Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

Published 09/04/2017
Copyright © 2017 SAE International
doi:10.4271/2017-24-0057

Model-Based Control of BMEP and NOx Emissions in a

Euro VI 3.0L Diesel Engine
Roberto Finesso, Omar Marello, Ezio Spessa, and Yixin Yang
Politecnico di Torino

Gilles Hardy
FPT Motorenforschung AG

ABSTRACT
A model-based approach to control BMEP (Brake Mean Effective Pressure) and NOx emissions has been developed and assessed on a
FPT F1C 3.0L Euro VI diesel engine for heavy-duty applications. The controller is based on a zero-dimensional real-time combustion
model, which is capable of simulating the HRR (heat release rate), in-cylinder pressure, BMEP and NOx engine-out levels. The
real-time combustion model has been realized by integrating and improving previously developed simulation tools. A new
discretization scheme has been developed for the model equations, in order to reduce the accuracy loss when the computational step is
increased. This has allowed the required computational time to be reduced to a great extent. The real-time combustion model has been
first calibrated and assessed at both steady-state and transient conditions, on the basis of experimental data acquired at the highly
dynamic test bench of ICEAL-PT (Internal Combustion Engines Advanced Laboratory - Politecnico di Torino), in the frame of a
research activity in collaboration with FPT Industrial. The model has then been used to realize a model-based control of BMEP and
NOx emissions. In particular, the controller provides the injected fuel quantity and the injection timing of the main pulse, for given
targets of BMEP and engine-out NOx levels. Finally, the developed controller has been tested on a rapid prototyping device (ETAS
ES910) through HiL (Hardware-in-the-Loop) techniques, and demonstrated to have real-time capability.

CITATION: Finesso, R., Marello, O., Spessa, E., Yang, Y. et al., "Model-Based Control of BMEP and NOx Emissions in a Euro VI 3.0L
Diesel Engine," SAE Int. J. Engines 10(5):2017, doi:10.4271/2017-24-0057.

INTRODUCTION The implementation of model-based controls has been made possible,

in the last few years, by the increasing computational performance of
Interest in model-based combustion control has increased in the last
modern ECUs.
few years [1]. A model-based approach for engine control can in fact
lead to several advantages compared to the traditional map-based one.
Therefore, the development of combustion models has become of
For example, the development of a model-based approach requires in
great interest.
general lower experimental effort compared to that required for the
calibration of the engine maps, also considering that a large number of
In general, engine simulation can be carried out with different
engine maps are implemented in modern ECUs (Engine Control
degrees of detail. The main simulation approaches include
Units) [1], which are even difficult to manage. Moreover, a model-
multidimensional, one-dimensional or zero-dimensional methods,
based control offers the possibility of optimizing the combustion
that are characterized by a different degree of detail and
process onboard, and to adjust the main calibration parameters during
computational effort.
the engine operation in real time. For example, this would allow the
management of the after-treatment system to be optimized, by setting
3D-CFD (Computer Fluid-Dynamics) methods [2, 3, 4, 5, 6, 7, 8, 9,
variable targets of exhaust temperatures and/or pollutant engine-out
10] and 1D-CFD approaches [11, 12, 13, 14, 15, 16] require a
emissions over a vehicle mission, and these targets may be achieved
computational time that is not currently suitable to develop model-
by adapting the main ECU variables in real-time. In addition, a
based combustion controls.
model-based approach may be capable of taking into account the
effects, on the combustion process, of the variability of the
Mean-value engine and combustion models [17-18] are capable of
environmental conditions (humidity, temperature, pressure) or the
simulating the combustion and emission formation processes with a
effects related to the engine transient operation (turbo-lag, EGR
good level of detail [18]. These methods offer the opportunity of
delays, …), without the need of implementing correction maps.

2288
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017) 2289

further decreasing the computational time in comparison to 1D-CFD The model has been first calibrated and assessed at both steady-state
approaches, while guaranteeing at the same time a good predictive and transient conditions over several speed/load ramps, on the basis
capability at steady-state and transient engine operating conditions. of experimental data acquired at the highly dynamic test bench of
Moreover, they are physically consistent, so that they do not require a ICEAL-PT (Internal Combustion Engines Advanced Laboratory -
high calibration effort, and at the same time their accuracy is still Politecnico di Torino), in the frame of a research activity in
acceptable outside the calibration range [18]. Therefore, they can be collaboration with FPT Industrial. Then, it has been inverted in order
considered as good candidates for the development of model-based to realize a model-based controller of BMEP and NOx emissions. In
control algorithms. A real-time combustion model, which belongs to particular, the controller is capable of providing the injected fuel
this category, has been developed by the authors in [18]. quantity and the injection timing of the main pulse, for given targets
of BMEP and NOx engine-out emissions.
Finally, artificial intelligence systems [19, 20, 21, 22, 23, 24], such as
support vector machines (SVM), genetic algorithms (GA) and Finally, the developed controller has been tested on a rapid
artificial neural networks (ANNs) constitute a last category of models prototyping device (ETAS ES910) through HiL (Hardware-in-the-
which are often used in the field of engine design and control. These Loop), and demonstrated to have real-time capability.
methods do not require the detailed physical knowledge of the
investigated process and are able to capture complex nonlinear
system behavior with relatively simple mathematical operations. ENGINE SETUP AND EXPERIMENTAL
Moreover, they are characterized by a very small computational time, ACTIVITY
so that they are good candidates for the development of model-based The experimental tests used for model calibration and validation were
control algorithms to be implemented in ECUs. However, their conducted on a FPT F1C 3.0L Euro VI diesel engine. The main
training usually requires a high number of experimental tests, and technical specifications of the engine are reported in Table 1.
their performance is usually not reliable outside the calibration range.
Table 1. Main technical specifications of the engine.

Contribution of the Present Study

It should be noted, from the previous background, that mean-value
models are among the best candidates for the development of
model-based combustion control algorithms, as they are physically
consistent, and generally require a low calibration effort and a low
computational time.

In this study, a previously developed real-time combustion model

[18] has been calibrated and validated for a FPT F1C 3.0L Euro VI
diesel engine, and has been used to develop a model-based controller
of BMEP and NOx emissions. The activity was carried out in the The engine (Fig. 1) is equipped with a short-route cooled EGR
frame of a research project in collaboration with FPT Industrial. system, in which the EGR valve is located upstream from the cooler.
A flap is installed in the exhaust pipe downstream the turbine, to
The real-time combustion model is capable of simulating the HRR control the temperature of the exhaust gas flowing to the
(heat release rate) and in-cylinder pressure, along with the related aftertreatment system and to allow high EGR rates to be obtained
metrics, such as MFB50, PFP (Peak Firing Pressure), IMEP (Indicated when the pressure drop between the exhaust and intake manifolds is
Mean Effective Pressure) and BMEP. In particular, the chemical not sufficiently high.
energy release has been simulated using an improved version [18, 25]
of the accumulated fuel mass approach, previously presented in [26,
27, 28, 29, 30]. A 3-zone thermodynamic model is also included to
simulate the in-cylinder burned gas temperatures. The latter model has
been enhanced in this paper with respect to previous versions, in order
to be able to account for the effect of intake air humidity. Engine-out
NOx emissions were simulated on the basis of a semi-empirical
correlation, that takes into account the in-cylinder burned gas
temperature, as well as MFB50 and additional metrics.

