
consumption. However, with lower emission limits and ducible for different engines sizes and types that have
World Harmonized Transient Cycle (WHTC) imposed by similar configurations. In the related literature review, the
emission legislations, the need for more advanced control studies in (Skogtjarn, 2002; Jung, 2003; Gambarotta et al.,
systems for diesel engines has emerged. Such restrictions 2011; Stamati et al., 2012; Lee et al., 2013) on airpath
also give rise to a higher number of sensors and actuators modeling, developed mean value models focused on only a
used. The rising number of control variables increases the limited number of states/outputs. Among those, Gambarotta
number of tests required for optimum engine calibration, et al. deal with airpath modeling, but do not deal with
which traditionally has been conducted by map based emission modeling; hence the effect of the airpath model
approaches and feedforward terms. The increasing per- on the emission model can not be investigated. Those
formance demands and associated rise in the number of studies do not perform a detailed airpath modeling, nor do
sensor/actuators make map based control very costly, both they validate transient performance under aggressive cycles,
in terms of time and economy. As a result, model based such as WHTC.
controller and calibration development has been gaining Studies taking a detailed physical based modeling and
increasing importance in automotive industry. To develop model tuning approach are quite rare (Wahlström, 2009;
model based controllers, it is necessary to build analytical Guzzella and Onder, 2010; Benz, 2010). However, those
models wherever possible for the engine airpath and studies make some simplifications to the highly complex
combustion model and develop identification schemes. nonlinear dynamics of the system; e.g. they either neglect
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 15

the throttle valve or assume it fully open at all operating map structure. This map depends on the current engine
points. The current control structures of ECUs in practice calibration and the system’s behavior under different
involve the operation of the throttle and EGR valve together; engine calibrations is not tested in these studies.
hence for a more realistic model, the throttle valve should Among other combustion studies, one can mention
be taken into consideration. To the authors’ best know- (Amsden et al., 1985; Hiroyasu, 1985; Johnson, 2009;
ledge, (Wahlstrom and Eriksson, 2010) is the only study Kirchen and Boulouchos, 2009; Kirchen et al., 2009;
that includes the throttle (THR) valve equations in the Maiboom et al., 2008; Stadlbauer et al., 2012; Zhang et al.,
detailed whole model, and takes the intercooler pressure 2007; Zhao, 2010) Most of these studies require detailed
into account among the model states. Development of the measurements, and are faced with difficulties in the
model also requires a parameter estimation process in determination of the system parameters; hence, some have
place; e.g. the study in (Stamati et al., 2012) focuses on required the design of special experiments for this.
only two airpath dynamics and designs an experiment for There are also statistical combustion models, which
identification. Studies in (Wahlström, 2009; Wahlstrom and make parameter identification somewhat simpler, but
Eriksson, 2010, 2011; Guzzella and Onder, 2010; Eriksson require a large amount of test data. Such models often yield
et al., 2009; Skogtjarn, 2002; Unver, 2013) generally use a satisfactory results for regions covered by the test data, but
least squares solver for parameter/model identification. not as much for regions covered via extrapolation. Physical/
The engine combustion model, on the other hand, has a chemical models do not require as much data as statistical
rather complicated dynamics as it combines different models and perform better than statistical models for
effects resulting from the interaction of fluid mechanics, extrapolated regions; however, they pose problems in para-
thermodynamics, and chemical processes. Hence, studies meter identification and real-time operation. Another issue
on engine combustion often take on detailed modeling with these model structures in the literature is that these
tasks and use special software developed for this purpose. models are based on a single injection strategy, while ECU
In the literature, three different approaches are adopted for calibration in the automotive industry mostly involves a
the modeling of the combustion system; i.e. empirical multi injection strategy. Additionally, a common trend in
(Hirsch et al., 2010; Hirsch et al., 2008; Zweigel et al., most combustion modeling studies is to perform the model
2013; Brahma et al., 2009), detailed phenomenological validation with the use of measured inputs, and not with the
(Hiroyasu, 1985), three-dimensional computational fluid use of physical variables generated by the airpath model as
dynamics CFD (Computational fluid dynamics) models is actually the case. This modeling and validation study
(Amsden et al., 1985). Generally crank angle based detailed addresses these issues by considering a multi injection
models in the former group of studies are computationally strategy, and proposes to integrate the airpath and com-
inefficient and use in-cylinder information, however, bustion models in cascade for a more realistic validation of
measurement/calculation of in-cylinder signals such as the models. Many studies in the literature focus on either
flame temperature is challenging. Additionally, identi- airpath or combustion modeling, hence studies taking this
fication of these detailed models is difficult. Therefore, cascaded validation approach are very few, if any.
these models are not suitable for transient simulations and As mentioned previously, airpath models developed for
controller design. Empirical combustion models are more control design are also very rare in the literature, with
suitable for this purpose. However, empirical model identi- (Wahlström, 2009; Wahlström and Eriksson, 2010, 2011)
fication generally requires a large number of measurement presenting the most convenient structures to this aim, and
data and offers poor extrapolation capability outside of the hence, are taken as bases for this study. Those studies do
identification data region. In order to identify the system not deal with the combustion model. However, for a
using an empirical model, transient modeling tests are used realistic engine model validation, the airpath and combus-
in (Brahma et al., 2009). Similar to most combustion tion models should be integrated, and taken into account as
modeling studies, (Zweigel et al., 2013) uses a simplified a whole; moreover, the combustion model inputs should
approach based on a single injection strategy for the involve airpath model output signals.
development of the model, while the commonly used In summary, the contributions of our study to the exist-
approach in the automotive industry is a multi-injection ing airpath model studies could be listed as follows:
strategy. (Arsie et al., 2004, 2007, 2012) takea multi injec- − Development of an engine model that integrates the
tion strategy into account in their combustion modeling airpath and combustion model,
studies. However, they do not use critical airpath states as − consideration of the valve (EGR, VGT and THR) actuator
inputs to the emission model, while these states are known dynamics in the airpath model,
to effect emissions. Additionally, none of these studies − while most studies use the least squares method for
validate the model by using aggressive transient cycles, parameter estimation, our study utilizes nonlinear estimation
such as WHTC. toolboxes of MATLAB and MATHEMATICA for a
As an alternative to existing models, mixed physics and more realistic performance, considering the highly non-
empirical model structures are presented in (Benz et al., linear nature of the system. The parameter tuning is also
2010; Aspriona et al., 2013), using a set-point relative base performed using a high volume of data.
16 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

− validation of the cascaded airpath and combustion models

as a whole,
− validation of the “whole model and variables using the
challenging WHTC for an extended duration; for model
validation, most of the above studies do not use this
cycle, which includes challenging transients, is more
realistic, and also demanded for new emission regulations.
Most studies present model validation results using
simpler cycles or short durations of more complex cycles.
The study in (Wahlstrom and Eriksson, 2010) uses the
WHTC for the validation of 5 model variables only.
− derivation of the system model with control inputs and
outputs, and with a structure that is convenient for use in Figure 1. Airpath system diagram.
a control loop, and for control design.
This paper is organized as follows. After the introduction
for all 3 valve positions and their angular speeds (ωegr, ωvgt,
in Section 1, airpath modeling and validation are presented
ωthr). Below are the state and control input vectors:
in Section 2. In this section, firstly model equations are
presented; next, the parameter estimation and validation x = ( pic pim pem XOim XOem ek θegr ω egr θvgt ω vgt θthr ω thr )T
procedure is discussed and implemented for the submodels
u = ( uinj_qty uegr uvgt uthr)T (1)
of the airpath model. Then the validation results of the
complete airpath model are given. In Section 3 the com- where uinj_qty, uvgt , uegr and uthr correspond respectively to
bustion model structure and validation results are presented injected fuel quantity, VGT blade position, EGR valve
with the conclusion is presented in Section 4. Next the position, and THR valve positions. Definitions of the
combustion model structure and validation and finally the notations can be found on the Nomenclature part and the
validation results of the integrated airpath and combustion structure of the model can be seen in Figure 1.
model are discussed. The conclusion and future directions The airpath model equations can be given as follows
are given in Section 5. based on Figure 1

