Professional Documents
Culture Documents
Modeling and Control of Retarder Using On/Off Solenoid Valves
Modeling and Control of Retarder Using On/Off Solenoid Valves
VIDAR STEINSLAND
XR-EE-RT 2008:007
Abstract
I appreciate the kindness of Mr. Fredrik Strååt for his suggestions and ad-
vices concerning the Retarder control unit, and would specially thank Mr.
Richard Riis for his assistance with the prototypes used for tests and exper-
iments.
I am also thankful to Mr. Sören Åberg, who guided me about the direction
of my thesis from the beginning, and assisted with practical experiments.
Finally, I wish to express my appreciation to the rest of the members of
NEST at Scania, and friends and family that have supported me in doing
this thesis.
1
Contents
1 Introduction 9
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Functional Description . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Actuation Requirements . . . . . . . . . . . . . . . . . . . . . 11
1.5 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 Retarder 15
2.1 Scania’s Retarder . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1 Retarder System Today . . . . . . . . . . . . . . . . . 15
2.1.2 Retarder System Using On/Off Solenoid Valves . . . . 16
2.2 Dead Volume . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Solenoid Valves . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Equipment in Experiments . . . . . . . . . . . . . . . . . . . . 19
2.4.1 ECU . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.2 Prototypes . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.3 Retarder . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4.4 Pressure Sensor . . . . . . . . . . . . . . . . . . . . . . 21
2.4.5 Software . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.6 Oscilloscope . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.7 Multimeter . . . . . . . . . . . . . . . . . . . . . . . . 22
3 Modelling 23
3.1 System Description . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 ECU Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3.1 Electrical Model . . . . . . . . . . . . . . . . . . . . . . 26
3.3.2 Magnetic Model . . . . . . . . . . . . . . . . . . . . . . 29
3.3.3 Mechanical Model . . . . . . . . . . . . . . . . . . . . . 29
3.3.4 Pneumatic Model . . . . . . . . . . . . . . . . . . . . . 31
3.4 Regulating Valve . . . . . . . . . . . . . . . . . . . . . . . . . 33
2
3.5 Model Summary . . . . . . . . . . . . . . . . . . . . . . . . . 35
4 Model Validation 37
4.1 Duty Cycle Limits . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2 Prototype 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2.1 Filling Characteristics . . . . . . . . . . . . . . . . . . 40
4.2.2 Ventilation Characteristics . . . . . . . . . . . . . . . . 40
4.2.3 Friction . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2.4 Validation . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 Prototype 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3.1 Filling Characteristics . . . . . . . . . . . . . . . . . . 48
4.3.2 Ventilation Characteristics . . . . . . . . . . . . . . . . 49
4.3.3 Validation . . . . . . . . . . . . . . . . . . . . . . . . . 51
5 Model Refinements 55
5.1 Parameter Tuning . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1.1 Air Gap . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1.2 Discharge Coefficient, Cd . . . . . . . . . . . . . . . . . 56
5.1.3 Force Balance . . . . . . . . . . . . . . . . . . . . . . . 57
5.1.4 Valve inlet orifice . . . . . . . . . . . . . . . . . . . . . 58
5.2 Time Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.3 Temperature Dependent Resistance . . . . . . . . . . . . . . . 59
5.4 System Identification . . . . . . . . . . . . . . . . . . . . . . . 62
6 Control Design 65
6.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.2 PID Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.2.1 Control Structure . . . . . . . . . . . . . . . . . . . . . 66
6.2.2 Model Based Control . . . . . . . . . . . . . . . . . . . 68
6.2.3 Tuned Controller . . . . . . . . . . . . . . . . . . . . . 68
6.2.4 Implementation . . . . . . . . . . . . . . . . . . . . . . 69
6.3 Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.3.1 Scheme 1 - Fill valve and Empty valve activated sepa-
rately for filling and venting . . . . . . . . . . . . . . . 70
6.3.2 Scheme 2 - Both valves activated simultaneously for
filling and venting . . . . . . . . . . . . . . . . . . . . . 71
6.4 Results - Scheme 1 . . . . . . . . . . . . . . . . . . . . . . . . 72
6.4.1 Simulations . . . . . . . . . . . . . . . . . . . . . . . . 72