The real-time combustion model used in this study has been improved
with respect to previous versions in terms of computational efficiency.
In particular, a new discretization scheme has been developed for the
in-cylinder pressure model, in order to reduce the accuracy loss when Figure 1. FPT F1C 3.0L Euro VI diesel engine installed on the highly dynamic
test bench at the Politecnico di Torino. The rapid prototyping device can be
the computational step is increased. This has allowed the required
observed on the right side.
computational time to be reduced to a great extent.
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

2290 Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017)

The test engine was instrumented with piezoresistive pressure variation of ±6 deg around the nominal values and a pf variation
transducers and thermocouples to measure the pressure and of ±20% around the nominal values were set.
temperature at different locations, such as upstream and downstream
from the compressor, from the turbine and intercooler, in the intake The developed control technique was tested on the engine over
manifold and in the EGR circuit. Thermocouples were also used to different load/speed ramps. Details on these ramps are reported in the
measure the temperatures in each intake and exhaust runners. “Results and discussion” section.
KISTLER 6058A high-frequency piezoelectric transducers were
fitted to the glow-plug seat to measure the in-cylinder pressure
time-histories, which were used to realize a pressure-based MFB50
control [1]. The in-cylinder pressure traces were corrected on the
basis of the intake pressure that was measured by means of high-
frequency KISTLER 4007C piezoresistive transducers, which were
located at the inlet runners of the cylinders.

All the experimental tests were carried out on the highly dynamic test
bed at ICEAL at the Politecnico di Torino. The test rig is equipped
with an ‘ELIN AVL APA 100’ cradle mounted AC dynamometer and
an ‘AVL KMA 4000’, with a reading accuracy of 0.1% over a
0.28-110 kg/h range, to continuously measure the fuel consumption.
An ‘AVL AMAi60’ system, consisting of three analyzer trains, was
used to measure the engine-out gaseous raw emissions. Two analyzer
trains were equipped with complete devices for the analysis of THC, Figure 2. Experimental tests used for the calibration of the real-time
CH4, NOx, and low as well as high CO, CO2 and O2, and were used combustion model.
to measure the intake and exhaust gas composition. All of the
abovementioned measurement devices were controlled by a PUMA REAL-TIME COMBUSTION MODEL
OPEN 1.3.2 automation system. In order to minimize the testing
The real-time combustion model that has been used for the
effort, the test bed environment was interfaced with AVL CAMEO
development of the BMEP/NOx controller includes the simulation of:
software to run intelligent engine calibration procedures on the basis
of the DoE (Design of Experiment) approach.
1. Chemical energy release: the approach is based on an enhanced
version [25] of the model previously presented by the authors,
An ETAS ES910 rapid prototyping device was used to realize
which was based on the accumulated fuel mass approach [30].
pressure-based and model-based controls of the combustion phasing
The input data of the model are the injection parameters, as well
(see [1]), and to test the real-time capability of the model-based
as the main thermodynamic conditions in the intake manifold
controller of BMEP and NOx which has been developed in this
and the engine operating parameters.
study. The main specifications of the ETAS ES910 device are
reported in Tab. 2. 2. In-cylinder pressure: the approach is based on the inversion
of a single-zone heat release model [31] which requires the
Table 2. Main specifications of the ETAS ES910 rapid prototyping device. net energy release as input; the latter is derived starting from
the predicted chemical energy release and estimating the heat
transfer between the charge and the walls. Polytropic evolutions
are assumed during the compression and expansion phases.
Several metrics, such as PFP and IMEP, can be extracted from
the simulated in-cylinder pressure.
3. Friction losses: the Chen-Flynn approach has been used to
Experimental Activity predict FMEP on the basis of the engine speed and peak firing
pressure; the simulation of friction losses allows BMEP to be
The experimental tests that have been considered in the present paper
evaluated starting from IMEP.
include steady-state tests and transient tests. The steady-state tests
were mainly used to calibrate the real-time combustion model. To this 4. Pumping losses: the pumping losses (PMEP) were simulated
aim, the following tests were considered (Fig. 2): on the basis of a semi-empirical correlation which takes into
account the intake and exhaust manifold pressure levels, as well
• A full engine map with baseline operating parameters, including as engine speed.
123 points. 5. NOx emission levels: an improved version of the semi-empirical
• EGR-sweep tests at fixed key-points, including 162 points. EGR correlation previously developed by the authors for a 2.0L
rate was varied from 0 to 50% by setting different levels of Euro 5 diesel engine and reported in [32] has been tuned and
trapped air mass with steps of 50 mg/cycle. validated for the 3.0L F1C Euro VI engine considered in the
present study.
• sweep tests of main injection timing (SOImain)/injection pressure
(pf) at fixed key-points, including 125 points. A SOImain
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017) 2291

The detailed description of the combustion model and the calibration

methodology is reported in [18]. However, a summary is provided
(4)
hereafter for the sake of clarity. Moreover, the correlations of the
model parameters which have been obtained from the model
recalibration on the F1C engine have also been reported.
(5)

Figure 3 reports the scheme of the real-time combustion model.

(6)

(7)

Table 3. Main equations of the Qch and Qnet models.

Figure 3. Scheme of the real-time combustion model.

Estimation of the Chemical Energy Release Qch and of

the Net Energy Release Qnet
The equations of the Qch and Qnet sub-models are reported in Table 3.
The reader may refer to [18] for further details related to these
equations.

The Qch and Qnet models were assessed for the steady-state conditions
reported in Fig. 2. Physically-consistent correlations were identified
In equations (1, 2, 3, 4, 5, 6, 7), ρSOI, ρSOC indicate the in-chamber
for the model calibration parameters, as a function of the in-chamber
densities evaluated at the start of injection or combustion,
thermodynamic quantities at SOI/SOC (start of injection/start of
respectively, and are expressed in kg/m3. The injection pressure pf is
combustion) and other engine variables, as follows:
expressed in bar, the engine speed n in rpm, the total injected fuel
quantity qf,inj (used as a load parameter) in mm3/cyc/cyl, the total
injected fuel quantity of the pilot shots qpil,tot in mm3/cyc/cyl and
finally the intake oxygen concentration O2 in %. Tint and pint indicate
(1)
the intake manifold temperature and pressure, respectively.

Estimation of the In-Cylinder Pressure

(2) The equations of the in-cylinder pressure sub-model are reported in
Table 4. The reader may refer to [18] for further details related to
these equations.
(3)
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

2292 Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017)

Table 4. Main equations of the in-cylinder pressure model. so that the pressure value at time instant ti is evaluated as follows:

(14)

The new approach for pressure discretization has allowed the

accuracy loss to be reduced to a great extent when the computational
step is increased.

