Air Charge Estimation in Turbocharged Spark Ignition Engines
Air Charge Estimation in Turbocharged Spark Ignition Engines
Air Charge Estimation in Turbocharged Spark Ignition Engines
Per Andersson
c 2005 Per Andersson
[email protected]
http://www.fs.isy.liu.se/
Department of Electrical Engineering,
Link
oping University,
SE581 83 Link oping,
Sweden.
Abstract
Turbocharged (TC) spark-ignited (SI) engines are popular as they combine
high power output with good fuel economy. Furthermore, their emissions can
be successfully reduced using a three way catalyst (TWC) provided that the air-
fuel ratio is precisely controlled. As the air-fuel ratio depends on the precision of
the air charge estimate, this thesis is devoted to the systematic improvement of
the cylinder air charge (CAC) estimation on TC SI engines. A second objective
is to provide the design engineer with a flexible framework for CAC estimation
that easily can be adapted to various engines.
The CAC is not measurable and it is therefore estimated using a model.
Part I concentrates on stationary conditions and examines existing air charge
estimation methods using engine experiments where the wastegate is opened
and closed. Measurements show that the existing methods are insufficient for
TC SI engines since the CAC depends on exhaust backpressure and charge
cooling from evaporating fuel. A new 2-parameter CAC model which accounts
for these effects is developed and the validation shows that the error is reduced
from 10% to 3%.
Part II deals with transient conditions and a dynamical component-based
model is developed for the gas flow systems of TC SI engines. The physical
structure of different TC SI engines is similar and these similarities are exploited
in the developed model. The division into components provides the basis for a
flexible framework that enables a straightforward adaption to various engines.
It is described how the model parameters are systematically fitted using an
engine map and maps from the turbocharger manufacturer. The accuracy of
the model is good and the stationary error is less than 10% on the intake side.
An observer that estimates the CAC, given available measurements, is sug-
gested. It is shown that the system is locally structurally observable from arbi-
trarily measured model states. Further, a specific combination of signals that
is most suitable for CAC estimation is pointed out. The developed observer is
based on the constant gain extended Kalman filter (CGEKF) and a systematic
method for selecting the design parameters in CGEKF filters is proposed. The
method only requires an engine map and the variance of the signals considered
for observer feedback. Several different combinations of observer feedback sig-
nals are studied and it is shown that the observer is capable of estimating the
model states. The design method is successfully tested on two different engines.
Finally, the developed model and observer is used for model-based air-fuel
ratio control. A TC SI engine is controlled by the proposed controller in real-
time and the transient deviations from = 1 are less than 7% in very rapid
throttle transients.
ii
iii
Acknowledgments
This work has been performed at Vehicular Systems and I am very grateful to
professor Lars Nielsen for letting me join his excellent group and for all the
support, discussions, and encouragement he has provided during this work.
First I would like to thank my supervisor Dr. Lars Lasse Eriksson, who is
greatly acknowledged for guiding me into the research field of mean value models
and for all careful proofreadings of research papers and this thesis. Thank you
Lasse for your patient guidance and encouragement during the project! Then, I
thank Dr. Erik Frisk for his help with LATEX and Emacs and for the interesting
discussions and proofreadings especially around the structural analysis, and
observer design chapters. Marcus Klein and Ylva Nilsson were kind to proofread
early versions of the manuscript.
This work would not have been possible without all the help from our re-
search engineers over the years Andrej Perkovic, Jan Brug ard, Mats Karlsson,
and Martin Gunnarsson who take good care of our research laboratory. Erik
Sunneg
ardh, Per Oberg, and Christer Rosenquist are also acknowledged as they
all have contributed in the development of our successful real-time control envi-
ronment. I thank Kristin Wiberg and Carolina Fr oberg as they made sure that
the administrative work always went smoothly. The staff of Vehicular Systems
is also acknowledged for creating a nice research atmosphere and amusing coffee
breaks.
I also thank GM-Powertrain Sweden AB and Mecel AB for show-
ing great interest, for the hardware supply, and support during the project.
The Swedish Agency for Innovation Systems (VINNOVA) is acknowledged for
their support through the center of excellence ISIS (Information Systems for
Industrial Control and Supervision).
Finally I thank my mother Karin, father Ragnar, and my brother Ulf for
their support and encouragement.
Link
oping, November 2005
Per Andersson
iv
Contents
1 Introduction 1
1.1 Precise Air-Fuel Ratio Control is Necessary . . . . . . . . . . . . 2
1.2 Air-Fuel Ratio Control Challenges . . . . . . . . . . . . . . . . . 3
1.2.1 Feedback and Feedforward Control is Necessary . . . . . . 3
1.2.2 Desired Properties: Low Cost and Fast Calibration . . . . 3
1.3 Problem Formulation and Solution Outline . . . . . . . . . . . . 4
1.4 List Of Publications . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Background 7
2.1 Why Focus on Air Estimation? . . . . . . . . . . . . . . . . . . . 7
2.1.1 Couplings Between Intake and Exhaust Side . . . . . . . . 8
2.1.2 The Necessity of CAC Prediction . . . . . . . . . . . . . . 8
2.2 Air Charge Estimation Methods . . . . . . . . . . . . . . . . . . 9
2.3 Measured Air-Mass Flow Principle . . . . . . . . . . . . . . . . . 10
2.4 Speed-Density Principles . . . . . . . . . . . . . . . . . . . . . . . 10
2.4.1 Filtering and Sensor Dynamics . . . . . . . . . . . . . . . 12
2.4.2 Side Effects of the Wastegate . . . . . . . . . . . . . . . . 12
2.5 Cylinder Air Charge Estimation Principle Selection . . . . . . . . 13
2.6 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6.1 Mean Value Engine Models . . . . . . . . . . . . . . . . . 16
2.6.2 Measuring Air-Fuel Ratio . . . . . . . . . . . . . . . . . . 17
2.6.3 Observers and Air-Fuel Ratio Control Aspects . . . . . . 17
v
vi
I Stationary 19
3 Speed-Density Methods and Different Wastegate Settings 21
3.1 Air Intake System Modeling . . . . . . . . . . . . . . . . . . . . . 22
3.1.1 Intake Manifold Pressure Dynamics . . . . . . . . . . . . 23
3.1.2 Air-Mass Flow Into the Intake Manifold . . . . . . . . . . 23
3.1.3 Port Air-mass Flow . . . . . . . . . . . . . . . . . . . . . 25
3.2 Intake Manifold Pressure Observers . . . . . . . . . . . . . . . . . 27
3.2.1 Port Air-mass Flow Estimation from Observed Intake Man-
ifold Pressure . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.2 Port Air-mass Flow Validation . . . . . . . . . . . . . . . 27
3.2.3 Observer Test Case Description . . . . . . . . . . . . . . . 28
3.3 Observer with Proportional Feedback . . . . . . . . . . . . . . . . 29
3.3.1 Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.2 Steady-State Properties . . . . . . . . . . . . . . . . . . . 31
3.4 Port Air-mass Flow Observer with Additive Offset in vol . . . . 32
3.4.1 Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.2 Steady-State Properties . . . . . . . . . . . . . . . . . . . 35
3.5 Observer with Air-Mass-Offset Estimation . . . . . . . . . . . . . 35
3.5.1 Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
II Dynamics 79
6 A Mean Value Model of a TC SI Engine 81
6.1 Model Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.1.1 Model Simplifications . . . . . . . . . . . . . . . . . . . . 84
6.1.2 Model Inputs . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.1.3 Model States . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.1.4 Data Sets for Parameter Estimation and Validation . . . 87
6.2 Incompressible Flow Restrictions . . . . . . . . . . . . . . . . . . 87
6.2.1 Air Filter Model . . . . . . . . . . . . . . . . . . . . . . . 88
6.2.2 Intercooler . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.2.3 Exhaust System . . . . . . . . . . . . . . . . . . . . . . . 92
6.3 Compressible Flow Restrictions . . . . . . . . . . . . . . . . . . . 92
6.3.1 Throttle Flow . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.3.2 Wastegate Flow . . . . . . . . . . . . . . . . . . . . . . . . 102
6.4 Engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.4.1 Port Air-mass Flow Model . . . . . . . . . . . . . . . . . . 102
6.4.2 Exhaust Mass-flow Model . . . . . . . . . . . . . . . . . . 104
6.4.3 Engine Torque . . . . . . . . . . . . . . . . . . . . . . . . 105
6.5 Turbocharger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.6 Compressor Model . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.6.1 Compressor Air-mass Flow Models . . . . . . . . . . . . . 108
6.6.2 Compressor Efficiency Model . . . . . . . . . . . . . . . . 112
6.6.3 Compressor Temperature Out . . . . . . . . . . . . . . . . 114
6.6.4 Compressor Torque . . . . . . . . . . . . . . . . . . . . . . 114
6.7 Turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.7.1 Turbine Mass-flow . . . . . . . . . . . . . . . . . . . . . . 116
6.7.2 Turbine Efficiency . . . . . . . . . . . . . . . . . . . . . . 118
6.7.3 Turbine Temperature Out . . . . . . . . . . . . . . . . . . 118
6.7.4 Turbine Torque . . . . . . . . . . . . . . . . . . . . . . . . 120
6.8 Model Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
viii
10 Conclusions 195
References 197
A Nomenclature 205
Introduction
Today, everyone want cars that are less harmful to the environment and have
better fuel economy. As a small engine use less fuel, it is easy to save fuel by
making the engine smaller. However, a small engine does not meet the power
expectation of the driver during e.g. overtakings. The maximum power of the
engine depends on the amount of fuel it can burn and this in turn depends on
the availability of air. Thus, more air is necessary to increase the engine power.
One way to increase the air to the engine is to add a turbocharger (TC) that
simply raise the pressure of the air. Hence, it is possible to combine the fuel
economy of the small engine with the maximum power of a bigger engine and
this concept is called downsizing supercharging (Guzzella et al., 2000).
Here spark ignition (SI) engines are considered as the emissions of carbon
monoxide (CO), hydro-carbons (HC), and nitrogen oxide (NOx ) can be suc-
cessfully reduced using a three way catalyst (TWC). Thus, downsized and
supercharged SI engines is a promising concept that gains a lot of attention in
Europe. For the TWC to be efficient it is necessary to accurately control the
ratio of air and fuel. The air-fuel ratio is the ratio of air-mass and fuel-mass in
the cylinder when the valves are closed. The mass of air inside the cylinder, the
cylinder air charge (CAC), is influenced by the driver pressing the accelerator.
The fuel is injected by electronically controlled valves, so called injectors, that
are located close to the intake valve. An electronic control unit (ECU) deter-
mines the mass of fuel to inject by first estimating the expected CAC. Hence,
a key component for precise air-fuel ratio control and thus lower emissions is
precise cylinder air-charge estimates. Improving the CAC estimates is not an
easy task and therefore the topic of this thesis.
1
2 Chapter 1. Introduction
Figure 1.1: The maximum allowed emissions for gasoline powered passenger cars in Europe.
From EURO-1 (1992) to EURO-5 (proposed, 2008) the maximum allowed emissions have been
more than halved, which requires more accurate engine control systems. Note that for Euro 1
and Euro 2, the limit is on the sum of the HC and NOx , while there are separate limits for
HC and NOx from Euro 3 and on .
To reach these low emissions, a three way catalyst (TWC) is used. Since the
TWC is only highly efficient in a narrow air-fuel ratio band, as shown in Fig-
ure 1.2, accurate air-fuel ratio control is necessary. In Figure 1.2 the TWC
efficiency is shown as a function of normalized air-fuel ratio . When there is
just enough air to fully oxidize all of the fuel is equal to one. If the mixture
is rich < 1, there are excess fuel and carbon monoxide is formed since there
is not enough oxygen to fully oxidize the fuel to carbon dioxide. On the other
side, for lean mixtures > 1, undesirable nitrogen oxides are instead formed.
Therefore, accurate air-fuel ratio control is essential (Heywood, 1988; Bauer
et al., 1996; Kiencke and Nielsen, 2000; Mondt, 2000).
1.2. Air-Fuel Ratio Control Challenges 3
CO
1 NO
x
HC
0.8
0.6
Efficiency []
0.2
Figure 1.2: The TWC conversion efficiency for stationary . Top: The conversion efficiency
is high only in a very narrow band around = 1. Especially the NOx conversion efficiency
drops rapidly for lean conditions.
already during the design phase. Thus, it is important to develop methods that
can help the design engineer to cut the development time.
1.2
65 with conventional
method.
[]
60 1.15
55 Transient 1.1
50 1.05
45 1
40 0.95
Figure 1.3: Measured air-fuel ratio at constant engine speed using a rapid step in throttle
where the CAC more than doubles. Two different controllers are used, one conventional
and the suggested controller from Chapter 9. Using the proposed model-based controller the
deviations from = 1 are negligible.
1.5 Contributions
Chapter 3 It is shown using measurements that the volumetric efficiency can
change up to 5% when the wastegate is opened on a TC SI engine. An
observer is proposed and deals with the varying volumetric efficiency that
estimates stationary correct CAC for different wastegate settings.
Chapter 7 It is shown that the mean value engine model from Chapter 6 is lo-
cally structurally observable using feedback from an arbitrarily measured
state. It is also shown that when 3 feedback signals are used then there
are only 6 out of 680 combinations that best observe the necessary states
for cylinder air charge estimation.
Background
This chapter has several objectives: Motivate why the focus is on air charge
estimation, describe the additional air estimation challenges on turbocharged
(TC) engines, and introduce common air-estimation methods. Last, this work
is placed among others in a selection of recent related research.
7
8 Chapter 2. Background
1.15
1.1
Change []
1.05
0.95
0.9
4.3 4.35 4.4 4.45 4.5 4.55 4.6 4.65 4.7 4.75 4.8
Time [s]
Figure 2.1: In this measured throttle step (tip-in) at a constant engine speed of 1800 RPM the
CAC (solid) is compared to the same CAC delayed one revolution. The deviation from one
is the resulting air-fuel ratio error if no prediction is used when the fuel-mass is determined.
Without prediction, there would be a 15% error in in this case!
the important quantity is the CAC at intake valve closing. This means that
there is an inherent delay that is constituted by the computation time, the
injection time, and the duration of the intake stroke. Typically this delay is
around one revolution.
During transients this time-delay becomes an issue and it is necessary to
predict the CAC. Here an example is used to show that the lack of prediction
results in large errors already for small throttle steps. To estimate the expected
air-fuel ratio error due to the delay, the CAC is compared to a one revolution
delayed CAC. The expected error in air-fuel ratio, if no prediction is used, is
the deviation from one in the calculated ratio. In this example, the expected
error is 15%, shown in Figure 2.1, but considerably larger errors occur for faster
transients or at lower engine speeds where longer prediction times are necessary.
Cylinder air charge can be estimated using an air-mass flow sensor and
divide the measurement by the engine speed. This strategy is particular
suitable for naturally aspirated engines with central fuel injection where
the fuel is injected at the throttle (Aquino, 1981).
Next, these air estimation principles are described with emphasis on the condi-
tions in port-injected TC-engines, and in Section 2.5 it is motivated why speed-
density methods are preferred. For a description of symbols and abbreviations,
please consult the nomenclature in Appendix A.
Inputs
0.04
Measured airmass flow is too large
due to intake system filling
0.02 Measured airmass flow at airfilter
Modeled port airmass flow
0
4 6 8 10 12 14 16
Expected Error in Airfuel Ratio
1.5
No stationary error in .
1
0.5
4 6 8 10 12 14 16
Time [s]
Figure 2.2: Simulated step response in throttle at a constant speed of 2000 RPM. Top:
Throttle is opened at 5 seconds and the wastegate opens slightly a few seconds later. Center:
Air-mass flow measured at the throttle compared to the port air-mass flow. During the tran-
sient, the air-filter mass-flow is considerably larger due to the intake system filling. Bottom:
The ratio between the air-flow at the air-filter and at the port. If this measured air-mass flow
would be used to determine injected fuel-mass, the error in would be 30% and the error
remains at a high level for several seconds.
4
x 10 Measured Intake Manifold Pressure
11
Measured
10 Mean value
Filtered
Pressure [Pa]
9
Filtered signal lags the
8 5% Deviations from mean value significantly
mean value in and the filter introduces a
7 measured signal more than 5% error during
the transients!
6
4.2 4.3 4.4 4.5 4.6 4.7 4.8
315
310
Figure 2.3: Rapid tip-in at a constant speed of 1800 RPM. Top: Measured intake manifold
pressure is subjected to noise caused by engine pumping and electronics. When the noise
is attenuated using a filter, the filtered signal lags the mean-value significantly. Bottom:
Measured and modeled intake manifold temperature during the transient. As the sensor has
a time constant of approximately 20 seconds, it misses the fast temperature rise captured by
the model.
Gives smaller transient error for tip ins/outs. See Figure 2.2.
Gives smaller transient error when the wastegate is opened. See Figure 2.5.
Has been shown to give valid transient cylinder air charge estimates on
NA-engines (Smith et al., 1999; Chevalier and M
uller, 2000).
Further, introducing a pressure and temperature observer can solve the speed-
density methods problems with noisy signals and slow sensors.
120
80 Exhaust Pressure
Intake Manifold Pressure
60
0 10 20 30 40 50 60
Time [s]
0 10 20 30 40 50
Time [s]
80
78
Stationary offset Stationary offset
76
74
0 10 20 30 40 50
Time [s]
Figure 2.4: An engine experiment at constant speed and air-mass flow, which shows that
the wastegate setting influences the volumetric efficiency. Top: Exhaust and intake manifold
pressure during opening/closing of the wastegate. Wastegate is opened at 9, 30, and 52 sec-
onds. Center: Measured air-mass flow, Wa . A throttle controller tries to maintain a constant
reference air-mass flow (dashed). As the wastegate is opened, the air-mass flow decreases
momentarily until the air-mass controller has opened the throttle. Bottom: Calculated and
estimated volumetric efficiency. Eq. (2.4) determines the estimated volumetric efficiency and
the calculated is determined using Eq. (2.3). There is a stationary offset between the mapped
and estimated volumetric efficiency when the wastegate is open.
2.6. Related Work 15
100
80
49 49.5 50 50.5 51 51.5 52 52.5 53
Airmass Flows
36
Massflow [g/s]
34
32 Portmass flow
Airfilter
Speeddensity
30
49 49.5 50 50.5 51 51.5 52 52.5 53
Estimation Errors
15
Measured airmass flow
Speeddensity
10
Error [%]
0
49 49.5 50 50.5 51 51.5 52 52.5 53
Figure 2.5: Simulated wastegate step at constant engine speed and the throttle controlled to
maintain constant air-mass flow. The engine model from Chapter 6 is used and the speed-
density air-mass flow is estimated using Eq. (2.2). Top: The wastegate opens at 50 seconds.
Center: The speed-density method captures the dynamics of the port air-mass flow well while
the measured air-mass flow underestimates the flow. Bottom: For the measured air-mass flow,
at the air-filter, the error is almost 15% while the speed-density principle gives an error of
only 2%.
16 Chapter 2. Background
Fuel Dynamics
One factor that influences the air-fuel ratio significantly especially for cold en-
gine conditions is fuel dynamics or fuel puddles. Extensive research on the fuel
dynamics has been performed, where it is commonly described as a two-path
flow (Aquino, 1981; Fozo and Aquino, 1988; Onder et al., 1998; Locatelli et al.,
2004). One part of the fuel from the injector enters the cylinder immediately
and the remaining part of the fuel is stored in a fuel film or a fuel puddle.
2.6. Related Work 17
From the puddle, some of the fuel evaporates and enters the cylinder together
with the fuel from the first part. It has also been argued that the identified
dynamics parameters are very sensitive to unknown time-delays, such as the
exact injection time (Henriksson and Nielsen, 1998).
TWC Control
Recent research shows that emissions can be further reduced using the fact that
the TWC can store a certain amount of oxygen to be used later in the emission
reduction (Brandt et al., 2000; Jones et al., 2000; Yasui et al., 2000; Jones,
2003; Auckenthaler, 2003; Balenovic et al., 2004). By utilizing knowledge of
the oxygen storage capacity together with active control of the oxygen entering
the TWC an HC reduction of over 30% and a 60% reduction of NOx has been
shown (Miyamoto et al., 2002). As the air contains oxygen, knowledge of the
air-fuel ratio and the air-mass flow into the TWC is important to describe the
oxygen level.
Stationary
19
3
Knowledge of port air-mass flow is important for accurate air/fuel ratio control
on port injected SI-engines. Unfortunately it is not directly measurable and
therefore several strategies to estimate it have been proposed, see for example
Hendricks et al. (1992); Chang et al. (1993); Shio and Moskwa (1996); Tseng
and Cheng (1999); Kotwick et al. (1999); Jankovic and Magner (1999). In
Chapter 1 it was established that speed-density methods are suitable for port
air-mass flow estimation. For speed-density methods accurate knowledge of the
intake manifold pressure is crucial and therefore several methods to estimate it
have been proposed.
Intake manifold pressure estimation is a well-studied topic for NA-engines.
On TC-engines there is an additional challenge as the turbocharger couples
the intake side and exhaust side and causes the volumetric efficiency to change
when the wastegate is operated, which was shown in Section 2.1.1. It is de-
sirable to open the wastegate at part load to reduce the pumping losses and
thereby improve fuel economy (Eriksson et al., 2002a). Changes in wastegate
opening will act as a non-measurable disturbance that influence the estimated
intake manifold pressure. Thus, this chapter focuses on estimation of intake
manifold pressure and port air-mass flow for different wastegate settings. In
this case only stationary conditions are considered as the port-air mass flow
then can be validated using any known flow in the intake system, such as a
measured air-mass flow or a modeled flow through a restriction. Here a model
of a restriction, the throttle, is used instead of an expensive air-mass flow meter
to determine the air-mass entering the intake manifold. An additional advan-
tage of the throttle model is that it estimates the mass-flow at the entry of the
21
22 Chapter 3. Speed-Density Methods and Different Wastegate Settings
intake manifold, compared to the air-mass flow sensor that is located far from
the intake manifold.
The port air-mass estimation problem is addressed in the following steps.
First two different port-air mass flow models based on the speed-density prin-
ciple from literature are described and then a third speed-density based model,
which combines the two previous models is presented. To each of the three
models an observer is developed. For the first two observers, solutions from the
literature have been used:
1. A nonlinear observer using proportional feedback from the estimation er-
ror (Hendricks et al., 1992).
2. A nonlinear observer using integration of the estimation error (Tseng and
Cheng, 1999).
3. Using a new port-air mass flow model and by combining the advantages
of the observers above; a third observer is developed that better suits the
conditions in a TC SI-engine with wastegate.
In the evaluation of the observers, the estimated port air-mass flow and the
estimated intake manifold pressure are compared to measured data at steady-
state. Thus, the strategy for port air-mass flow estimation relies on:
A fast pressure sensor in the intake manifold pim . A fast sensor has a
higher bandwidth than in this case the intake manifold pressure dynamics.
Here a fast Kristal 4295A2 pressure sensor is used which has a cut off
frequency of 2 kHz (Kri, 1997).
Measured intake manifold temperature Tim , pressure pic and temperature
Tic before the throttle.
Throttle plate angle .
Measured engine speed N .
Models of parts of the intake system.
For a description of the subscripts and symbol names, please see the nomencla-
ture in Appendix A.
three different port air-mass flow models will be described in Section 3.1.3. Each
of the models are then used in the observer design and the observers capability of
describing port air-mass flow and intake manifold pressure for different settings
of the wastegate are investigated.
dpim
= Kim (Wat (, pim , pic , Tic ) Wcyl ) (3.1)
dt
The port air-mass flow Wcyl is described by one of the models in Equations (3.3,
3.4, 3.5). The flow into the manifold is described by the throttle model,
Eq. (3.2).
Temperature dynamics is also present in the intake manifold during large
pressure transients (Chevalier et al., 2000). In this study, temperature dynamics
is neglected as the focus is on stationary conditions and stationary effects.
1 1
2
1
2
+1 +12
otherwise
The flow through the throttle depends not only on the effective area Aeff but
also on the pressure ratio through Eq. (3.2d). The effective area Aeff () is a
product of the area A () and discharge coefficient Cd (). Here, this product is
modeled as in Nyberg and Nielsen (1997) but using a different parameterization.
The parameters in Eq. (3.2c) are fitted in a least squares sense to mapped engine
data. In the left column of Figure 3.1 the result of the modeled Aeff () is shown.
A systematic relative error is present in the bottom left plot of Figure 3.1.
The relative error is positive for large which indicates that Aeff () could be
slightly improved by including , which is supported in Krysander (2000). This
is further addressed in Chapter 6.
In the lower right plot of Figure 3.1 it is shown that the stationary air-mass
flow into the intake manifold is the same as the measured air-mass flow 6%
for most points. To decrease the effect of throttle model errors, observers for
throttle air-mass flow have been proposed in e.g. Jensen et al. (1997). Here no
throttle air-mass observer has been added as the focus is on the behavior when
the wastegate is operated.