6.4.2 Tests on Prototypes . . . . . . . . . . . . . . . . . . . . 73
6.5 Control Improvements . . . . . . . . . . . . . . . . . . . . . . 75
6.5.1 Anti-Windup . . . . . . . . . . . . . . . . . . . . . . . 76
3
6.5.2 Improved control using prediction . . . . . . . . . . . . 76
6.5.3 Improved control using non-linear control . . . . . . . . 79
6.6 Comparison Between Scheme 1 and Scheme 2 . . . . . . . . . 80
A Appendix 89
A.1 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
A.1.1 Fill valve and Ventilation valve are both activated . . . 90
4
List of Figures
4.1 Filling the dead volume with a duty cycle of 100 %, i.e. the
valves are fully open and there is a maximum flow through the
valves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Venting the dead volume with a duty cycle of 82 % . . . . . . 41
4.3 Current in the coil using a duty cycle of 75 % . . . . . . . . . 43
4.4 Current in the coil using a duty cycle of 40 % . . . . . . . . . 44
4.5 The modeled pressure and the measured pressure when 82 %
duty cycle has been applied to the fill (top) and empty (bot-
tom) valve separately . . . . . . . . . . . . . . . . . . . . . . . 45
4.6 Pressure in the dead volume when filling valve and venting
valve are both applied a duty cycle of 82 % . . . . . . . . . . . 46
4.7 Pressure in the dead volume when filling valve and venting
valve are both applied a duty cycle of 50 % . . . . . . . . . . . 47
5
4.8 Modeled and real pressure when a scheme of different duty
cycles have been used as input to the fill and vent valve) valve
separately . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.9 Filling (left) and ventilating (right) the dead volume with and
orifice of 1.0 mm . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.10 Filling (Left) and ventilating (Right) the dead volume with an
orifice of 1.3 mm . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.11 Filling (Left) and ventilating (Right) the dead volume with
and orifice of 1.9 mm . . . . . . . . . . . . . . . . . . . . . . . 50
4.12 Inserted orifice of 1.0 mm in the inlet port of Prototype 2 . . . 51
4.13 Filling and Ventilation Verification with 1.0 mm orifice, 75 cm3
volume, and 82 % applied duty cycle to fill and vent valve
separately . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.14 Filling and Ventilation Verification with 1.0 mm orifice, 75 cm3
volume, and 60 % applied duty cycle to fill and vent valve as
seen in the figure . . . . . . . . . . . . . . . . . . . . . . . . . 53
6
6.8 The actual error and the predicted error in the next sample
(top). Control Signal when prediction is introduced (bottom). . 77
6.9 The prediction included in the control design. . . . . . . . . . 78
6.10 Error in previous, actual and next sample, based on the pre-
diction calculation. . . . . . . . . . . . . . . . . . . . . . . . . 79
6.11 Reulting PID-controller using Scheme 2 . . . . . . . . . . . . . 80
7
List of Tables
4.1 Duty cycle limits for valves to start to open and for valves
to keep fully open when a pressure force equal the system
pressure helps to open the valves . . . . . . . . . . . . . . . . 39
4.2 Duty cycle limits for valves to start to open and for valves to
keep fully open when no pressure force is acting on the valves 40
4.3 Filling from Patm to 0.26Psup and ventilation from 0.88Psup to
0.71Psup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4 Filling from Patm to Psup and ventilation from Psup to Patm . . 51
8
Chapter 1
Introduction
1.1 Background
The Retarder is an integrated component in Scania’s trucks’ braking system,
mounted directly on the shaft at the end of the gear box. The Retarder is an
aid for reducing the speed without constant use of the regular service brake
and the exhaust brake. The actual retarder system uses a proportional valve
to control an air flow that determines an oil pressure in the Retarder, which
results in a braking torque. The proportional valve is constructed in such
way that it is very sensitive to temperature and vibrations, and has a strong
influence on the internal friction and calibration. Proportional valves are
quite expensive and experience has shown that they are affected by so called
hysteresis. Using two on/off valves, a fill and a ventilation valve, to regulate
the braking torque, the costs can be reduced and some of the disadvantages
9
affecting the dynamics can be eliminated. A former concept study performed
at Scania [16], investigating different approaches for controlling the Retarder,
confirms the advantages with the valves and concludes that a concept using
on/off valves would be the most convenient and robust method. This thesis
is based on the former research.