A comparison of the model accuracy (quantified by the RMSE, i.e.,

root mean squared error) when using the new scheme (i.e., Eq. (14))
With reference to table 4, the in-chamber pressure was evaluated, during and the old one (i.e., Eq. (12)) is reported in Table 5, for different
the combustion interval, on the basis of the inversion of a single-zone values of the computational step. It can be seen that the new
heat release model, which can be summarized as follows [31]: discretization scheme leads to a virtually negligible loss of accuracy
in terms of IMEP and BMEP prediction, when the crank angle
integration step (CAstep) is increased to 0.5 or 1 deg, starting form a
(8) nominal value of 0.1 deg.
where Qht represents heat transfer and Cv the specific heat at constant
Table 5. Comparison between the values of RMSE for the main quantities of
volume. In Eq. (8) the last term is usually rewritten using the
the real-time combustion model, when using the old (a) and new (b) pressure
ideal-gas law and assuming a constant mass m, so that: discretization schemes. Engine map tests are considered.

(9)

where γ=cp/cv. The last differential in Eq. (9) is typically rewritten in

terms of pressure and volume differentials, and the following
differential equation is commonly proposed in the literature [31]:

(10)

In the previous versions of the combustion model proposed by the

authors [25, 30], the pressure differential was evaluated on the basis
of Eq. (10), as follows:

(11)

which had been then discretized according to the following scheme:

In Eq. (14) the isentropic coefficient γ=cp/cv was set constant and
equal to 1.37.

The following correlations were identified for the pressure model

(12) parameters, which are functions of the intake manifold
thermodynamic conditions and of the engine load and speed:
The new approach proposed in this study starts directly from Eq. (9)
and is based on the following discretization scheme:
(15)

(13) (16)
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017) 2293

(17)

The in-cylinder pressure trace has been simulated over the (21)
compression and combustion phases only, and this allows the gross
IMEP (i.e., IMEP360) to be estimated. Pumping losses were
evaluated by means of a dedicated correlation, which is reported in
the next section.
(22)
The simulation of the in-cylinder pressure traces also allows
peak-firing pressure (PFP) to be evaluated.

Estimation of PMEP and FMEP

(23)
The following correlation was identified to evaluate PMEP for the
engine considered in this study:

(18) (24)

The Chen-Flynn approach [33] was adopted to estimate FMEP. Equations (21, 22, 23, 24) were derived considering all the data
related to the engine map, EGR-sweep and SOImain/pf sweep tests, as
The experimental values of FMEP were evaluated as the difference it was verified that this led to satisfactory results for all datasets.
between the experimental values of the net IMEP and the measured Moreover, at the beginning, all the main engine variables related to
values of BMEP, as follows: NOx formation were included in the correlations, and a sensitivity
analysis was carried out in order to exclude the least influential ones.

(19) In order to avoid discontinuities when applying the NOx model, Eqs.
(21, 22, 23, 24) have been calibrated using overlapping calibration
The values of IMEP of one of the four cylinders were taken as being
datasets. In other words, each of the four calibration datasets also
representative of all the cylinders, due to the low cylinder-to-cylinder
included tests characterized by injected quantities and speed levels
dispersion.
exceeding the breakpoints (i.e., qf,inj = 45 mm3, n = 1900 rpm).

The following correlation was identified to evaluate FMEP for the

It can be seen that the temperature of the burned gases evaluated at
engine considered in this study:
MFB50 (i.e., Tb,MFB50) has been used. The temperature of the burned
gases was evaluated by means of the real-time thermodynamic
three-zone model that is described in the next section. The evaluation
(20)
of the Tb,MFB50 term, in general, requires the evaluation of the
where the engine speed is expressed in rpm and PFP is expressed in in-cylinder pressure, which is an input of the thermodynamic model.
bars.
Moreover, it can be seen that the MFB50 parameter has also been
The squared correlation coefficient R between the predicted and
2 used in the correlation. It was verified that the use of MFB50 leads to
experimental values of FMEP is the order of 0.9. a significant increase in the NOx prediction accuracy for all the
experimental datasets.

Estimation of the NOx Emissions

Three-Zone Thermodynamic Model
The engine-out NOx emissions were evaluated starting from the
semi-empirical model developed by the authors in [32]. That A refined version of the real-time three-zone thermodynamic model
correlation has been re-tuned for the engine considered in the present presented in [34] has been used in order to evaluate the temperature of
study, on the basis of the experimental tests reported in Fig. 2. the burned gases. In particular, the model has been refined by taking
into account the effect of the intake air humidity. The 3-zone
The following correlations were obtained from the tuning of the NOx thermodynamic model is based on the splitting of the in-chamber
emission model: content into a vapor-fuel zone (f), an unburned gas zone (u), made up
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

2294 Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017)

of fresh-air, residual gas and EGR, and a burned gas zone (b) obtained Table 7. Summary of the coefficients used in Eq. (25) to estimate the
from a globally stoichiometric combustion process. Energy and mass enthalpies of the burned zone (b), unburned zone (u) and fuel zone (f). In the
conservation equations are applied to each zone (see Table 6). table, Habs indicate the absolute humidity, Xr,tot the total residual + EGR ratio
in the combustion chamber, λ the relative air-to-fuel ratio
Table 6. Energy and mass conservation equations of the 3-zone
thermodynamic model.

The burned gas zone is considered to be made up of CO2, H2O, O2,

N2, O, H, OH and NO, and the dissociation effects are therefore taken
into account for accurate calculation of the burned gas temperatures.
It has been verified in [34] that second-order polynomial correlations
are able to accurately describe the specific enthalpy of the different
zones, as follows:

(25)

The values of the coefficients of the enthalpy terms were improved

with respect to those reported in [34]. In particular, in previous papers
the correlations for the enthalpy of the burned and unburned gases
were derived assuming a value of the intake air humidity Habs = 10
gv/kga, and the effect of humidity on NOx emissions was taken into
account by adopting the standard correction formulas. In this paper,
the effect of humidity has been directly taken into account in the
definition of the enthalpies, as shown in Table 7, in order to improve
the physical consistency of the approach.
MODEL-BASED CONTROL OF BMEP
AND NOX
The heat transfer terms in the equations reported in Tab. 6 were A model-based approach to control BMEP and NOx emissions has
estimated by means of a convective and a radiative contribution, as been developed in this study. The controller is based on the inversion
explained in [34], the first one being modeled by the Woschni of the real-time combustion model reported in the previous section, in
correlation. order to predict the values of the injected fuel quantity qf,inj and of the
injection timing of the main pulse, i.e., SOImain, that allows desired
On the basis of the procedure reported in [34], the energy and mass targets of BMEP and engine-out NOx emissions to be reached. SOImain
conservation equations are then discretized considering finite time has been selected as the control variable for the NOx emissions
intervals Δt = ti - ti-1. It is thus possible to explicitly derive the control due to its capability to realize a cycle-by-cycle engine
temperature of the three zones by solving the resulting second-order response, while the EGR rate (whose dynamics is slower) has been
polynomial equations for the three zones. kept constant and equal to the setpoint value. The model was inverted
by adopting an iterative procedure, in which the first run was based on
The unknowns can therefore be solved in closed form without an the initial assumption of the qf,inj and SOImain control variables and a
iterative procedure, and this allows a very low computational effort to cycle-based integral control was applied to adjust the control variable
be obtained. value in order to attain convergence of the target variables (i.e.,
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017) 2295