Measured
500 Modeled 0.1
Aeff() [mm2]
400 0.08
300 0.06
200 0.04
100 0.02
0 0
0 10 20 30 40 0 0.02 0.04 0.06 0.08 0.1
Throttle angle [deg] Measured MassFlow [kg/s]
5 5
Model error [%]
0 0
5 5
10 10
0 0.2 0.4 0.6 0.8 1 0 0.02 0.04 0.06 0.08 0.1
Pressure Ratio [] MassFlow [kg/s]
Figure 3.1: Throttle model validation. Left column: Comparison of measured and calculated
Aeff (). The fit is within 5% for most points. Right column: Validation of the throttle
mass-flow. For flows larger than 20 g/s the error is less than 6%.
When throttle air-mass flow is determined through Eq. (3.2a) both the pres-
sure pic and the temperature Tic before the throttle are required. Fortunately,
3.1. Air Intake System Modeling 25
measurements can be used as the dynamics of pic and Tic are considerably slower
than pim , due to the substantially larger volume of the system before the throt-
tle and to the slower dynamics of the turbocharger. The measurements of pic
and Tic are also subjected to less pumping noise.
pim Vd N
Wcylts (N, pim , Tim , vol ) = (vol (N, pim ) + vol ) =
Rim Tim nr
pim Vd N pim Vd N
vol (N, pim ) +vol (3.4)
Rim Tim nr Rim Tim nr
| {z }
Wcyl std
The original purpose of the additive offset vol was to compensate for slow
day-to-day variations and it is therefore assumed to be more slowly varying than
other dynamics. By proper selection of vol this approach is suited to adapt to
26 Chapter 3. Speed-Density Methods and Different Wastegate Settings
90
80
v [%]
70
60 Measured
Model
50
20 40 60 80 100 120 140 160
Intake Manifold Pressure [kPa]
Relative Error
20
15
10
[%]
10
20 40 60 80 100 120 140 160
Intake Manifold Pressure [kPa]
the changing volumetric efficiency for different wastegate settings. Here vol
is estimated using an observer.
pim Vd N pim Vd N
Wcylts = vol (N, pim ) + vol (3.5)
Rim Tim nr Rim Tim nr
| {z } | {z }
Wcyl std CAC, Air-mass offset
If CAC can be estimated using for example an observer, the port air-mass
flow can be written as a sum of the air-mass expected from Eq. (3.3) and the
in-cylinder air-mass offset CAC.
Table 3.1: The table shows the different observer configurations with states, port air-mass
flow model, and which state(s) that are affected by the feedback.
The following measurement signals are used by the observers: Engine speed
N, pressure before the throttle pic , temperature before the throttle Tic , throttle
plate angle , intake manifold pressure pim , and intake manifold temperature
Tim .
Next, the validation method is described together with the test case that is
used for validation. Then the three observers are described and evaluated in
Sections 3.3, 3.4, and 3.5.
analysis, based on the throttle mass-flow Eq. (3.2a), is therefore made to de-
termine whether the observed intake manifold pressure can be used or not to
determine the throttle air-mass flow.
a+1
2
Wat (,pim ,pic ,Tic ) 2 pim
+ ppim
a a
(a + 1)
pim pim pic ic
S = Wat (,pim ,pic ,Tic )
= 2 a+1
pic a
2a ppim pim a
pim
ic
+ pic
pim
As the sensitivity only depends on the pressure ratio pic = means that S is
reformulated into a function of the pressure ratio .
a +1 2
(a + 1) a 2 a
S() = a +1 2
(3.7)
2a a a
Eq. (3.7) is valid for pressure ratios above the critical pressure crit :
1
2
crit
+1
For pressure ratios below crit the sensitivity function is constant and zero
which means that the throttle air-mass flow is insensitive to changes in intake
manifold pressure. Thus the observed intake manifold pressure can only be
used to determine the air-mass flow into the intake manifold for pressure ratios
below crit . Figure 3.3 shows S() and it can be seen that |S()| > 1 for
pressure ratios > 0.75. This means that even small errors in the observed
intake manifold pressure results in a biased ; if > 0.75 where |S()| > 1
and thus the error in the estimated throttle air-mass flow increase! Therefore,
the observed intake manifold pressure is not suitable to determine the throttle
air-mass flow unless it converges to the measured intake manifold pressure. In
addition, when the observers are validated, the stationary error in estimated
port air-mass flow is determined
cyl (N, pim , Tim , . . .) Wat (, pim , pic , Tic )
W
the measured intake manifold pressure is used to calculate Wat . This guarantees
an observer independent validation of the stationary port air-mass flow error.
Sensitivity []
4
6
8
10
12
0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95
0.5
1.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Pressure ratio over throttle
Figure 3.3: Throttle air-mass flow sensitivity to intake manifold pressure for different pressure
ratios = ppim . Top: For high pressure ratios the throttle air-mass flow is very sensitive to
ic
intake manifold pressure errors. Bottom: For low pressure ratios, below = 0.53, the flow is
insensitive to changes in intake manifold pressure as S() = 0.
speed. The test data, shown in Figure 3.4, is measured at 3100 RPM and at
BMEP=5.3 bar. This operating point represents a medium engine speed at
part load where the pumping losses are reduced when the wastegate is opened.
In addition, all data have been low-pass filtered using a zero-phase filter to
suppress noise. During the experiment the air-mass flow is held constant by a
throttle controller. This is clearly visible in the center of Figure 3.4.
Wastegate steps of high amplitude are achieved using a manual control de-
vice instead of the production vacuum actuator. When the wastegate is opened
and closed there are air-mass flow transients at 5 seconds and 21 seconds. What
is furthermore interesting is that the mapped volumetric efficiency does not
match the calculated when the wastegate is open. This shows up as a 5%
steady-state difference between the mapped and measured volumetric efficiency
in the bottom of Figure 3.4. The cause is that the volumetric efficiency is sen-
sitive to changes in residual gas mass, which depends on the pressure ratio ppem
im
120
Wastegate open
100
Exhaust Manifold, p
e
80 Intake Manifold, p
i
60
0 5 10 15 20 25 30
0.03 18
0 5 10 15 20 25 30
Measured
84 Modeled
5% Stationary error!
82
80
78
76
74
0 5 10 15 20 25 30
Time [s]
Figure 3.4: For observer validation the following measured engine data is used: Wastegate
opening and closing at constant air-mass flow, constant speed 3100 RPM, and 5.3 bar brake
mean effective pressure. Top: Exhaust pressure drops when the wastegate is opened at time
5 and it increases again when the wastegate is closed at time 21 seconds. Center: Air-mass
flow is controlled to a constant value by changing the throttle plate angle and transients in
the air-mass flow are only present after wastegate setting changes. Bottom: The volumetric
efficiency increases for open wastegate. At most there is a 5% steady-state error, compared
to the mapped value.
3.3. Observer with Proportional Feedback 31
pim Vd N
Wcyl = Wcylstd (N, pim , Tim ) = vol (N, pim ) (3.8)
Rim Tim nr
is given by Eq. (3.3), and the mass flow into the intake manifold is given by
Eq. (3.2a). The resulting observer is then:
d
pim
= Kim Wat (, pim , pic , Tic ) Wcylstd (N, pim , Tim ) + Kobs (pim pim )
dt
(3.9)
In Eq. (3.9), the CGEKF design methodology is used to determine the feedback
gain Kobs .
3.3.1 Tuning
When Kobs is determined in Eq. (3.9), the following inputs to the design method
are necessary: First, the variance of the state noise pim has to be determined.
In Hendricks et al. (1992) the engine pumpings in the intake manifold pressure
is considered as state noise. This is here determined by taking the variance of
the measured intake manifold pressure in the design points. The second input
is the variance of the measurement signal noise in pim . The variance of the
measurement noise is determined as in Jensen et al. (1997) by measuring the
intake manifold pressure with the engine off but with ignition and dynamometer
on. Further no correlation between the measurement and the state noises is
assumed. As the system is nonlinear, it is linearized in its stationary points
and Kobs is determined for each stationary point, where the stationary points
(N, pim ) are given by a measured engine map. The result is a table of Kobs
where the current state of the engine (N, pim ) are used as lookup keys. The
current gain Kobs is then determined by taking the nearest neighbor.
For air/fuel control the latter is most interesting. Two different feedback gains
are used to better show how the gain influences the estimated intake manifold
pressure and the estimated port air-mass flow. The first gain is referred to as
Design gain and is determined according to the systematic method described
in Section 3.3.1. The design gain K = 41. As the intake manifold pressure
only changes 2 kPa the feedback gain K is constant for the different wastegate
settings in the test case. To illustrate the behavior of a lower feedback gain, a
32 Chapter 3. Speed-Density Methods and Different Wastegate Settings
second gain is used and referred to as Low gain. The Low gain is simply
set to 1.
The results of running the observer with Low and Design gain are pre-
sented in Figure 3.5. In the center plot it is clear that the pressure estimation
error is inversely proportional to the feedback gain. That is to get a low pressure
estimation error a higher feedback gain should be used. However, the primary
interest is the port air-mass flow which is estimated using Eq. (3.3) and here
the difference
Wat (, pim , pic , Tic ) Wcylstd (N, pim , Tim )
is shown for the test data set in the bottom of Figure 3.5. For closed wastegate
there is no stationary error as the modeled volumetric efficiency is very accurate.
When the wastegate is open, there is a stationary error in the estimated port
air-mass flow. The port air-mass flow error is higher for the design gain, while
the intake manifold pressure error is higher for the low gain.
This can be explained by the fact that when there is an error in the modeled
volumetric efficiency, which it is when the wastegate is open, this error propa-
gates to the observed intake manifold pressure. Thus, given a pressure error the
corresponding error in port air-mass flow can be determined by setting the left
hand side of Eq. (3.9) to zero and solve for the stationary pressure difference
(pim pim ):
Kim
(pim pim ) = Wat (, pim , pic , Tic ) Wcylstd (N, pim , Tim ) (3.10)
| {z } Kobs | {z }
6= 0 6= 0
The equation above is nonlinear and to show how the pressure estimation error
and port air-mass error depends on observer gain an example where a known
error in volumetric efficiency is used. The result, shown in Figure 3.6, is that the
error in estimated port air-mass flow increases with the feedback gain. That is if
a low error in estimated port air-mass flow is desired then a low gain should be
used. As expected, the estimated intake manifold pressure error decreases with
increasing feedback gain. When there is an error in the volumetric efficiency, the
proportional feedback gain is a compromise between accurate intake manifold
pressure estimates (higher gain) or accurate port air-mass flow (lower gain).
Hence observers relying on proportional feedback for only the intake manifold
pressure state are not suitable for port-air mass estimation in turbocharged
engines where the volumetric efficiency changes with wastegate setting.
80 p
im
60
0 5 10 15 20 25 30 35
72 Small
Pressure [kPa]
68
Low gain feedback (K=1)
66 Design gain feedback
Measured
64
0 5 10 15 20 25 30 35
Figure 3.5: Results for observer relying on proportional feedback Eq. (3.9). Top: The waste-
gate is open between time 5 seconds and 21 seconds. Center: For closed wastegate, where the
description of vol is correct, the stationary error in the observed intake manifold pressure is
independent of the feedback gain. When the wastegate is open the error is inversely propor-
tional to the feedback gain. Bottom: In the estimated port air-mass flow there is a stationary
error present when the wastegate is open. For the design gain the error is 2%.
34 Chapter 3. Speed-Density Methods and Different Wastegate Settings
Absolute Error
3
Pressure [kPa]
Port airmass flow [g/s]
2
Absolute error
1
0 100 200 300 400 500 600 700 800 900 1000
Observer gain []
Figure 3.6: Calculated error in observed intake manifold pressure and estimated port-air
mass flow using the proportional feedback observer when there is a 5% error in the volumetric
efficiency. Engine operating conditions, N = 2000 RPM, = 14 , and pim 55 kPa. Top:
Relative error in pressure and port air-mass flow. It is clear that the port air-mass flow error
is lower for small gains and that the intake manifold pressure estimation error decreases with
feedback gain. Bottom: The absolute pressure estimation error goes to zero for high gains
while the port air-mass flow error increases with the gain.
3.5. Observer with Air-Mass-Offset Estimation 35
possible bias in vol by integrating the pressure error. Given this bias the
steady-state error in the intake manifold pressure estimate can be removed. In
the observer equations the port air-mass flow model is Eq. (3.4) and the air-
mass flow into the intake manifold is estimated by Eq. (3.2a). Intake manifold
pressure dynamics is given by Eq. (3.1). The port air-mass flow is given by:
p V N
im d
Wcyl = Wcyl ts N, pim , Tim , vol = vol (N, pim ) + vol
Rim Tim nr
and the mass flow into the intake manifold is given by Eq. (3.2a). The observer
estimates vol in a similar manner as in Tseng and Cheng (1999).
d
pim
= Kim Wat , pim , pic , Tic , vol Wcylts (N, pim , Tim ) (3.11a)
dt
2
vol (N, p
im ) + vol N Vd
dvol 1
= (pim pim ) (3.11b)
dt L1 Rim Tim Wat (, pim , pic , Tic ) nr
In Eq. (3.11a) no feedback from the measured intake manifold pressure is used
and the feedback gain in Eq. (3.11b) is from Tseng and Cheng (1999) where L1
is a tuning parameter.
3.4.1 Tuning
The convergence rate of the vol estimation in Eq. (3.11b) is controlled by a
scaling factor L1 . No systematic tuning method for L1 is presented in Tseng
and Cheng (1999) and here L1 was simply set to one.
80 p
im
60
0 5 10 15 20 25 30 35
Estimated Intake Manifold Pressure for Observer With Integrated Offset in vol
74
Integrated offset in vol
Pressure [kPa]
70 No stationary error
68
66
0 5 10 15 20 25 30 35
0
No stationary error No stationary error
1
3
0 5 10 15 20 25 30 35
Time [s]
Figure 3.7: Results for the observer that estimates an additive offset in the volumetric effi-
ciency vol . Top: The wastegate is open between time 5 and 21 seconds. Center: Observed
intake manifold pressure tracks the measured intake manifold pressure excellently for station-
ary conditions. Bottom: The estimated port air-mass flow is equal to the estimated air-mass
flow through the throttle.
3.5. Observer with Air-Mass-Offset Estimation 37
manifold pressure nor to the correct port air-mass flow when the wastegate
setting is changed. The second observer, which uses an augmented model,
offers stationary correct estimates of the intake manifold pressure and of the
port air-mass flow. Here a new observer is suggested that uses feedback of the
estimation error to both the pressure state and to an augmented state compared
to the previous observers, which only used feedback to one state.
A change in wastegate position results in a changed port air-mass flow. In
the port air-mass flow model described by Eq. (3.6) this change is described
using the in-cylinder air-mass-offset CAC. Now denote the estimated in-
[ If CAC
cylinder air-mass-offset CAC. [ is assumed to be slowly varying,
it can be estimated together with the intake manifold pressure, Eq. (3.1), using
[ is then used to determine the port
integration of the estimation error. CAC
air-mass flow. The observer equations are based on Eq. (3.1) where Wcyl is
given by Eq. (3.6)
and the air-mass flow into the manifold is given by Eq. (3.2a).
d
pim
=
dt
Expected air mass Offset
z }| {
z }| {
pim Vd N N [
Kim vol (N, pman ) + CAC Wat (, pim , pic , Tic ) +
Rim Tim nr nr
| {z }
Air mass flow to cylinder
[
dCAC
= K2 (pim pim ) (3.12b)
dt
By applying the systematic tuning method given by the CGEKF-theory the
feedback gains K1 and K2 in Eq. (3.12) are determined.
3.5.1 Tuning
The observer gains, K1 and K2 , are determined by linearizing Eq. (3.12) in
stationary points given by an engine map and applying Constant Gain Extended
Kalman filtering technique as in Section 3.3.1. To determine the gains, two
noise covariance matrices are required: the state variance and the measurement
variance. Using the same method as in Section 3.3 the intake pressure state
noise variance and measurement noise variance are determined.
The variance of the in-cylinder air-mass offset state noise is determined by
assuming that it is the change in cylinder air charge caused by the exhaust
38 Chapter 3. Speed-Density Methods and Different Wastegate Settings
manifold pressure changes. Thus, a model of the cylinder air charge with an
explicit exhaust manifold pressure is necessary and therefore the model from
Chapter 4 is used:
1
pim 1 rc ppem
im
CAC(pim , Tim , pem ) = C1 Vd
Rim Tim 1 + A 1 rc 1
( F )
s
Here pim 0 and Tim 0 means the conditions during closed wastegate and pem (t) is
the varying exhaust manifold pressure as the wastegate is opened and closed.
To determine the variance of CAC for various exhaust manifold pressures
it is necessary to have knowledge of the exhaust manifold pressure when the
wastegate opens and closes. Here, two methods are considered to determine
the exhaust manifold pressure. First, the pressure can be measured during
wastegate steps. Second, as the wastegate is controlled by a slow pneumatic
actuator, the exhaust manifold pressure is assumed to vary sinusoidally when
the wastegate is opened and closed. Its peak-to-peak amplitude is the exhaust
pressure difference caused by an opening and closing of the wastegate. Here the
second method is chosen.
Further, the intake manifold pressure and air-mass-offset are assumed to
be independent. The resulting K1 and K2 are stored in a table. In the table
(N, pim ) are used as lookup keys in the same manner as described in Section 3.3.
3.5.2 Properties
The result of applying the test case is shown in Figure 3.8. When the state
derivatives are zero, i.e. stationary conditions, two important features of this
observer are shown. First, it estimates the same intake manifold pressure as
measured. Second, there is no error in the estimated port-air mass flow.
To prove that there is no bias in the pressure nor any port air-mass flow error;
start by considering that the following relation holds for stationary conditions:
d
pim [
dCAC
= =0
dt dt
[
Now, as ddtCAC
= 0 this means that the observed intake manifold pressure
equals the measured:
K1 (pim pim )
3.6. Results 39
d
pim
=
dt
Expected air mass Offset
z }| {
z }| {
pim Vd N N [
Kim vol (N, pman ) + CAC Wat (, pim , pic , Tic ) +
Rim Tim nr nr
| {z }
Air mass flow to cylinder
K1 (pim pim ) = 0
| {z }
=0,as pim =pim
3.6 Results
Port air-mass flow estimation using speed-density methods that relies on ob-
served intake manifold pressure has been studied on a turbocharged SI-engine
with wastegate. The challenge is that the wastegate influences the non-measured
exhaust manifold pressure that governs the residual gas mass in the cylinder and
thus the volumetric efficiency. Therefore, the volumetric efficiency will change
with the wastegate setting and as speed-density methods rely on accurate de-
scriptions of the volumetric efficiency this results in incorrect port-air mass flow
estimates. It has been shown using measurements that the volumetric efficiency
can change by 5% when the wastegate is opened.
Stationary conditions are chosen as it is only possible to validate the port
air-mass flow for these conditions. Here, the modeled throttle air-mass flow
has been used to validate the estimated port air-mass flow. As the throttle
model uses observed intake manifold pressure it is important that the observed
intake manifold pressure converges to measured pressure in order to determine
the true flow, especially for higher intake manifold pressures which is shown in
Figure 3.3.
Three different observers with feedback from measured intake manifold pres-
sure were studied. In the observer evaluation, it was demonstrated that it is
not possible to estimate the system state for the observer that relies solely on
feedback to the pressure state. There, the feedback gain is a trade-off between
pressure convergence or accurate stationary port air-mass flow estimation. A
40 Chapter 3. Speed-Density Methods and Different Wastegate Settings
p
em Wastegate closed
Wastegate open
100
pim
50
0 5 10 15 20 25 30 35
70
Observer
Measured
65
0 5 10 15 20 25 30 35
Time [s]
5
x 10 Estimated Mass Offset
5
CAC [kg]
5
0 5 10 15 20 25 30 35
Time [s]
Relative Port AirMass Flow Error
Mass flow error [%]
1
0
No stationary error No stationary error
1
2
3
0 5 10 15 20 25 30 35
Time [s]
Figure 3.8: Results for the observer with air-mass-offset estimation. First: The wastegate is
open between time 5 and 21 seconds. Second: Observed intake manifold pressure tracks the
measured intake manifold pressure excellently for stationary conditions. Third: The increase
[ shows that more air is inducted into the cylinder, which is the expected result when
in CAC
the wastegate is opened. Fourth: The estimated port air-mass flow is equal to the estimated
air-mass flow through the throttle for stationary conditions.
3.6. Results 41
high gain results in fast pressure convergence but a large error in estimated port
air-mass flow. Better steady-state port air-mass flow estimates are achieved us-
ing observers with an additional state. The additional state represents an offset
in the volumetric efficiency or an offset from the expected cylinder air-mass.
The estimated offset can be compared to an I-part in a PI-controller and is here
able to compensate for model errors caused by for example a change in volumet-
ric efficiency. For the two observers with offset estimation, the intake manifold
pressure converges to measured intake manifold pressure and they estimate the
same port air-mass flow as the modeled air-mass flow into the intake manifold.
Therefore these observers are better suited for port air-mass flow estimation in
TC SI-engines.
42 Chapter 3. Speed-Density Methods and Different Wastegate Settings
4
Development of a Control
Oriented CAC Model
Today, the demands for lower emissions and improved fuel economy have re-
sulted in an increased popularity of TC SI-engines. An advantage of TC SI
engines is their ability to produce high torques as the turbocharger increases
the intake manifold pressure and thus increases the cylinder air charge (CAC).
A higher CAC allows more fuel to be injected and more torque is produced
by the combustion. For high torque output conditions, the engine starts to
knock and the ignition is retarded to reduce the knock. Unfortunately, this
exhaust temperature increase can destroy the TWC and the turbine. To cool
the exhaust gases, fuel enrichment is therefore necessary. Further, at these high
torques, it is important to avoid that the maximum torque to the transmission
is exceeded. As torque sensors are expensive, models of engine torque based on
CAC is used instead. However, at high loads where fuel enrichment is present
the traditional CAC-models can give an up to 10% error that propagates to an
error in estimated torque. Here the error is reduced down to only 3% by model-
ing the additional charge cooling during fuel enrichment at high loads that also
significantly improve the torque estimates.
On SI-engines, it is possible to improve the fuel economy by opening the
wastegate at part load. In the previous chapter, it was shown that the exhaust
manifold pressure and the cylinder air charge are influenced when the wastegate
is operated. As the emissions are reduced using a three way catalyst which is
only effective if the ratio of air and fuel is accurately controlled. Precise CAC
estimates are therefore required at part load. To maintain low emissions at
part load it is therefore desirable to find an improved CAC model for air-fuel
ratio control that includes exhaust manifold pressure. Further, the model is
43
44 Chapter 4. Development of a Control Oriented CAC Model
used to determine the operating conditions where the CAC is most sensitive
to exhaust manifold pressure changes. It is also used to quantify how large
exhaust manifold pressure changes that can occur in order not to exceed a
pre-determined maximum deviation from stoichiometric.
In this chapter, a speed-density based CAC model for a warmed up TC
SI-engine is developed in two steps. First, a conventional speed density CAC-
model is augmented with a model component that improves CAC estimates
during fuel enrichment. Then, an exhaust manifold pressure dependent model
component is developed. Finally both model components are merged and the
total model, which has only two parameters, is validated using a 2.3 liter tur-
bocharged SAAB 95 engine with wastegate. Thus, the model covers a wider
range of the TC SI engines operating region compared to models based on vol-
umetric efficiency. Finally, it is shown how the proposed CAC model improves
torque estimates based on the estimated CAC during fuel enrichment and for
different exhaust manifold pressures.
Now, how does the volumetric efficiency change when the air-fuel ratio changes?
As the fuel vapor volume changes with the air-fuel ratio, it is natural to start by
considering this effect on the volumetric efficiency. If it is assumed that all fuel
enters the cylinder as vapor, which is reasonable for a warmed up engine, then
the volumetric efficiency for quasi-stationary conditions is inversely proportional
to the volume of fuel vapor (Heywood, 1988, p. 210). Therefore, the volumetric
efficiency should decrease when the air-fuel ratio decreases.
1
voltheor 1 (4.3)
1+
( FA )s
In the top of Figure 4.1 this decrease is illustrated. However, when volumetric
efficiency is measured for rich conditions, shown in the bottom of Figure 4.1,
the opposite phenomenon occurs. This increase can not be explained by other
4.1. CAC and Air-Fuel Ratio 45
[]
0.94
vol
Quasi stationary
Quasistationary decreases with decreasing
0.93 vol
theor
0.92
0.91
0.9
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
0.95
[]
vol
0.9
Measured
0.85
Measured vol increases with decreasing
meas
0.8
0.75
Figure 4.1: Volumetric efficiency for the conditions where the mixture is rich ( < 1). Top:
Theoretically the quasi-stationary volumetric efficiency should decrease with decreasing as
the fraction of fuel vapor increases. Bottom: The estimated volumetric efficiency vol from
measurements with various air-fuel ratios . Volumetric efficiency vol increases significantly
for low .
man
Q
Intake manifold
Injector
Throttle (pim , Tim )
Tcyl Wcyl
in
Intake valve
(pim , Tcyl )
Cylinder
Q valve and cylinder
Figure 4.2: A schematic of the intake manifold with throttle and injector located close to
the intake valve. To make a thermodynamical model of Tcylin only a fraction of the intake
manifold is considered. The studied control volume is marked as a dotted box. In the control
volume parts of the injected fuel, xe , evaporates. The remaining fuel evaporates, outside of the
control volume, on the intake manifold walls using energy from Q man and do not contribute
to the temperature drop.
transfers are shown. It is assumed that parts of the injected fuel vaporize by
taking energy from the surrounding air and thus cooling the charge. The rest of
the fuel, which hits the wall as liquid, is vaporized as the intake manifold wall
heats it. When it is time to induct the air-fuel mixture all fuel has evaporated
and the charge temperature is Tcylin .