2. DEVELOPMENT OF AIRPATH MODEL

The model development process in this study not only aims

a configuration suitable for model based control, but also a
structure that captures the system dynamics with as few
parameters as possible. Among numerous modeling studies
(2)
in the literature, those yielding a structure adequate for
control are very few and can be cited as (Wahlström, 2009;
Wahlstrom and Eriksson, 2010), In the derived model
configuration below, different features of these models
have been modified and integrated to represent the engine
system in the best possible way, further developing a
structure that could be used in the development of model-
based controllers. In this process, ideal gas constant of air In the equations above:
Rair and ideal gas constant of exhaust gas Reg are assumed as
Rair Rair Reg
not changing with temperature variations. kic = ------- ; k = ------- ; k = ------- (3)
Vic im Vim em Vem
2.1. Airpath Model Equations where, Rair and Vic are ideal gas constant and intercooler
In order to obtain a model convenient for the development volume. Reg, Vem, are respectively the corresponding ideal
of model based control scheme, the dynamics of the gas constant of exhaust gas, exhaust manifold volume and
actuators in the airpath should be taken into consideration. Vim is input manifold volume.
The airpath model consists 12 states; namely, the inter-
KtkKbk n2ak- Kak rak 2
1- KatkEnak
cooler, intake and exhaust manifold pressures (pic, pim, pem), ak1 = ------------------- ; ak2 = ------------
- ; dk = ----- T ( t ); bk = ------------------ (4)
Rak Jak Jak Jak Lk RakJak
oxygen concentrations at intake and exhaust manifolds
(XOim, XOem), turbine blade kinetic energy (e), EGR valve The valve mechanisms are often controlled by a DC motor.
position, VGT blade position and THR valve positions The k sub index depends on the egr, vgt or thr actuators and
(θegr, θvgt, θthr). 0 and θmax indicate fully closed and fully open should be replaced by those actuator abbreviations,
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 17

respectively. As for the Rak, Kbk, Ktk motor parameters, they

are the related motor resistance, emf constant, and torque are constants. As for kc10, kc11, kc20, kc21, kc30 kc31 coefficients,
constant, respectively. E indicates the constant DC power these will be determined via Wc values obtained through
supply; as for Kak, Jak, nak and rak, these are parameters engine dynamometer tests. The estimation of the kci
related to the mechanical components of the valves, and are parameters is critical for the compressor mass flow rate
respectively, the valve return spring constant, total inertia, calculations. The condition of ψc > kc1/kc3 gives rise to the
gear turn ration, and eccentric drive diameter (performing compressor being in the unstable region. Also, to avoid
the conversion of rotary motion to translation). incorrect calculations at higher compressor speeds, it is
Using the relationship between the kinetic energy and required to fulfil the following condition (Benz, 2010).
turbine angular velocity, and organizing the equation of
kc1 + kc2. kc3 < 0 (10)
motion with respect to kinetic energy, a simpler relation-
ship is obtained as below, also avoiding the complications The fuel flow mass rate as a function of engine speed can
due to division by ωt: be given as below:
Defining the kinetic energy of the turbocompressor as,
10–6
Wf = --------- uf ne ncyl , Wf = kf ne uf (11)
1 120
e = --- Jt ω 2t ; Jt : inertia of turbo system (5)
2 –6
10 -
where kf = -------- n and is constant and ne is engine speed.
The equation of motion of the turbocompressor can be 120 cyl
Wt turbine mass flow rate is calculated as follows
derived as,
(Heywood, 1988).
pem-
(6) Wt = kt ---------- fΠ ( Πt )fvgt( θvgt ) (12)
Tem t
Here,
Wc and Wthr in (2) are mass flow rates of the compressor and
throttle valve, respectively, while Wegr and Wt are the EGR
mass flow rate and turbine mass flow rate, respectively, and
Wf is the fuel mass flow rate:
(13)

In above equation the Avgtmax term shows the maximum

turbine area faced by exhaust gas flowing through the
turbine. Measurements prove that the corrected mass flow
rate depends on the Πt and θvgt. When the pressure ratio
through the turbine is decreased, because the flowing gas
(7) velocity reaches the sonic velocity, turbine mass flow starts
to choke. The below equation represents the situation with
only one parameter.
Substituting these terms in Wt ,

(14)

Defining fc(.) in (7), the relationship can be expressed in a

significantly simplified form as in (8), which will also be (15)
beneficial for control design.
Wc = kc fc ( pic , e ) (8)
Similar to (7), the turbine mass flow rate can be given in
where the closed form below:
Rc - , kWc = 2c pair T amb Wt = kt ft ( pem , pt , Tem , θvgt ) (16)
ka = ----------------------------- --------------------
-,
Kair Rair Tamb R2c
where kt is calculated at (13) and cf1, cf2, cvgt1, cvgt2 will be
pamb πR-3c ( k – 1 )/k estimated via Wt values measured through the dynamo-
kc = ----------------- , k = ( pic/pamb ) air air (9)
Rair Tamb PR
18 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

meter tests. be calculated as below (Wahlstrom and Eriksson, 2010).

With a compressible flow assumption,
(22)

where
(17)
(23)

EGR mass flow rate : Substituting (23) into (22),

(18)
(24)
Aegr,max is the maximum effective area of the EGR valve.
Also, feg estimates the effective area of the EGR valve as a
function of EGR valve position (θegr). The flow function is
symbolized with ψegr and feg given with the following which, in closed form could be expressed as,
equation. Wthr = kthr fthr ( pic , pim , Tim , θthr ) (25)
Here, kthr is calculated at (23); Kair, bthr1, bthr2, athr1 and athr2
(19) will be estimated through measurements taken via the
dynamometer tests.
The cylinder mass flow rate can be defined as the total
mass flow rate flowing into the cylinders and hvol gives the
Finally, by defining the feg effective area, all parameters to
volumetric efficiency at (26).
calculate the EGR mass flow are fully determined.
(26)

where

(27)

Given in closed form,

kei is calculated at (27); cvol1, cvol2, cvol3 can be estimated via
Wegr = kegr fegr ( pic , pim , pem , Tem , θegr ) (20)
measurements. Hence, Wei can be given in closed form as
Here, kegr is calculated at (19); Keg, begr, aegr1 and aegr2 below:
coefficients can be determined via measurements through
(28)
the dynamometer tests.
The throttle mass flow is modeled as a compressible The turbine efficiency is used to model the power distri-
flow at constant entropy. The critical pressure ratio when bution of the turbine. This efficiency calculation is based
the throttle flow reaches the speed of sound is given with on the temperature difference between the turbine inlet and
the equation below. outlet. However, heat losses may result in inaccurate
turbine efficiency calculations. To overcome this problem,
another efficiency term, ηtm is widely used in the literature
(Heywood, 1988). This efficiency term is based on the
steady state power balance (Benz, 2010).
(21)