10
The valves convert the electrical signals provided by the Electronic Con-
trol Unit (ECU), connected to the connectors, to a pneumatic pressure in
the volume between the on/off valves and the regulating valve. The pressure
affects the regulating valve, which determines the braking torque in the Re-
tarder. A pressure supply, Psup , which is the source for building pressure in
the volume, is seen to be connected to the inlet port of the filling valve while
the air drainage is seen to be connected to the outlet port of the ventilation
valve. The pressure in the drainage equals the atmospheric pressure, Patm .
Due to security reasons, the ventilation valve is thought to be designed as
normally open1 . For experiments and tests on bench in this work, a valve
that is normally closed2 is used, and is due to limitations on the physical
models available for use. The ventilation valve is illustrated as seen down to
the right in the regulating function in the figure.
11
and requirements on the regulating function will be considered, and is sum-
marized in the following specification.
General
• A change in the input (control signal) must result in a change in the
output (air pressure)
Regulating function
• The function shall include one pressure sensor and two 2/2 on/off
valves 3 where the filling valve shall be normally closed and the venti-
lation valve normally open4 .
• An output volume, Vch , which includes the volume in the chamber, the
volume in the valve housing, and the actuator volume in the regulating
valve, has to be decided (50 − 125 cm3 ).
• Desired pressure, Pch , should be reached within a time of Treq for all
valid conditions with a tolerance of ± Ptol . As a reference for the
results, Treq is for this work equal to 50 time units/samples.
12
1.5 Notation
Notation
ECU
ε Power Supply Volt
UP W M Pulse Width Modulated Signal V
D Duty Cycle %
RM Measure Resistance Ω
Electrical
Lc Inductance in coil H
Rc Resistance in coil Ω
i Current in coil A
N Number of turns in coil -
Magnetic
µ0 Permeability in air Vs/Am
Aa Area of armature m2
lg Length of air gap in valve m
xp Position of armature in valve m
lg,of f Air gap length when the valve is closed m
lg,on Air gap length when the valve is opened m
Mechanical
ma Mass of armature kg
ks Spring coefficient N/m
b Viscous friction coefficient Ns/m
Fprs Pressure force N
Fpld Preload force N
Fk Spring force N
Fb Force, viscous friction N
Fsf Static friction N
13
Thermodynamics
Cd Discharge coefficient -
Cd,f ill Discharge coefficient fill valve -
Cd,vent Discharge coefficient vent valve -
k = Cp /Cv Specific heat ratio in air -
Cv Specific heat capacity at constant volume J/KgK
Cp Specific heat capacity at constant pressure J/KgK
Rgas Gas constant -
Ao Area of orifice m2
Tair Temperature of supply air K
M Mach number -
mch Mass of air in chamber kg
mf ill Mass of filling air kg
mvent Mass of venting air kg
Vch Chamber volume m3
Vr,max Max actuating volume in reg. valve m3
d Inlet and outlet diameter of valves m
Psup Supply pressure Bar
Patm Atmospheric pressure Bar
Pch Chamber pressure Bar
Pu Upside pressure Bar
Pd Downside pressure Bar
14
Chapter 2
Retarder
In this chapter the Retarder will be described further, and the equipment
used for experiments will be presented and explained.
15
magnetic field will appear between the armature and the iron core in the
valve. This will result in a magneto motive force which will affect the ar-
mature causing it to move proportional to the current. Air will pass to the
cylinder or to the drain depending on the armature’s position. According
to the air pressure in the cylinder the plunge will move, and an oil pressure
causing a braking torque will be created.
16
connected. This volume is denoted the dead volume and will be used as a
reference to the total volume in the retarder through the thesis. The little
amount of air capacity in the actuating volume (21cm3 ), makes it difficult for
the controller to regulate the pressure. To increase the performance of the
controller, extra chamber volume has been inserted. The total volume has
yet not been decided, but is a part of the parameters that will be investigated
and decided, in order to fulfill the requirements on filling time, ventilation
time, and control performance.
• Solenoid
• A moveable armature
The solenoid, seen as the gray boxes with a cross inside in Figure 2.1,
consists of a coil of wire wounded in the form of a cylinder. The coil covers
the movable armature, which is mounted on a spring that keeps the valve in
its initial position. The valve is activated by applying a current to the coil,
causing the armature to move away from its initial position.
Pos
X-dir
17
As it appears from the figure, the interior of a solenoid valve is quite com-
plex. It contains four subsystems that all are related to each other; electrical,
magneto-dynamic, mechanical and fluid dynamical. When current is applied
to the system, a magnetic field is induced around the coil contributing to
a magnetic force. The magnetic force tries to overcome the counteracting
forces, i.e. spring force and friction forces, resulting in opening or closing of
the valve, depending on if the valve is normally open or closed.