The inducted mixture temperature Tcylin is modeled using the first law of
thermodynamics. This law states that the energy is conserved, that is the
energy of the entering fluids into the control volume plus heat must equal the
energy of the exiting fluid. The dotted box in Figure 4.2 indicates the studied
control volume. Its volume is considerably smaller than the volume of the intake
manifold and it is therefore assumed that the air-mass flows into and out of the
control volume are the same, Wcyl . The temperature of the air-mass flow into
the control volume is Tim and heat transfer to the gas inside of the control
volume is not considered during fuel enrichment. The temperature of the flow
out of the control volume is Tcylin . When the liquid fuel with temperature Tfinj is
injected into the control volume a fraction of it, xe on mass basis, evaporates and
in the process consumes energy (heat) from the air-mass flow. The remaining
fuel hits the intake manifold wall and is vaporized by heat transfer from the
wall to the fuel and after the vaporization it has the same temperature as the
surrounding gas, that is Tcylin . A simplified schematic of the fluid and heat
transfer flows of the studied control volume are shown in Figure 4.3. The fuel
W
flow is Wf = Acyl and hf v is the heat required to vaporize the liquid. For
( F )s
additional nomenclature, please see Appendix A.
4.1. CAC and Air-Fuel Ratio 47
Air
Air
Wcyl cpa Tcylin
Wcyl cpa Tim xe Wf vaporizes when injected
and cools the fluid:
Liquid fuel Fuel vapor
xe hf v Wf
Wf cf Tfinj Wf cpf Tcylin
Figure 4.3: A thermodynamical model of the control volume in Figure 4.2. Fluid flows in and
out of the volume are shown with their respectively temperature. It is shown that a fraction
xe , on mass basis, of the fuel evaporates in the control volume and consumes energy. All of
the fuel enter the cylinder as vapor.
When the energy into the control volume is determined, the air related term
is considerably larger than the fuel related term:
cf Wcyl
cf Wf Tfinj = A
Tfinj
F s
cf 2cpa
2cpa Wcyl 2
cf Wf Tfinj A
Tfinj = A
cpa Wcyl Tfinj
F F
s
| {z } s
0.15...0.261
As the factor in front of cpa Wcyl Tfinj is less than 0.3 together with that the
fact that the fuel temperature Tfinj is in the same range as the air temperature
means that the fuel related term Wf cf Tfinj can be neglected. The energy flow
into the control volume can thus be summarized as:
Ein = Wcyl cpa Tim + Wf cf Tfinj Wcyl cpa Tim
| {z } | {z }
Air Fuel
Next, the energy necessary for the fuel vaporization in the control volume is
determined. Here, a fraction xe of the injected fuel vaporizes and consumes
energy Evap . This evaporation term is non-negligible as hf v = (380 . . . 500)cpa
(Bauer et al., 1996, p. 238).
Evap = xe hf v Wf
When the the energy flow out of the control volume Eout is determined, the fuel
related term Wf cpf Tcylin is neglected using the same argument as for Ein .
Eout = Wcyl cpa Tcylin + Wf cpf Tcylin Wcyl cpa Tcylin
The energy balance becomes:
Ein Evap = Eout
Wcyl cpa Tim xe hf v Wf = Wcyl cpa Tcylin
48 Chapter 4. Development of a Control Oriented CAC Model
Eq. (4.4) describes the effect of charge cooling over all the engines operat-
ing regime as a function of Tim , xe , and , where all but xe are known. The
fraction of evaporated fuel, xe , is described by the evaporation curve and for
temperatures above 30 C it is almost linear in temperature (Guzzella and On-
der, 2004, p. 56). In Figure 4.4 stationary measurements of the intake manifold
temperature is shown as a function of 1 and it is clear that Tim 310 can be
approximated by a linear function of 1 . Further it can be seen that the in-
take manifold temperature is above 30 C when fuel enrichment is present and
thus the fraction of evaporated fuel can be expressed as xe = xe1 1 = xe ( 1 ).
Now, the temperature of the charge entering the cylinder can be expressed as
335
330
325
[K]
320
im
T
315
310
305
Measured
Fit
300
1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 1.45 1.5
Scaled 1/ []
1
Figure 4.4: For rich mixtures the intake manifold temperature is affine in
.
1
a function of Tim and :
1 xe ( 1 )hf v
Tcylin (Tim , ) = Tim A
(4.5)
F c
s pa
As fuel enrichment normally is only present at high loads, that are high intake
manifold pressures, only the additional cooling compared to stoichiometric is
4.1. CAC and Air-Fuel Ratio 49
considered. Now, the additional cooling, for non stoichiometric conditions, can
then be expressed as:
1 1
TAdditional cooling = Tcylin Tim , , xe
1 1 hf v 1
Tcylin Tim , , xe = x (4.6)
e1
cpa FA 2
=1
| {z }
s
Constant = C1
h A
In Eq. (4.6) the factors xe1 cpf v F s are lumped together to one parameter C1 .
a
The next step is to include the charge cooling model in a CAC-model, where it
describes the additional temperature drop in intake manifold temperature for
rich mixtures. The total model is then validated using measured CAC-data.
1
Tim,new = Tim TAdditional cooling = Tim C1 (4.8)
2
By inserting Eq. (4.8) in Eq. (4.7) the augmented CAC-model becomes:
Vd
CAC = (a1 pim + a0 ) (4.9)
Rim Tim C1 1
2
3
x 10 Estimated vs Measured CAC
5
Estimated=Measured
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
3
x 10
Relative Error for Estimated CAC at High Loads
0.05
No Charge Cooling by Fuel Evaporation
0.1 With Charge Cooling by Fuel Evaporation
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Measured CAC [kg] 3
x 10
Figure 4.5: Top: Measured CAC for stationary engine data is compared to estimated CAC
with charge cooling, Eq. (4.9), and estimated CAC without charge cooling, Eq. (4.7). When
enrichment is present, the model without charge cooling by fuel evaporation gives up to 10%
too low estimates. Bottom: With the modeled effect of charge cooling by fuel evaporation
the error in CAC is substantially reduced from 10% down to 23% for higher CACs.
The additional temperature drop for rich mixtures caused by charge cool-
ing at high loads can be compensated using an augmented temperature
model.
330
320
Temperature [K]
310
300
280
0.5 1 1.5 2 2.5 3 3.5 4 4.5
Measured CAC [kg] 3
x 10
Figure 4.6: Measured intake manifold temperature and modeled temperature with the charge
cooling model. The data is measured at stationary conditions. The maximum modeled
temperature drop is 32K or 10%.
120
80 Exhaust Pressure
Intake Manifold Pressure
60
0 10 20 30 40 50 60
Time [s]
0 10 20 30 40 50
Time [s]
80
vol
78
Stationary offset Stationary offset
76
74
0 10 20 30 40 50
Time [s]
Figure 4.7: An engine experiment performed at constant speed and constant air-mass flow.
Top: Pressure changes in exhaust system, and intake manifold pressure during manual open-
ing/closing of the wastegate. Center: Measured air-mass flow, Wa . As the wastegate is
opened the air-mass flow decreases momentarily while the air-mass controller opens the throt-
tle. Bottom: Mapped and calculated volumetric efficiency using Eq. (4.2). The calculated
volumetric efficiency is only valid at stationary conditions which are marked with ellipses.
When the wastegate is open the mapped volumetric efficiency is incorrect. The change is
(82 78)/78 5% which would result in a 5% CAC error.
4.2. CAC and Exhaust Pressure Changes 53
2. Given a desired accuracy of the estimated CAC: How large exhaust mani-
fold pressure changes can be tolerated with an exhaust manifold pressure
independent model?
The models purpose is primarily CAC estimation for air-fuel ratio control where
it is possible to include the effects of the exhaust manifold pressure. Here the
modeled CAC is based on an estimate of the inducted volume of air Va . On
turbocharged engines with fix valve timing it easy to estimate the volume of
inducted air, Va as the valve overlap is small which enables a simplified gas
exchange process to be used. As the valve overlap is small, the backflow is
zero. In Figure 4.8 a simplified gas exchange model illustrates how the volume
of inducted air Va is estimated by subtracting the residual gas volume and
the volume of evaporated fuel Vf from the total volume at intake valve closing
(IVC). In the model IVC occurs at bottom dead center (BDC) and exhaust
valve closing (EVC) occurs at top dead center (TDC).
Va
z }| {
1
rc pim
pem e
pim pim 1
CAC = Va = C Vd (4.10)
Rim Tim Rim Tim vol 1 + A 1 rc 1
( F )
s
54 Chapter 4. Development of a Control Oriented CAC Model
At IVC
Figure 4.8: An illustration of the idealized induction model to determine the volume of in-
ducted air Va . Left: Prior to the inductions starts, at exhaust valve closing (EVC), the residual
gases at pressure pem fill the clearance volume Vc . Right: At the end of the intake stroke,
when the intake valve is about to close, the volume is Vd + Vc and the residual gases have
expanded from pem to pim . At IVC the pressure is pim in all of the volumes. The volume of
each component is schematically shown separated by dashed lines and Va = Vd + Vc Vr Vf .
The total model given by Eq. (4.11) includes both the exhaust manifold pressure
and also the effect of fuel enrichment at high engine loads which was described
in the previous section. In the model, there are only two tuning parameters and
these are the pumping parameter Cvol and the effect of charge cooling by fuel
evaporation C1 .
Validation of the CAC model is performed in two steps: First it is validated us-
ing stationary measured data and then for data with different exhaust manifold
pressures.
4.2. CAC and Exhaust Pressure Changes 55
1. That the model including exhaust manifold pressure has the same ac-
curacy, for a nominal wastegate setting, as models based on volumetric
efficiency such as Eq. (4.9).
2. That the model including exhaust manifold pressure better describes CAC
when the wastegate is opened, or when it is not in its nominal setting,
than models based on mapped/parameterized volumetric efficiency such
as Eq. (4.9).
In the validation, a different dataset has been used compared to the tuning data.
The validation data is measured three years later than the tuning dataset and
with a different gasoline quality. For both datasets, the wastegate is controlled
to a nominal opening by the ECU, which means that it is closed for most of the
points.
In Figure 4.9 it is shown that the exhaust manifold pressure dependent
model shows about the same high accuracy as the model using parameterized
volumetric efficiency. In terms of root mean square error, the pem dependent
model has an error of 1.8 103 compared to 1.9 103 for the parameterized
volumetric efficiency. As the pem dependent model has one parameter less, the
stationary accuracy is very good.
Here, measured engine data with different exhaust manifold pressures are used
to test how well the models describe the change in CAC. It has not been possible
to measure the transient CAC and therefore the estimates are compared to
measured air-mass flow after the air filter. This means that the comparison
in Figure 4.10 is only valid for stationary conditions due to the filling and
emptying of the intake system. The locations of the stationary conditions when
the wastegate is open have been marked using ellipses. Both models have been
tuned for nominal wastegate settings, which in this case is closed wastegate.
In Figure 4.10 the wastegate is opened and closed several times to vary
the exhaust manifold pressure. The pem dependent model Eq. (4.11) better de-
scribes the CAC change than the parameterized volumetric efficiency, Eq. (4.9).
As the exhaust manifold pressure dependent CAC model is able to better de-
scribe CAC for conditions where it has not been tuned it is study how sensitive
the CAC is to the exhaust manifold pressure changes.
56 Chapter 4. Development of a Control Oriented CAC Model
3
x 10 CAC Validation
3.5
2.5
Modeled CAC [kg]
1.5
RMS exh. press. dep : 1.8e03
RMS parameterized nvol : 1.9e03
1
Figure 4.9: Validation of the exhaust manifold pressure dependent CAC model using station-
ary data. It is compared to a model using parameterized volumetric efficiency and measured
data. The accuracy of Eq. (4.11) is in the same magnitude as the model using parameterized
volumetric efficiency, Eq. (4.9).
There are two properties of the sensitivity function Eq. (4.12) that makes the
results from the analysis general. First, it does not include any specific engine
parameters that require tuning to measured data. Second, it only depends on
the pressure ratio ppem
im
.
The minus sign in Eq. (4.12) shows that an increase in exhaust manifold
pressure decrease CAC. In Figure 4.11 the sensitivity function is evaluated for
4.2. CAC and Exhaust Pressure Changes 57
120
110
Pressure [kPa]
90
80
70
0 10 20 30 40 50 60
3
x 10 CAC Comparison
1.5
1.4
1.3
0 10 20 30 40 50 60
5
x 10 CAC Error
0 10 20 30 40 50 60
Time [s]
Figure 4.10: Validation using wastegate opening/closing steps. Top: Intake manifold pressure
and exhaust manifold pressure varies depending on whether the wastegate is open or closed.
Center: The estimated CAC compared to measured CAC. To better study the estimated
change in CAC when the wastegate opens the stationary errors for closed wastegate have
been removed for both models by adding a constant offset. The model based on parameterized
volumetric efficiency underestimates CAC when the wastegate is open. Bottom: The CAC
error of the exhaust manifold pressure dependent model is smaller.
58 Chapter 4. Development of a Control Oriented CAC Model
five fix intake manifold pressures. From the figure it is clear that the CAC is
most sensitive to changes in pem for low intake manifold pressures, that is part
load conditions. This is a very important observation since at part load, where
the engine is run stoichiometricly, it is vital to make precise CAC-estimates.
For part load conditions, it is also most beneficial to improve the fuel economy
by opening the wastegate that changes the exhaust manifold pressure. There-
fore, to estimate the expected CAC change when the wastegate is opened it is
necessary to determine the magnitude of the exhaust pressure drop.
0.1
p = 200 kPa
im
p = 95 kPa
Sensitivity []
im
0.3
0.4
pim = 65 kPa
0.5
CAC is more sensitive to
p changes for low p
em im
0.6
pim = 45 kPa
0.7
100 150 200 250 300 350
Exhaust Manifold Pressure [kPa]
Figure 4.11: The figure shows that CAC decreases with increasing exhaust manifold pressure
for a given intake manifold pressure. CAC is almost insensitive to pem -changes at high intake
manifold pressures.
150
100
50
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
Airmass flow [kg/s]
10
Exhaust manifold pressure can drop more
15
than 20% when the wastegate is opened.
20
25
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
Airmass flow [kg/s]
Figure 4.12: Stationary exhaust manifold pressures for different air-mass flows with open and
nominal wastegate. Top: The exhaust manifold pressures can successfully be described using
third order polynomials in air-mass flow. Bottom: The exhaust manifold pressure drop when
the wastegate is fully opened in percent as a function of air-mass flow. It is calculated using
the difference between the fitted lines in the top plot. For this engine the exhaust manifold
pressure drops more than 20% when the wastegate is opened.
The base for this investigation is that the necessary accuracy of air-fuel ratio
controllers is 2% to 3% (Kiencke and Nielsen, 2000, pp. 69) during stoichio-
metric conditions. In Figure 4.13 it is shown that during part load, where it is
motivated to open the wastegate to improve the fuel economy, the CAC changes
up to 5% for the points in the ellipsis. This means that a CAC-estimator with-
out knowledge of exhaust manifold pressure is not sufficiently accurate, that is
controllers based on mapped volumetric efficiency.
60 Chapter 4. Development of a Control Oriented CAC Model
4
CAC Change [%]
0
10 20 30 40 50 60 70 80 90 100
Intake Manifold Pressure [kPa]
Figure 4.13: Estimated CAC change when the wastegate goes from nominal to fully open for
mapped engine data. By applying the pressure drop fraction from Figure 4.12 the change in
CAC is estimated using Eq. (4.12). At part load there can be an up to 5% CAC increase.
Example 4.1
The engine is running at part load with pim =60 kPa and the maximum accept-
able CAC error is 3%. The maximum exhaust manifold pressure drop that can
be accepted and still keep the CAC change lower than 3% is found by following
the 3% line in Figure 4.14. The maximum pemdrop is estimated to 20 kPa.
4.3. The Proposed CAC Model Improves Torque Estimates 61
40
drop
em
Maximal p
30
20
10
0
20 30 40 50 60 70 80 90 100
Intake manifold pressure [kPa]
Figure 4.14: Estimated maximal pemdrop until the CAC changes more than the specified
limit.
With active wastegate control at part load the pressure drop when the
wastegate is opened results in a CAC change of up to 5%. This is not
properly described by an exhaust manifold pressure independent CAC
model. Thus, a CAC model with included exhaust pressure dependency
is necessary for accurate CAC estimation on TC engines with wastegate.
These results are general as the sensitivity analysis, Eq. (4.12), does not include
any engine specific parameters that require tuning to measured data.
Figure 4.15 shows the estimated Tqcs when t1 and t2 has been fitted to mea-
sured CAC. This simple model is used to show that the modeled torque from
250
m
N
0
30
d
an
200
0
n
ee
tw
Torque [Nm]
be
AC
C
150
in
e
fin
af
is
ue
rq
100
To
50
Measured
Model (with validity region)
0
0.5 1 1.5 2 2.5 3 3.5 4
Measured CAC [kg/cycle] x 10
3
Figure 4.15: Left: Measured torque and modeled as a function of measured air-mass/cycle
(CAC). The model is valid for a wide range of the engines operating region.
estimated CAC using the proposed CAC model gives better torque estimates
when compared to CAC estimates from a conventional speed-density method.
The conventional CAC model is represented by Eq. (4.7), which parameters are
tuned for stoichiometric conditions and nominal wastegate setting. This model
is compared to the proposed model Eq. (4.11) for two conditions: First when
fuel enrichment is present and then for part load operation with open wastegate.
In the comparison, the parameters t1 and t2 are fix, i.e. not re-tuned for each
model.
30
20
Torque error [Nm]
10
10
20
180 200 220 240 260 280 300
Measured torque [Nm]
Figure 4.16: Absolute torque model errors using Eq. (4.13) and CAC models with and without
charge cooling. Using the proposed CAC model, Eq. (4.11), with charge cooling and exhaust
manifold pressure dependency the torque estimates are significantly improved.
4.4 Results
Cylinder air charge (CAC) estimation have been studied for turbocharged SI-
engines with emphasis on two topics: Fuel enrichment at high loads and varying
exhaust manifold pressure. The objective is to investigate whether standard
models based on volumetric efficiency can be used or how improvements can be
made to find a good model for control and diagnosis.
At high loads where rich air-fuel ratios are used, the additional fuel influences
CAC and standard models give an error of up to 10%. The error is caused
by the charge cooling effect that the fuel has when it evaporates and thus
64 Chapter 4. Development of a Control Oriented CAC Model
10
15
20
0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2
Measured CAC [kg/cycle] x 10
3
Figure 4.17: Engine torque estimation error for stationary data with open wastegate. It is
clear that the torque error for the exhaust pressure dependent CAC model is considerably
lower.
increases the charge density. The charge cooling effect is modeled and only
one additional parameter is introduced by the model. A standard CAC model
is then augmented with the charge cooling model. With the augmented CAC
model, the estimation error at rich conditions is reduced from 10% down to 3%.
CAC depends on the exhaust manifold pressure and the exhaust manifold
pressure can drop up to 20% when the wastegate is opened. Standard volumetric
efficiency based CAC models do not capture this exhaust manifold pressure
dependency. A CAC model that includes exhaust manifold pressure is therefore
examined and it shows good agreement with measured data even for operating
conditions where it has not been tuned. Using the model a CAC sensitivity
analysis is performed and it can be concluded that:
Exhaust manifold pressure influences CAC at most during part load con-
ditions.
It is necessary to include the exhaust manifold pressure in the CAC model
to estimate CAC with an error less than 3% at part load.
When the exhaust manifold pressure dependent CAC model is augmented
with the charge cooling model, the total model is able to describe CAC with
changing exhaust manifold pressure as well as it describes CAC during fuel
enrichment. This was illustrated using a torque model based on estimated
CAC. With the suggested CAC model the estimated errors at high loads and
with open wastegate were significantly reduced. As the model is able to estimate
CAC over a wide range of operating conditions it is therefore highly suitable
for CAC estimation for control and diagnosis of turbocharged SI-engines.
5
Leaks before the first oxygen sensor can result in increased emissions and a
defective wastegate can cause expensive damages to the turbine if it exceeds its
maximum speed. All of these issues are related to the exhaust manifold pressure.
Knowledge of the exhaust manifold pressure is therefore an important tool for
diagnosis of the turbine, wastegate, and for leakage detection before the first
oxygen sensor (Andersson and Eriksson, 2002). Normally, the exhaust manifold
pressure is not measured due to the high temperatures in the exhaust system
and the extra cost of an additional sensor. Thus, it is desirable to estimate the
exhaust manifold pressure using information from available sensors. A challenge
is that all engine sensors, except the oxygen sensor(s), are located on the intake
side.
Exhaust manifold pressure and temperature have successfully been esti-
mated on naturally aspirated (NA) engines using mean value models (Maloney
and Olin, 1998; Fons et al., 1999). These models rely on the assumption that the
exhaust manifold pressure is the result of a mass-flow through a fix restriction.
In NA engines the flow restriction on the exhaust side can be can be accu-
rately modeled using this assumption (Eriksson et al., 2001). On TC-engines
the wastegate acts as a variable restriction, which means that the models suc-
cessfully used for NA-engines can not be applied directly for exhaust manifold
pressure estimation. Further, the position of the wastegate is not fully known
which makes exhaust manifold pressure estimation a challenging task.
To estimate the absolute exhaust manifold pressure on a turbocharged en-
gine a two-component model is introduced. The model is based on a separation
of the exhaust manifold pressure into two parts, one nominal which is a function
65
66 Chapter 5. Exhaust Manifold Pressure Estimation
of the air-mass flow and one offset from the nominal pressure. No additional
sensors in the exhaust system are needed by the estimator after calibration.
Only the following sensors are required: ambient pressure pa , air-mass flow Wa ,
intake manifold pressure pim and temperature Tim , and air-fuel ratio . These
signals are available on many production engines.
Air filter
1111
0000
Intercooler
0000
1111
1111
0000 Compressor
Throttle Turbine Shaft
pem Turbine
Wcyl
pim , Tim Waste-
Engine gate
Intake Exhaust
Manifold Manifold
Catalyst
Figure 5.1: Sensors and actuators on the engine. Only the air-mass flow after the air-filter
is measured, Wa . For stationary conditions the port air-mass flow Wcyl equals Wa . Exhaust
manifold pressure pem depends among others on the wastegate setting. On the exhaust side
there is only one sensor, the oxygen sensor .
Turbine
pem Muffler
Wastegate pa
+
Nominal setting
The engine pumps a constant mass-flow
Figure 5.2: Hera a resistor analogy is used to show the partition of the exhaust manifold
pressure. The wastegates effect on the exhaust manifold pressure is like a variable resistor
with a nominal setting. When the wastegate is operated the exhaust manifold pressure
changes around its nominal value.
68 Chapter 5. Exhaust Manifold Pressure Estimation
For this polynomial the error is less than 4% and hence it is a good approxima-
tion of pem nom . Figure 5.3 shows the fitted polynomial and measured exhaust
manifold pressure as a function of measured air-mass flow.
For low air-mass flows it is possible to improve the accuracy by including
a second order term as Bergstrom and Brug ard (1999) did. A second order
polynomial describes the exhaust manifold pressure very accurately for closed
wastegate. However, the wastegate opens for higher loads, which means that
the second order polynomial can not be used to describe the nominal exhaust
manifold pressure.
Vd
CAC(pim , Tim , ) = (a1 pim + a0 ) (4.9)
| {z } Rim Tim C1 1
2
vol pim
When the engine is not running with nominal wastegate setting, the exhaust
manifold pressure is changed pem from nominal pressure and this changes
the residual gas mass. A different mass of residual gases will influence the
volumetric efficiency and thus the inducted CAC(pim , Tim , ). It will therefore
be a difference between measured and modeled CAC when the engine is not
running with a nominal wastegate setting. The difference is called CAC and
for stationary conditions, where Wat equals the measured Wa , it is calculated
by rearranging Eq. (3.12a):
nr
CAC = Wa CAC(pim , Tim , ) (5.3)
N
An illustration is given in Figure 5.4, where a step in wastegate setting is used
to show the change in volumetric efficiency and difference between measured
and modeled CAC, CAC. The experiment in Figure 5.4 shows that pem
information is present in the measured signals on the intake side. Next, a
method is proposed where CAC is used to estimate pem .