BMEPtgt and NOxtgt). The iterative procedure stops when the parameter was identified. In particular, the value of was
difference between the predicted values and required values of the limited to the [0.15-2] range, and was varied, iteration by iteration, as
target variables fall below the predefined thresholds εBMEP and εNOx. a function of the sign of the error between two consecutive iterations,
according to the following method:
More in detail, the model inversion has been carried out according to
the following procedure. Target values of BMEP and NOx, i.e.,
, are set for a given cycle ‘j’. The first model run
is carried out using the nominal engine map values for the qf,inj and
SOImain control variables.

The predicted values of BMEP and NOx for the generic iteration ‘i’,
i.e., , are compared with the target values (30)
.The errors between the target and the actual
With reference to the correction of qf,inj, a similar correction scheme
values for the iteration ‘i’ are then estimated as follows:
has been used:

(26)

(31)
(27)
where is a BMEP-to-fuel sensitivity factor which is derived
SOImain is then corrected on the basis of NOx error, while qf,inj is from the average engine fuel conversion efficiency, and is a
corrected on the basis of BMEP error and a new iteration is carried modulation factor, with is defined as follows:
out. This choice is justified by the fact that the sensitivity of BMEP to
qf,inj is much greater than the sensitivity to SOImain, and the sensitivity
of NOx to SOImain is greater than the sensitivity to qf,inj (at least when
the BMEP error is small and qf,inj is not far from the value which
allows the BMEP target to be obtained). It should be noted that the
combustion model is obviously able to estimate the combined effects
of SOImain and qf,inj variations on both BMEP and NOx emissions (32)
during a given iteration, therefore the choice of controlling BMEP
The strategy for the modulation of the factors
using qf,inj only and of controlling NOx using SOImain only just affects
was inspired by a previous closed-loop MFB50 controller, which was
the number of iterations needed to achieve convergence and does not
presented by the authors in [1]
affect the accuracy in the estimation of BMEP and NOx levels.

Figure 4 shows the model inversion scheme.

With reference to the SOImain correction scheme, a preliminary
estimation of the sensitivity of NOx emissions with respect to SOImain
is needed. To this end, Eqs. (21, 22, 23, 24) are used to estimate how
NOx emissions change when setting an MFB50 variation with respect
to a reference value, keeping constant all the other parameters (i.e.,
Tb,MFB50, O2, n, qf,inj). A NOx sensitivity factor with respect to MFB50
variation, i.e., , is calculated as follows:

(28)

The value of SOImain for the subsequent iteration (i+1) is then

obtained adopting the following correction scheme:

(29)

where is a modulation factor that was introduced in order to

optimize the response of the controller and to guarantee stable Figure 4. Flow chart of the BMEP/NOx controller.
operations. An optimal strategy for the definition of the
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

2296 Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017)

It should be noted that, in this study, the variables n, pint, Tint, DT, qpil values of engine calibration parameters, but also when EGR, injection
and the air mass flow rate were obtained from the engine sensors timing and injection pressure are varied at fixed operating condition.
(i.e.., these variables were available from the ECU), while Habs and This confirms its physical consistency.
pexh were obtained from test bench sensors. With reference to the
EGR mass flow rate, the experimental value derived from intake
manifold CO2 concentration was used for model calibration and
validation in steady-state conditions. Instead, a look-up table was
used for the transient simulations, which was built using the nominal
EGR rate values obtained from the steady-state engine map tests.

RESULTS AND DISCUSSION

Assessment of the Real-Time Combustion Model:
Steady State Tests
First, the accuracy of the real-time combustion model has been
assessed for the steady state tests reported in Fig. 2. All the results
shown in this section were obtained adopting a crank angle
computational step of 0.1 deg. However, it was shown in Table 5 that
the loss of accuracy is virtually negligible when increasing this step
a.
to 1 deg. The adoption of the latter step is associated to a
computational time that is compatible for real-time applications.

Figure 5 reports the predicted vs. experimental values of MFB50,

while Fig. 6 reports the predicted vs. experimental values of PFP,
IMEP720 (i.e., net IMEP) and BMEP, and finally Fig. 7 reports the
predicted vs. experimental values of engine-out NOx emissions, for
all the tests reported in Fig. 2. The prediction accuracy of each
model has been quantified by the squared correlation coefficient
(R2) and by the Root Mean Squared Error (RMSE), which are
reported in each figure.

b.

Figure 5. Predicted vs. experimental MFB50 values for the steady-state tests
reported in Fig. 2.

With reference to MFB50 prediction, it can be seen in Fig. 5 that the

RMSE of the physical model is of the order of 0.9 deg and that the
dispersion of the predicted values remains in an acceptable range for
all the test categories. With reference to Fig. 6, it can be noted that a c.
very accurate prediction of PFP is obtained (RMSE=2.5 bar), as well
as of IMEP720 (RMSE=0.18 bar) and of BMEP (RMSE = 0.15 bar). Figure 6. Predicted vs. experimental values of PFP (a), IMEP720 (b) and
The model is robust not only for the engine map tests with nominal BMEP (c) for the steady-state tests reported in Fig. 2.
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017) 2297

The main model results have been reported in Figs. 9, 10, 11, 12. In
particular, the figures report the predicted and experimental values of
MFB50 and PFP (Fig. 9a, 9b), of BMEP (Fig. 10), of the
instantaneous engine-out NOx emissions (Fig. 11) and of the
cumulated engine-out NOx emissions (Fig. 12). In all the charts, the
experimental values have been reported in blue color, while the
results of the model have been reported in red color. The values of
RMSE are also reported at the top of each graph.

Figure 7. Predicted vs. experimental values of engine-out NOx emissions for

the steady-state tests reported in Fig. 2.

With reference to the prediction of NOx emissions (Fig. 7), the

RMSE is of the order of 35 ppm, and the model accuracy is good for
all the considered test categories.

Validation of the Real-Time Combustion Model:

Transient Tests
The real-time combustion model has then been validated in transient a.
conditions. The results reported in this section are related to six sets
of up/down speed/load ramps of different duration. Figure 8 reports
the time histories of the engine speed and torque for the six analyzed
sets of ramps. The engine speed variation range was between 1600
rpm and 2500 rpm, while the engine torque variation range was
between 55 Nm and 215 Nm. The duration of the first three sets of
ramps is 5s, while the duration of the remaining three sets of ramps is
3s. It can be seen in Fig. 8 that the first test is constituted by a
ramp-up and a ramp-down of both engine speed and torque, the
second test is constituted by a ramp-up and a ramp-down of torque at
fixed engine speed, while the third test is constituted by a ramp-up
and a ramp-down of speed at fixed engine torque. The three sets of
ramps have then been repeated by reducing the ramp duration to 3s.

b.