(29)

Using (21) and assuming that the there is no reverse flow In (29) above, ηtm,max, vb_opt and vb denote maximum turbine
through the throttle valve, the pressure ratio between the efficiency, turbine blade speed at maximum turbine
inlet and outlet of the throttle valve can be calculated. After efficiency and turbine blade speed, respectively. In the
defining the pressure ratio, the throttle mass flow rate can literature, ηtm,max and vb_opt are obtained as a function of VGT
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 19

position (θvgt) as below:

(36)
And finally, assuming constant values for Tamb and pamb and
by substituting Wc in (8) into (36),
(30)

(37)
The turbine pressure ratio, Πt, in (30) can be given as Here kPc = kc cpairTamb and is constant. a1, a2, a3 parameters
Πt = pt /pem . In the equation above, Rt corresponds to the can be determined via measurements again.
turbine blade radius. Furthermore, cpeg and cveg are Pc can be given in closed form as below:
respectively indicating the heat capacities of the exhaust
Pc = kPc fPc ( pic , e ) (38)
gas at constant pressure and volume, and Keg is the ratio of
these heat capacities cpeg /cveg . Finally, to re-organize the exhaust manifold oxygen con-
( Keg – 1 )/Keg centration, XOem in (2), the relationship below will be used.
Pt ηm = ηtm Wt cpeg Tem ( 1 – Πt ), (31)
Wei XOim - Wei XOim −Wf ( O/F )-s
λO = ---------------------- , X = ------------------------------------------- (39)
Substituting (29), (30) into (31); Wf ( O/F )s Oe Weo
Here, λO repsents the oxygen/fuel ratio in the cylinders,
while (O/F)s corresponds to stoichiometric ratio between
oxygen and fuel.
In diesel engines, to decrease the soot emission, the
condition of λO > 1 should be fulfilled. The developed
model takes this condition into consideration, and the
oxygen concentration flowing out the cylinders is calculated
as unburned oxygen ratio. XOe denotes the oxygen concent-
ration of the air going out from the cylinders to exhaust
manifold oxygen concentration. The XOeWeo product can be
(32)
re-organize using (39) as below:
Substituting Wt in (16) into (32) and expressing it in closed
XOe Xoe = Wei XOim −Wf ( O/F )s
form:
Pt ηm = fm ( pt , pem , Tem , θvgt )kt ft ( pem , pt , Tem , θvgt ) WeiXOim-
XOe Xoe = Wei XOim − ---------------- (40)
λO
= kt fPt( pt , pem , Tem, θvgt ) (33) ·
By substituting (39) and (40) into XOem( t ) in (2);
where kt is calculated at (13); b0, b1, b2, b3, n0, n1, n2, n3 can
be estimated through measurements.
While calculating the Pc compressor power, ηc com- (41)
pressor efficiency is used. The compressor pressure ratio
between the inlet and outlet is derived with the equation
below:
and using Wei in (28) and Wf in (11) in the above, we obtain,
pic-
Πc = --------
pamb
2c π (42)
ηc =ηcmax − [a1( Wc – Wcopt )2 + a2( Πc – 1 )
c
+ 2a3 ( Wc – Wcopt ) ( Πc – 1 ) π ] (34)
The compressor power: Tim and Tem values will be derived below:
The exhaust manifold temperature model includes the sub
Wc cpair Tamb- (Kair – 1)/Kair
Pc = ------------------------ ( Πc – 1) (35) models of exhaust pipe heat loss model and exhaust gas
ηc
temperature model flowing out of the cylinders.
Substituting (34) in (35), Pc is obtained as The temperature drop at the exhaust manifold can be
20 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

calculated as a function of mass flow rate flowing out the

cylinders. (48)

Derivation of Tem:
The exhaust manifold temperature model includes the sub The above cylinder out temperature model can not give
models of exhaust pipe heat loss model and exhaust gas directly the exhaust manifold temperature due to heat
temperature model flowing out the cylinders. Te cylinder losses on the exhaust pipe. Temperature drop at exhaust
out temperature is calculated based on the Seiliger cycle manifold can be calculated as a function of mass flow rate
that is given in detail in (Heywood, 1988; Wahlström, flowing out the cylinders.
2009). htot ⋅ π ⋅ Apipe
--------------------------------
W eo c pe
Tem = Tamb + ( Te – Tamb )e (49)

Finally, in (49), Tamb, htot and Apipe denote the ambient air
(43)
temperature, total heat transfer coefficient and effective
pipe area causing the temperature drop.
To calculate the turbine out temperature, the following
turbine efficiency based equation is used.
ηsc is a compensation constant for cycles that are not ideal
( Keg – 1 )/Keg
and xcv indicates the consumed fuel quantity during the Tt = Tem −Tem ( 1 – Πt )ηtm (50)
constant volume combustion. The below equation denotes
the pressure ratio between the inlets and outlets of the By using equation (32) with the equation above,
cylinders. Πe = pem/pim Tt = Tem −ftm ( pt , pem , Tem, θvgt )/cpeg (51)
Using the intermediate variables and definitions of the
Seigler cycle, the cylinder out temperature can be calculated (degistirildi)
with numerical iterations. For each kth iteration below The intake manifold temperature Tim is assumed to be equal
equations are calculated so as to starting from xr,0 ve T1,0 to intercooler outlet temperature (Wahlstrom and Eriksson,
initial values. According to Seiliger cycle pressure ratio 2010), hence
between post-combustion and pre-combustion xp is given Tim = max ( Tcool,ic , Tcool,ic ηic + ( 1 – ηic )Tc ) (52)
with following equation.
In (52) Tc, Tcool,ic corresponds to the compressor outlet
(44) temperature and intercooler coolant temperature. In this
study, intercooler efficiency, ηic is calculated as a function
of Tc and Wc (Unver, 2013):
In (43), cpair and cvair symbolize the constant pressure and
constant volume heat capacities while Kair is the ratio of ηic = nic0 + nic1Wc + nic2Tc + nic3W2c + nic4T2c (53)
these heat capacities cpair/cvair.. Furthermore, the specific The compressor outlet temperature is calculated as a
heat energy, qin in (43) is defined as; function of compressor efficiency (34) and given with the
equation below (Heywood, 1988).
(45) (K – 1 )/K
Tamb( Πc air air – 1 -)
Tc = Tamb + --------------------------------------------- (54)
ηc
After suction stroke and fuel gas mixture, the temperature
at the intake valve is calculated as below (Wahlström, The turbine out pressure calculation is also added to the
2009), model to avoid re-parameterisation of all turbo-compressor
model in case of different exhaust system configurations.
(46) Thus, in case of different exhaust system configurations, it
The remaining gas ratio xr in (46) is defined as; is only required to update parameters of this sub model.
Heg0 and Heg1 indicate the pressure drop coefficient and the
turbine out pressure pt, is calculated with the following
(47)
equation.
( Heg0 + Heg1 Wt )W2t T-t
At the Seiliger cycle, the volume ratios of post-combustion pt = pamb + --------------------------------------------- (55)
pt
and pre-combustion gasses are defined in following equation.
Finally, by substituting the equations in (2), into (8, 16, 21,
25, 28, 33, 38, 43) and by using the re-organized (6) and
(42) relationships, the airpath model is derived in the
following control-oriented form:
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 21

(56)