CAC
CAC(pim , pem nom + pem , Tim , ) = CAC(pim , pem nom , Tim , ) + pem
pem
(5.4)
5.2. Exhaust Pressure Model 69
200
Pressure [kPa]
150
100
Measured
Estimated
50
0 0.02 0.04 0.06 0.08 0.1 0.12
Relative Error in Estimated Exhaust Manifold Pressure
4
Relative Error [%]
4
0 0.02 0.04 0.06 0.08 0.1 0.12
Measured Exhaust Manifold Pressure Compared to Estimated for Low Mass Flows
140
130
Pressure [kPa]
120
110
100 Measured
Estimated
90
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
Air Mass Flow [kg/s]
Figure 5.3: Top: Exhaust manifold pressure during mapping and fitted first order polynomial
in air-mass flow. Center: The relative error of the polynomial fitting is less than 4 percent
and largest for low mass flows. Bottom: For low air-mass flows a second order polynomial
would fit better.
70 Chapter 5. Exhaust Manifold Pressure Estimation
100
80
Exhaust Pressure
Intake Manifold Pressure
60
0 10 20 30 40 50 60
85
v [%]
80
75
Mapped
Calculated
70
0 10 20 30 40 50 60
0 Stationary Stationary
2
0 10 20 30 40 50 60
Time [s]
Figure 5.4: Steps from nominal to fully open wastegate at constant engine speed. A controller
in the ECU tries to maintain constant air-mass flow. Top: When the wastegate has opened,
the exhaust manifold pressure drops significantly but the intake manifold pressure drops only
slightly. Center: Comparison of mapped and calculated volumetric efficiency. Eq. (2.3) was
used to determine the calculated volumetric efficiency. Both are equal for nominal wastegate
but the volumetric efficiency increases stationary when the wastegate is open. Bottom: For
open wastegate a there is a stationary offset between the measured and the expected or
modeled CAC. The difference between them are called CAC
5.2. Exhaust Pressure Model 71
After some substitutions, it will be shown that the linearized model can be
inverted and yield an explicit expression for pem .
For nominal exhaust manifold pressure, the cylinder air-charge is described
well by Eq. (4.9) that is independent of pem :
Further, the cylinder air charge CAC(pim , pem nom + pem , Tim , ) in Eq. (5.4) is
also measured and this term is therefore replaced by the measurement.
nr
CAC(pim , pem nom + pem , Tim , ) = Wa
| {zN}
Measured
By inserting the relation above together with Eq. (4.9) into Eq. (5.4) the result
is:
nr CAC
Wa = CACpem nom (pim , Tim , ) + pem
| {zN} pem
Measured
When the terms are rearranged and the relation from Eq. (5.3) is applied, the
result is that CAC
pem pem must equal CAC:
nr CAC
Wa CACpem nom (pim , Tim , ) = pem (5.5)
| N {z } pem
CAC
CAC
Now, if pem is monotone and non-zero it is possible to solve Eq. (5.5) for pem :
CAC
pem = CAC
(5.6)
pem
1.75
1.8
[kg/(sPa)]
1.85
1.9
em
CAC/ p
1.95
2.05
2.1
As all factors above are non-zero and do not change sign the partial derivative
will always be monotone and negative. Therefore, the inverse given by Eq. (5.6)
always exists. Also, note that CAC
pem , for a given , only depends on the ratio
pem CAC
pim . In Figure 5.5 pem has been evaluated for points in an engine map and as
the variations around the mean value of CAC CAC
pem are small a constant Ke = pem
will be assumed. The constant Ke had to be slightly increased compared to the
theoretical value of 1.9 109 . Instead the value Ke = 4 109 was used,
which results in the following simple pem model:
First the temperature of the air-fuel mixture is calculated according to Eq. (4.8)
and then the air-mass offset CAC, Eq. (5.3), is calculated. The nominal
exhaust pressure is determined by Eq. (5.2), the exhaust manifold pressure
5.3. Exhaust Pressure Estimator Validation 73
offset by Eq. (5.8) and finally inserted into Eq. (5.1) which yields pem .
1
Taf (Tim , ) = Tim C1
2
vol pim
z }| {
nr (a1 pim + a0 ) Vd
CAC (Wa , N, pim , Tim , ) = Wat
N RTaf (Tim , )
pem nom (Wa ) = pa + k1 Wa + k2
pem (CAC) = Ke CAC
pem (Wa , CAC) = pem nom (Wa ) pem (CAC)
Important second order effects, such as heat transfer, and valve overlap etc. are
taken into account by pem nom .
Figure 5.6 where the intake manifold pressure is below ambient and the air-mass
flow is controlled to 45 g/s during the experiment.
The controller succeeds in maintaining the desired air-mass flow even when
the wastegate is open. The nominal exhaust manifold pressure is expressed
as a polynomial in air-mass flow and since the air-mass flow is constant, the
estimated exhaust manifold pressure is also constant. When the wastegate is
opened, there is an offset from the nominal exhaust manifold pressure. This
offset called pem can be estimated by inverting the linearization of CAC with
respect to deviations from nominal exhaust manifold pressure. When the esti-
mated pem are applied, the drops in exhaust manifold pressure are captured.
During transients there are over shoots in the estimated exhaust manifold pres-
sure, which are results of a using a static intake model.
More results of applying the method described by Equations (5.2, 5.8) are
shown in Figure 5.7 (low brake mean effective pressures, bmep) and top of
Figure 5.8 (high bmep) where different settings of the wastegate are used in four
different operating points. The absolute exhaust manifold pressure is described,
in most cases, within 5%.
In the bottom of Figure 5.7 it is shown that even when there is a bias in the
estimated pem the exhaust manifold pressure change is still correct. The bias
in the estimated exhaust manifold pressure change pem is caused by errors
in the modeled volumetric efficiency. This shows that even though there are
errors the estimated pressures the estimator still gives useful exhaust manifold
pressure information.
nom
130
Air Transient
Stationary Air
0.05
Mass Flow
Mass Flow [kg/s]
0.045
10 20 30 40 50 60
Time [s]
Figure 5.6: Top: Measured exhaust manifold pressure data compared to estimated pressure
with and without pem information. The polynomial description gives accurate estimates for
nominal wastegate setting, but overestimates the pressure for open wastegate. With the pem
information, Eq. (5.8), it is also possible to describe the pressure for open wastegate. The
estimated exhaust manifold pressure follows the measured within 5%. Bottom: When the
wastegate is operated there is a short transient in the air-mass flow. Between the transients
the air-mass flows are constant, which have been marked.
76 Chapter 5. Exhaust Manifold Pressure Estimation
135
130
Pressure [kPa]
125
120
115
Measured p
em
110 Estimated p +p
em em
nom
105 Polynomial p
em
nom
100
0 20 40 60 80 100 120 140 160
110
Pressure [kPa]
105
Measured p
em
100 Estimated p +p
em em
nom
Polynomial p
em
nom
95
0 20 40 60 80 100
Time [s]
Figure 5.7: Measured exhaust manifold pressure (solid) compared to nominal pressure (dash-
dotted), and estimated using the in-cylinder air-mass offset information (dashed). When the
additional pressure offset pem is added, the estimates are improved for open wastegate. In
the bottom plot there is a stationary error in pem , but pem captures that there is a change
in exhaust manifold pressure.
5.3. Exhaust Pressure Estimator Validation 77
160
150
140
Pressure [kPa]
130
90
0 50 100 150 200
Time [s]
150
140
Pressure [kPa]
130
100
0 20 40 60 80 100 120
Time [s]
Figure 5.8: Measured exhaust manifold pressure (solid) compared to nominal pressure (dash-
dotted), and estimated using the in-cylinder air-mass offset information (dashed). In the
bottom plot the air-mass flow through the engine drops when the wastegate is opened and
thus also the pem nom . However it is still necessary to add the exhaust manifold pressure offset
pem to better describe the exhaust manifold pressure.
78 Chapter 5. Exhaust Manifold Pressure Estimation
5.4 Results
On turbocharged spark-ignition engines with wastegate the absolute exhaust
manifold pressure can not be estimated using a simple function of the air mass
flow. This as one of the three parts in the exhaust system, the wastegate can
not be described as a static restriction while the two other components the
muffler and the turbine can be described as a static functions. To resolve this
a two-stage exhaust manifold pressure estimator is proposed. It first estimates
the nominal exhaust manifold pressure, which is a static function of the ambient
pressure and the air-mass flow. Then, changes from nominal exhaust manifold
pressures cause the in-cylinder air-mass-offset CAC to be non-zero. Using the
estimated CAC the pem (CAC) can then be estimated and added to the
nominal exhaust manifold pressure and thus a better estimate of the exhaust
manifold pressure is produced when the wastegate is not in its nominal setting.
The estimator is validated using steps in wastegate for various speeds and
loads for two cases:
1. When the air-mass flow can be maintained before and after the step in
wastegate.
2. When it is not possible to maintain the same air-mass flow since the
opening of the wastegate causes the boost pressure to drop too much.
In neither case does the nominal exhaust manifold pressure describe the exhaust
manifold pressure sufficiently accurate. It is therefore necessary to consider the
varying restriction caused by the wastegate. By adding the estimated pressure
offset pem (CAC) caused by the wastegate, the absolute exhaust manifold
pressure estimates are improved. Accurate pem estimates are produced, using
sensors from the intake side only, provided that there is a precise description of
the volumetric efficiency. Here the precision of the estimated absolute exhaust
manifold pressure has been within 5% for most cases.
Part II
Dynamics
79
6
Engine control systems are growing more complex as the demands for low emis-
sions and good fuel economy increase. To facilitate the controller design process,
a physically based model of the engine is desirable. Additionally, the model can
serve as a virtual engine during the initial tests. For these purposes, a highly
suitable model class is mean value engine models (MVEMs) which describe the
average behavior of the engine over one to several thousands of engine cycles, see
eg. (Aquino, 1981; Hendricks and Sorensen, 1990; Powell et al., 1998a; M uller
et al., 1998; Eriksson et al., 2002b).
The objective of this chapter is to present a component based MVEM to-
gether with systematic methods to determine its parameters. Here the word
component based reflects that the component structure of the engine is pre-
served. Typical components are air-filter, intercooler, and throttle that are
connected by pipes or manifolds. One benefit of this modeling strategy is that
every component can be separately identified and then the engine model is
built using the separate components. Each component is described in terms of
equations, constants, parameters, states, inputs, and outputs.
The intended use of the component based engine model is to form a basis
of a pressure, temperature and mass-flow observer along the air-path of a tur-
bocharged engine. The first application is in Chapter 7 where the model is used
to determine observability of the model and further which signals that are most
suitable for observer feedback. An observer is then designed in Chapter 8 and
the resulting observer is used in Chapter 9 to achieve accurate air/fuel control.
Other applications of the TC spark ignited (SI) engine model are diagnosis,
prediction, estimation of non-measured signals, and turbocharger control.
81
82 Chapter 6. A Mean Value Model of a TC SI Engine
All plots shown in this chapter have been slightly scaled in order not to
reveal the true engine parameters. As both the modeled and measured values
are scaled by the same scaling factor the scaling does not influence the relative
error. Therefore, the model quality is best reflected using the relative error.
Ground Ground1
Temperature of downstream flow
m_flow down
Q in
T up
T down
mFlow up
mFlow down
Q in
T up
T down
T flow
m flow
mFlow up
mFlow down
T flow
m flow
Control Volume:
Control volume:
effective area
Pipe between Restriction:
Restriction: Intake manifold
airfilter and Throttle
Air filter
compressor
T down
p down
T down
p down
T up
p up
T up
p up
p
T
p
1e5 Temperature to next restriction
Ambient
pressure
Figure 6.1: Illustration of modeling methodology with control volumes in series with restric-
tions. From the left there is one restriction (air-filter) which produces a mass-flow into a
control volume (the pipe between the air-filter and the throttle). The mass-flow out of the
control volume is governed by a restriction (throttle), which in turn is followed by another
control volume (intake manifold).
W_ic
Q in
T up
T_c
Q in
T up
T flow
m*_c
Tq_c
m flow
Q in
T up
Intercooler
T down
T down
mFlow up
T down
Incompressible
mFlow up
Air filter
mFlow up
Air filter
mFlow down
Compressor Restriction and
T_fwd_flow [K]
mFlow down
mFlow down
Incompressible Control Compressor Control Temperature Intercooler
Restriction Volume Volume model Control Volume
T_cool [K]
p_up
T_up
T_down
p_down
T
p
p up
T up
T down
p down
p_Raf
T_Raf
w_tc
p_Rc
T
p
T
p
C 6
T_ic [K]
6.1. Model Structure
1 T_cool
p_af [Pa]
T_c p_c 4
T_c [K]
2 5 T_ic p_ic
T_af p_af T_af [K] p_ic [Pa]
17
3 Throttle W_th [kg/s]
p_c [Pa] Compressible
Restriction
Ambient Conditions
p up
T up m flow W_th
p_amb
Tq_c MATLAB
1 effective area
w_tc Function
5 u_th [deg] T down
Tq_t deg2area T flow
p_amb [Pa]
p down
6 TC dynamics
T_amb [K] Intertia with friction
T_amb
13 Intake
Manifold
w_tc [rad/s] Control Volume
mFlow up
T
T up
Q in
T down Ground2
10 p
12 T_em [K] rad/s > U/min mFlow down 18
2 w_e
T_es [K] W_cyl [kg/s]
3 n_e 30/pi
11 T_t p_t u_wg [0...1] T_em p_em
p_es[Pa] 7
9 p_i m[Pa]
p_em[Pa]
4 p_im
lambda []
T
p
p up
T up
w_tc
p_es
T_es
u_wg
w_e
p_em
T im
p im
T_im
T_em
T
p
p em
T down
p down
lambda
with a shadowed backdrop are blocks where dynamics (states) are present.
Control Volume
T_turb
m_es
Tq_t
mFlow up
T up
Q in
T down
mFlow down
m flow
T flow
mFlow up
T up
Q in
T down
mFlow down
mFlow e
T ti
Tq_e
air Flow
1 W_e_fg
conv
W_es 1
Ground6 Ground5 Terminator
W_e
Note that the modeling methodology with control volumes between restrictions is implemented
without exceptions. In the center the turbocharger shaft dynamics block is shown. The blocks
85
Figure 6.2: The mean value engine model implemented in Simulink using MVEM-Library.
86 Chapter 6. A Mean Value Model of a TC SI Engine
All gases are considered to be ideal and there are two sets of thermody-
namic properties:
1. Air on the intake side has gas constant Ra and the ratio of specific
heats is a .
2. Burned (exhaust) gas has gas constant Reg and the ratio of specific
heats is eg .
State Description
paf Pressure after air-filter
Taf Temperature after air-filter
pcomp Pressure after compressor
Tcomp Temperature after compressor
pic Pressure after intercooler
Tic Temperature after intercooler
pim Intake manifold pressure
Tim Intake manifold temperature
pem Exhaust manifold pressure
Tem Exhaust manifold temperature
pt Pressure after turbine
Tt Temperature after turbine
TC Turbocharger speed
6.2. Incompressible Flow Restrictions 87
Tuning Data
An engine map with 343 points has been used to determine the engine model
parameters. Additional data from the turbocharger manufacturer have been
used together with engine specific parameters such as volumes and geometry.
In addition, a data set where the engine is run with constant throttle angle is
used to determine the throttle model parameters.
Validation Data
This data set has 68 points spread over the entire operating region. It was
measured 3 years after the tuning data set. In the meantime, the research lab-
oratory was redesigned and the status of the engine changed. Here is a list
of modifications made to the engine compared to its status when the tuning
data was measured: a replaced air-mass flow sensor which gave slightly dif-
ferent readings, a modified exhaust system (different back-pressure), modified
intercooler installation, different pressure and temperature sensors, and a new
dynamometer. The following additions were also made to the engine: crank
case ventilation and compressor by-pass valve. The validation data set was
measured using a summer grade gasoline while the tuning data was measured
using a winter grade gasoline.
As the pressure drop p is known together with the temperature before the
restriction, the mass-flow W is determined by inverting Eq. (6.1):
q
p
qpbefore HTbefore , p > plin
W = pbefore
p
0 p plin (6.2)
HTbefore plin ,
0, p < 0
For pressure drops p plin , the mass-flow is linearized, to reduce the derivative
W
p for small p as:
W
lim =
p0 p
The parameter H is determined using the method of least squares, given mea-
surements of p, pbefore , and Tbefore . Next tuning and validation of the pa-
rameter H in Eq. (6.1) is shown for the following incompressible restrictions:
air-filter, intercooler and exhaust system.
Parameters
h 2 2i Inputs Output
Haf Pa s
Kkg2
Ta [K] Waf [kg/s] Eq. (6.3)
plinaf [Pa] pa [Pa]
paf [Pa]
6.2.2 Intercooler
The intercooler is an incompressible restriction where the pressure drop follows
Eq. (6.4). One major difference from the air-filter is that the temperature of the
flow out of the intercooler is not equal to the temperature of the gases flowing
into the intercooler. The temperature of gases flowing out of the intercooler is
instead described by Eq. (6.5a).
6.2. Incompressible Flow Restrictions 89
Tuning: Air Filter Pressure Drop Model Validation: Air Filter Pressure Drop Model
8000 8000
Pressure drop [Pa]
6000 6000
4000 4000
2000 2000
0 0
0 0.05 0.1 0 0.05 0.1
Air mass flow [kg/s] Air mass flow [kg/s]
Figure 6.3: Air filter pressure head loss model described by the parameter Haf . Left: The
model describes the pressure drop very well for all operating points. Right: The model
describes the pressure drop within 10% for higher mass flows when the laminar mass flow
meter LFE 3 is disconnected. When it is connected the model predicts a too low pressure
drop which is correct as it is tuned without the LFE3.
90 Chapter 6. A Mean Value Model of a TC SI Engine
Tuning: Intercooler Pressure Drop Model Validation: Intercooler Pressure Drop Model
Measured Measured
14000 Modeled 14000 Modeled
12000 12000
10000 10000
Pressure drop [Pa]
6000 6000
4000 4000
2000 2000
0 0
0 0.05 0.1 0 0.05 0.1
Air mass flow [kg/s] Air mass flow [kg/s]
Figure 6.4: Intercooler pressure head loss model described by the parameter Hic . Left: Model
fit for the tuning data is better than 10% for high air-mass flows. Right: For the validation
data the fit is better than 20% for high flows.
Pressure Drop
Intercooler pressure head loss is described by
Tcomp Wic 2
pic = pcomp pic = Hic (6.4)
pcomp
where Wic is the air-mass flow through the intercooler. This kind of intercooler
pressure drop model is also suggested in (Watson and Janota, 1982, pp. 321).
A validation of modeled data against measured is shown in Figure 6.4.
Temperature Change
The intercooler is a cross-flow heat exchanger with unmixed flows. Here the
temperature change is considered and the heat transfer is modeled for stationary
conditions by a regression model (Eriksson et al., 2002b):
390 390
380 380
Modeled temperature [K]
360 360
350 350
340 340
330 330
320 320
320 340 360 380 320 340 360 380
Measured temperature [K] Measured temperature [K]
Figure 6.5: Intercooler temperature model. Left: For the tuning data the fit is within 1%.
Right: The model predicts the temperature with an average error of less than 2% for low to
medium mass-flows and the maximum error is 8% that occurs for high loads.
that makes sure that the temperature out of the intercooler is not lower than
the cooling air.
In Figure 6.5 a validation of the temperature model is shown. For the tuning
data, the maximum error is within 1%, but for the validation data the error is
larger. The large maximum error in the validation data is probably caused by a
lower flow of cooling air though the intercooler. The lower mass-flow of cooling
air can be a result of the changed intercooler fan installation.
The intercooler model inputs, parameters, and outputs are listed in Ta-
ble 6.2.
Parameters
h 2 2i Inputs Outputs
Hic Pa s
Kkg2
Tcomp [K] Wic [kg/s] Eq. (6.2)
plinic [Pa] paf [Pa] Ticout [K] Eq. (6.5a)
a0 , a1 , a2 , a3 [-] pcomp [Pa]
Tcool [K]
Wcool [kg/s]
2
Tesin Wes
pt = pt pa = Hes (6.6b)
pt
As measurements of the temperature into the exhaust system were available,
these were used when the pressure head-loss parameter Hes is determined. A
validation of modeled data against measured is shown in Figure 6.6. The model
fit can be improved by including a linear term, corresponding to dArcys law for
head losses (Eriksson et al., 2002b) but this has not been done, as the absolute
error is small.
The exhaust system model inputs, parameters, and outputs are listed in
Table 6.3.
Parameters
h 2 2i Inputs Output
Hes Pa s
Kkg2
Ta [K] Wes [kg/s] Eq. (6.6b)
plines [Pa] pt [Pa] Tesin [K] Eq. (6.6a)
Tem [K]
Tt [K]
Wt [kg/s]
Wwg [kg/s]
4
x 10 Tuning x 10
4 Validation
7 Measured 7 Measured
Modeled Modeled
6 6
5 5
Pressure drop [Pa]
3 3
2 2
1 1
0 0
0 0.05 0.1 0.15 0 0.05 0.1 0.15
Mass flow [kg/s] Mass flow [kg/s]
Figure 6.6: Exhaust system pressure head loss. Left: For the tuning data there is a systematic
error present, which could be reduced by augmenting the model using a linear term. Right:
The fit for the validation data shows the same trends as the tuning data.
pafter
= pbefore , pafter < pbefore
(6.7b)
1, otherwise
r 2
+1
2
1
() = s (6.7c)
1
2 +1
1
2
1
2
+1 2
+1
+1
0 < +1 2
1 +1
() = () 2
< lin (6.7d)
(lin )
+1
To prevent dWd from reaching for pressure ratios close to 1, Eq. (6.7c) is
linearized for pressure ratios above lin . In the sections below, the flow through
the throttle is described first followed by the flow through the wastegate.
The normalized1 Cd (A(), )(), presented in Figure 6.7, follow one curve,
as the area dependency is canceled in the normalization, which suggests that
Cd can be separated into two functions:
By inserting Eq. (6.10) into Eq. (6.7a) and rearrange the factors two new func-
tions appear. One function that depends on the pressure ratio and one that
depends on :
pic
W (, pim , pic , Tic ) = () Cd () Cd (A ()) A () (6.11)
Ra Tim | {z }| {z }
Function of Function of
the experiment.
96 Chapter 6. A Mean Value Model of a TC SI Engine
12 Ath=100 mm2
A =125 mm2
th
10 A =150 mm
2
th
6
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
Normalized Cd() []
0.9 2
A =70 mm
th
Figure 6.7: Measurements with different pressure ratios over the throttle for four fix throt-
tle areas. Top: The maximum deviation in throttle angle is less than 0.5 degree and the
median deviation is 0.2 degree for each sweep in pressure ratio . Bottom: The normalized
Cd (A(), )() follow one curve which shows that the area dependency of Cd (A(), )()
is canceled and thus it can be factorized into Cd (A())Cd ()().
6.3. Compressible Flow Restrictions 97
0.95
Normalized Cd()() []
0.9
Measured critical
pressure
0.85
0.8
0.75
0.7 70 mm2
100 mm2
125 mm2
0.65
150 mm2
Theoretical ()
0.6
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Pressure ratio =pim/pic
Figure 6.8: Normalized ()Cd () and modeled compressible flow function (), Eq. (3.2d).
The modeled () is shown as a dashed line. It has the same shape but the modeled critical
pressure (the knee) appears at a too high pressure ratio.
In the first model, presented in Hendricks et al. (1996), the ordinary ()-
function is replaced by:
1 p
1 p2 , pcrit
() = pn (6.12)
1, < pcrit
p p 1
p1 2 1
pcrit = (6.13)
p2
q
pn = ppcrit
1
ppcrit
2
(6.14)
There are two tunable parameters p1 and p2 in Eq. (6.12), which were deter-
mined using the Matlab function lsqcurvefit.
The second model is based on the fact that the critical pressure is a function
of . By replacing the physical value of in Eq. (6.7d) with a value determined
using optimization to better describe the measured ()Cd (). The optimiza-
tion was carried out using the Matlab function lsqcurvefit and the new was
chosen in a way such that the difference between the measured product and the
modeled product of () = Cd ()() was minimized on the experimental
data in Figure 6.8.
In Figure 6.9 a comparison is shown between (), (), and (). For
pressure ratios between 0.4 and 0.7, () shows a slightly better behavior over
(). However as the differences between them are small, approximately 1%,
98 Chapter 6. A Mean Value Model of a TC SI Engine
0.95
0.9
Measured critical
Normalized Cd()() []
pressure
0.85
0.8
0.75
2
Ath=70 mm
2
Ath=100 mm
0.7
A =125 mm2
th
A =150 mm2
th
0.65 Original ()
()
() where =1.91
0.6
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Pressure ratio p /p
im ic
Figure 6.9: Improved versions of the compressible flow restriction function (). Two models
are shown together with measured data and the original function Eq. (6.7d). Both new
models show a considerable improvement compared to the original function (dashed). All
fitted functions have been normalized to one at sonic flows. The () with = 1.91 reduces
the error from 10% down to less than 3%.