Figure 9. Predicted and experimental trends of MFB50 (a) and PFP (b) for the
analyzed transient test.

With reference to the prediction of the MFB50 and peak-firing

pressure, it can be seen in Fig. 9 that the model accuracy is
satisfactory, as the RMSE values are of the order of 0.8 deg and 1.8
bar, respectively. This suggests that the prediction of the heat release
and of the in-cylinder pressure trace is accurate over the considered
transient tests.

Figure 8. Engine speed and torque as a function of time for the analyzed
transient test.
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

2298 Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017)

a. Predicted NOx trace is not filtered

a.

b. Predicted NOx trace has been filtered

b.
Figure 11. Predicted and experimental trends of engine-out NOx emissions for
Figure 10. Predicted and experimental trends of BMEP for the analyzed the analyzed transient test. a: the original predicted NOx trace is reported. b:
transient test. a: raw experimental BMEP trace is reported. b: the engine the predicted NOx trace has been filtered using a time constant of 5s.
inertial term was subtracted from the raw experimental BMEP trace.

With reference to the prediction of BMEP (Fig. 10), a preliminary

comparison between the predicted and raw experimental trends
revealed some discrepancies (see Fig. 10a), especially over the speed
ramps at constant torque (highlighted with a red circle in Fig. 10a),
and the values of RMSE was of the order of 0.55 bar. It was found
that these BMEP peaks represent the inertial contribution due to the
engine speed variation. Therefore these terms are not related to fuel
injection. If the inertial term (i.e., J dω/dt, where J is the engine
moment of inertia and ω is the engine angular speed) is subtracted
from the raw experimental BMEP trace, a better agreement is
observed (see Fig. 10b) between the experimental and predicted
trends of BMEP, as the RMSE value decreases to a value of 0.28 bar.

Figure 12. Predicted and experimental trend of cumulated engine-out NOx

emissions for the analyzed transient test.
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017) 2299

With reference to the prediction of the instantaneous engine-out NOx BMEP/NOx Controller
emissions (Fig. 11), a preliminary comparison between the predicted
and the experimental trends also revealed some discrepancies (see Steady-State Conditions
Fig. 11a). In particular, the experimental trend seems to be much The controller of BMEP and NOx is based on the inversion of the
smoother than the predicted one, as though it was a result of a time real-time combustion model, according to the scheme reported in Fig. 4.
filtering. In order to check this effect, the predicted NOx trend has In particular, the controller is able to predict the values of the injected
been filtered over time using the Matlab function ‘filter’. It was found fuel quantity qf,inj and of the main injection timing SOImain that allow
that a time constant τc=5s leads to the best agreement between the desired targets of BMEP and engine-out NOx emissions to be reached.
predicted and experimental engine-out NOx trends, as can be seen in
Fig. 11b (RMSE = 28 ppm). The physical reason of the time filtering This approach has been validated at steady-state conditions over the
of the experimental NOx trace may be due to the mixing of the engine map tests and SOImain/pf sweep tests shown in Fig. 2. EGR-
exhaust gases that occurs in the pipes between the engine and the sweep tests has not been considered, as the NOx variation in those
exhaust gas analyzer of the test bench. Therefore, the NOx emission tests was obtained through EGR variation and not SOImain variation. In
trend obtained from the measurement of the exhaust gas analyzer particular, the experimental values of BMEP and NOx emissions were
seems not to be representative of the actual dynamics which occurs in set as targets of the controller, and the values of qf,inj and SOImain
the exhaust manifold of the engine. In order to check this effect, a predicted by the controller were compared to the real actuated values.
NOx sensor with high frequency response (not available for the Constant values of qf,inj and SOImain were set as initial conditions. The
considered tests) should be installed in the engine. results are shown in Figs. 13 and 14. In particular, Fig. 13 reports the
predicted vs. experimental values of injected fuel mass, while Fig. 14
Figure 12 reports the comparison between the experimental and reports the predicted vs. experimental values of SOImain.
predicted trends of the cumulated engine-out NOx emissions, where
the predicted trace was obtained using the filtered predicted NOx
instantaneous emissions (i.e., those reported in Fig. 11b). A good
agreement is observed, as the cumulated error is of the order of 0.5%.

Sensitivity Analysis to Input Variables

A sensitivity analysis has been carried out, for the engine map
steady-state tests, in order to check the effect of deviations in the
actual/state input variables with respect to the nominal values. The
results are reported in Table 8. In particular, Table 8 reports, for each
model outcome, the deviation in the values of the RMSE with respect
to the baseline case, in which the nominal values of the input
variables are used.

Table 8. Sensitivity analysis of the main model outcomes with respect to

deviations in the input state variables. The table reports, for each model
outcome, the deviation in the values of the RMSE with respect to the baseline a.
case, in which the nominal values of the input variables are used. x: virtually
no effect.

b.
Figure 13. Inverted model: predicted vs. experimental values of injected fuel
mass for the engine map tests (a) and SOImain/pf sweep tests (b) shown in Fig.
2. The experimental values of BMEP were set as targets.
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

2300 Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017)

The inverted model is highly accurate in estimating the injected The sensitivity of NOx variation with respect to SOImain variation has
quantity, as the RMSE values are of the order of 0.7 mg/stroke and been calculated for the experimental SOImain/pf sweep tests shown in
0.5 mg/stroke for the engine map tests and SOImain/pf sweep tests, Fig. 2, and the results are reported in Fig. 15 for a high-load key-
respectively (Fig. 13). point (Fig. 15a) and a low-load key-point (Fig. 15b). In particular, the
figure reports the values of ΔNOx/ΔSOImain as a function of ΔSOImain,
With reference to the estimation of SOImain, the inverted model that is, the difference between the actual value of SOImain during the
accuracy is good, as the RMSE values are of the order 1.4 deg and sweep test and the nominal value of SOImain for the given key-point.
0.75 deg for the engine map tests and SOImain/pf sweep tests, The results indicate a much larger sensitivity of NOx with respect to
respectively (Fig. 14). However, with reference to the engine map ΔSOImain for the higher-load point (up to 60-80 ppm/deg), and a
tests (Fig. 14a), a larger dispersion is shown in some regions of the smaller sensitivity for the low-load key-point (10-20 ppm/deg).
map. It was verified that the larger dispersion occurs especially at Moreover, it can be seen in Fig. 15 that the sensitivity of NOx is not
lower loads. This larger dispersion may be explained as follows. symmetrical with respect to a positive or negative variation of
First, at lower loads the absolute values of NOx emissions are SOImain: an advance in injection timing has a larger impact on NOx
generally low (some tens of ppm), and therefore the relative error of variation than a delay.
the NOx model becomes significant. Second, at lower loads the
sensitivity of NOx variation with respect to SOImain variation is much The previous analysis suggests that a SOI-based control of NOx
lower than for higher loads. Both effects have an impact on the emissions is more effective for medium-high load conditions than for
accuracy in the values of SOImain predicted by the inverted model. low-load conditions.

a. a.

b. b.