2.2. Airpath Model Equations optimizations achieved on the MATLAB environment.

To complete the modeling process, model parameters must Also, hybrid techniques including genetic algorithms or
be determined, and model validation must be performed in different initial points are utilized to prevent optimization
the next step. For these processes, measurement data function from local minimums around the optimization
collected from an engine dynamometer at Ford OTOSAN initialization points.
company’s premises has been used. After connecting the It should be noted that the input-output measurements
engine to the dynamometer, the pedal is span through 0 ~ used for the modeling process are taken in steady state
100 for each speed value and upon reaching the steady- from a Fulel Loop test. However, the cycle used for model
state, data is recorded. This test is known as the Airpath validation should also include transient state. In this study
Modeling Test (Fuel Loop/Engine Mapping Test). The for the validation of the airpath, the WHTC test is used,
derived torque-speed characteristic is presented in red in while the validation of the combustion model is performed
Figure 2. While performing the pedal-speed test, measure- with the Design of Experiment (DOE) test. In Figure 2 , red
ments of other physical variables, such as, pressure, mass bubbles represent the dynamometer test data used for the
flow, temperature, emission values etc. are also recorded to airpath modeling/tuning process (via fuel loop test). The
be used in the modeling process. green points demonstrate the dynamometer test data used
To determine the parameters for each submodel, the for combustion modeling/tuning (DOE), while the blue
associated input and output variables for each submodel are points show the dynamometer test data used for airpath,
determined from the collected data. With the use of these combustion, integrated airpath+combustion validation/pre-
values and consideration of the model linearities and diction (WHTC). This figure also indicates data collected
nonlinearities, appropriate parameter estimation algorithms for the modeling process (depicted in red) over a wider
have been implemented on Matlab and Mathematica. operation region during steady-state from Fuel Loop test,
During parameter estimation, the cost functions are defined together with the more effectively used data over a
as squared differences of the measured and modeled vari- narrower operation region during WHTC and DOE tests.
ables. Thus, the parameter estimation problem is treated as For the evaluation of developed models, two sets of
an optimization problem aiming to minimize this cost results have been presented throughout the study: results
function for the estimated parameters. for Modeling Test and results for the Validation Tests. For
Physically meaningful parameters, like turbine blade the former group of results, the actual measurement values
radius, parameter ranges are utilized as a constraint for the for a given model output (measured from the steady-state
22 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

Figure 3. Modeling and validation results of compressor

Figure 2. Data collected for airpath modeling (Fuel Loop
mass flow rate.
Test), combustion modeling (DOE Test) and validation
(WHTC Test) processes.
figure is for modeling and the lower one is for validation.
test), and the estimated value of the same model output The mean absolute relative errors are %7.54 and %5.70 for
(based on the substitution of associated measured input modeling and validation test. The results are presented in
values into the derived model) are plotted together for Table 1.
comparison; as for the validation results in the latter group,
these are obtained by plotting the actual and estimated Parameter Estimation and Validation of Compressor effici-
values of the given model variable during a WHTC or ency and Compressor Outlet Temperature
DOE run for the airpath and combustion mode, respectively. The experimental values of compressor efficiency is
The WHTC validation has been performed for each calculated by using the compressor efficiency maps com-
submodel, for the complete airpath model, and for the ing from the compressor supplier. In the supplier data, the
cascaded airpath and combustion model. efficiency values are given as dependent on compressor
For all modeling processes in this study, the relative pressure ratio (pic/pamb) and compressor mass flow rate (Wc).
errors are calculated with the below equations for steady By using the griddata function of MATLAB, the supplier
state and transient tests: data converted to a map accepting the pressure ratio and
compressor mass flow rate as its inputs. The required
ymeas,ss ( i )−ymodel,ss ( i )-
REss ( i ) = -------------------------------------------- inputs of the map is fed from the engine fuel loop test and
1- N
--- Σ y (i) the experimental data for compressor efficiency is estimated
N i = 1 meas,ss
easily. The efficiency model inputs and outputs are given
ymeas,tra ( j )−ymodel,tra ( j -)
REtra( i ) = ----------------------------------------------- (57) below.
1- N
--- Σ y (j)
N j = 1 meas,ss

2.2.1. Submodel tuning

Parameter Estimation and Validation of Compressor Mass
Flow Model
Inputs : pic, pamb, Tamb, ωt
Outputs : Wc
Parameters : Rc, kc11, kc10, kc21, kc20, kc31, kc30
The parameters are predicted using the NMinimize func-
tion of the Mathematica program. The compressor mass
flow rate is one of the most important variables for the
airpath modeling. As it is explained in the modelling section
above, ki terms are tuned to ensure that the compressor is
working in its stable region.
The above figure shows the modeling and validation Figure 4. Modeling and validation results of compressor
results for compressor mass flow rate model. The upper efficiency.
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 23

Inputs : pic, pamb, Wc

Outputs : ηc
Parameters : cpair, Wcopt, πcopt, a1, a2, a3, ηcmax
Modeling and validation results are given in the figure
below. The mean absolute relative errors for modeling and
validation are %1.54 and %2.86.
Another way to prove the validity of the compressor
efficiency model is by making a comparison of the
measured compressor outlet temperature and calculated
temperature based on the efficiency. In temperature
calculations, estimated parameters for compressor effici-
ency are used as they are. The inputs and outputs of the
model are fed from the engine mapping to estimate model
parameters given below.
Inputs : pic, pamb, Wc, Tamb
Outputs : Tc Figure 6. Modeling and validation results of turbine mass
flow rate.
Additionally, a first order filter is used at the output of the
temperature model as rapid changes are observed in
errors for modeling and validation are %3.17 and %5.98.
validation results because of the highly transient charac-
teristic of the WHTC test. To predict the filter coefficients,
Parameter Estimation and Validation of Turbine power
the tfest function of the MATLAB “System Identification
and outlet temperature
Toolbox“ is used. Finally, the mean absolute relative errors
During parameter estimation, it is assumed that the turbine
for modeling and validation are %1.94 and %1.84.
power is equal to compressor power for steady state
running condition. Compressor power is calculated by
Parameter Estimation and Validation of Turbine mass flow
using the parameters that are estimated for compressor
rate
efficiency. The inputs and outputs of the model is fed from
The inputs of the turbine mass flow rate model are fed from
the engine mapping test and the following parameters are
the fuel loop test and below parameters are estimated.
estimated.
Inputs : Pamb, pem, Tem, θvgt
Inputs : Tamb, pamb, pic,Wc, Tem, pem, pt, Wt, ωt, θvgt
Outputs : Wt
Outputs : Pt
Parameters : Kt, cvgt1, cvgt2, cf1, cf2, Avgtmax
Parameters : Rt, b0, b1, b2, b3, n0, n1, n2, n3
The NMinimize function of the Mathematica is used for
The NMinimize function of the Mathematica is used for
parameter estimation. During prediction of the Avgt,max, a
parameter estimation and the mean absolute relative error
plausible range is given as a constraint to the optimization
for modeling is %3.94.
function. As a consequence, the mean absolute relative
The error calculation for a transient WHTC test is not

Figure 5. Modeling and validation results of compressor

outlet temperature. Figure 7. Modeling and validation results of turbine power.
24 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