6.3. Compressible Flow Restrictions 99
for pressure ratios between 0.4 and 0.7, () is chosen as it has only one
parameter and its structure is well known.
For this engine a throttle area model has been developed with the following
appearance (Nyberg and Nielsen, 1997):
Cd ()A() = A1 1 cos(a2 2 + a1 + a0 ) + A0 (6.16)
The Matlab function lsqcurvefit was used to determine the model parameters
and the resulting area estimates are shown in Figure 6.10. Points with pressure
ratios close to one have been omitted during the tuning as the () function
is very sensitive to small changes in , caused by for example pressure sensor
errors.
There is a large discrepancy between the model error for the tuning data and
the model errors for the validation data. Therefore the cause of the increased
error was investigated and was traced to a slight change in the parameters
A0 and A1 . This was verified by re-tuning the throttle area parameters using
the validation data and there was only a significant change in two parameters
A0 (30% increase) and A1 (7% decrease). The cause of the parameter change is
most likely the exchange of air-mass flow sensor and the pressure sensor before
the throttle. These sensors were replaced in the time between the measurements
of the tuning and the validation data.
Cd()A() [m ]
2
2
0.8 0.8
0.6 0.6
0.4 0.4
d
0.2 0.2
0 0
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
Pressure ratio pim/pic Pressure ratio pim/pic
Relative error []
0.05 0.05
0 0
0.05 0.05
0.1 0.1
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
Pressure ratio p /p Pressure ratio p /p
im ic im ic
Figure 6.10: Validation of throttle area model. In the left column, the tuning data is shown
and to the right is the validation data shown. Top: Modeled and estimated area from mea-
surements using Eq. (6.15). Bottom: Relative error in the throttle area model. For the tuning
data the model error is less than 5% for most of the points. In the validation the errors are
larger but this is explained by a change in the throttle area parameters A0 and A1 due to
different sensors.
0.05 0.05
0 0
0 0.05 0.1 0.15 0 0.05 0.1 0.15
Measured airmass flow [kg/s] Measured airmass flow [kg/s]
0.15 0.15
Relative error []
Relative error []
0.1 0.1
0.05 0.05
0 0
0.05 0.05
0.1 0.1
0 0.05 0.1 0.15 0 0.05 0.1 0.15
Measured airmass flow [kg/s] Measured airmass flow [kg/s]
Figure 6.11: Measured and modeled air-mass flow past the throttle. Left: Tuning data. Right:
Validation data. Bottom: Relative error in air-mass estimations as a function of measured
air-mass flow. There is a systematic relative error in the validation data that is caused by
offsets in the throttle area parameters A0 and A1 . By re-fitting A0 and A1 the error has the
same shape as for the tuning data, which show that these parameters have changed slightly
due to changed sensors.
102 Chapter 6. A Mean Value Model of a TC SI Engine
6.4 Engine
The engine produces a port air-mass flow, exhaust-mass flow, temperature of
the latter, and also torque to the crank shaft:
Port air-mass flow Mass-flow from the intake manifold into the cylinder.
Torque The engine torque model includes pumping and friction effects.
Both the exhaust manifold mass-flow and the engine torque are modeled as
instantaneous, that is the delay caused by the cycles in the four stroke engine
is neglected.
Modeled [kg/s]
0.1 0.1
0.05 0.05
Model Model
Ideal Ideal
0 0
0 0.05 0.1 0.15 0 0.05 0.1 0.15
Measured [kg/s] Measured [kg/s]
0.1 0.1
Relative error []
Relative error []
0.05 0.05
0 0
0.05 0.05
0.1 0.1
Figure 6.12: Modeled and measured port air-mass flow where the wastegate is controlled by
the ECU. In the left column, the tuning data is shown and to the right the validation data
is shown. Top: The ideal air-mass flow refers to when the modeled flow and the measured
flow are equal. Bottom: The accuracy of Eq. (6.17) is better than 10% for most points.
The model has largest relative errors for low flows where small absolute errors result in large
relative errors.
Here the port air-mass flow model with included exhaust manifold pressure
dependency and fuel charge cooling effect is used from Chapter 4:
1
1 rc ppem N
Vd
im
Wcyl = pim C1
1+ A 1 rc 1
( F )s 1
Rim
Tim C2 2
120
| {z }
Charge cooling
(6.17)
The model has only two engine specific parameters, and those are the gain
parameter C1 , which describes the engine pumping capabilities, and the effect
of charge cooling by fuel evaporation C2 . The parameters are determined using
the Matlab-function lsqnonlin.
The high accuracy of Eq. (6.17) is shown in the top of Figure 6.12 for sta-
tionary data where the engine is running with the wastegate controlled by the
ECU. For most points the error is less than 10% even though a different gaso-
line quality has been used in this validation compared to the tuning data set.
ter 4, Figure 4.10. The port-air mass flow model inputs, parameters, and out-
puts are listed in Table 6.6.
Parameter
Inputs Output
A
F s [-] Wcyl [kg/s] We [kg/s] Eq. (6.18)
[-]
Using the first law of thermodynamics, for stationary conditions, the heat
transfer from the gas inside the exhaust manifold can be superpositioned as a
temperature drop on the gas entering the exhaust manifold. This technicality
simplifies the modeling and is easy to implement in the model. The temperature
model is based on the stationary model presented in Eriksson (2002), where the
modeling starts by describing the temperature out of the engine. By assuming
a nominal and spark-timing, the temperature is modeled as an affine function
of mass flow:
We
Te = Te0 + Temax (6.19)
Wemax
The heat transfer from the gas is then superpositioned on this temperature:
htot Dem lem
Temin = Ta + (Te Ta )e Wairfuel cpeg (6.20)
HTexh1 HTexh2
4 We 7 1
hexhint = HTexh0 exh (6.21)
exh ncyl Dem 10 Dem
1 1
htot = Dem lem ncyl (eg 1) 1 1 (6.22)
hexh + hexh We Re eg
int ext
A validation of the total model, Equations (6.19, 6.20), is shown in Figure 6.13,
where the fit for stoichiometric conditions is within 20% for all of the tuning
data points and within 25% for the validation data points. The error decreases
rapidly with increasing load and for part load conditions, the error is less than
5% which is about the same accuracy as the other models. Therefore this model
is sufficiently accurate.
The parameters are determined using the Matlab-function lsqcurvefit. To
determine the thermal conductivity and dynamic viscosity, the polynomials from
Cho et al. (1997) were evaluated at 1100K. The exhaust mass flow temperature
model input, parameters, and output are listed in Table 6.8.
BMEP Vd
Tqcs = (6.23)
4
BMEP = IMEP PMEP FMEP (6.24)
60 2 qHV min(, 1)
IMEP = Wcyl A
e (6.25)
N F s
Vd
PMEP = pem pim (6.26)
N
mps = 2 Stroke (6.27)
60
106 Chapter 6. A Mean Value Model of a TC SI Engine
Table 6.8: Exhaust mass flow temperature model input, parameters, and output. We is
determined using Eq. (6.18). exh and exh were determined at 1100 K.
1100 1100
1000 1000
[K]
900 900
em
Modeled T
800 800
700 700
600 600
1 1
<1 <1
500 500
400 600 800 1000 400 600 800 1000
Measured Tem [K] Measured Tem [K]
Figure 6.13: Measured and modeled turbine inlet temperature. The model has been tuned for
stoichiometric conditions ( = 1) and the non stoichiometric conditions are only shown for
the sake of completeness. Left: For the tuning data the maximum error is 20% that occurs
for low loads. Right: The model predicts up to 25% too high temperatures for low exhaust
temperatures but the error decreases to a 5% error for part load conditions.
6.5. Turbocharger 107
r
75
FMEP =
1000 Bore
0.464 + 0.0072mps
1.8
bl 105 + |0.0215BMEP
{z } (6.28)
BMEPWcyl 60n r
N V CTq1 +CTq2
d
The friction model originates from ETH, see e.g. Soltic (2000). In the torque
model above all parameters are fitted using the method of least squares in two
steps. First the parameters CTq1 and CTq2 are determined, followed by the
two remaining parameters and e . During the parameter tuning only air-fuel
ratios close to stoichiometric have been used. The resulting model is compared
to measured data in Figure 6.14, where it is shown that the model describes
the torque to within 20% for all but very low loads. By including the effects
of fuel enrichment and retarded spark advance it is possible to further improve
the torque model.
In Table 6.9 a summary of the torque model inputs, parameters, and output
are listed.
Table 6.9: Torque model inputs, parameters, and output. The port-air mass-flow input Wcyl
is given by Eq. (6.17).
6.5 Turbocharger
The turbocharger consists of three components:
Compressor It produces the mass-flow through the compressor and its tem-
perature. To produce the air-mass flow a driving torque is required that
also is modeled. See Section 6.6.
Turbine It generates the turbine mass-flow, its temperature, and torque deliv-
ered to the turbine shaft. See Section 6.7.
108 Chapter 6. A Mean Value Model of a TC SI Engine
450
300
400
350 250
Modeled torque [Nm]
250
150
200
150 100
100
50
50
0 0
0 100 200 300 400 500 0 100 200 300 400
Measured torque [Nm] Measured torque [Nm]
Figure 6.14: Validation of measured and modeled crank shaft torque. Left: Modeled and
measured torque for the tuning data. Here the model overestimates low torques. Right: The
torque is described within 20% for the validation data.
Connecting axle Connects the turbine and compressor. This part models the
dynamic behavior of the turbocharger speed. See Section 6.8.2.
this simple model gives similar precision as the model by Jensen et al. (1991)
over the most interesting region of the compressor map. Advantages of the
simpler model are:
Good description of the air-mass flow in the low load region where the
engine operates most of the time.
Simple tuning.
Dc
Ucomp = TC (6.30)
2
The compressor air-mass flow model by Jensen, A.F, Sorenson, Houbak, and
Hendricks (1991) is based on a parameterization of comp1 using comp , the
110 Chapter 6. A Mean Value Model of a TC SI Engine
Table 6.10: Jensen and Kristensen compressor air-mass flow model inputs, parameters, and
output. The blade speed Ucomp is determined using Eq. (6.30).
0.45
0.4
0.35
comp []
0.3
0.25
0.2
0.15
Measurements from manufacturer
Model
0.1
0.7 0.8 0.9 1 1.1 1.2 1.3 1.4
comp []
Figure 6.15: comp and comp follows a quarter of an ellipsis. Here measured and modeled
comp are shown as a function of comp .
Using Eq. (6.31) with comppar = comp2 from Eq. (6.39), the air-mass flow
through the compressor is determined. In Eq. (6.39) the head parameter comp
is calculated using Eq. (6.32). To ensure that Eq. (6.39) is not complex comp
must fulfill:
! 1a
1 16 K1 cpa Taf 2 Dc 2 TC 2 K1 a
comp
16 K1 cpa Taf
1
a
1 16 K1 cpa Taf + 2 Dc2 TC
2
K1 a
(6.40)
16 K1 cpa Taf
In Table 6.11 a summary of the compressor air-mass flow model inputs, param-
eters, and output are listed.
3
NTC= 195000
Pressure ratio comp=pcomp/paf
2.5
NTC= 169000
2 NTC= 143000
NTC= 104000
1.5
1
0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22
Airmass flow [kg/s]
Figure 6.16: Two different compressor air-mass flow models are compared to measured data
from the manufacturer. For low to medium compressor speeds both models give accurate
estimates of air-mass flow through the compressor. At higher compressor speeds and high
mass-flows, none of the models give good agreements with measured data as the models do
not include choke. The speed lines show corrected compressor speed. Due to lack of data the
same data has been used for tuning and validation.
In Figure 6.16 the two models are compared to measured data from the tur-
bocharger manufacturer. Both models show similar accuracy for low to medium
turbocharger speeds but the Jensen and Kristensen model gives slightly better
agreement at higher compressor speeds. However it is considerably harder to
tune the Jensen and Kristensen model as it has six parameters and the objective
function has several local minima. The parameterized dimensionless numbers
method does not require any manual input to determine the correct parameters
and the model gives the same accuracy for lower compressor speeds. As the
engine operates most of the time in the part load region where both models
give the same accuracy, the model using parameterized dimensionless numbers
is chosen to model compressor air-mass flow.
0.06
0.04
Relative error in c []
0.02
0.02
0.04
0.06
1 1.5 2 2.5 3 3.5
Pressure ratio []
Figure 6.17: The modeled and measured compressor efficiency for the data supplied by the
manufacturer (the same data have been used for tuning and validation). The accuracy is
within 8% and the error is largest for low pressure ratios. For high pressure ratios the error
is less than 2%.
location of the maximum efficiency, max , pressure ratio, corrected air-mass flow
(Heywood, 1988, p. 262), and three tunable parameters, a11 , a12 , and a22 :
p
T0 /Tstd
Wcompcorr = Wcomp (6.41a)
p0 /pstd
a11 a12
Q = (6.41b)
a12 a22
Wcompcorr Wcompmax
= p (6.41c)
1 + comp 1 compmax
comp = max T Q (6.41d)
Equations 6.41c and 6.41d require the maximum efficiency max and its loca-
tion in terms of corrected mass-flow Wcompmax and pressure ratio compmax . It
is not probable that this point is included in the measured data supplied by the
manufacturer and therefore max , Wcompmax , and compmax are considered as
parameters which are determined using the Matlab-function lsqcurvefit. Given
the maximum efficiency and its location the parameters a11 , a12 , a22 are given
by the method of least squares. To make sure that the efficiency is not com-
plex, the pressure ratio must fulfill comp 1. The resulting model validation
is shown in Figure 6.17.
114 Chapter 6. A Mean Value Model of a TC SI Engine
Table 6.13: Summary of inputs, parameter, and output of the temperature model of the air
leaving the compressor. The compressor efficiency comp is determined using Eq. (6.41d).
550
500
Model [K]
450
400
350
350 400 450 500 550 600
Measured [K]
Figure 6.18: Validation of compressor temperature out model using the validation data set
instead of the compressor map. The average error is less than 3% and the maximum error is
8%.
6.7 Turbine
Four sub-models describe the turbine:
The parameters k1 and k2 in Eq. (6.45) are determined using the Matlab-
function lsqcurvefit and the validation is shown in Figure 6.19. The relative
error is less than 5% for the data supplied by the manufacturer and for the
measured validation data, the fit is good considering the uncertainties regarding
the temperature before the turbine.
In Table 6.15 a summary of the turbine mass-flow model inputs, parameters,
and output are listed.
0.014
0.012
0.01
0.008
0.01
0.008
0.006
0.004
Figure 6.19: Validation of corrected turbine mass flow. Top: Modeled data is compared to
measured data from the manufacturer and the error is less than 5% for all points. Bottom:
Measured data in the engine lab is compared to modeled data for operating points where the
wastegate is closed. The measured corrected mass-flow is higher than the modeled data as the
measured temperature did not reach stationary conditions at low loads and at higher loads
the engine knocks and retard the ignition timing which increases the temperature and thus
the corrected mass flow.
118 Chapter 6. A Mean Value Model of a TC SI Engine
To make sure that the BSR is not complex the following inequality must hold:
1 eg
1 eg
<1
t
Given the BSR the turbine efficiency is modeled using two parameters. The first
parameter is the maximum efficiency tmax and the second is its corresponding
blade speed ratio BSRtmax .
!
BSR BSRtmax 2
t = tmax 1 (6.48)
BSRtmax
The turbine efficiency model parameters are determined using the Matlab-
function lsqcurvefit. In Figure 6.20 the efficiency model is validated using data
supplied by the manufacturer and it shows a very good fit. A summary of the
turbine efficiency model inputs, parameters, and output are listed in Table 6.16.
In Figure 6.21 the model is validated to measured temperature after the turbine.
The model underestimates the temperature with at most 10%.
A summary of the turbine temperature out model inputs, parameters and
output are listed in Table 6.17.
6.7. Turbine 119
Turbine Efficiency
0.925
Measurement from manufacturer
Model
0.92
0.915
0.91
Efficiency []
0.905
0.9
0.895
0.89
0.885
0.88
0.65 0.7 0.75 0.8 0.85 0.9
Blade Speed Ratio
Figure 6.20: Top: Validation of efficiency as a function of blade speed ratio using the tuning
data supplied by the manufacturer. Here the fit is within 1%.
900
850
800
Modeled Tt [K]
750
700
650
600
550
500
500 550 600 650 700 750 800 850 900 950 1000
Measured Tt [K]
Figure 6.21: The temperature is described within 10%. This model can only be validated
in operating points where the wastegate is closed as the mass-flow through the wastegate is
unknown.
t Wt cpeg Tem 1 t eg
T qt = (6.50)
TC
In Table 6.18 a summary of the model inputs, parameters, and output are listed.
Receiver filling and emptying All control volumes are modeled with filling
and emptying dynamics using two states: pressure and temperature.
Turbocharger speed The connecting axle between the turbine and compres-
sor has dynamics as it is driven by the turbine torque and loaded by the
compressor and friction.
dT 1 1
= (Win cv (Tin T )) + R (Tin Win T Wout ) + Q =
dt mcv mc
v
f dT p, T, Tin , Win , Wout , m, R, V, cv , Q (6.51b)
dt
dp RT mR dT
= (Win Wout ) + = f dp p, T, Win , Wout , m, R, V, f dT
dt V V dt dt dt
(6.51c)
The locations of the control volumes are described in Table 6.19.
Control volume between restrictions
Name Restriction 1 Restriction 2
af air-filter compressor
comp compressor intercooler
ic intercooler throttle
im throttle cylinder (intake valve)
em cylinder (exhaust valve) turbine
t turbine exhaust system
In Table 6.20 the inputs, parameters, and states/outputs are listed for each
control volume.
Here the volume parameter is determined using known engine geometry,
and measurements of the pipe volumes etc. To validate the control volume it is
necessary to know the flows into and out of the volume. Since there is only one
mass-flow sensor, this is not possible. Therefore, the control volume dynamics
are validated during the total model validation.
122 Chapter 6. A Mean Value Model of a TC SI Engine
dTaf 1
= (Waf cva (Ta Taf )) +
dt maf cva
1
Ra (Ta Waf Taf Wcomp ) + Q (6.53a)
maf cva
6.9. Summary of Model Equations 123
paf Vaf
maf = (6.53c)
Ra Taf
q
pa paf
pa paf > plin
qpa Haf Taf ,
Waf = pa pa paf (6.53d)
plin , 0 pa paf plin
Haf Ta
0, pa paf < 0
v
u
u a 1
!2
u a
1
u 1 min K1 cpa Taf 1comp , 1
u 2
2 Ucomp
paf 2 t
Wcomp = D Ucomp
Ra Taf 4 c K2
(6.53e)
Dc
Ucomp = TC (6.53f)
2
dTcomp 1
= Wcomp cva Tcompout Tcomp +
dt mcomp cva
1
Ra Tcompout Wcomp Tcomp Wic + Q (6.54a)
mcomp cva
pcomp Vcomp
mcomp = (6.54c)
Ra Tcomp
Tcompout = Taf +
0 1
a 1
B a C
Taf @comp 1A
2 q 3T 2 q 3
T0 /Tstd T0 /Tstd
6 Wcomp Wcomp 7 a11 a12 6 Wcomp Wcomp 7
max 4 5 4 5
p p0 /pstd max a12 a22 p p0 /pstd max
1 + comp 1 comp 1 + comp 1 comp
max max
(6.54d)
q
pcomp pic
pcomp pic > plin
qpcomp Hic Tcomp ,
Wic = pcomp pcomp pic (6.54e)
Hic Tcomp
plin , 0 pcomp pic plin
0, pcomp pic < 0
124 Chapter 6. A Mean Value Model of a TC SI Engine
dTic 1
= (Wic cva (Ticout Tic )) +
dt mic cva
1
Ra (Ticout Wic Tic Wat ) + Q (6.55a)
mic cva
pic Vic
mic = (6.55c)
Ra Tic
p
Wat = ic () Aeff () (6.55d)
Ra Tic
pim
= (6.55e)
pic
r 2
+1
2
1
() = s (6.55f)
1
2 +1
1
2
1
2
+1 2
+1
+1
0 < +1 2
1 +1
() = () 2
< lin (6.55g)
(lin )
+1
dTim 1
= (Wat cva (Tic Tim )) +
dt mim cva
1
Ra (Tic Wat Tim Wcyl ) + Q (6.56a)
mim cva
pim Vim
mim = (6.56c)
Ra Tim
1
1 rc pem
pim N
Wcyl = pim C1 Vd (6.56d)
1+ 1 rc 1 Rim Tim C2 1
2 120
( F
A
) s
dTem 1
= We cveg (Temin Tem ) +
dt mem cveg
1
Reg (Temin We Tem (Wt + Wwg )) + Q (6.57a)
mem cveg
pem Vem
mem = (6.57c)
Reg Tem
!
1
We = Wcyl 1+ A
(6.57d)
F s
h D l
We tot em em
Temin = Ta + (Te0 + Temax Ta )e Wairfuel cpeg (6.57e)
Wemax
HTexh1 HTexh2
4 We 7 1
hexhint = HTexh0 exh (6.57f)
exh ncyl Dem 10 Dem
1 1
htot = Dem lem ncyl (eg 1) 1 1 (6.57g)
hexhint + hexhext
W e R e eg
( q
Wt =
pem k1
Tem
1 kt 2 , kt 2 1 (6.57h)
0, otherwise
126 Chapter 6. A Mean Value Model of a TC SI Engine
pem
Wwg = p em (em ) Aeffwg (uwg ) (6.57i)
Reg Tem
pt
em = (6.57j)
pem
s
2 eg +1
2eg
eg 1 eg eg
em (em ) = v ! (6.57k)
u eg 1
u 2 +1
t 2eg 2 eg 1
2 eg
eg 1 eg +1 eg +1
eg+1
0 < em eg2+1
eg
1
eg+1
em (em ) = (6.57l)
em (em ) < em lin
2 eg
em (lin )
eg +1
dTt 1
= (Wt + Wwg )cveg (Tesin Tt ) +
dt mt cveg
1
Reg (Tesin (Wt + Wwg ) Tt Wes )) + Q (6.58a)
mt cveg
pt V t
mt = (6.58c)
Reg Tt
q
pt pa
pt pa > plin
qpt Hes Tt ,
Wes = pt pt pa (6.58d)
plin , 0 pt pa plin
Hes Tt
0, pcomp pic < 0
cpeg Ttout Wt + cpeg Tem Wwg Tt Wt + Tem Wwg
Tesin = = out (6.58e)
cpeg (Wt + Wwg ) Wt + Wwg
6.10. Total Model Validation 127
eg 1
Ttout = Tem Tem 1 t
eg
Dt 2
2 u
v TC
! BSRtmax
u
1eg
t2cp Tem 1( 1 ) eg
eg t
tmax 1 (6.58f)
BSRtmax
dTC
ITC = Tqt Tqcomp TC frictionTC (6.59a)
dt
cpa Wcomp (Tcompout Taf )
Tqcomp = (6.59b)
TC
eg 1
t Wt cpeg Tem 1 t eg
T qt = (6.59c)
TC
Two PI-controllers are introduced to handle the throttle and wastegate. One
fast that controls the throttle angle to achieve a desired (measured) air-mass
flow and a second slower controller that controls the wastegate uwg in order to
get the same pressure after the intercooler as in the measured data set. The
latter controller is disabled for low boost pressures as on the modeled engine.
These controllers are used when the stationary model validation is performed
by simulating the system in Figure 6.2.
Results of the stationary validation are shown in the following figures with
measured values on the x-axis and simulated values on the y-axis with one
exception for the wastegate opening that is shown as a function of air-mass
flow.
The inputs to the model are shown in Figure 6.22. As the throttle angle was
determined using a controller it is encouraging to see that the relative error is
less than 5% for all but very low and high pressure ratios over the throttle.
The modeled and measured pressures are shown in Figure 6.23 and the relative
errors are shown in Figure 6.24. The pressure errors are generally small, only a
few percent, except for the intake manifold pressure where there is a mean error
of -5%. To investigate the causes remember that, the throttle plate is controlled
to achieve the measured flow as measured in the validation data. Further, there
is a new air-mass flow sensor compared to when the tuning data was measured.
Now, since the intake manifold pressure is the most important input to the port
air-mass flow model, model errors in the port air-mass flow will propagate to
mostly the intake manifold pressure. Additionally the pressures are controlled
on the intake side to measured pressure, at least after the intercooler, which
reduce their errors when the wastegate is open.
Small, but systematic, underestimations of the pressures on the intake side
are present when the wastegate is closed. On the exhaust side, these errors are
present as a small negative trend in especially pt .
Figure 6.25 shows the modeled temperatures and the relative errors are shown
in Figure 6.26. Here all temperatures on the intake side are modeled within
6% even though no heat transfer is modeled in the control volumes. On the
exhaust side the temperatures are described within 22%. The temperature
after the turbine is overestimated partly due to the large error in the exhaust
manifold temperature.