Valves are divided into different groups, according to their function and use.
The most common valves are 2/2 valves, 3/2 valves and 5/2 valves. As seen
in Figure 2.2, a 3/2 valve has three ports (inlet port, outlet port, and a drain)
and two states (on and off), thereby the name.
Figure 2.2: Normally closed 3/2 valve. Unaffected (left) and affected (right)
The available valves used in this work are 3/2 valves where the drain (port 3)
has been sealed and operates therefore as a 2/2 valve. Observe that figure 2.2
shows a valve that opens or closes by pressing and releasing a button. This
is a mechanical valve. However, the principle is the same for solenoid valves,
but they are instead activated using a current.
18
Figure 2.3: Symbolic sketch of a normally closed 3/2 valve, unaffected (right)
and affected (left), where Port 1 is the inlet port, Port 2 is the outlet port
and Port 3 is the drain.
2.4.1 ECU
The valves’ operating conditions are established by a PWM-scheme generated
by the ECU. Available outputs from the ECU relevant for the work are
current, pressure. The ECU can also be used to control other electrical
systems for the trucks. The ECU runs in a frequency fECU , i.e. the control
unit samples only once every period, TECU .
PWM signal
The PWM signal is a periodic square form signal in which the frequency and
the duty cycle can be chosen, and is shown below in Figure 2.4.
The duty cycle, D, is defined by the ratio, D = Ton /T , where Ton is the time
for which the signal is high and T is the period time. Ton is limited to be in
the interval [0, T ]. The duty cycle is often also referred to in percentage, D
19
T - Period
High
D=65%
Low
Td
High
D=25%
Low
∈ [0%, 100%]. Unless the current in the coil is not at its maximum, it will
continue to increase as long as the PWM is high. If the PWM goes low, the
current will start to discharge. When experimenting on the two prototypes,
the duty cycle and the frequency for the filling and venting valve can be
chosen separately, and is available as variables in the calibration software
tool used2 . This is an advantage since the choice of the duty cycle and the
frequency of the PWM signal for the two valves probably will affect the
filling and venting times in different ways and will be future parameters in
the control design.
2.4.2 Prototypes
During the Master’s Project two prototypes have been available for experi-
ments and validation use. The first prototype is only usable for early tests
and verification of the model. It has a fixed volume and can only be used
on bench. Prototype 2 can be mounted on the real Retarder in a truck, con-
necting the outlet pressure to the regulating valve, also making it possible to
verify the controller and its performance in the real system. Norgren Herion
3/2 valves with a sealed drain, which was described in Section 2.3, have been
used in both prototypes.
Prototype 1
Prototype 1 consists of a pressure supply inlet, a chamber with a fixed vol-
ume, Vch = 100cm3 , two on/off valves, each for filling and venting, and a
2
Gredi KleinKnecht [2]
20
pressure sensor. A given PWM signal can be used as input to the valves,
which will partly or fully open the valves depending on the duty cycle and
frequency of the PWM signal. Experiments can be done either with pressure
supply or without. If only the electrical, magnetic and mechanical part of
the valve is to be studied it is convenient to start with experiments where
no pressure supply is connected. When the pressure supply is connected, the
complete dynamics can be studied. The valves are equipped with a fixed
orifice of 1.9 mm.
Prototype 2
This prototype is equipped with the same valves as are used in Prototype
1, with the default orifice diameter of 1.9 mm. Smaller orifices can be in-
troduced to get a smaller air flow, and are inserted into the valve housing
on the inlet port to the valve. An extra orifice with diameter 1.0 mm has
been inserted into the housing as default, and can be seen in figure 4.12. It
is not meaningful to use smaller orifices, because orifices less than 0.8 mm
can cause problems with dirt.
The fixed volume of the chamber is 51 cm3 . Taking into account the vol-
ume in the regulating valve (21 cm3 ) and the connecting channels (2.8 cm3 )
the total dead volume in the prototype is approximately 75 cm3 . However, in
Prototype 2, extra volume can manually be added to the chamber if needed.
The total volumes available for use are 75 cm3 , 100 cm3 and 125 cm3 . This
is convenient when the regulating requirements are to be examined. A large
volume takes longer time to fill and ventilate, which is a drawback for filling
and venting time constraints, but is easier to regulate than a smaller volume
just because of this.