Figure 8. Modeling test results of turbine outlet temperature. Figure 9.Turbocharger model modeling test results of
turbine speed.
meaningful due to assumption of power equality at steady
state operation. However, to observe the model response, Compressor and turbine parameters are not estimated again
validation result is given in the Figure 7. while estimating the Jt and kJt. After estimating the
Similar to the compressor outlet temperature, turbine parameters, it is observed that the kJt has a minimum effect
outlet temperature is calculated by using the parameters on the results and it is assigned as 0.
estimated during the turbine power modeling. It is seen that The inputs and outputs of the model are taken from the
the modeled turbine outlet temperature is always more than engine mapping test recorded as transient. The reason for
the measurements. This observation is directly related with using transient data is that the model includes integrators,
the measurement position on the engine. Due to instrumenta- thus a transient engine mapping test is used for estimation
tion constraints, temperature measurements are collected at of the following parameters.
a slight distance from the turbine outlet. However, even at
Inputs : Pic, Tamb, pamb, θvgt, Tem, pem, pt
this small distance, piping heat loses are becoming effec-
Outputs : Wc, Wt, ωt, Tc, Tt
tive on the measurements due to high temperatures. As a
Parameters : Jt, kjt
last retouch, model results are multiplied with a constant
and then filtered with a first order filter due to the same After parameter estimation, a steady state test is applied on
reasons mentioned in the compressor outlet temperature the coupled turbocharger model with the inputs coming
parameter tuning. Model inputs and outputs are fed from from the measurements. During simulations, it is observed
the engine mapping test and turbine outlet temperatures are that 10 second is a sufficient time for all testing points to
calculated. reach steady state. Furthermore, first order filters which are
utilized at the temperature outputs are bypassed during
Inputs : Tem, pem, pt, ωt, θvgt
steady state testing and they are used in their original form,
Outputs : Tt
during transient WHTC validation test. For the model
For turbine outlet temperature, the modeling and validation performance, modeling and validation test results are given
mean absolute relative errors are calculated as %3.99 and in Figure 9.
%4.74 and results are plotted with in the Figure 8.
Parameter Estimation and Validation of Cylinder mass
Parameter Estimation and Validation of Turbocharger flow rate
inertia and turbine speed The cylinder mass flow rate can be defined as the total
To calculate the turbine speed, it is required to run turbine mass flow rate flowing into the cylinders. For parameter
and compressor models as coupled with their earlier estimation, engine fuel loop test is used. However, due to
predicted parameters. Therefore, turbine and compressor the difficulty of EGR flow measurement, the points where
models are combined for tuning of the turbine inertia (Jt) the EGR is closed has been used for parameter estimation.
and turbine friction coefficient (kJt) parameters. For the Thus, cylinder mass flow rate is assumed to be equal to
tuning of Jt and kJt, it is aimed to model precisely the Wc, Wt measured compressor flow rate. Increasing number of
and wt. For turbine inertia tuning, the inertia value coming closed EGR points and their uniform distribution through
from the turbine supplier is utilized as an initial value on testing data improve the model precision for the open EGR
MATLAB “Simulink Design Optimization – Parameter conditions. Model inputs and outputs are fed from the
Estimation Toolbox”, so a precise prediction is achieved. engine mapping test to estimate model parameters given
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 25

Figure 10. Modeling and validation results of cylinder Figure 11. Modeling test results of exhaust manifold
mass flow rate. temperature.

below. with these assumptions. The modeling and validation results

are illustrated in the Figure 11.
Inputs : Tim, pim, ne
Because the exhaust manifold temperature is calculated
Outputs : Wei
with iterative methods, a first order filter is utilized during
Parameters : cvol1, cvol2, cvol3
validation testing. The parameters of the filter are estimated
Cylinder mass flow rate results can be seen in Figure 10. by using the tfest function of MATLAB “System Identi-
For this case, with available data having been collected in fication Toolbox” and mean absolute relative errors for
EGR open status, it is not reasonable to use the WHTC test. modeling and validation are %3.68 and %5.1.
Hence, to evaluate the model, a specific EGR closed
transient test is used and the results are plotted above. The Parameter Estimation and Validation of EGR mass flow
mean absolute relative errors for modeling and validation rate
are %1.02 and %1.39. Because exhaust gas has high temperature and does not
spread homogeneously through the pipes, it is not possible
Parameter Estimation and Validation of Cylinder out and to measure EGR flow by using the sensors utilized for
exhaust manifold temperature compressor mass flow rate measurements. In the literature,
Because of the non-existence of temperature sensor at even if there are techniques using venture type sensors, the
cylinder out, model parameters are estimated to capture engine in the consideration is not equipped with such
exhaust temperature. Model inputs and outputs are fed instrumentation, hence, it was not possible to use that
from the engine mapping test and model parameters given method. Another approach for calculating the EGR mass
below are estimated. flow rate is utilizing the CO2 measurements taken from
intake and exhaust manifolds. This approach was also not
Inputs : Tamb, Tim, pim, pem, Weo, Wf, Wei
possible for our case as the intake manifold of our engine is
Outputs : Tem
not equipped the CO2 sensors. In this study, to calculate the
Parameters : xcv, ηsc, htot, Apipe
EGR mass flow rate, cylinder mass flow rate model is
As can be seen in the above list, Wei is one of the inputs of used, for which associated parameters are estimated first:
the model. In literature, often the test data for EGR closed
Wegr = Wei, calculated − Wc, measured (58)
is used in parameter estimation. However, in this study, Wei
is calculated from cylinder mass flow rate model. By this Using the above relationship, a virtual calculation channel
way, all fuel loop data is used for exhaust manifold has been created for the purpose of parameter estimation.
temperature model. NMinimize function of the Mathematica The inputs and outputs of the model are fed from the
is used for parameter estimation. Plausible minimum and engine mapping test and the following parameters are
maximum parameter ranges are used in the optimization estimated.
function as constraints. Also, only one iteration step is used
for tuning purposes while calculating exhaust manifold Inputs : Tim, pim, Tem, pem, ne, θegr
temperature. Furthermore, the initial values of xcv and T1 are Outputs : Wegr
accepted as 0 and 300. The parameter estimation is done Parameters : Aegrmax, aegr1, aegr2, begr1
26 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

Figure 13. Modeling test results of THR mass flow rate

(Throttle loop test).

the model for changing THR positions. The model inputs

and outputs are fed from the THR mapping test and the
parameters below are estimated.
Inputs : pic, pim, Tim, θthr
Figure 12. Modeling test results of EGR mass flow rate.
Outputs : Wthr
Parameters : Athrmax, athr1, athr2, bthr1, bthr2
MATLAB optimization functions are utilized to estimate
Due to the points discussed above about EGR tuning,
the model parameters. Also, a first order filter is used to
MATLAB optimization functions are used for parameter
achieve accurate results at transient WHTC simulations.
estimation. The estimation process yields a mean absolute
After parameter estimation, model performance is tested
relative error of %1.35.
with modeling and validation data. The mean absolute
After parameter estimation, it is concluded that the tuned
relative errors for modeling and validation are calculated as
model with THR step test is adequate to demonstrate the
%18.15 and %20.5.
effect of THR position on the intake manifold pressure.
Parameter Estimation and Validation of Throttle mass flow
Parameter Estimation and Validation of Exhaust System
rate
Aftertreatment devices coupled with engine are used to
Different from most studies in the literature, this study
meet present emission legislations. However, the process of
models the THR flow. In most studies, the throttle valve is
modeling and validation is performed without this instru-
often considered open during normal engine operation,
mentation due to the development process. Thus, the ex-
hence, no model is derived for the throttle mass flow rate.
haust system considered as a pipe starting from the turbine
However, current control strategies in practice decrease the
outlet. So a pressure drop model is created for the present
pressure at intake manifold by closing the THR. The
exhaust system configuration. The model inputs and outputs
decrease in pressure at the intake manifold accelerates the
are fed from the engine mapping test and the following
gas mass flowing through the EGR. Hence, for a realistic
system model, this dynamics should be taken into con-
sideration. The fuel loop and WHTC data are not used for
modeling and validation, because the THR is fully opened
at each data points. Therefore, a steady state throttle step
test is designed for modeling purpose. At several engine
speed-torque points, the throttle position is made to span
different values between closed to open and from opened to
closed, so a pressure drop at intake manifold is expected to
be observed. Moreover, EGR valve is kept as closed and
VGT is fixed to its base map values. The important point
here is that THR mass flow rate is used while calculating
the intake manifold pressure. Therefore, THR parameters
will affect the modeling results even if the THR is fully
opened, THR parameter estimation is done in 2 steps. As
can be seen from the throttle equations, for %100 THR
position, Athrmax fthr product is constant due to fthr. So, at first
step, the product Athrmax fthr is tuned in order to model the Wc.
At second tuning step, Athrmax fthr is taken as constant for Figure 14. Modeling test results of turbine out pressure
%100 THR position and the throttle step test is used to tune drop.
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 27