6.10. Total Model Validation 129
Simulated [deg]
Simulated
Simulated
5000 50 1
0 0 0.5
0 5000 10000 0 20 40 0.5 1 1.5
Measured Measured [deg] Measured
Simulated
Simulated
100 320
0.5
99 300
0 98 280
0 0.1 0.2 98 100 102 250 300 350
Air Mass Flow [kg/s] Measured Measured
80
Relative Error [%]
60
40
Large errors for high pressure ratios
20
Figure 6.22: Stationary model validation. Here the inputs are shown for each operating point.
Bottom: The relative error in modeled throttle angle compared to measured. Except for high
pressure ratios, the errors are only a few percent.
130 Chapter 6. A Mean Value Model of a TC SI Engine
4 p 5 pcomp
x 10 af x 10
9.5 2.5
9 2
Simulated
Simulated
8.5 1.5
8 1
7.5 0.5
7.5 8 8.5 9 9.5 0.5 1 1.5 2 2.5
Measured 4
x 10 Measured 5
x 10
5 pic 4 pim
x 10 x 10
2.5 15
2
10
Simulated
Simulated
1.5
5
1
0.5 0
0.5 1 1.5 2 2.5 0 5 10 15
Measured 5
x 10 Measured 4
x 10
5 pem 5 pt
x 10 x 10
3.5 2
3 1.8
Simulated
Simulated
2.5 1.6
2 1.4
1.5 1.2
1 1
1 1.5 2 2.5 3 1 1.2 1.4 1.6 1.8
Measured x 10
5 Measured x 10
5
Figure 6.23: Stationary model validation of the pressure states. The maximum error on the
intake side is 12% and 10% on the exhaust side.
6.10. Total Model Validation 131
paf pcomp
0.5
2
Rel. Err [%]
0.5 8
8 8.5 9 1 1.5 2
Pressure [Pa] x 10
4 Pressure [Pa] x 10
5
pic pim
0
0
2 2
Rel. Err [%]
4
4
6
6
8
8 10
12
1 1.5 2 4 6 8 10 12 14
Pressure [Pa] x 10
5 Pressure [Pa] x 10
4
pem pt
10
6
8
Rel. Err [%]
4
6
2
4
2 0
1 1
0 <1 2 <1
T T
af comp
325 600
320 550
315
Simulated
Simulated
500
310
450
305
300 400
295 350
290 300 310 320 330 300 400 500 600
Measured Measured
Tic Tim
390 400
380 390
370 380
Simulated
Simulated
360 370
350 360
340 350
330 340
320 340 360 380 400 340 360 380 400 420
Measured Measured
Tem T
t
1200 1200
1 1
<1 <1
1000 1000
Simulated
Simulated
800 800
600 600
400 400
400 600 800 1000 400 600 800 1000
Measured Measured
13 T T
x 10 af comp
1.5
1
1
0.5 0.5
Rel. Err [%]
1
0.5
1.5
1 2
T T
ic im
2
1 0
0 1
Rel. Err [%]
1
2
2
3
3
4 4
5 5
6
6
340 360 380 360 370 380 390 400 410
Temperature [K] Temperature [K]
Tem Tt
20
20
15
15
10
Rel. Err [%]
10 5
5 0
5
0 1 1
<1 <1
10
500 600 700 800 900 500 600 700 800 900
Temperature [K] Temperature [K]
Figure 6.26: Relative errors for the modeled temperatures. On the exhaust side the temper-
atures are overestimated for high temperatures as the temperature model lacks the cooling
effect when fuel enrichment is present.
134 Chapter 6. A Mean Value Model of a TC SI Engine
Figure 6.27 shows the resulting turbine speed and relative error in turbine speed.
The turbocharger speed is shown separately as this is a key component that
connects the intake side to the exhaust side. Turbocharger speed is described
within 10% for most points at medium to high turbocharger speeds. At lower
speeds, the model underestimates the turbine speed.
5
x 10 Turbocharger Speed [RPM]
2.5
2
Simulated
1.5
0
Rel. Err []
10
20
Closed wastegate
Open wastegate
30
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2
Measured [RPM] x 10
5
Figure 6.27: Validation of turbocharger speed. In the plots it is shown where the controller
has opened the wastegate as these point are marked using circles.
xi
pa
xi
pa
6.10. Total Model Validation 135
N p
4 Taf 5p Tc
x 10 af x 10 c
315 500
Temperature [K]
Temperature [K]
Pressure [Pa]
Pressure [Pa]
2050 10 2
310 450
2000 1.5
9 305 400
1950 1
8 300
0 10 20 0 10 20 0 10 20 0 10 20 0 10 20
5p p
5 T p5
x 10 ic x 10 im im
x 10 em
2 2 Temperature [K] 380
Pressure [Pa]
Pressure [Pa]
Pressure [Pa]
3
Angle [deg]
40 360
1.5
1 340 2
20
320
1 1
0 0 300
0 10 20 0 10 20 0 10 20 0 10 20 0 10 20
T 5p T
4
em
x 10 t t
x 10 tc
1.5 900 2
Temperature [K]
Temperature [K]
Pressure [Pa]
Speed [rad/s]
800
1.4 800
700 1.3 1
700
600 1.2
600 0
0 10 20 0 10 20 0 10 20 0 10 20
Time [s] Time [s] Time [s]
Figure 6.28: Dynamic model validation using large throttle steps at 2000 RPM. Solid lines
are the models and the dash-dotted (gray) are measured data. Initially a few seconds are
necessary for the model to converge as it starts from a fix initial state. During the experiment,
the wastegate did not open or just opened slightly as the model accurately describes pressures
on the intake side just before the tip-outs. Also, the model captures the fast temperature
dynamics in the intake manifold, which is important for air/fuel ratio control. Further, it also
suppresses noise in the measured pressure signals.
6.11. Modeling Methodology Applied to a Different Engine 137
N p4 T p4 T
x 10 af af
x 10 c c
9.05 303 10 385
Temperature [K]
Temperature [K]
Pressure [Pa]
Pressure [Pa]
1900 9 380
302 9.5
8.95 375
1800 301
8.9 9 370
1700 8.85 300 365
0 5 10 0 5 10 0 5 10 0 5 10 0 5 10
p4 p4 T p5
x 10 ic x 10 im im
x 10 em
10 380 1.8
Temperature [K]
15
Pressure [Pa]
Pressure [Pa]
Pressure [Pa]
8
Angle [deg]
360 1.6
9.5 6 340 1.4
10
320 1.2
4
9 300 1
5
2 280 0.8
0 5 10 0 5 10 0 5 10 0 5 10 0 5 10
Tem p
5 Tt tc
x 10 t
700 1.26
Temperature [K]
Temperature [K]
Pressure [Pa]
Speed [rad/s]
1.24 6000
700
1.22
600 4000
1.2 650
1.18 2000
500 1.16 600
0 5 10 0 5 10 0 5 10 0 5 10
Time [s] Time [s] Time [s]
Figure 6.29: Dynamic model validation at low load using medium throttle steps at 1800 RPM.
Solid lines are the model and the dash-dotted (gray) are measured data. Initially a few seconds
are necessary for the model to converge as it starts from a fixed initial state. Here it is shown
that even for small steps, the model captures the mean value of the pressure. Further, the
time constant of the turbocharger speed is correct even though the speed is underestimated.
N p4 T p4 T
x 10 af af
x 10 c c
1800 9.05 9.4 400
Temperature [K]
Temperature [K]
Pressure [Pa]
Pressure [Pa]
1600 9 9.2
310 390
1400
8.95 9 380
1200 305
1000 8.9 8.8
370
800 8.85 300 8.6
0 10 20 0 10 20 0 10 20 0 10 20 0 10 20
p4 p4 T p5
x 10 ic x 10 im im
x 10 em
10 400 1.8
Temperature [K]
15
Pressure [Pa]
Pressure [Pa]
Pressure [Pa]
9.4
Angle [deg]
1.6
10 1.4
9.2 5 350
1.2
5 9 1
0 300 0.8
0 10 20 0 10 20 0 10 20 0 10 20 0 10 20
Tem p
5 Tt tc
x 10 t
650 1.3 750
Temperature [K]
Temperature [K]
Pressure [Pa]
Speed [rad/s]
600 700 4000
Figure 6.30: Dynamic model validation using engine speed steps between 900 and 1600 RPM.
Solid lines are the model and the dash-dotted (gray) are measured data.
in that the temperature out of the intercooler had not settled properly for
low mass-flows. Therefore, the parameter values from the B235R intercooler
temperature model was reused as the intercoolers are similar.
To compare the quality of the models, the root mean square of the stationary
model error in percent is used. The results are summarized in Table 6.22. On
the B207R, the errors on the exhaust side are slightly larger except for pt . This
is partly caused by the turbine efficiency where the maximum efficiency is speed
dependent and this is not captured by the model. Also, the turbocharger speed
error is lower for the B207R which is result of that the wastegate opens earlier
on this engine and thus enables the modeled wastegate controller. During the
stationary validation the same type of wastegate controller was used as for the
B235R but with a lower activation pressure. It is therefore concluded that the
model and tuning methodology works for engines of the same physical structure.
6.12. Results 139
RMS Error
State B235R B207R
paf 0.3 0.2
Taf 0 0
pcomp 3.1 2.1
Tcomp 1.2 2.1
pic 3.8 2
Tic 2.1 3.1
pim 5.9 3.9
Tim 3.3 2.5
pem 3.8 7.5
Tem 9.4 10.6
pt 2.1 0.6
Tt 8.2 16.3
TC 21.5 13.7
Table 6.22: Root mean square of the relative errors for the B235R and B207R engine models.
6.12 Results
A component based mean value engine model of a turbocharged SI engine is
presented. The model is suitable for observer design, diagnosis, and control.
For each component, it is described how to determine its parameter(s). To
determine the parameters measured engine data, engine geometry, and data
from the turbocharger manufacturer are required. During the modeling work,
it was observed that the throttle model needs improvement which was realized
by introducing a non-physical value of the ratio of specific heats in the incom-
pressible flow model. Further it is shown that the discharge coefficient in the
compressible flow model can be separated into two factors, one that depends on
the throttle area and one that depends on the pressure ratio.
The model has been validated in several steps. First, each component is
validated and then the total model is validated both for stationary conditions
and for different transients. The stationary validation of the total model shows
that the accuracy on pressures and temperatures on the intake side is better
than 5% for most of the points. In addition, the model is able to predict
turbocharger speed within 5% for medium to high turbocharger speeds. Further,
the model attenuates pumping fluctuations in the measured pressure signals. In
step response experiments the model has the same time constants as the physical
engine. In addition, a sensitivity analysis to changes in ambient conditions was
performed and none of the modeled states are particularly sensitive to changes
in ambient pressure or temperature.
Finally it can be concluded that the component based modeling methodology
is successful in describing the air-system dynamics of a turbocharged SI-engine
and the parameter tuning method makes it easy to tune the model to different
engines.
140 Chapter 6. A Mean Value Model of a TC SI Engine
7
In the previous chapter, a mean value engine model was presented which in-
cludes descriptions of the intake and exhaust side. Using this model it is possible
to estimate pressures, temperatures, mass-flows, and turbocharger speed. How-
ever, the estimates can be improved if some signals can be measured on the
real process. On TC SI-engines, several signals are normally measured such as
pressures before and after the throttle, temperatures in various locations and
the air-mass flow. When a models estimates of states are improved by feedback
from measured signals, it is called an observer. Using an observer it is possi-
ble to estimate signals that otherwise would have been necessary to measure.
This enables the number of sensors to be reduced which means that the cost of
sensors is lower. Further, the observer can be used to create redundant infor-
mation to provide sensor diagnosis opportunities etc. All of this requires that
the sensor(s) used for observer feedback are carefully chosen.
Observers require feedback from one or more sensors and an important ques-
tion that arises during the design phase is: Given a limited number of sensors,
what sensors or sensor configurations are the best choices? This question is
addressed in this chapter. A challenge is that the number of possible sensor
combinations virtually explode when more than one sensor is used, which is
illustrated in the following example. Here there are 17 possible sensor sources.
When two sensors are used there are 17 2 = 136 combinations and if three sen-
17
sors are used then there are 3 = 680 combinations. Therefore, a systematic
method is proposed to reduce the number of possible candidates and aid the
selection of sensors.
A necessary condition in observer design is that the system is observable from
141
142 Chapter 7. Sensor Selection for Observer Feedback
the selected feedback signal(s). It is shown that the system is locally structurally
observable from any measured state or function of measured states such as air-
mass flows. Thus, there are a vast number of possible sensor configurations.
The observed system has 13 states and the primary objective of the model
is air-fuel control that relies on proper cylinder air charge estimates. When a
systematic method to select sensors for observer feedback is proposed, the fact
that only a subset of the states are required for cylinder air charge estimation
is taken into account.
1
0
0
1
Air flow meter, Waf , Wcomp 0
1
0
1 paf , Taf
pic , Tic
Wic pcomp , Tcomp 0
1
Intercooler
111111
000000
000000
111111 Compressor
111111
000000
pem ,Tem
Throttle Turbine shaft, TC
Wth
Turbine
Wcyl
Figure 7.1: A schematic of a turbocharged SI-engine. Arrows pointing out of the engine
indicate possible sensor placements. Arrows inside the engine indicate air-mass flows, of which
the air-mass flow into the cylinders Wcyl is of particular interest for air/fuel ratio control.
7.1.1 Inputs
Inputs to this model are engine speed N , throttle plate angle , the air/fuel
ratio , opening of the wastegate uwg , together with ambient conditions such
as pressure pa and temperature Ta .
Thus it depends only on one state, the pressure after the air-filter. For the other
possible air-mass flows in the intake side the state dependencies are as follows:
Compressor air-mass flow: Wcomp = g(paf , Taf , pcomp , TC )
Intercooler air-mass flow: Wic = g(pcomp , Tcomp , pic )
Throttle air-mass flow: Wth = g(pim , pic , Tic )
7.2 Observability
Before an observer is designed, it must be determined whether the system is
observable and as the system is nonlinear, this is not an easy task. One method
to show that the system is at least locally observable is to linearize the equation
system in Section 7.1.2 in stationary points and then use linear theory to deter-
mine observability. Given that fx (x, u) is the partial derivative of f (x, u) w.r.t.
x, (x0 , u0 ) is a stationary point, then the linearized system matrices A, B, C
and D are defined as follows:
A = fx (x0 , u0 ) B = fu (x0 , u0 )
C = gx (x0 , u0 ) D = gu (x0 , u0 )
The linearized system can now be written as
x = Ax + Bu
7.2. Observability 145
where the matrix O without index means = n, i.e. O = On . Then the system
is observable if and only if O has full column rank (Kailath, 1980, p. 81). The
smallest that makes the system observable is called the observability index.
Unfortunately, the linearized system matrix A is ill-conditioned due to large
differences in time-constants in the engine dynamics. This makes it hard to
numerically determine the observability index using Eq. (7.3) as it, for example
when one feedback is used, involves taking A to the power of 12.
Using this definition, the observability index for a subset of the states can
be defined.
rank O = n1 + rank O2
The structural rank of a matrix can be efficiently computed using graph the-
oretical algorithms for matchings in bipartite graphs. In Matlab, the structural
rank is computed using the dmperm command.
Eqn. paf Taf pcomp Tcomp pic Tic pim Tim pem Tem pt Tt TC
Eq. (7.1a) X X X 0 0 0 0 0 0 0 0 0 X
Eq. (7.1b) X X X 0 0 0 0 0 0 0 0 0 X
Eq. (7.1c) X X X X X 0 0 0 0 0 0 0 X
Eq. (7.1d) X X X X X 0 0 0 0 0 0 0 X
Eq. (7.1e) 0 0 X X X X (X) 0 0 0 0 0 0
Eq. (7.1f ) 0 0 X X X X (X) 0 0 0 0 0 0
Eq. (7.1g) 0 0 0 0 X X X X X 0 0 0 0
Eq. (7.1h) 0 0 0 0 X X X X X 0 0 0 0
Eq. (7.1i) 0 0 0 0 0 0 X X X X X 0 0
Eq. (7.1j) 0 0 0 0 0 0 X X X X X 0 0
Eq. (7.1k) 0 0 0 0 0 0 0 0 X X X X X
Eq. (7.1l) 0 0 0 0 0 0 0 0 X X X X X
Eq. (7.1m) X X X 0 0 0 0 0 X X X 0 X
Using Theorem 7.1 it can be verified that the linearized engine model is
structurally observable from any single sensor signal. Thus the system is locally
structurally observable using one feedback from Y and as Y includes all states
the system is locally structurally observable with feedback from any measured
state. This result is valid for all TC SI-engines with the same structure.
148 Chapter 7. Sensor Selection for Observer Feedback
Consequently, for the total system the lower bound1 of the observability
index depends on the number of system states. As slow sensors introduce ad-
ditional states, it is better to use fast sensors. Further, a specific combination
of feedbacks that would give a minimized observability index can not be found
using the test given by Theorem 7.1 as it only gives a true or false result. To
determine the best combination it is necessary to know if more than the min-
imum number of differentiations is required. Therefore, it is only possible to
give a lower bound of the observability index through Eq. (7.4).
Table 7.1: Observability index for the subset x1 = {pim , Tim , pem } using three feedback
signals.
Index Signal 1 Signal 2 Signal 3
2 pim Tim pem
3 pic pim Tim
3 pim Tim Wth
3 pim pem Tem
4 pim pem pt
4 Tim pem Tem
Only one combination with index 2. The obvious selection to measure the
signals in x1 is the best choice, even though it involves augmenting the
system by one additional state for the temperature sensor dynamics. A
practical aspect is that the exhaust manifold pressure is hard to measure.
In the cases above, the highest power of A is three, and therefore the numerical
problems are avoided. Thus, the analytical method is able to produce a very
short list of good candidates. As the underlying engine model has physically
based parameters, it is reasonable to assume that this result is valid for all TC
SI-engines with the same structure.
7.5 Results
The problem of selecting signals for observer feedback has been studied. Using
a structural method, it is possible to show that the studied system is locally
structurally observable from any measured state or combination of states. A
metric, observability index, is used to aid the selection of what signals that
are most suitable for observer feedback. Two scenarios are considered: The
entire system and a subset of the system. In the latter case the selected subset
of states are those involved in the CAC estimation but the methodology is
generally applicable.
In the first scenario, observability of the entire system, it is best to use
sensors that do not require the system to be augmented with sensor dynamics.
When the system is augmented with sensor dynamics the minimum number of
required differentiations increase and hence the observability index.
In the second scenario, when the application cylinder air charge estimation
is considered, only a fraction of the state space is required to determine the
150 Chapter 7. Sensor Selection for Observer Feedback
cylinder air charge. Therefore, the observability index of this subset of states
is used to evaluate which feedback signal and/or combination of signals that
observes this subset best. When three sensors are used only 6 configurations
of 680 possible reach an observability index less or equal to 4. One unique
combination is best and reaches the observability index 2. It is the obvious
combination of measuring pim , Tim , and pem . Even though it requires that
the system is augmented with one additional state to describe the temperature
sensor dynamics it is the best choice. In addition, the normally measured pic ,
pim , and Tim is a very good combination with observability index 3.
Thus, the proposed method provides the observer designer with valuable
information during the sensor selection. An advantage is that only the model
equations are necessary inputs to the method. To test the observability of the
entire system it is sufficient to know only the structure. As most TC SI engines
are similar in structure, the observability results are valid for all engines of the
same structure.
8
151
152 Chapter 8. Air-System Observer Design
next section. Then the chapter continues with a more detailed description of
the observer design.
120
Pressure [kPa]
100
80
Measured
60 Model
3 4 5 6 7 8 9
315
Temperature [K]
310
305 Model captures
300 temperature dyn.
295 Measured
Model
290
3 4 5 6 7 8 9
Measured
200 Measured filtered
Pressure [kPa]
Model
150
100
3 4 5 6 7 8 9
Time [s]
Figure 8.1: Model compared to measured data during a very rapid throttle step at constant
speed at 1800 RPM. Top: The model reduces noise in the intake manifold pressure signal.
Center: Transient behavior of the intake manifold temperature is captured by the model but
it suffers from a bias before the transient as heat transfer to the gas in the intake manifold is
not modeled. Bottom: Measured exhaust manifold pressure compared to modeled. There is
a bias in the estimated mean exhaust manifold pressure.
154 Chapter 8. Air-System Observer Design
reduce the state biases and thus improve the CAC estimates it is necessary to
introduce feedback from measured signals ym .
An observer is to be designed using the known inputs u, measured outputs
ym , and the non-linear process model x = f (x, u) given by Eq. (8.1). In the ob-
server, the observed states are called x
and the dynamics of x are described by
the non-linear system dynamics f ( x, u) together with feedback from the differ-
ence between the measured signals ym and the modeled measurement signals y.
The observer structure used in this chapter is thus:
= f (
x x, u) (ym y)
x, u) + K(
y = g(
x, u)
weighing between only two design inputs Q and R, which describe the accuracy
of the model and the measurements respectively. The design is performed in
points spread across the operating region of the engine to take the varying
process dynamics, varying quality of the model, and varying quality of the
measurements into account.
1. Read the current design point from the engine map, that is the following
values: N, Wa , pic
3. Store the resulting design point (x0 , u0 ) in a table for later use in the
linearization.
Repeat the process for all remaining entries in the engine map.
Model Linearization
The stationary point of the model were determined in the previous section and
stored in a table according to step 3. Now, for every stationary point (x0 , u0 ),
the model is linearized:
A = fx (x0 , u0 ) B = fu (x0 , u0 )
C = gx (x0 , u0 ) D = gu (x0 , u0 )
Above, the notation fx (x0 , u0 ) refers to the Jacobian of f (x, u) with respect to
the states x evaluated in (x0 , u0 ). The linearized model is in Section 8.2.8 when
the Kalman gains are determined.
4
x 10 Measurement Uncertainty R
9
Deviations from mean value are considered
8.9
as measurement uncertainty
8.8
8.7
8.6
Pressure [Pa]
8.5
8.4
8.3
8.2
8.1
8 Measured
Meanvalue
7.9
3.3 3.4 3.5 3.6 3.7 3.8
Time [s]
Figure 8.2: The measurement uncertainty R describes deviations from the desired measure-
ment which is the mean value. The measurements are considered as independent of the process
and here the measurement uncertainty R is defined as the variance of the measured signal.
The variances of the measured signals are preferably stored during the engine mapping.
1 X 2
N
Ri = y measi ymeasi [n]
N 1 n=1
where i is the i:th measured signal. The uncertainties of the measured signals
are assumed to be independent, which results in a diagonal R matrix.
Further, the matrix R must be positive definite and therefore the elements
are bounded from below to make sure that the matrix is not singular. The
threshold is individually set for each signal to one of the lower measured values.
Examples of thresholds used are: Rpim 2 105 and Rpic 104 .
100
0
0 50 100 150 200
Measured (mapped) [kPa]
Model Uncertainty Q
200
100
0
0 50 100 150 200
Measured (mapped) [kPa]
Figure 8.3: An example illustrating how the model uncertainty parameter is determined
for stationary operating points. Top: Here the measured intake manifold pressure and the
modeled intake manifold pressure is shown. Bottom: The resulting model uncertainty Qpim =
(pim meas pim )2 is shown.
K = P CR1 (8.2)
P = 0 = AP + P AT P C T R1 CP + Q
Observer feedback gains can now be pre-computed offline for each design
point given by the engine map. Later, when the observer is run, a simple table
look-up of the closest feedback gain is used. Look-up keys in the gain switch-
ing are engine speed and intake manifold pressure, as these two are important
parameters for the system dynamics.
sufficient to use an engine map that has been extended with the variance of
the measurement signals that are considered for observer feedback. This design
choice is supported by the fact that it has been tested on two different engines
and no manual tuning of the resulting gains has been necessary, which supports
that the method as suitable for its purpose.
Below, an example is given that shows a potential drawback of the proposed
method to select the model uncertainty. In models where the state equations
have strong dependencies on surrounding states, the use of a diagonal Q causes
the selected uncertainty measure to propagate errors in one state equation to the
surrounding equations. This is illustrated in the following synthetic example,
where there is no error in the x1 state equation but there may be an error
present in the x2 state equation:
x 1 = x1 + k1 x2
x 2 = (k2 + k2 )x2 + u
Given that there
is nomodel error, i.e. k2 = 0, the stationary point will be
u0 and x0 = k2 u0 , uk20 . When the model error k2 is non-zero the stationary
k1
Section 8.2.3, the pressure after the intercooler and the intake manifold pressure.
Given these signals, the observer gains were determined in 67 design points
spread across the operating region of the engine.
First properties of some selected feedback gains are discussed followed by an
observer evaluation using measured engine data. The evaluation shows how the
observer estimates the necessary states in the CAC model and as an example
estimates of the not normally measured turbocharger speed is shown.