2.4.3 Retarder
A full scale retarder has been available for experiments on Prototype 2. In-
cluded in the retarder is the rotor and stator, and the regulating valve. It
has on the other hand not been connected to the gear box and no oil has
been present in the retarder. The effect of the oil pressure on the regulating
valve has therefore been neglected in the regulating valve model.
21
but is represented by the ECU in relative pressure related to the atmospheric
pressure, i.e 0 bar on the output corresponds to 1 bar in absolute pressure.
Durability of the pressure sensor and surrounding temperature affect its pre-
cision. Operating temperature is -40 to 135 ◦ C and the sensor requires a
supply voltage of 5 ± 0.1 V to work properly.
2.4.5 Software
”Gredi Kleinknecht” is a calibration tool for use with the ECU. It includes
functions to display, record and evaluate simultaneously acquired ECU in-
ternal and process data [2]. In this work Gredi Kleinknecht has been used in
experiments to acquire data such as current and pressure. The data has been
exported to Matlab where it easily can be examined. From Gredi, internal
parameters in the ECU can be set, such as input to the valves used in the
prototypes. Among the inputs that have been possible to vary are the PWM
duty cycle and the frequency.
Matlab and Simulink has been used in the modeling and simulation of the
system.
2.4.6 Oscilloscope
A Fluke 45 Dual Display Multi meter was used to examine the dynamics of
the ECU and the electrical part of the valves.
2.4.7 Multimeter
To examine the electrical circuit’s dynamics, a basic multimeter, Fluke 75,
manufactured by John Fluke has been used.
22
Chapter 3
Modelling
23
formed on the prototype and a model is built from measurements of inputs
to and outputs from the system by adapting the model properties to the real
system’s properties.
This chapter will describe the modeling of an electro-pneumatic system.
First physical modeling will be described, where mathematical relations for
the system are derived from known physical relations. Eventually system
identification will be mentioned and handled briefly.
System Pressure
PWM
Fill Valve
PWM
Vent Valve
Oil
Drain
The resulting air pressure in the dead volume and the oil pressure in the
retarder balances a regulating valve which position determines the braking
torque in the retarder. An overview of the complete system and its different
models is shown in Figure 3.2.
In this section the mathematical model for each part is separately derived
and explained in details. Eventually, all parts are combined to a system and
the complete mathematical model is expressed in state space equations.
1
See Section 2.2 for description of the dead volume
24
Regulating
ECU Valve Chamber
Valve
+ +
e +
-
UPWM e +
-
UPWM
- -
Figure 3.3: ECU circuit when the PWM is low (left) and when the PWM is
high (right)
The switching behavior of the PWM can be lethal for the electronics in the
ECU if a free wheel diode is not included on its output. If this is the case,
when either of the valves are activated, the solenoid valve is charged with
current. In periods when the PWM is low, the current discharges from the
valve’s coil and results in heating and worst case damaging the ECU and its
components. Because of this, the retarder control unit (ECU) is constructed
with free wheel diodes. The diode is connected in parallel with the electrical
drives on the ECU output, as shown in Figure 3.3 (left). The diode makes
the energy stored in the coil to be discharged in a closed circuit consisting
of the valve’s electrical components and the freewheel diode, preventing the
current to be absorbed by the ECU.
A drawback having the free wheel diode is the delayed current discharge
in the coil. When the PWM signal is deactivated or set to zero, because of the
diode, there will still be current in the circuit, resulting in a delayed closing
time of the valves. The time constant for the current discharge is determined
by the coil resistance and is hard to affect. Examining the system’s dynamics,
the delays have to be considered, and a model of the ECU is convenient.
25
According to Section 2.4.1, the output signal from the ECU, and input
to the valves, can be expressed as
½
high for t < DT ;
UP W M = (3.1)
low for DT < t < T .
where D is the duty cycle and T is the period of the PWM voltage. The
period depends on the frequency of the PWM signal. Using a frequency of
e.g. 100 Hz and a duty cycle of 60 %, the PWM voltage will be high for
6 ms and low for 4 ms during a total period of 10 ms. When the PWM is
high, UP W M equals the voltage supply e, and opposite, when the PWM is
low, UP W M equals 0 V.