Figure 14. Inverted model: predicted vs. experimental values of SOImain for Figure 15. ΔNOx/ΔSOImain as a function of ΔSOImain for the experimental
the engine map tests (a) and SOImain/pf sweep tests (b) shown in Fig. 2. The SOImain/pf sweep tests shown in Fig. 2. Negative values of ΔSOImain indicate
experimental values of NOx emissions were set as targets. that SOI is anticipated with respect to the nominal SOImainvalue for the
considered key-point.
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017) 2301

Transient Conditions Number of Iterations Required for Model Inversion

With reference to transient conditions, only the BMEP controller could The number of iterations required for model inversion depends on the
be tested over the ramps shown in Fig. 8. In particular, the initial conditions for qf,inj and SOImain and on the error thresholds
experimental values of BMEP were set as targets, and the predicted εNOX, εBMEP (see Fig. 4). With reference to the results reported in Figs.
fuel quantities were compared to the experimental ones. The SOImain 13, 14, 15, 16, error thresholds of 0.1 bar for BMEP and 20 ppm for
values were instead kept equal to the nominal values derived from the NOx emissions were set to stop the iterative process. It was verified
engine maps. In fact, the NOx controller could not be tested, as the that, on average, 3-4 iterations are required if the initial conditions of
measured NOx emissions trends were found not to be representative qf,inj and SOImain are derived from the nominal engine-map values,
of the actual engine-out levels (see Fig. 11), due to a smoothing effect while about 6 iterations are required if the initial conditions of qf,inj
that may be related to the mixing of the exhaust gases in the pipes and SOImain are set as constant values. Therefore, it may be
between the engine and the gas analyzer of the test bench. Therefore, advantageous to set the nominal engine map-based values as initial
the measured NOx emission trends could not be set as targets for the conditions in order to minimize the number of required iterations.
NOx controller. The results of the BMEP controller are shown in Fig.
16, which reports the predicted vs. experimental values of the injected
fuel mass (Fig. 16a), as well as the comparison between the
HiL and Computational Time
experimental BMEP values, which were set as targets, and the The BMEP/NOx control was developed in Matlab/Simulink
predicted BMEP values obtained from model inversion. Fig. 16a environment and was then implemented on a rapid prototyping (RP)
shows that the BMEP controller accuracy is good, as the RMSE value device (i.e., ETAS ES910), through ETAS Intecrio software, in order
of the injected fuel quantity is equal to 1.1 mg/stroke, while Fig. 16b to perform Hardware-in-the-Loop testing and to check the
confirms that the BMEP targets have been achieved over the transient. computational time required by the control. The RP device was
coupled to a real-time engine emulator represented by an NI PXI
device equipped with a real-time engine model (see [1]).

The aim of this phase was to test the real-time capability of the
controller, for the subsequent implementation on the real engine. The
rapid prototyping activity on the engine is currently ongoing and the
results will be shown in a future paper.

The results of the HiL activity, in terms of computational time

required by the controller on the ETAS ES910 device per iteration,
are shown in Table 9.

In particular, Tab. 9a reports the average computational time required

for the calculation of the main outcomes of the real-time combustion
model per iteration, when implemented on the ETAS ES910 RP
device, along with the RMSE values. A crank angle integration step
of 1.0 deg was used. The reported times are progressive (e.g., the
a.
time required to evaluate NOx also includes the time required to
evaluate MFB50 and pressure-related metrics).

It should be noted that this analysis has been carried out considering
the engine map tests only, therefore the RMSE values are slightly
different from those reported in Figs. 5, 6, 7, which refer to the whole
experimental dataset.

It can be seen that the required computational time to evaluate the

heat release and pressure-related metrics is quite low, i.e., about 350
μs per iteration. The total required computational time to estimate
NOx is instead of the order of 1.4 ms per iteration, due to the 3-zone
thermodynamic model that is time consuming.

A sensitivity analysis was made in order to verify the impact of using

a larger computational step for the 3-zone thermodynamic model, by
keeping the computational step of the heat release and pressure
b.
models equal to 1.0 deg. The results are shown in Table 9b, and
Figure 16. a: predicted and experimental values of the injected fuel mass when indicate that it may be possible to adopt a step of 4 deg without any
testing the BMEP controller over the ramps shown in Fig. 8. b: predicted and
experimental values of BMEP over the same ramps.
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

2302 Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017)

accuracy loss in the estimation of NOx emissions (at least at metrics, such as MFB50, IMEP, BMEP) as well as the NOx engine-
steady-state conditions), and this leads to a reduction in the total out levels. A 3-zone thermodynamic model has been used to estimate
required computational time to a value of about 700 μs per iteration. the burned gas temperatures, which are required by the NOx model.

Table 9. a: average computational time required for the calculation of the main The work has been carried out in the frame of a research activity in
outcomes of the combustion model, when implemented on the ETAS ES910 RP collaboration with FPT Industrial, and the experimental data were
device. The reported times are progressive (e.g., the time required to evaluate acquired at the highly dynamic test bench of ICEAL-PT (Internal
NOx also includes the time required to evaluate MFB50 and pressure-related Combustion Engines Advanced Laboratory - Politecnico di Torino).
metrics). b: average computational time required for the calculation of the NOx
emissions, using different integration steps in the 3-zone thermodynamic model.
First, the real-time combustion model has been calibrated and
The analysis was made considering engine map tests.
assessed at both steady-state and transient conditions, over some
speed/load ramps. It was verified that the model provides an accurate
estimation of the combustion metrics and NOx emissions, as the
values of RMSE (root mean squared error) are of the order of 0.9 deg
for MFB50, 0.18 bar for IMEP, 0.15 bar for BMEP and 35 ppm for
NOx emissions, at steady-state conditions. The accuracy does not
change significantly in transient operation.

The model has then been inverted in order to realize the BMEP/NOx
controller. The latter has been tested at steady-state conditions, using
the experimental values of BMEP and NOx emissions as targets. The
predicted values of injected quantity and injection timing of the main
pulse have been compared with the experimental ones, and a good
accuracy was found. In particular, the RMSE value of the injected
quantity was of the order of 0.5-0.7 mg/stroke, while the RMSE value
of the main injection timing was of the order of 0.7-1.4 deg.

Finally, the controller has been implemented on a rapid prototyping

device (ETAS ES910) and tested in Hardware-in-the-Loop in order to
check its real-time capability. It was found that a computational time
of 700 μs per iteration can be achieved using a computational step of
1 deg for the pressure model and of 4 deg for the 3-zone
thermodynamic model, without any significant deterioration in the
model accuracy with respect to the baseline case, in which an
integration step of 0.1 deg is adopted. This result was also achieved
by adopting a new numerical discretization scheme for the in-cylinder
pressure model, which allowed the computational step to be relaxed
with a virtually negligible loss of accuracy in the calculation of the
pressure-related metrics and NOx emissions.