Feedback gains from measured pic to pic and pim In the top left corner
of Figure 8.4, it can be seen that the gain is initially increasing for air-
mass flows up to 40 g/s and then stabilize at a lower level. Recall that
the model uncertainty in the intercooler pressure state is higher for low
air-mass flows, which was seen in Figure 6.24. For low air-mass flows,
the measured pic shows low measurement uncertainty. The combination
of model and measurement uncertainty results in high feedback gains to
the pic state itself for low air-mass flows. For higher air-mass flows, above
40 g/s the model uncertainty is lower as the wastegate opens; the observer
therefore relies more on the model and as a result the gain stabilizes at a
lower level. Also, the feedback from pic do influence pim through the cross
coupling to the pim state. This is shown in the bottom left corner, where
the feedback from pic to pim increases for mass-flows up to 40 g/s as the
model quality is less good for these low mass-flows where the wastegate is
inactive.
With the proposed method to determine the model and measurement uncer-
tainty, the resulting observer feedback gains can be applied without any manual
tuning. This is not always the case, Jensen et al. (1997) reports that using a
different definition of model Q and measurement uncertainties R the resulting
observer gains had to be manually reduced.
8.3. Design Example 163
80
1
Gain []
Gain []
60
40
0.5
20
0 0
0 50 100 150 0 50 100 150
Air mass flow [g/s] Air mass flow [g/s]
80
1
Gain []
Gain []
60
40
0.5
20
0 0
0 50 100 150 0 50 100 150
Air mass flow [g/s] Air mass flow [g/s]
Figure 8.4: Four feedback gains have been selected to illustrate how the resulting gains reflect
the weighing of model and measurement uncertainty for the design example. The gains are
shown as a function of air-mass flow. Top left: The gain from measured pic is higher for
mass-flows less than 40 g/s where the model errors are larger. Top right: The gain from
measured intake manifold pressure is lower as the measured signal pimmeas is noisier than
picmeas . Bottom left: Here the coupling between feedback signals and other states is shown.
The observer tries to increase intake manifold pressure when the estimated pressure after
the intercooler is too low. Bottom right: Here it is shown how the observer uses pressure
measurements to improve the turbocharger speed state estimate.
164 Chapter 8. Air-System Observer Design
P = AP + P AT P C T R1 CP + Q (8.3)
120 315
Temperature [K]
Pressure [kPa]
310
100
Excellent noise 305 Model captures
80 reduction temperature dyn.
300
60 295
290
4 6 8 4 6 8
130 140
Pressure [kPa]
Pressure [kPa]
120 130
120
110
110
100
4 6 8 4 6 8
Time [s] Time [s]
Turbocharger Speed TC
9000 Measured
Model
8000 Observed
Speed [rad/s]
7000
6000
5000
4000
3 4 5 6 7 8 9
Time [s]
Figure 8.5: Observed and measured states during a throttle tip-in at constant engine speed
of 1800 RPM. Note that the legend shown in the bottom plot applies to all plots. Top left:
The observer reduces pumping noise. Top right: The intake manifold temperature is only
slightly improved as the model lacks heat transfer to the gas in the intake manifold. Center:
The observer reduces the stationary errors in pem and pic . Here the measured pem has been
zero-phase low pass filtered to improve clarity. Compare it to Figure 8.1. Bottom: Observed
turbocharger speed TC is very close to the measured compared to the open loop simulation.
166 Chapter 8. Air-System Observer Design
Stationary
100 Nonstationary
0.5
80
0.4
Gain []
Gain []
60
0.3
40
0.2
20 0.1
0 0
4 6 8 10 4 6 8 10
Time [s] Time [s]
Figure 8.6: Kalman feedback gains for the stationary and non-stationary Kalman filter during
a throttle transient at constant speed. When the non-stationary observer feedback gains have
converged, they are in the same magnitude as the stationary feedback gain.
cooler, which is the same as used in the previous design example. The filters
are evaluated using a step in throttle at constant speed.
First, selected feedback gains from the stationary and non-stationary Kal-
man-filters are discussed. Selected feedback gains are: the gain from measured
intake manifold pressure to the intake manifold pressure state and the feedback
gain from measured pressure after the intercooler to the intercooler pressure
state. These gains are shown in Figure 8.6. At a first glance, it may seem
like that there are large differences in feedback gains for the pressure after the
intercooler state, but the gains converge to values close to the stationary in
a few seconds. It can be seen that the non-stationary Kalman gains change
like step responses when the stationary feedback gain switches. This as the
measurement and state uncertainties only are available for discrete operating
points given by the engine map. When the stationary observer switches to
another feedback gain, the non-stationary observer starts to use different model
and measurement uncertainties that influence the resulting feedback gain. In
addition, the gains for the non-stationary filter did not converge to the same
value as the stationary gains as the system is not linearized in the same operating
points. The stationary filter is linearized in discrete points given by the engine
map while the non-stationary system is linearized in the current operating point
that results in a slightly different A-matrix. As the A-matrix is differs the
resulting feedback gain Knon stat will be different from Kstat which is seen in
Figure 8.6.
Second, hardly any differences are visible in the state estimates, shown in
8.5. Using Different Signals for Observer Feedback 167
the top of Figure 8.7. Therefore, the differences between them are shown at
the bottom of the figure. For the pressure after the intercooler, the maximum
pressure difference is less than 1.5%. The maximum difference between the
intake manifold pressure state estimates is less than 2.5%. As the differences
between the stationary and non-stationary Kalman filter are small this is good
motivation for the assumption of stationary error covariance P .
pic , pim A standard configuration were two commonly available pressure sen-
sors are used for observer feedback.
pim This setup is shown to indicate that the observer can be used for diagnosis
of in this case pic . For diagnosis it is necessary to create redundant infor-
mation that does not rely on the signals that should be diagnosed. Here
the observer estimates the pressure after the intercooler pic only using
feedback from measured intake manifold pressure.
TC This example shows how the turbocharger speed, which is a not normally
measured signal, can be used to improve both the port air-mass flow and
estimated pressures.
Three different test cases are used: A modest throttle transient at 2500 RPM,
a large throttle transient at 2000 RPM, and an engine speed transient at part
load. Results of the different feedback combinations are shown for each test
case in Figure 8.8, Figure 8.9, and Figure 8.10. The results for each observer
are summarized below:
The standard configuration with feedback from pic and pim estimates the
pressures, air-mass flows, and also the turbocharger speed well for all test
cases.
1.5
12
1.45
1.4 11
1.35
10
Pressure [Pa]
Pressure [Pa]
1.3
9
1.25
1.2 8
1.15 Stationary 7
1.1 Nonstationary
6
1.05
1 5
4 6 8 10 4 6 8 10
Time [s] Time [s]
500 800
600
0 400
200
Pressure [Pa]
Pressure [Pa]
500 0
200
1000 400
600
1500 800
1000
2000 1200
4 6 8 10 4 6 8 10
Time [s] Time [s]
Figure 8.7: State estimates compared from a stationary and non-stationary Kalman filter
during a throttle transient at constant speed. The legend in the top left also applies to the
top right plot. There are no differences that can be detected by visual inspection between
the state estimates provided by the stationary and non-stationary observer. Bottom: State
estimate differences between stationary and non-stationary. During the transients there are
small differences of less than 2.5%.
8.5. Using Different Signals for Observer Feedback 169
higher loads. For lower loads, such as in Figure 8.8 and the speed tran-
sient in Figure 8.10 the feedback from measured intake manifold pressure
alone is insufficient to create accurate estimates of pic and also of TC .
This indicates that when the observer is to be used for diagnosis it is
necessary to select an operating region where the observer gives accurate
estimates. Here the observer can be used to create redundant information
for diagnosis of, in this case, the pressure after the intercooler sensor for
conditions such as in Figure 8.9.
In addition, it can be seen that using any of the observers, the estimated air-
mass flow at the air-filter is close to the measured. This indicates that it might
be possible to replace the air-mass flow sensor by the observer and thereby
achieve a cost reduction. The port air-mass flow estimates Wcyl are also close
to each other regardless of observer feedback. For stationary conditions, Wcyl
can be compared to the measured Waf and it can be seen that the difference
between them are small. This shows that the observers estimate CAC well.
170 Chapter 8. Air-System Observer Design
4 p 5 p
im ic
x 10 x 10
12 1.4
10 1.3
Pressure [Pa]
Pressure [Pa]
8 1.2
6 1.1
4 1
2 0.9
3 4 5 6 3 4 5 6
Waf Wcyl
Massflow [kg/s]
Massflow [kg/s]
0.04 0.04
0.03 0.03
0.02 0.02
0.01 0.01
0 0
3 4 5 6 3 4 5 6
tc
10000
Meas.
pic & pim
8000
Speed [rad/s]
p
im
6000
tc
No feedb.
4000
2000
3 4 5 6
Time [s]
Figure 8.8: A modest throttle transient at a constant speed of 2500 RPM. The legend applies
to all plots. In this case the intake manifold pressure alone is not sufficient to describe neither
the turbocharger speed TC nor the pic accurately.
8.5. Using Different Signals for Observer Feedback 171
5 p 5 p
im ic
x 10 x 10
2.5 2.5
2
2
Pressure [Pa]
Pressure [Pa]
1.5
1.5
1
1
0.5
0 0.5
1 2 3 4 5 1 2 3 4 5
W W
af cyl
Massflow [kg/s]
Massflow [kg/s]
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0 0
1 2 3 4 5 1 2 3 4 5
4 tc
x 10
1.5
Meas.
1 pic & pim
Speed [rad/s]
p
0.5 im
tc
0 No feedb.
0.5
1
1 2 3 4 5
Time [s]
Figure 8.9: A large throttle transient at a constant speed of 2000 RPM. The legend applies
to all plots. All tested signals are well described regardless of observer feedback.
172 Chapter 8. Air-System Observer Design
4 p 5 p
im ic
x 10 x 10
10 1.2
9 pim
1.15
Pressure [Pa]
Pressure [Pa]
8
1.1
7 No fb.
1.05
6
5 1
2 4 6 8 2 4 6 8
Waf Wcyl
0.03 0.03
Massflow [kg/s]
Massflow [kg/s]
Engine Speed
tc
7000 2600
2400
6000
Speed [rad/s]
Speed [RPM]
2200
5000
2000
4000
1800
3000 1600
2 4 6 8 2 4 6 8
Time [s] Time [s]
Figure 8.10: An engine speed transient at part load. The legend applies to all plots. Here it
can be seen that the observer relying only on the intake manifold pressure feedback is not able
to accurately describe neither the turbocharger speed nor the pressure after the intercooler.
For the other cases the observer dynamics is correct but there are stationary biases in pic
when pim or TC are used for observer feedback.
8.6. Results 173
8.6 Results
A systematic and automatic method is proposed for a basic TC SI observer
design. The observer relies on a mean value engine model and the observer is
designed using the CGEKF framework. Here the contribution is a systematic
method to select the design parameters model and measurement uncertainty for
the CGEKF framework.
The purpose of the model uncertainty is to describe the accuracy of the
model in each design point. Here the stationary error of the model in each design
point is used to quantify the model uncertainty. The measurement uncertainty
is given by the variance of the measured signals that are considered for observer
feedback. As input to the design method, it is sufficient to supply an engine map
that has been extended with the variance of the signals considered for observer
feedback. A benefit of the proposed design parameter selection method is that
the resulting CGEKF gains can be applied without any manual tuning. Thus
the method results in a good basic observer design.
Design examples are used to illustrate the benefits of the observer. It is
shown how the observer is suitable for its primary objective of estimating air-
mass flow to the cylinder for air/fuel ratio control as it provides excellent noise
suppression and does not introduce a large phase shift. Other possible applica-
tions are:
1
pim C1 Vd rc pem e
pim
CAC = (9.1)
1+ A1
(rc 1) Rim Tim C2 1
2
( F )s
The cylinder air charge (CAC) estimates thus depend on the following signals:
pim , Tim , pem , and . However, as only stoichiometric operation is considered
in this chapter, can be regarded as a constant.
175
176 Chapter 9. Observer Based Feedforward Air-Fuel Ratio Control
Cylinder air charge estimation using Eq. (9.1) poses several challenges:
The intake manifold pressure pim is subjected to noise as seen in Figure 9.1.
Most of the noise originates from the deterministic engine pumpings which
can be suppressed by a filter with a cut-off frequency selected in such
way that the pumpings are suppressed at idle. A side-effect is the non-
negligible filter lag which in the example in Figure 9.1 results in a more
than 5% error in the CAC estimate.
4
x 10 Measured Intake Manifold Pressure
11
Measured
10 Mean value
Filtered
Pressure [Pa]
9
Filtered signal lags the
8 5% Deviations from mean value significantly
mean value in and the filter introduces a
7 measured signal more than 5% error during
the transients!
6
4.2 4.3 4.4 4.5 4.6 4.7 4.8
315
310
Figure 9.1: Rapid tip-in at a constant speed of 1800 RPM. Top: Measured intake manifold
pressure is subjected to noise caused by engine pumping and electronics. Noise attenuation
using a causal filter results in a significant lag compared to the mean-value. Bottom: Measured
and modeled intake manifold temperature during the transient. As the sensor has a time
constant of approximately 20 seconds, it misses the fast temperature rise captured by the
model.
Example 9.1
Assume that the necessary prediction time is the sum of the following events:
The induction stroke corresponds to half a revolution.
The time to inject the fuel is between one millisecond and up to fully open
injectors. Typical injections last a few milliseconds.
The time necessary to calculate the fuel injection time.
Communication delays are present when the computations are performed
on external hardware.
This sums up to approximately the time of one revolution.
In Figure 9.2, a measured step in throttle at constant speed shows that the
expected error in is approximately 15%. Here ideal prediction was simulated
by translating the modeled port air-mass flow in time corresponding to the time
of one revolution. By taking the ratio of the curves, an estimate of the expected
error in is available.
Cylinder Air Charge (CAC)
600
550
500
CAC [mg]
450
400
350
300 CAC
CAC delayed one revolution
250
4.3 4.35 4.4 4.45 4.5 4.55 4.6 4.65 4.7 4.75 4.8
1.15
1.1
Change []
1.05
0.95
0.9
4.3 4.35 4.4 4.45 4.5 4.55 4.6 4.65 4.7 4.75 4.8
Time [s]
Figure 9.2: A measured step in throttle at constant engine speed 1800 RPM where the port
air mass flow (solid) is compared to a translated, corresponding to perfect prediction one
revolution ahead, port air-mass flow. The difference between the curves is the resulting error
if no prediction is used when the fuel mass is determined and in this case it would result in a
15% error in .
1
Driver command alpha
alpha
x_hat_pred
In1
In1 Out1 In1 Out1
p_im In2 Out
x_hat In Out
Out In Out In Out Out In Out In In
In2 Out2
In2
Out
Out2
In3 u
Predictor
u
Airfilter Control Volume Compressor Control Volume
Airfilter Compressor
Intercooler Control Volume
Intercooler
Throttle Control Volume
Intake manifold
Observer
p_ic
Out
In
Turbine shaft CAC
Out
Estimator
In
Feedforward controller
Feedback controller
Figure 9.3: An overview of a turbocharged SI-engine where the couplings between the intake
and exhaust side are explicitly shown through the connecting shaft and the cylinder air charge
(CAC). The observer estimates an initial condition to the predictor which estimates the system
state one revolution ahead. The predicted state is used to estimate CAC (feedforward).
Measured air-fuel ratio is used as feedback to remove stationary errors. In the figure the
letters CV means control volume, which represents a volume that connects two components.
The control volume have states for pressure and temperature.
et al. (2000) for NA engines. But here turbocharged (TC) engines are consid-
ered, which have a more complex air-system where the turbocharger connects
the intake side to the exhaust side through the turbine shaft and the CAC has
a dependency on the exhaust back-pressure. Therefore, a more sophisticated
controller, and observer, is necessary.
For TC-engines, a traditional control structure is suggested, with one feedfor-
ward part for transient control and one slower feedback controller. The feedback
controller maintains = 1 at stationary conditions using a PI-controller with
feedback from a wide-band oxygen-sensor. This chapter focuses on design of
the advanced feedforward controller that predicts CAC at intake valve closing.
An overview of the controller is shown in Figure 9.3.
alpha [deg]
N [RPM]
N
alpha_wg
u
lambda u
x
K(yg(x,u)) x
K(yg(x,u))
p_a
T_a f(x,u)+K(yg(x,u))
Inputs
N [RPM]
K
K
Matrix
Multiply p_im [Pa] U(E) U
g(x,u)
U(E) U
is shown that the system is locally structurally observable using feedback from
arbitrary state(s). Therefore, the selection is based on which sensors that is
available. On the engine in the laboratory of Vehicular systems the following
sensors are available: Air-mass flow after the air filter, pressure in the intake
manifold, pressure before the throttle, and intake manifold temperature. Pres-
sure sensors are preferred for their fast dynamics and the following two sensors
are chosen as feedback sources: Pressure after the intercooler pic and intake
manifold pressure pim .
The observer gains are then generated offline using the systematic design
method in Chapter 8 and gain switching is then performed online to select a
pre-computed K. This means that the online computational power demand
is low. The gains were determined in 45 different stationary points for engine
speeds between 1200 RPM and 4700 RPM and intake manifold pressures be-
tween 40 kPa up to 160 kPa.
Figure 9.5: The duration of the injection depends on the inducted cylinder air charge. Pre-
diction is necessary as the injection has to be finished before the induction stroke starts. The
required prediction time is the sum of the injection time and the time for the fuel calculation
and in this case the transport delay on the CAN-bus.
Predictor Design
Prediction is implemented by simulating the dynamic system f (x, u) using a
numerical integration method.
Z tpred
(tpred ) =
x f (x, u(t0 ))dt
t0
This type of problem is referred to an initial value problem (Ascher and Pet-
zold, 1998). To solve initial value problems (IVP), it is necessary to provide a
description of the system dynamics and initial values of the states. Here, the
system dynamics is described by Eq. (9.2) that also is used for the observer
and the initial state values are given by the observer. Also, the simulation re-
quires knowledge of future inputs but here this is replaced by setting the inputs
constant to the value at t0 . Next, the prediction end time tpred , integration
method, and integration step size are chosen.
First, the prediction horizon tpred has to be selected. It is determined by
the sum of the computation time, injection time, and the time of the intake
stroke. In the tests performed here the injection time is shorter than half a
revolution (between 3 and 12 ms), the intake stroke is half a revolution, and
1
the computation time 400 as the controller is run at 400 Hz. Additionally, in
this implementation the computations are performed on an external PC which
means that the transfer time over the CAN bus has to added. The PC sends the
calculated injection time with 400 Hz and the ECU reads the CAN-bus every
1
160 seconds. The timing diagram is shown in Figure 9.5 and the total prediction
time is:
1 60 1 1
tpred t0 = + + + tinj
2N 400 160
In Figure 9.6 the delays in the crank angle domain are shown for the injection,
computation, and transmission. As the injection time depends on the inducted
CAC which is not known means that the injection time is not known either.
One method is to iterate and solve for the prediction time or use a worst case
scenario based on the longest possible injection time. Instead a compromise is
made by taking a mean value of the minimum delay and maximum delay for the
182 Chapter 9. Observer Based Feedforward Air-Fuel Ratio Control
400
200
0
6000
5000 15
4000 10
3000
2000 5
Engine 1000 0
Speed
[RPM] t [ms]
inj
Transport and Computational Delays
350
300
Duration [deg]
250
200
150
100
50
1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000
Engine Speed [RPM]
Figure 9.6: Crank angle duration of the injection and the computation and transmission
delays. For engine speeds between 1500 and 3000 RPM the sum of the delays are between 90
to 360 degrees.
studied operating regime. For engine speeds between 1500 and 3000 RPM the
required prediction time is between 270 and 540 degrees and the mean of those
is approximately the time of one revolution. The prediction time is therefore
fixed to the time of one revolution. Given the required prediction time, the
state at tpred can be determined by simulating x = f (x, u) with the assumption
that the input u is constant between t0 and tpred .
Second, a numerical integration method have to be selected and here an
explicit one step method, Runge-Kutta 4 (RK4), was chosen with an integration
1
step size of step size of 400 .
Given the predicted states x (tpred ), the CAC at tpred is estimated:
1
pem (tpred ) e
pim (tpred )C1 Vd rc pim (tpred )
[ pred ) =
CAC(t
1+ A 1
(rc 1)Rim Tim (tpred ) C2 1
2
( F )s
e = ( ref ) (9.3a)
Z
1
y = KFB (e + e dt) (9.3b)
TI
9.3. Experimental Setup 183
1 0.25 1
lambda_meas Out1
K_FB Saturation
2 6 1
s
lambda_ref
1/Ti Integrator
In this case the output y is limited and also the integrator is limited in order to
avoid wind-up. The selection of KFB and T1I are not crucial and the following
values were chosen using step response experiments: KFB = 0.25 and T1I = 6.
Figure 9.7 shows the Simulink implementation, note that the output has been
limited to 10.5. From Figure 9.3 it can be seen that the output of the controller
is multiplicative, which is standard in air-fuel ratio control as most errors are
multiplicative.
\ Z
CAC(t pred ) 1
mf = A
K FB ( 1) + ( 1)dt (9.4)
TI
| F s
{z } | {z }
Feedforward Feedback
0000000
1111111
0000000
1111111
1111111111111111
0000000000000000 Actuators
00000000000000001111111
0000000
1111111111111111
Sensors
Figure 9.8: The PC is running a realtime operating system (Linux with the RTAI patch) and
communicates with the ECU over a CAN-bus. The ECU is responsible for sending sensor
readings to the PC and also to inject the determined amount of fuel.
4
2 2.5 3 3.5
Measured Engine Speed
2060
Speed [RPM]
2040
2020
2000
2 2.5 3 3.5
Time [s]
Figure 9.9: Here the high repeatability of the engine test cell is shown. The repeatability is
excellent as it is hard to distinguish any differences between the four measurements.
80
Pressure [kPa]
70
60
Transient
50
40
Figure 9.10: Top: Intake manifold pressure. Bottom: Conventional speed density control,
with filtered intake manifold pressure, results in a 27% error. With a disabled low-pass filter
the error was still 14%. Using the observer without prediction reduces the error to a 10% lean
transient in , shown as a dash-dotted line. When the suggested observer is used together
with prediction of the CAC one revolution ahead, the error decreases to approximately 3%.
9.4. Experimental Results 187
9.4.4 Validation
Above, the controller was been tested at 2000 RPM in one operating point. To
validate the controller for a larger operating region, the following tests have
been done:
1. Test the same rapid medium load step at 1500 RPM and 3000 RPM. This
shows that the controller is able to control the air-fuel ratio accurately at
different speeds without modifications.
2. A larger throttle step, where the wastegate opens has also been made for
three different speeds: 1500, 2000, and 3000 RPM.
80 80
Pressure [kPa]
Pressure [kPa]
60 60
40 40
p p
im im
20 20
5 10 15 20 25 30 5 10 15 20 25 30
1.05 1.05
[]
[]
1 1
0.95 0.95
0.9 0.9
5 10 15 20 25 30 5 10 15 20 25 30
Time [s] Time [s]
Figure 9.11: Rapid load transients at 1500 RPM and 3000 RPM with the suggested observer
based controller. Top: Intake manifold pressure. Bottom: Measured air-fuel ratio. During
transients the resulting errors in are less than 7% for both operating points.
1
enable wg
1
wg_pos [0,1]
2 ref Product
p_ic_ref
u
3 y
p_ic
PIcontroller
Figure 9.12: The wastegate controller is of a PI-type with K = 5 105 and T1 = 10. The
I
controller opens the wastegate when the observed pressure after the intercooler is higher than
the measured. There is also an on/off functionality to make sure that the wastegate is closed
for low boost pressures.
9.4. Experimental Results 189
100
80
60
40
4 4.5 5 5.5 6 6.5 7 7.5 8
2000 RPM 40 > 120 kPa
120
pim [kPa]
100
80
60
40
4 4.5 5 5.5 6 6.5 7 7.5 8
3000 RPM 40 > 120 kPa
120
pim [kPa]
100
80
60
40
4 4.5 5 5.5 6 6.5 7 7.5 8
Time [s]
Figure 9.13: A larger load step at constant engine speed. In all cases the transient consists
of two parts. First the rapid increase in intake manifold pressure up to ambient which takes
approximately 2 to 4 revolutions. It is followed by a slower increase in pressure that depends
on the time required for the turbocharger to reach sufficient boost pressure.
most of the reduction of the error in originates from the introduction of the
observer. When the prediction is added to the observer a further reduction of the
maximum error is possible and did not reach more than 6% in the transients..
14% error
1 4% error
12% error
1 5% error
5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 6
Time [s]
Figure 9.14: Measured for three different controllers using a large rapid load transient at
three different speeds. The legend shown in the top plot applies to all plots. For all speeds
there is a lean transient in . As in the previous tests the introduction of the observer is able
to reduce the error in with 50% compared to pure speed-density CAC estimation. When the
predicting step is enabled a further reduction is possible and the maximum error is reduced
to 6%.
9.4. Experimental Results 191
The effect of the fuel pressure regulator settling time. Depending on type
this can vary between a few milliseconds up to 0.5 second (Hubbard, 2002).
Model Errors
As CAC is only measurable for stationary conditions a test of the feedforward
component, observer with prediction, was made by disabling the -feedback
controller. The result is shown in Figure 9.15 and note that regardless of engine
speed there is a small lean transient after each tip-in. This transient error can
be caused by several phenomenon such as:
The prediction time should vary with the injection time, but here a fix
prediction of one revolution is used.