3.3 Valve
With UP W M as an input signal to the valves, they can be controlled by
changing the duty cycle and the frequency of the PWM signal. As shown in
Figure 3.4 the modeling of the valves consists of an electrical, a magnetic,
a mechanical and a pneumatic part. Each part will be handled separately
where the mathematical expressions are derived.
UPWM Electrical
Magnetic
FM
Mechanical
xp
Pneumatic &
m
26
two cases has to be considered when the electrical model is derived; Case 1
when the PWM is high and Case 2 when the PWM is low.
Case 1 - Energizing:
i R
+ VR - -
e +
VL L
-
+
Figure 3.5: Electrical circuit when the PWM signal is set high
When the PWM is high no current will flow through the diode. The diode
can be considered an open circuit. The power supply, e, works as the source
and energizes the solenoid, as seen in Figure 3.5. According to Kirchhoff’s
Voltage Law (KVL), the sum of all voltage drops equals zero. Using KVL,
the mathematical equation for the first case can be expressed as:
e − Ri − VL = 0 (3.2)
As seen in (3.2), the voltage drop over the resistance is given as VR = Ri,
while the voltage drop over the inductance is more complex. As an armature
exists inside the coil, an electro motive force (emf) is induced when the
armature starts to move and is due to a change in the inductance. The
voltage drop over the inductance can according to [6] be expressed as
dΦB d di dL
VL = N = (Li) = L + i (3.3)
dt dt dt dt
where N is the number of turns in the coil, and ΦB is the magnetic flux
density in the solenoid.
di dL
e − Ri − L −i =0 (3.4)
dt dt
27
R
-
+ VR - -
VD VL L
+ +
Figure 3.6: Electrical circuit when the PWM signal is set low
Case 2 - Discharging:
When the PWM is low, no power supply e is connected to the circuit. The
charged solenoid will operate as a source and current will flow in a loop
through the diode until the energy has been absorbed in the circuit. This is
called the discharging case, or Case 2. Using Kirchhoff’s Voltage Law, the
mathematical expression can be derived:
di dL
−L −i − Ri − Vd = 0 (3.5)
dt dt
As the armature moves, the effective air gap and the amount of iron core in
the solenoid will change, affecting the magnetic field inside the solenoid. A
movement of the armature will also result in a change in the inductance. The
inductance values for the on and off position of the armature, Lon and Lof f ,
are provided by the valve manufacturer. Since the magnetic field is hard to
measure, the inductance has been approximated as a linear function of the
armature’s position, xa , and is given by
Lof f − Lon
L(xa ) = Lof f + xa (3.6)
lg,on − lg,of f
where xa is the armature’s position, and lg,on and lg,of f is the air gap inside
the solenoid in the on- respective off-position, xa,on and xa,of f .
It should be mentioned that the air gap is maximum in the armature’s off-
position, and minimum in the armature’s on-position. When the armature
is at rest, i.e. the armature is in the off-position, xa equals zero, but lg > 0.
The air gap decreases as the armature opens, and should not be confused
with the armature’s position.
28
3.3.2 Magnetic Model
Due to the current in the coil, a magnetic field will appear inside the solenoid
and result in a magnetic force affecting the armature. According to [5], the
change in the magnetic force can be expressed as
dFM B 2 dAp
= (3.7)
dt 2µ0 dt
where B is the magnetic field, µ0 is the permeability in air, and Ap = constant
is the cross sectional area of the armature.
To make the model as easy as possible, the solenoid has been approximated
to be very long. Then, inside a long solenoid, the magnetic field is given
by [7] as
Ni
B = µ0 (3.8)
lg
where N is number of turns in the coil, and lg is the air gap inside the
solenoid. The valves are constructed such that the air gap is never zero.
This would then result in an infinite large magnetic field. Different air gaps
for the on and off position of the armature have been provided by the valve
manufacturer and can give information on how much the armature moves in
total. As a function of the armature’s position, the air gap is expressed as
Combining (3.7), (3.8), and (3.9), the magnetic force can eventually be ex-
pressed as
µ0 Ap N 2 i2
FM = (3.10)
2(lg,of f − xa )2
29
assumed to be acting in the positive direction rather than counteracting the
armature’s motion. Newton’s Second Law yields, and together with assumed
and provided information, the mechanical model for the valve can be derived
as
Pressure force
According to [7] a force due to hydrostatic pressure is given by
F = pA (3.12)
As long as the pressure on the inlet port of the valve is bigger than the
pressure on the outlet port of the valve, a pressure force will help lifting the
armature in the valve. The resulting pressure force Fprs for each valve, can
be modeled as the difference between the upside (inlet port) pressure and the
downside (outlet port) pressure of the valve, which in mathematical terms is
expressed as
d
Fprs = Fu − Fd = π( )2 (Pu − Pd ) (3.13)
2
where d is the diameter of the affected body area, Fu is the force affecting
the upside, Fd is the force affecting the downside, Pu is the pressure on the
upside, and Pd is the pressure on the downside.