FUTURE WORK The developed controller will be tested on the engine installed at the
The BMEP/NOx controller is going to be tested on the engine test bench in the near future, through rapid prototyping.
installed at the test bench through rapid prototyping, and the results
will be shown in the near future. Moreover, the controller will be
further developed by including additional control variables, such as
REFERENCES
1. Finesso, R., Marello, O., Misul, D., Spessa, E. et al., "Development and
EGR and boost pressure. Assessment of Pressure-Based and Model-Based Techniques for the
MFB50 Control of a Euro VI 3.0L Diesel Engine," SAE Int. J. Engines
10(4):2017, doi:10.4271/2017-01-0794.
SUMMARY/CONCLUSIONS 2. Lucchini, T., D'Errico, G., Onorati, A., Frassoldati, A. et al., "Modeling
Non-Premixed Combustion Using Tabulated Kinetics and Different
A model-based approach to control BMEP (Brake Mean Effective Fame Structure Assumptions," SAE Int. J. Engines 10(2):593-607, 2017,
Pressure) and NOx emissions has been developed and assessed on a doi:10.4271/2017-01-0556.
3. Maghbouli, A., Lucchini, T., D'Errico, G., Izadi Najafabadi, M. et al.,
FPT F1C 3.0L Euro VI diesel engine for heavy-duty applications. The "Numerical Investigation of PPCI Combustion at Low and High Charge
controller provides the injected fuel quantity and the timing of the Stratification Levels," SAE Technical Paper 2017-01-0739, 2017,
main pulse, for given targets of BMEP and engine-out NOx levels. doi:10.4271/2017-01-0739.
4. Fontanesi, F., Giacopini, M., “Multiphase CFD-CHT optimization
The controller is based on the inversion of a zero-dimensional of the cooling jacket and FEM analysis of the engine head of a V6
real-time combustion model, which is capable of simulating the HRR diesel engine”, Applied Thermal Engineering 52 : 293-303, 2013, doi:
(heat release rate) and in-cylinder pressure (along with the related 10.1016/j.applthermaleng.2012.12.005.
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017) 2303

5. Jemni, M.A., Kantchev, G., Abid, M.S., “Influence of intake manifold 23. Brusca, S., Lanzafame, R., and Messina, M., "A Combustion Model for
design on in-cylinder flow and engine performances in a bus diesel ICE by Means of Neural Network," SAE Technical Paper 2005-01-2110,
engine converted to LPG gas fuelled, using CFD analyses and 2005, doi:10.4271/2005-01-2110.
experimental investigations.”, Energy 36:2701-2715, 2011, doi: 24. Brahma, I., He, Y., and Rutland, C., "Improvement of Neural Network
10.1016/j.energy.2011.02.011. Accuracy for Engine Simulations," SAE Technical Paper 2003-01-3227,
6. Lee, C.H., Reitz, R.D., “CFD simulations of diesel spray tip penetration 2003, doi:10.4271/2003-01-3227.
with multiple injections and with engine compression ratios up to 25. Finesso, R., Spessa, E., Yang, Y., Alfieri, V. et al., "HRR and MFB50
100:1”, Fuel 111:289-297, 2013, doi: 10.1016/j.fuel.2013.04.058. Estimation in a Euro 6 Diesel Engine by Means of Control-Oriented
7. Perini, F., Dempsey, A., Reitz, R., Sahoo, D. et al., "A Computational Predictive Models," SAE Int. J. Engines 8(3):1055-1068, 2015,
Investigation of the Effects of Swirl Ratio and Injection Pressure on doi:10.4271/2015-01-0879.
Mixture Preparation and Wall Heat Transfer in a Light-Duty Diesel 26. Chmela, F. and Orthaber, G., "Rate of Heat Release Prediction
Engine," SAE Technical Paper 2013-01-1105, 2013, doi:10.4271/2013- for Direct Injection Diesel Engines Based on Purely Mixing
01-1105. Controlled Combustion," SAE Technical Paper 1999-01-0186, 1999,
8. Rakopoulos, C.D., Kosmadakis, G.M., Pariotis, E.G., “Investigation doi:10.4271/1999-01-0186.
of piston bowl geometry and speed effects in a motored HSDI diesel 27. Egnell, R., "A Simple Approach to Studying the Relation between Fuel
engine using a CFD against a quasi-dimensional model”, Energy Rate Heat Release Rate and NO Formation in Diesel Engines," SAE
Conversion and Management 51:470-484, 2010, doi: 10.1016/j. Technical Paper 1999-01-3548, 1999, doi:10.4271/1999-01-3548.
enconman.2009.10.010.
28. Ryan, T. and Callahan, T., "Homogeneous Charge Compression
9. Mobasheri, R., Peng, Z., Mirsalim, S.M., “Analysis the effect of Ignition of Diesel Fuel," SAE Technical Paper 961160, 1996,
advanced injection strategies on engine performance and pollutant doi:10.4271/961160.
emissions in a heavy duty DI-diesel engine by CFD modeling”,
29. Ericson, C., Westerberg, B., Andersson, M., and Egnell, R., "Modelling
International Journal of Heat and Fluid Flow 33:59-69, 2012, doi:
Diesel Engine Combustion and NOx Formation for Model Based
10.1016/j.ijheatfluidflow.2011.10.004.
Control and Simulation of Engine and Exhaust Aftertreatment Systems,"
10. Cipolla, G., Vassallo, A., Catania, A., Spessa, E. et al., "Combined SAE Technical Paper 2006-01-0687, 2006, doi:10.4271/2006-01-0687.
application of CFD modeling and pressure-based combustion diagnostics 30. Catania, A.E., Finesso, R., Spessa, E., “Predictive Zero-Dimensional
for the development of a low compression ratio high-performance diesel
Combustion Model for DI Diesel Engine Feed-Forward Control”,
engine," SAE Technical Paper 2007-24-0034, 2007, doi:10.4271/2007-
Energy Conversion and Management. 52(10):3159-3175, 2011,
24-0034.
doi:10.1016/j.enconman.2011.05.003.
11. Lešnik, L., Iljaz, J., Hribernik, A., Kegl, B., “Numerical and 31. Heywood, J.B., “Internal Combustion Engine Fundamentals”, McGraw-
experimental study of combustion, performance and emission Hill Intern. Editions, 1988.
characteristics of a heavy-duty DI diesel engine running on diesel,
biodiesel and their blends”, Energy Conversion and Management 32. D’Ambrosio, S., Finesso, R., Fu, L., Mittica, A., Spessa, E., “A Control-
81:534-546, 2014, doi: 10.1016/j.enconman.2014.02.039. Oriented Real-Time Semi-Empirical Model for the Prediction of NOx
Emissions in Diesel Engines”, Applied Energy 130:265-279; 2014,.doi:
12. Kéromnès, A., Delaporte, B., Schmitz, G., Le Moyne, L., “Development 10.1016/j.apenergy.2014.05.046
and validation of a 5 stroke engine for range extenders application”,
Energy Conversion and Management 82:259-267, 2014, doi: 10.1016/j. 33. Chen, S. and Flynn, P., "Development of a Single Cylinder Compression
enconman.2014.03.025. Ignition Research Engine," SAE Technical Paper 650733, 1965,
doi:10.4271/650733.
13. Baratta, M., Finesso, R., Kheshtinejad, H., Misul, D. et al., "Use of an
Innovative Predictive Heat Release Model Combined to a 1D Fluid- 34. Finesso, R. and Spessa, E., "Real-Time Predictive Modeling
Dynamic Model for the Simulation of a Heavy Duty Diesel Engine," of Combustion and NOx Formation in Diesel Engines Under
SAE Int. J. Engines 6(3):1566-1579, 2013, doi:10.4271/2013-24-0012. Transient Conditions," SAE Technical Paper 2012-01-0899, 2012,
doi:10.4271/2012-01-0899.
14. Baratta, M., Finesso, R., Misul, D., and Spessa, E., "Comparison
between Internal and External EGR Performance on a Heavy Duty
Diesel Engine by Means of a Refined 1D Fluid-Dynamic Engine
Model," SAE Int. J. Engines 8(5):1977-1992, 2015, doi:10.4271/2015- CONTACT INFORMATION
24-2389.
15. Baratta, M., Kheshtinejad, H.,, Laurenzano, D., Misul, D., et al., Prof. Ezio Spessa
“Modelling aspects of a CNG injection system to predict its behavior IC Engines Advanced Laboratory
under steady state conditions and throughout driving cycle simulations”,
Journal of Natural Gas Science and Engineering 24:52-63, doi: Dipartimento Energia, Politecnico di Torino
10.1016/j.jngse.2015.03.010. c.so Duca degli Abruzzi, 24 - 10129 Torino (Italy)
16. Montenegro, G., Onorati, A., Piscaglia, F., and D'Errico, G., "Integrated phone: +39-011-090.4482
1D-MultiD Fluid Dynamic Models for the Simulation of I.C.E. Intake
and Exhaust Systems," SAE Technical Paper 2007-01-0495, 2007, [email protected]
doi:10.4271/2007-01-0495.
17. Rakopoulos, C. and Giakoumis, E., "Review of Thermodynamic Diesel
Engine Simulations under Transient Operating Conditions," SAE DEFINITIONS/ABBREVIATIONS
Technical Paper 2006-01-0884, 2006, doi:10.4271/2006-01-0884.
18. Finesso, R., Spessa, E., and Yang, Y., "Development and Validation of ANN - Artificial neural network
a Real-Time Model for the Simulation of the Heat Release Rate, In-
Cylinder Pressure and Pollutant Emissions in Diesel Engines," SAE Int. BMEP - Brake Mean Effective Pressure
J. Engines 9(1):322-341, 2016, doi:10.4271/2015-01-9044. CA - crank angle
19. Gani, E. and Manzie, C., "Indicated Torque Reconstruction from
Instantaneous Engine Speed in a Six-cylinder SI Engine Using CFD - Computer Fluid-Dynamics
Support Vector Machines," SAE Technical Paper 2005-01-0030, 2005,
doi:10.4271/2005-01-0030. cp - specific heat at constant pressure
20. Asik, J., Peters, J., Meyer, G., and Tang, D., "Transient A/F Estimation cv - specific heat at constant volume
and Control Using a Neural Network," SAE Technical Paper 970619,
1997, doi:10.4271/970619. DT - Dwell-time
21. Bennett C., Dunne J.F., Trimby S., Richardson D., “Engine cylinder ECU - Engine Control Unit
pressure reconstruction using crank kinematics and recurrently-trained
neural networks”, Mechanical systems and signal processing 85:126- EGR - Exhaust Gas Recirculation
145, 2016, doi: 10.1016/j.ymssp.2016.07.015.
22. Yusuf Cay, “Prediction of a gasoline engine performance with EOC - end of Combustion
artificial neural network”, Fuel 111:324-331, 2013, doi: 10.1016/j.
fuel.2012.12.040.
EOI - end of injection
 Downloaded from SAE International by Birmingham City Univ, Sunday, August 19, 2018