parameters are estimated: Parameter Estimation and Validation of Intercooler

In the literature, the intercooler temperature is taken as a
Inputs : Pamb, pem, Tt, Wt
state variable and calculated from ideal gas equation with
Outputs : pt
integration. However, in this study, it is assumed that this
Parameters : Heg0, Heg1
temperature does not have a considerable impact on model
The parameter estimation of the model is achieved by dynamics in flows and pressures, so it is not included
using the NMinimize function of the Mathematica. among the model states and calculated with algebraic
Because the pressure drop did not calculated with a equations. In fact, the most effective parameters while
differential equation, a first order filter added to the model calculating the intercooler outlet temperature are inter-
during transient WHTC validation. The parameters of the cooler coolant mass flow rate and intercooler coolant
filter are estimated using tfest function of MATLAB temperature. Due to lack of these measurements, mass flow
“System Identification Toolbox”. The resulting modeling rate effect is neglected and intercooler coolant temperature
and validation simulations yield mean absolute relative is accepted as equal to environment temperature. The
errors of %2.48 and % 1.12 respectively. inputs and outputs are fed from the engine mapping test
and the following parameters are estimated:
Parameter Estimation and Validation of EGR Cooler
Inputs : Tc, Tcool, Wc
The EGR cooler temperature is not used as input to the
Outputs : Tic
other submodels. In ideal case, the EGR cooler outlet
Parameters : Nic0, nic1, nic2, nic3, nic4
temperature should be modeled to care its effect while
modeling the intake manifold temperature. However, in The NMinimize function of the Mathematica is used for
this study, this effect is ignored due its minor effects on the the estimation. Also, a first order filter is added to the
results. The inputs and outputs are fed from the engine model whose parameters are estimated with the tfest
mapping test and the following parameters are estimated: function of MATLAB for transient WHTC validation
simulations. The validation and modeling simulations
Inputs : Tem, Tcool, Wegr
provide following results:
Outputs : Tegr
The mean absolute relative errors for modeling and
Parameters : negrc0, negrc1, negrc2
validation results are %0.54 and %0.68. To increase intake
The NMinimize function of Mathematica is used for para- manifold temperature model accuracy, the intercooler
meter estimation. The EGR mass flow rate is calculated by temperature parameter estimation can be realized by using
using equation (26). Tcool temperature is the engine coolant EGR closed points and EGR cooler outlet temperature
temperature which is assumed as 90 Celsius. After effect can be included to the model.
parameter estimation, the following results are obtained for
steady state fuel loop test and WHTC validation test.
While running the WHTC simulation, a first order filter
is added to the model similar to all temperature models.
The modeling and validation simulations yield mean
absolute relative errors of %3.44 and %2.95, respectively.

Figure 16. Modeling test results of Intercooler outlet

temperature.

2.2.2. Complete airpath model and comments

The complete airpath model is created by matching the
inputs and outputs of all submodels explained above. While
Figure 15. Modeling test results of EGR cooler outlet creating the complete model, outputs of the submodels are
temperature. connected with the bus creator block of Simulink. Then all
28 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

Figure 17. Complete airpath model structure.

outputs are connected to only one bus. Bus usage simplifies Inputs : ne, pamb, Tamb, uinj_qty, θvgt, θegr, θthr
the model creation for such a big Simulink block. At this Outputs : Wc, Wt, ωt, pic, pim, pem, pt, XOem, Tc, Tim, Tem, Tt
stage, only the following control volumes are left for para-
Simulation results are obtained for steady state modeling
meter estimation. However, parameter values coming from
and transient validation tests as can be seen in Figures 18 ~
engine design are utilized as they are and there was no need
27. During transient simulations first order filters are added
for additional parameter estimation/additional parameter
to the model due to the mentioned reasons above. A
estimation were not needed.
detailed view of the Simulink model of the airpath is given
Parameters : Vic, Vim, Vem with the figure below.
As can be seen in Figure 18 ~ 27, the obtained model
Based on this information, following inputs, which come
outputs are compatible with the measured values. The
from the engine mapping measurements, are fed to the
mean absolute relative errors of the modeling and valida-
complete engine model and and following outputs are
tion tests are summarized in Table 1 for sub-models and
compared with the measurements.
complete airpath model. In literature, the performance of

Figure 18. Complete model modeling results of compressor Figure 19. Complete model modeling results of turbine
mass flow rate. mass flow rate.
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 29

Figure 20. Complete model modeling results of turbine Figure 23. Complete model results of exhaust manifold
speed. pressure.

Figure 21. Complete model modeling results of Intercooler Figure 24. Complete model modeling results of turbine
pressure. outlet pressure.

Figure 22. Complete model modeling results of Intake Figure 25. Complete model modeling results of compressor
manifold pressure. outlet temperature.
30 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

Table 1. Airpath model error rates.

Modeling mean Validation mean
absolute relative error absolute relative error
Variable [%] [%]
Sub model All model Sub model All model
Wc 7.54 3.33 5.70 7.41
Wt 3.18 3.25 5.99 7.30
ψt - 4.12 - 7.85
Tc 1.95 1.86 1.86 2.47
Tt 3.99 6.80 4.74 8.10
Tem 3.68 4.14 5.10 5.43
pic - 7.90 - 8.04
pim - 3.83 - 6.96
pem - 3.41 - 7.35
Figure 26. Complete model modeling results of exhaust pt 2.49 2.61 1.04 1.50
manifold temperature. Xoem - 5.96 - 8.40