4
x 10 Transient at 1500 RPM
10 1.125
pim
9 1.1
Pressure [Pa]
8 1.075
7 1.05
6 1.025
5 1
4 0.975
3 0.95
0 2 4 6 8 10
4
x 10 Transient at 2000 RPM
10 1.125
p
9 im 1.1
Pressure [Pa]
8 1.075
7 1.05
6 1.025
5 1
4 0.975
3 0.95
0 2 4 6 8 10
4
x 10 Transient at 3000 RPM
10 1.15
pim
9 1.125
Pressure [Pa]
8 1.1
7 1.075
6 1.05
5 1.025
4 1
3 0.975
0 2 4 6 8 10
Time
Figure 9.15: Measured data with prediction enabled but with the feedback controller disabled
for three different speeds. The accuracy of the model is measured using the difference between
before the step and at the end of the step. The important result here is that the stationary
error of the model is small at 1500 and 2000 RPM. For 3000 RPM there is a large difference
in stationary, approximately 10%, error before and after the step that have to dealt with by
the -feedback controller. Finally, note that there is a lean transient after each tip-in.
9.4. Experimental Results 193
the requested throttle area as input instead. As no model of the throttle servo
was available, this approach has not been implemented. Another approach was
applied in Chevalier et al. (2000) where the throttle actuation was delayed in
order to know the throttle position in advance. This was not followed here as
it introduces an undesirable torque delay.
The expected error in from the assumption of a constant throttle area
during the prediction is that there would be a lean transient during tip-ins
and a rich transient during tip-outs. In Figure 9.11, there are systematic lean
transients during tip-ins and small rich transients during the tip-outs. However,
these can also be caused by fuel dynamics, which is the next topic.
8
x 10 Throttle Area Port Airmass Flows
0.025
MassFlow [kg/s]
1.5
Area [mm2]
0.02
1 Wc
0.015
Ideal prediction of Wc
Measured Area
0.5 0.01 RK4 Prediction of Wc
Translated Area
2.35 2.4 2.45 2.5 2.55 2.6 2.35 2.4 2.45 2.5 2.55 2.6
300 0
Area change [%]
250
10
Error [%]
200
150 20
100 30
Wc without prediction
50
40 RK4 Prediction of Wc
0
2.35 2.4 2.45 2.5 2.55 2.6 2.35 2.4 2.45 2.5 2.55 2.6
Time [s] Time [s]
Figure 9.16: The prediction is carried out with the assumption of fix throttle area and the
prediction time corresponds to one revolution or 30 ms at 2000 RPM. Left column: Here the
magnitude of the area change is shown for a tip-in. The constant area assumption results in
a short error of up to 300%. Right column: Estimated port air-mass flows. The predicted
mass-flow using RK4 and the constant area assumption results in a short error of up to 40%
which is considerably smaller than the throttle area error!
Fuel Dynamics
Fuel dynamics is related to that not all fuel vaporizes during injection as a
fraction of the injected fuel fp forms a liquid film on the intake manifold wall.
Fuel dynamics is therefore commonly modeled as a puddle which mass is
modeled using one state mfp and Eq. (9.6). This film evaporates with a time
constant of fp and then enters the cylinder as vapor together with the fuel that
was not deposited in the puddle, Eq. (9.7).
dmfp 1
= fp Wfinj mfp (9.6)
dt fp
1
Wfc = (1 fp )Wfinj + mfp (9.7)
fp
This phenomena has been studied and for port injected engines in for example
Fozo and Aquino (1988) where the effect is most pronounced for cold engines.
According to Fozo and Aquino (1988) the effect is around 5% during a slow load
194 Chapter 9. Observer Based Feedforward Air-Fuel Ratio Control
transient and warmed up engine but others claim higher values. An example is
shown in Guzzella and Onder (2004, p. 55) where the identified fp for a fully
warmed up 1.8 dm3 engine can reach values close to 50% for high loads. The
time constant fp varies between 0.2 seconds and up to 0.5 seconds according
to the same reference for the 1.8 dm3 engine. Similar values are described in
Simons et al. (1998). The parameters are engine dependent and the values
shown here is only a guideline.
The effect of fuel dynamics would be a lean transient during the tip-in and
a rich transient during the tip-out. In the measured data, there are systematic
lean and rich transients at tip-ins and tip-outs respectively but this is the same
effect that can be expected by the unknown inputs during the prediction. Hence,
the effect of fuel dynamics can not be excluded and it is possible that the
introduction of fuel dynamics can improve the air-fuel ratio control further.
9.5 Results
TC SI engines have a complex air-system that makes precise air-fuel ratio con-
trol a challenging task. Accurate air-fuel ratio control relies on precise cylinder
air charge estimates. To estimate the CAC, knowledge is necessary of the in-
take manifold pressure and temperature together with the exhaust pressure.
Especially transient control requires the fast intake manifold dynamics to be
taken into account. Therefore an observer based feedforward control structure
is proposed.
Two controllers are used, one feedforward and one feedback. The feedback
controller cancels stationary errors using feedback from measured . The feed-
forward controller consists of an observer and a predictor that both are based
on a mean value engine model. The observer estimates the system state us-
ing feedback from measured pressure after the intercooler and intake manifold
pressure. The estimated system state is used as an initial condition to the pre-
dictor, which then estimates the system state one revolution ahead. Given the
predicted system state it is now easy to estimate the CAC. In engine experi-
ments, the deviation from stoichiometric is reduced from almost 30% down to
less than 7% over a wide range of engine speeds. Most of the improvement
originates from the introduction of a model-based observer.
10
Conclusions
195
196 Chapter 10. Conclusions
Per Andersson and Lars Eriksson. Cylinder Air Charge Estimator in Tur-
bocharged SI-Engines. Society of Automotive Engineers, 2004. SAE Technical
Paper No. 2004-01-1366.
U.M Ascher and L.R Petzold. Computer Methods for Ordinary Differential
Equations and Differential-Algebraic Equations. siam, 1998. ISBN 0-89871-
412-5.
197
198 References
J.W. Grizzle, J.A. Cook, and W.P. Milam. Improved Cylinder Air Charge
Estimation for Transient Fuel Ratio Control. In Proceedings of the American
Control Conference, pages 15681573, Baltimore, June 1994.
Lino Guzzella and Alois Amstutz. Control of diesel engines. IEEE Control
Systems, pages 5371, October 1998.
Tomas Henriksson and Lars Nielsen. Modeling effects due to varying com-
mand signal timing. In IFAC Workshop Advances in Automotive Control
(Preprints), 1998.
References 201
Per B. Jensen, Mads B. Olsen, Jannik Poulsen, Elbert Hendricks, Michael Fons,
and Christian Jepsen. A New Family of Nonlinear Observers for SI Engine
Air/Fuel Ratio Control. In Electronic Engine Controls, SP-1236, pages 91101,
1997. SAE Technical Paper No. 970615.
James C. Peyton Jones. Modeling Combined Catalyst Oxygen Storage and Re-
versible Deactivation for Improved Emissions Prediction. 2003. SAE Technical
Paper No. 2003-01-0999.
Uwe Kiencke and Lars Nielsen. Automotive Control Systems. For Engine,
Driveline, and Vehicle. Springer-Verlag, 2000. ISBN 3-540-66922-1.
Allan J. Kotwick, John D. Russell, Ross Pursifull, and Don Lewis. An air
meter based cylinder air charge estimator. In Electronic Engine Controls 1999,
SP-1419, pages 8190, 1999. SAE Technical Paper No. 1999-01-0856.
J. David Powell. Engine Control Using Cylinder Pressure: Past, Present, and
Future. Journal of Dynamic Systems, Measurements, and Control, 115:343
350, June 1993.
Ross Pursifull, Allan J. Kotwick, and Sulgi Kong. Throttle Flow Characteriza-
tion. In Modeling of SI Engines, SP-1511, pages 149161, 2000. SAE Technical
Paper No. 2000-01-0571.
Tung-Cing Tseng and Wai K. Cheng. An adaptive air/fuel ratio controller for
SI engine throttle transients. In Electronic Engine Controls 1999 (SP-1419),
pages 3748, 1999. SAE Technical Paper No. 1999-01-0552.
Robert W. Weeks and John J. Moskwa. Transient Air Flow Rate Estimations
in a Natural Gas Engine Using a Nonlinear Observer. 1994. SAE Technical
Paper No. 940759.
Nicholas Wickstr
om and Stefan Byttner. Robust Tuning of Individual Cylin-
ders AFR in SI Engines with the Ion Current. In Electronic Engine Controls
2005, SP-1975, 2005. SAE Technical Paper No. 2005-01-0020.
Yuji Yasui, Shusuke Akazaki, Masaki Uenzo, and Yoshihisa Iwaki. Secondary
O2 Feedback Using Prediction and Identification Type Sliding Mode Control.
In Electronic Engine Controls, SP-1501, pages 169178, 2000. SAE Technical
Paper No. 2000-01-0936.
A
Nomenclature
A variable specifying the type (presser, temperature, etc.) and subscript are
used to describe the location. Example: paf refers to the pressure after the
air-filter.
Variable Symbol
Pressure p
Temperature T
Mass-flow W
Efficiency
Pressure ratio
Angle
Torque Tq
Diameter D
Volume V
MAss m
Subscripts are used in the following way to indicate the location of the described
variable:
205
206 Appendix A. Nomenclature
Location Subscript
Ambient or air a
Exhaust gas eg
Air-filter af
Compressor comp
Intercooler ic
Intake manifold im
Cylinder cyl
Exhaust manifold em
Turbine t
Wastegate wg
Crank shaft cs
Displacement d
Before before
After after
All pressures are assumed to be static pressures and all temperatures are
assumed to be stagnation temperatures as this is the output of the sensors used
(Ekroth and Granryd, 1991).
Other symbols used are:
Symbol Description
A () Throttle area
Aeff () A fitted function to the measured product of area and
discharge coefficient
A
F s Stoichiometric air/fuel ratio
Cd () Discharge coefficient for the throttle
cv Specific heat at constant volume
k1 Constant in polynomial for nominal exhaust manifold
pressure
k2 Constant in polynomial for nominal exhaust manifold
pressure
k Scaling factor to calculate air and fuel mass given air
1
mass, k = 1 + A
( F )s
Ke Constant in equation for exhaust manifold pressure dif-
ference
Kim Filling/emptying constant Kim = Rim Tim
Vim for the intake
manifold
Kinj Maximum delivered fuel mass per second
Kobs Feed-back gain from pressure estimation error in ob-
server using proportional feed-back to the pim state
K1 Feed-back gain from pressure estimation error in the 2-
state observer to the pim state
K2 Feed-back gain from pressure estimation error in the 2-
[ state
state observer to the CAC
207
R J
Gas constant, 8.31 [ moleK ]
pem nom Nominal exhaust manifold pressure
pem Exhaust manifold pressure difference from nominal
pressure
pem Mean of exhaust manifold pressure difference from nom-
inal pressure
pim Estimated intake manifold pressure
rc Compression ratio of the engine
Rc Specific in cylinder gas constant at intake valve closing
Rim Specific gas constant in the intake manifold
T1 Temperature of charge (air, fuel, and residual gases) at
start of compression
Tr Temperature of residual gases
t0 Time in seconds for the injector needle lift or start of
prediction time
tinj Time in seconds where the injector is open
Vc Clearance volume
Vr Volume of residual gases
Wa Measured air-mass-flow
Wcyl Air-mass-flow to cylinder
Wcylstd Air mass flow to cylinder using mapped volumetric effi-
ciency
Wcylts Air mass flow to cylinder using mapped volumetric effi-
ciency with estimated offset vol
xr Residual gas fraction
Throttle angle
c
Ratio of specific heats cvp
vol Volumetric efficiency
vol Offset in volumetric efficiency
vol Estimated offset in volumetric efficiency
208 Appendix A. Nomenclature
Air-fuel ratio is the composition, on mass basis, of air and fuel in the cylinder
when the intake valve has closed. Denote the mass of air by ma , the mass of
ma
fuel by mf , and the air-fuel ratio is then m f
. In most cases the normalized
air-fuel ratio is used, that is the air-fuel ratio divided by the stoichiometric
A
ratio. The stoichiometric air-fuel ratio F s
describes the ratio of air and fuel,
on mass basis, where it is just enough air to fully oxidize the fuel. A typical
stoichiometric reaction of air and a fuel is shown below.
b c
Ca Hb Oc + a + (O2 + 3.77N2 )
| {z } 4 2
Fuel | {z }
Air
| {z }
Reactants
b b c
aCO2 + H2 O + a + 3.77N2
2 4 2
| {z }
Products
209
210 Appendix B. Definition of Air-Fuel Ratio
Two other common definitions are lean and rich mixture. For lean mixtures
there are excess air, > 1, and in rich mixtures there are more fuel than the
available air can oxidize, < 1. At stoichiometric conditions the normalized
air-fuel ratio is one.
C
Instead of using energy balance as in (Taylor, 1994, pp. 510), which lacks pa-
rameters that can be tuned for a specific engine, the volume of inducted air Va
is estimated (Eriksson et al., 2002a) and the CAC is hence:
pim Va
CAC = (C.1)
Rim Tim
Using the fact that the ideal gas law can be expressed like
nRT RT
V = = (n1 + + nn )
p p
RT
(V1 + + Vn ) = (n1 + + nn )
p
RT
Vi = ni 1in
p
This results in that the gases in the cylinder can be divided into volumes of air,
fuel, and residual gases, which is illustrated to the right of Figure 4.8.
The volume of air Va and evaporated fuel Vf is simply estimated by subtract-
ing the volume that the residual gases occupies at intake valve closing (IVC)
from the total volume.
V a + Vf = V +V Vr
| {z } | d {z }c
Volume of air and fuel Total cylinder volume
211
212 Appendix C. Derivation of pem Dependent CAC Model
from volume Vc which gases occupies at exhaust manifold pressure pem to the
volume Vr which they occupy at IVC and intake manifold pressure pim :
1
pem
Vr = Vc
pim
Here the residual gas volume at EVC is assumed to be constant and equal to
the clearance volume Vc . This is reasonable as the engine is not equipped with
any cam phasing device and also by the fact that volumes change only slightly
around TDC. The remaining volume, after the expansion of the residual gases
to intake manifold pressure, is then compensated for the volume of the fuel vapor
(all fuel is assumed to enter the cylinder as vapor). The volume of inducted air
is then Va :
1
pem
rc pim
Vaf = Vd
rc 1
1
Va = 1 Vaf
1+ ( F
A
) s
Here it will also be assumed that the fresh mixture of air and fuel Taf has
1 Here m denotes the measured CAC. This is a deviation from the standard nomenclature
a
to keep the notation in the energy balance consistent.
213
214 Appendix D. Alternative Derivation of pem
R
pem nom CAC Taf k (D.6b)
Vr
given the volume of the residual gases is Vr . As the term to the left of pem nom
in Eq. (D.6b) is linear in pim and changes in pim are small when the exhaust
manifold pressure changes, these terms cancel each other which results in:
!
1 R
pem = CAC Taf 1 + A
(D.7)
F s |{z}Vr
| {z }
Ke
k
For a motivation of the small change in pim when the wastegate is opened see
for example Figure 5.4. In the equations above Vr is constant as the engine
D.2. Summary of Exhaust Pressure Calculation Process 215
does not have variable valve timing. For an ideal otto cycle Vr = Vc but a real
engine has valve overlap and heat transfer, therefore Vr is not necessarily equal
to Vc . In the final exhaust manifold pressure model, the parameters Vr and R
are lumped together to Ke . Ke is determined from measured wastegate steps
and in the studied region, from 1800 to 3100 RPM and brake mean effective
pressures from 5 to 12 bar, a fixed Ke 2.4 103 kgK
Pa
was used.
Further support for the assumption of constant Ke is given when the CAC
sensitivity to changes in exhaust manifold pressure, Eq. (4.12), is studied. In
Figure D.1, the sensitivity is shown for operating points with nominal wastegate
setting over the entire operating region. The result is plotted as a function of
intake manifold pressure and it is clear that the sensitivity is small (0.2 to
0.1) for intake manifold pressures above 50 kPa.
0.06
0.1
0.12
0.14
0.16
0.18
0.2
50 100 150 200
Intake manifold pressure pim [kPa]
Figure D.1: Sensitivity to exhaust manifold pressure changes are small over the operating
region of the engine. However for low intake manifold pressures the magnitude of the sen-
sitivity increases which suggest that the exhaust pressure estimation can be improved by
incorporating the intake manifold pressure.
offset by Eq. (D.7) and finally inserted into Eq. (5.1) which yields pem .
1
Taf (Tim , ) = Tim C1
2
vol pim
z }| {
nr (a1 pim + a0 ) Vd
CAC (Wa , N, pim , Tim , ) = Wat
N RTaf (Tim , )
pem nom (Wa ) = pa + k1 Wa + k2
1
k = 1+ A
F s
pem (Wa , CAC, k, Tim ) = pem nom (Wa ) Ke kTim CAC
Important second order effects, such as heat transfer, and valve overlap etc. are
taken into account by pem nom .
E
Experimental Setup
1. Initially a SAAB B235R was used for the wastegate experiments, to de-
velop the mean value engine model, and the observer design methodology.
2. A modified SAAB B207R, was used for the real-time control experiments.
217
218 Appendix E. Experimental Setup
CAN
00000000000
11111111111
Sensors Firewire
CAN
Sensors Sensors Measurement
Engine
ECU System
HP E1433A
Actuators HP E1415A
Axle
Torque Sensor
Dynamometer
X-ACT
Figure E.1: Experimental setup. The two computers, measurement system and X-ACT are
located in the control room. The engine, dynamometer and engine control system (ECU) are
in the engine test cell.
The additional temperature sensors are listed in Table E.2 and the additional
pressure sensors are listed in Table E.3. Further sensor signals are listed in
Table E.4. Note that the laminar air-mass flow sensor is not always connected.
The approximate location of the sensors are shown in Figure E.2.
Nr Location Designation
1 Ambient pressure pa
2 Before compressor paf
3 After compressor pcomp
4 After intercooler pic
5 Intake manifold pim
6 Exhaust manifold pem
7 After Turbine pt
Table E.3: Available pressure sensors. All sensors except for the ambient pressure are high
performance Kistler Kristall sensors (Kri, 1997). The ambient pressure is measured through
the laminar flow meter LFE3, which is equipped with an ambient pressure sensor.
Dynamometer
Table E.4: Miscellaneous sensors. The second air-mass flow sensor is a laminar air-mass flow
sensor.
Air filter
Wa
pic , Tic pcomp , Tcomp
1111
0000 Intercooler
0000
1111
0000
1111 Compressor
pem , Tem
Throttle Turbine Shaft
Turbine
pim , Tim pt , Tt
Waste-
Engine gate
Intake Exhaust
Manifold Manifold
Catalyst
Figure E.2: Engine schematic with approximate sensor locations. Note that the wastegate
position is not measurable.
Manufacturer Schenck
Model Dynas2 220
Maximum power 220 kW
Maximum torque 450 Nm
Max speed 9500 RPM
Table E.6: B207R: Available temperature sensors. The thermocouples are of type K and
3 mm in diameter.
Nr Location Designation
1 Ambient pressure pa
2 Before compressor paf
3 After compressor pcomp
4 After intercooler pic
5 Intake manifold pim
6 Exhaust manifold pem
7 After Turbine pt
Table E.7: B207R: Available pressure sensors. All sensors except for the ambient pressure are
high performance Kistler Kristall sensors (Kri, 1997).
Dynamometer
An asynchronous Schenck Dynas3 250 dynamometer is fitted to the engine.
With this type of dynamometer it is possible to either brake or supply torque
222 Appendix E. Experimental Setup
to the engine. The later is used to start the engine and gives the possibility to
simulate downhill driving.
The dynamometer is controlled via a user interface called X-ACT. From
X-ACT the engine speed and engine throttle position are controlled. It is also
possible to control the engine speed and throttle position from a computer via a
serial interface (RS-232). This is done during for example the engine mapping.
Dynamometer data are listed in Table E.9.
Manufacturer Schenck
Model Dynas3 250
Maximum power 250 kW
Maximum torque 480 Nm
Max speed 10000 RPM
Isetta Communicates with the ECU over a CAN-bus. From Isetta it is possible
to set and read data in the ECU. Isetta is used mainly during the real-time
experiments to enable/disable controllers within the ECU.
Saab Controls the measurement system and the dynamometer. See Section E.2.1.
HP E1415A
E1415A can measure up to 64 channels with a maximum sampling frequency
of 2000 Hz and it features a built-in self calibration. The raw measurement
signals are adapted to the instrument through signal conditioning hardware.
This instrument can be equipped with up to eight different signal conditioning
modules, where each covers a maximum of eight channels. The installed signal
conditioning modules are listed in Table E.10.
To measure the resistance over the PT200 elements the current source mod-
ule is used and the voltage over the element is measured using the so called
4-wire principle. As the voltage and the current are known it is easy to calcu-
late the resistance and hence the temperature.
HP E1433A
This instrument can measure eight channels simultaneously using separate A/D
converters at a sampling frequency of up to 196 kHz. It can further perform
measurements in the crank angle domain. Here E1433A is used to measure the
noise levels which are used in the observer design in Section 8.2.
sfun_comedi_data_read 1 auto
(???) INPUTS sfun_rtai_scope N
Throttle Area Data Type Conversion8 Wcomb [mg/c]
Gain2 y
volt2area
CAC No Prediction
N
N [RPM] auto y
Wcomb [mg/c]
(???)
Data Type Conversion1 Predicted CAC Euler 1
Engine Speed
N
0 auto Wcomb [mg/c]
(???) y
Data Type Conversion2
wastegate Predicicted CAC RK4
1
1 predHorizon
lambda ref
Observer/Predictor
105e3
p_a Inputs
States Feedback
293
T_a N
enableObs
Activate data
un_comedi_data_re
lambda, kanal 3
sfun_comedi_data_read K auto
(???)
p_ic, kanal 0 Gain3 Data Type Conversion3
lambda_meas sfun_rtai_scope
1
auto y gain
sfun_comedi_data_read K (???)
Selector3 lambdaRef
p_im, kanal 1 Gain6 Data Type Conversion4 ref
FuelFactor LAMBDA CONTROLLER
3 K 1
SelectSource t_inj0
Gain5
t_inj
sfun_rtai_scope Time [us]
Selector
0 Injection Time
LambdaControllerOnOff LambdaController
MEASURED FEEDBACK
p_im [Pa]
Out1 K
T [C] auto T_im [K]
(???) Gain7 Multiport
T Air Inlet Data Type Conversion6 Switch
Speed density CAC
273
offset
Figure E.3: The implemented Simulink schematic that was compiled and run in hard real-time
on Bristol.
to a data acquisition card is also supported through the Comedi project. Com-
pared to the initial implementation from Rosenquist (2003), the environment
was updated to use RTAI 3.1 and also the ability to communicate with the ECU
over a CAN-bus was added. To access the ECU additional Simulink-blocks were
developed, and an example of a simulink schematic used for control is shown in
Figure E.3. For details on the system, please see Table E.11.
E.2.3 Measurements
In this thesis two kinds of measurements have been performed:
Engine Mapping
The engine mapping is an automated process where the engine speed and throt-
tle position are controlled by the computer Saab. When the engine has run a
pre-determined time in each point, the measurement is started and the signals
are measured for 5 to 10 seconds and then the mean values are stored in a table.
Three engine maps were measured:
1. An engine map of the B235R with 343 points is measured between 800 and
4300 RPM. These measurements are used for tuning of the mean-value
models.
2. A small engine map of the B235R with different fix throttle areas. The
pressure ratio over the throttle was varied by changing the engine speed.
These measurements are used when the parameter(s) in the compressible
flow model in Section 6.3.1 are determined.
3. An engine map of the B235R in 69 points for validation of the mean
value model. This map is also used in the observer design in Chapter 8.
In the observer design methodology it is also necessary to supply signal
noise data. Therefore, an additional time-based measurement is performed
immediately after the mean-value measurement. There the instrument is
re-initiated to take a fast measurement at 1000 Hz on all channels. The
duration of the noise measurement corresponds to 100 cycles. The map
covers engine speeds from 1000 to 6000 RPM.
4. An engine map of the B207R in 45 points including noise measurements.
These data are used to tune the engine model and in the observer design
in Chapter 9.
Transient Experiments
In the transient experiments, signals are measured over time with a fix sam-
pling frequency. The dynamic experiments are performed with four different
objectives:
1. Measurements to study how the intake side is influenced by changes in
exhaust manifold pressure caused by an opening of the wastegate. A
sampling frequency of up to 1000 Hz is used and the alias filters were
disabled to reduce the phase lag. This measured data is mainly used in
Chapter 3.
2. Measurements using steps in wastegate are made to study how information
on the intake side can be used to estimate the exhaust manifold pressure
in Chapter 5. As the stationary behavior is most interesting here the data
is sampled at 10 Hz.
226 Appendix E. Experimental Setup
227
228 Index