2304 Finesso et al / SAE Int. J. Engines / Volume 10, Issue 5 (December 2017)

EVO - Exhaust Valve Opening Qfuel - chemical energy associated with the injected fuel
FMEP - Friction Mean Effective Pressure Qht - heat transfer between the charge and the walls
FPT - Fiat Powertrain Technologies Qht,glob - global heat transfer between the charge and the walls over
h - Specific enthalpy the combustion interval

Habs - absolute humidity of the air Qnet - net heat release

HL - lower heating value of the fuel qf,inj - total injected fuel volume quantity per cycle/cylinder

HiL - Hardware-in-the-Loop qpil - injected fuel volume quantity of the pilot injection

HRR - Heat Release Rate qpil,tot - total injected fuel volume quantity of the pilot injections

ICEAL-PT - Internal Combustion Engines Advanced Laboratory at R2 - squared correlation coefficient

the Politecnico di Torino RMSE - root mean square error
IMEP - Indicated Mean Effective Pressure SOC - start of combustion
IMEP360 - Gross Indicated Mean Effective Pressure SOI - electric Start Of Injection
IMEP720 - Net Indicated Mean Effective Pressure SOImain - electric Start Of Injection of the main pulse
IVC - Intake Valve Closing SVM - Support vector machine
K - combustion rate coefficient t - time
m - mass; compression phase polytropic coefficient T - temperature
mf,inj - total injected fuel mass per cycle/cylinder Tb,MFB50 - temperature of the burned gas zone at MFB50
m’ - expansion phase polytropic coefficient tgt - target
ṁf ,inj - fuel injection rate Tint - intake manifold temperature
MFB50 - crank angle at which 50% of the fuel mass fraction has V - volume
burned VGT - Variable Geometry Turbocharger
n - engine rotational speed
O2 - intake charge oxygen concentration GREEK SYMBOLS
p - pressure εBMEP - BMEP error threshold to stop iterations for model inversion

pexh - exhaust manifold pressure εNOx - NOx error threshold to stop iterations for model inversion

pf - injection pressure γ = cp/cv - specific heat ratio

PFP - Peak firing pressure ρ - density

pint - intake manifold pressure ρSOI - in-chamber ambient density evaluated at the SOI instant

pIVC - In-cylinder pressure at IVC ρSOC - in-chamber ambient density evaluated at the SOC instant

pil - pilot injection τ - ignition delay coefficient

PMEP - Pumping Mean Effective Pressure τmain - ignition delay coefficient of the main pulse

q - injected fuel volume quantity τpil - ignition delay coefficient of the pilot pulse

Qch - chemical heat release

Qf,evap - energy associated to fuel evaporation

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or
otherwise, without the prior written permission of SAE International.

Positions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE International. The author is solely responsible for the content of the paper.