3. COMBUSTION MODEL

The diesel engine combustion model taken into consideration

in this study involves three outputs: NOx, PM and engine
shaft torque. In literature, detailed phenomenological, CFD
and empirical models are used for modeling of the engine
combustion. Phenomenological and CFD models genereally
use in-cylinder information and these crank angle based
models are computationaly inefficient additional to their
parameter identitification difficulty. The combustion model
created in this study is an empirical model and takes its
inputs from ECU control outputs and airpath model states
instead of in-cylinder signals. Because the created airpath
model does not generate crank angle information as an
output, a regression model is found to be more appropriate
Figure 27. Complete model modeling results of exhaust in this case as the modeling approach. In addition, its
manifold oxygen rate. capability of capturing effects of multi-injection strategy is
another important reason why a regression model is
adopted in this study.
the models is usually studied with short transient tests.
However, it is not faced with a study giving the all 3.1. Model Structure
validation results for WHTC test. In this study, combustion model uses 8 inputs that are
Also, it is known that the measurement delay for exhaust considered to be correlated with its outputs; NOx, PM and
manifold oxygen rate is high. As a result of validation tests, engine torque. With the assumption of one pilot and one
4 second sensor delay is observed and added as an offset to main, injection stategy is used for engine calibration;
the model. Furthermore, the other important factor increas- combustion model can be defined as;
ing the model precision is robustness of measurement
systems utilized at FORD-OTOSAN. These equipments NOx f1( xcomb )
are all well calibrated and measured data covers a broad PM = f2( xcomb )
range of working points of modeled engine. TQ f3( xcomb )
The validation of the entire model yields a mean value of
xcomb = [ne uinj_main uinj_pre usoi_main uinj_pre Wc pim prail ]
all absolute relative errors for all measured outputs to be
%8.4; hence, it is demonstrated that the model derived in uinj_main, uinj_pre, usoi_main, usoi_pre, prail are quantities of main and
this study demonstrates a more accurate performance than pilot injection, start of injection angles of main and pilot
the one in (Wahlstrom and Eriksson, 2010). injection, and rail pressure, respectively.
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 31

A DOE test is performed to estimate the parameters of

the combustion model. The test points are obtained based
on the space filling approach by varying the inputs to span
the 8-dimensional space. Data is collected after steady state
is reached in terms of temperature and emission values. A
total of 3200 data points are collected for this objective.
The required data is collected by scanning through the
reference points of ECU around the speed and torque
characterized operating points. The references are timings
of main injection and pilot injection, their quantities, rail
pressure, mass air flow and manifold absolute pressure.
DOE test points distribution over engine speed and engine
torque can be seen in Figure 2.
For the derivation of the combustion model, statistical
modeling approach is used. A polynomial model structure
is utilized for modeling. The polynomial includes the 1st
and 2nd order products and summations of the inputs with Figure 29. NOx validation and modeling results without
their coefficients, which are estimated by using the regstats airpath model.
function of the MATLAB. Terms with negligible effects on
the results are eliminated by calculating the p-values of modeling results while the lower ones present the WHTC
each polynomial term. As a result, the model order is reduced validation results. Combustion model validation test is
to include only the most effective polynomial terms. performed in two steps. In the first step, all model inputs
A switch case modeling approach is used for modeling are fed to model directly from measurement data. In the
the idle operating mode accurately. The main injection fuel second step, combustion model is integrated with airpath
quantity is calculated approximately from the measurements model and uses airpath model states (Wc and pim) as one of
for idle engine operation and used for switching between its inputs. The remaining combustion model inputs (except
two regression models whose parameters are estimated airpath model states) are still fed from measurement data.
separately from the DOE test. Combustion model structure Figures which belong to the first step can be seen in
can be seen in Figure 28. The delay and filter part is an Figures 29 ~ 34.
approximation for sensor delays/transfer functions which is Mean absolute relative errors of NOx are calculated as
valid for the transient testing. The delay sources of the %18.2 for modeling and %12.0 for validation test.
system can be listed as; The WHTC test did not include the PM signal due to
issues with the microsoot device. As a solution, the
• Mass air flow sensor and transportation delays,
• Manifold absolute pressure sensor delay,
• Rail pressure sensor delay and,
• Emission sensor delay.
Tuning of delay and filter parameters are done with first
100 seconds of WHTC test, but validation of the com-
bustion model is realized with complete WHTC test.

3.2. Results
The figures below include the modeling and validation Figure 30. Engine out PM modeling results without airpath
results together. The figures in the upper section depict the model.

Figure 31. Engine out PM WHTC results without airpath

Figure 28. Combustion model structure. model.
32 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

%2.3 for modeling and %8.8 for validation test.

The NOX and TQ output validation results for the
integrated airpath-combustion model can be seen in Figure
33 and Figure 34.
For this case, mean absolute relative errors are calculated
as %15.0 and %10.7 for NOx and TQ, respectively. Com-
bustion model results are summarized in Table 2.

4. CONCLUSION

This study develops an integrated airpath and combustion

model for the engine that, wherever possible, is based on
analytical relationships and can be used in the design of
model based control schemes. The combined model is
specifically designed with control inputs and outputs in a
form that allows for the use of the model in a control loop
Figure 32. Engine torque validation and modeling results and for controller design.
without airpath model. The approach taken in this study is similar to that of
(Wahlström, 2009; Wahlstrom and Eriksson, 2010, 2011)
in the development of the airpath model, which are among
the few studies in the literature deriving detailed physical
models geared towards controller design. Some contribu-
tions of this study to the above mentioned literature are the
inclusion of the emission model and its validation together
with the developed airpath and combustion model. The
WHTC based validation has been performed for all system
states outside of PM due to test limitations mention above.
Figure 33. Engine torque with airpath model. To further enhance the model, this study also takes the
actuator dynamics of the valves into account, develops a
model for the combustion system considering a multi-
injection strategy and integrates the airpath with the
developed combustion model. Additionally, to capture the
nonlinearities of the system in the best possible way, non-
linear software tools are used for parameter estimation. As
another contribution to diesel engine modeling and valida-
tion, after the individual validation of the airpath sub-
models and the full model individually, the combined
Figure 34. Engine out NOX with airpath model. airpath and combustion models are validated in cascade for
a more realistic evaluation of the developed model. The
Table 2. Combustion model error rates. validations are performed for long durations of WHTC,
Modeling (DOE) Validation (WHTC) unlike most studies that do not use this challenging cycle,
mean absolute mean absolute or use it for a very short interval, or for the validation of
relative error [%] relative error [%] just a few variables. The parameter estimation process is
also designed to have acceptable error rates not only in for
Model NOx Torque NOx Torque the subsystems, but also for the fully integrated model and
Without 18.2 2.3 12.0 8.8 WHTC based validation of the model.
airpath From the airpath error rates presented in Table 1, it can
With be seen that almost all estimated outputs of the airpath
airpath - - 15.0 10.7 model follow the measured values within a 8.4% relative
mean absolute error band (both in the modeling and
validation process), which is better than most other studies
validation was conducted via the comparison of the in the current literature. The results for the polynomial
calculated PM (from the dynamic model) with the static based combustion model yield a mean absolute relative
PM values obtained from the DOE test, resulting in a mean error of approximately 15%, while the engine torque estimation
absolute relative error of 35.4%. is almost below 10%. Hence, it can be concluded that the
Mean absolute relative errors of TQ are calculated as implemented approach and derived engine model can be
 MODELING AND VALIDATION OF TURBOCHARGED DIESEL ENGINE AIRPATH 33

used in the design of model-based advanced engine control Eriksson, L., Wahlström, J. and Klein, M. (2009). Physical
strategies and hardware in the loop simulation (HILS) Modeling of Turbocharged Engines and Parameter Identi-
based control prototyping. Further improvements to the fication, Automotive Model Predictive Control: Models,
model and design of the engine controller are the subject Methods and Applications. Springer. Berlin. 59–79.
for our next studies. Gambarotta, A., Lucchetti, G. and Vaja, I. (2011). Real-
time modelling of transient operation of turbocharged
ACKNOWLEDGEMENTS−This work was supported by the diesel engines. Proc. I.Mech.E., Part D: J. Automobile
Scientific and Technological Research Council of Turkey Engineering, 225, 0954-4070.
(TUBITAK Evrena)-110E117. This study is also supported by Guzzella, L. and Onder, C. H. (2010). Introduction to Model-
FORD-OTOSAN A.S. and Istanbul Technical University ing and Control of Internal Combustion Engine Systems.
Mechatronic Research Center (ITU-MEAM).
Springer-Verlag. Berlin.
REFERENCES Heywood, J. (1988). Internal Combustion Engine Funda-
mentals. McGraw-Hill New York.
Ahmed, F. (2013). Modeling, Simulation and Control of the Hirsch, M., Oppenauer, K. and del Re, L. (2010). Dynamic
Air-path of an Internal Combustion Engine. Ph.D. Dis- Engine Emission Models, Automotive Model Predictive
sertation. Universite de Technologie de Belfort-Montbeliard. Control-Models, Methods and Applications, LNCIS 402.
Belfort. Springer-Verlag. 402, 73−87.
Amsden, A., Ramshaw, J., O’Rourke, P. and Dukowicz, J. Hirsch, M., Alberer, D. and del Re, L. (2008). Grey-box
(1985). Kiva: A Computer Program for Two- and Three- control oriented emissions models. The Int. Federation
Dimensional Fluid Flows With Chemical Reactions and of Automatic Control Proc. 17th World Cong., Seoul,
Fuel Sprays. Los Alamos National Laboratory, Technical Korea, 8514−8519.
Report No. LA-10245-MS. Hiroyasu, H. (1985). Diesel engine combustion and its
Arsie, I., Di Genova, F., Pianese, C., Sorrentino, M., Rizzo, modeling. Proc. COMODIA Symp. Diagnostics and Model-
G., Caraceni, A., Cioffi, P. and Flauti, G. (2004). Develop- ing of Combustion in Reciprocating Engines, Tokyo,
ment and identification of phenomenological models for Japan, 53–75.
combustion and emissions of common-rail multi-jet Johnson, T. (2009). Review of diesel emissions and control.
diesel engines. SAE Paper No. 2004-01-1877. Int. J. Engine Research 10, 5, 275–285.
Arsie, I., Pianese, C. and Sorrentino, M. (2007). Effects of Jung, M. (2003). Mean-value Modelling and Robust Control
control parameters on performance and emissions of of the Airpath of a Turbocharged Diesel Engine. Ph.D.
HSDI diesel engines: Investigation via two zone modell- Dissertation. University of Cambridge. Cambridge.
ing. Oil & Gas Science and Technology 62, 4, 457−469. Kirchen, P. and Boulouchos, K. (2009). Development and
Arsie, I., Di Leo, R., Pianese, C. and De Cesare, M. (2012). validation of a phenomenological mean value soot
Combustion noise and pollutants prediction for injection model for common-rail diesel engines. SAE 2009 World
pattern and EGR tuning in an automotive common-rail Cong. & Exhibition, Detroit, Michigan, USA.
diesel engine. IFAC Workshop on Engine and Powertrain Kirchen, P., Obrecht, P. and Boulouchos, K. (2009). Soot
Control, Simulation and Modeling (E-COSM’12), Rueil- emission measurements and validation of a mean value
Malmaison, France. soot model for common-rail diesel engines during
Aspriona, J., Chinellatob, O. and Guzzella, L. (2013). A transient operation. SAE Int. J. Engines, 2, 1663–1678.
fast and accurate physics-based model for the NOx Lee, B., Jung, D., Kim, Y. W. and Nieuwstadt, M. (2013).
emissions of Diesel engines. Applied Energy, 103, 221– Thermodynamics-based mean value model for diesel.
233. Trans. ASME, J. Engineering for Gas Turbines and Power,
Benz, M. (2010). Model-based optimal emission control of 135, 1−9.
diesel engines. Ph.D. Dissertation. Retrieved from http:/ Maiboom, A., Tauzia, X. and Hétet, J. F. (2008). Experimental
/e-collection.library.ethz.ch/eserv/eth:1197/eth-1197- study of various effects of exhaust gas recirculation
02.pdf (EGR) on combustion and emissions of an automotive
Benz, M., Onder, C. H. and Guzzella, L. (2010). Engine direct injection diesel engine. Energy 33, 1, 22−34.
emission modeling using a mixed physics and regression Patil, K., Molla, S. and Schulze, T. (2012). Hybrid vehicle
approach. Trans. ASME, J. Engineering for Gas Turbines model development using ASM-AMESim-simscape co-
and Power 132, 4, 042803-1-11. simulation for real-time HIL applications. SAE Paper
Bonnet, B. (2007). Matching of Internal Combustion Engine No. 2012-01-0932.
Characteristics for Continuously Variable Transmissions. Skogtjarn, P. (2002). Modelling of the Exhaust Gas Temper-
Ph.D. Dissertation. Cranfield University. Bedfordshire. ature for Diesel Engines. LiTH-ISY-EX-3379, M.S.
Brahma, I., Sharp, M. and Frazier, T. (2009). Empirical Thesis. Department of Electrical Engineering. Linkoping
modeling of transient emissions and transient response University. Linkoping.
for transient optimization. SAE Int. J. Engines 2, 1, Stadlbauer, S., Alberer, D., Hirsch, M., Formentin, S. and
1433−1443. del Re, L. (2012). Evaluation of virtual NOx sensor
34 B. UNVER, Y. KOYUNCUOGLU, M. GOKASAN and S. BOGOSYAN

models for off road heavy duty diesel engines. SAE Int. Linkopings.
J. Commercial Vehicles 5, 1, 128−140. Wahlstrom, J. and Eriksson, L. (2010). Modeling of a Diesel
Stamati, I., Telen, D., Logist, F., Van Derlinden, E., Hirsch, Engine with Intake Throttle, VGT, and EGR. Linkoping
M., Passenbrunner, T. and Van Impe, J. (2012). Optimal University. Linkoping. Report: LiTH-ISY-R-2976.
experiment design for calibrating an airpath model of a Wu, H., Wang, X., Winsor, R. and Baumgard, K. (2011).
diesel engine. SNE Simulation Notes Europe 22, 3-4, Mean value engine modeling for a diesel engine with
121−128. GT-power 1D detail model. SAE Paper No. 2011-01-
Unver, B. (2013). Modeling Diesel Engines, Development 1294, 1−19.
and Application of Model Based Airpath and Emission Zhang J., Gao, S. and Jiang, F. (2007). A diesel engine real
Controllers. Ph.D. Dissertation. Istanbul Technical time NOx emission simulation system based on RTW
University. Istanbul. and VxWorks. SAE Paper No. 2007-01-0025.
Wahlstrom, J. and Eriksson, L. (2011). Modelling diesel Zhao, H. (editor) (2010). Advanced Direct Injection Com-
engines with a variable-geometry turbocharger and ex- bustion Engine Techniologies and Development: Volume
haust gas recirculation by optimization of model 2: Diesel Engines. Woodhead Publishing Ltd. Cambridge.
parameters for capturing non-linear system dynamics. UK.
Proc. Institution of Mechanical Engineers, Part D, J. Zweigel, R., Albin, T., Hesseler, F.-J., Jochim, B., Pitsch,
Automobile Engineering 225, 7, 960−986. H. and Abel, D. (2013). Greybox modeling of the diesel
Wahlström, J. (2009). Control of EGR and VGT for Emission combustion by use of the scalar dissipation rate. Euro-
Control and Pumping Work Minimization in Diesel pean Control Conf. (ECC), 3961−3966.
Engines. Ph.D. Dissertation. Linkopings University.