Viscous friction
The viscous friction can be modeled according to [7] as
Fb = bv = bẋa (3.14)
Spring force
The spring force is given by [7] as
F k = k s xa (3.15)
30
Preload force
The spring is preloaded a certain length, lpld , which is provided by the valve
manufacturer, and corresponds to a preload force, Fpld . The preload force is
given by
Fpld = ks lpld (3.16)
The pneumatic model is used to describe the air and heat transfer, and
the expansion and compression relations. Air is a compressible fluid which
behavior differs from that of a perfect gas. As a pressure supply of Psup is
used, which is considered as low pressure (p ≤ 1.0M P a), the deviations from
an ideal gas could be neglected. This is not necessarily true since the Re-
tarder can be subjected to extreme temperatures, where temperature has an
influence on the dynamics. However, a controller based on pressure feedback
can hopefully quickly compensate for errors due to temperature influence,
and the temperature variations have therefore not been considered in the
early work. The heat due to compression or expansion is minimal and has
also been neglected. This is due to an almost constant dead volume and will
be explained further in the next section.
Using the common law of gas [8], the pressure in the dead volume can be
expressed as
mch Rgas Tair
Pch = (3.17)
Vch
where Pch is the pressure [Bar], Vch is the total dead volume [m3 /kg], Rgas is
the gas constant [J/(kgK)], Tair is the temperature in the dead volume [K]
and mch is the mass of the air [kg].
The pressure change in the dead volume can now be derived by differen-
31
tiating left and right side of (3.17), and gives
where it can be seen that the heat and the volume change have been ne-
glected.
According to [8], the mass flow for valves, pipes, couplings, filters, etc. can
be calculated according to
Cd Pu A0 Pd
m˙ch = p Ψ( )ζ (k) (3.19)
Rgas Tair Pu
where s
k+1
2k k−1
ζ (k) =
k+1
Pd
s 1 if Pu
≤ Pcr
Pd 2 k+1
Ψ( ) = P P
( P d ) k −( P d ) k Pd
Pu
u u
k+1 if Pcr < Pu
≤1
k−1 2
2
( k+1 )
k−1
The outlet area depends on the armature’s position. If the armature is closed,
no air will flow through the valve. As soon as the armature opens, the outlet
area gets bigger. The smallest outlet orifice area can be expressed as
Ao = πd0 xa (3.20)
where d0 is the diameter of the inlet orifice, and is illustrated in Figure 3.7.
Remember that the outlet area cannot be bigger than the area of the inlet
orifice. If this is the case, the outlet area has to be saturated to be
µ ¶2
d0
Ao = π (3.21)
2
32
Armature
Air Out
A0
Air in
Pcr specifies the critical pressure ratio for a component, e.g. a valve, and
depends on the shape of the orifice in the component. If the ratio between
the upside and downside, also called the pressure ratio, is less than the criti-
cal pressure ratio, the flow is called a critical flow. For pressure ratios higher
than the critical pressure ratio, the flow is called an under-critical flow. For
a sharp edged orifice the value of Pcr is given by the equation [8]
2 k−1k
Pcr = ( ) ≈ 0.528 (3.22)
k+1
The orifice in real pneumatic components often have a different shape and it
is not unusual that there are series of orifices that reduce the critical pressure
ratio. Therefore the Pcr -value is always less than 0.528 for real pneumatic
components.
The net airflow to the chamber can be seen as the change in the air mass in
the dead volume, and can be expressed as the mass flow into minus the mass
flow out of the dead volume.
33
air
oil
The plunge moves according to the air pressure in the dead volume and the
oil pressure in the retarder, which is connected to the outlet port. Present
in the valve are friction forces and spring forces that counteract the plunge.
Figure 3.8 shows the regulation valve at its maximum stroke, where the pos-
itive direction is defined as a movement of the plunge from air side to the
oil side. The maximum expansion when regulating has been calculated to
be sufficient small compared to the total dead volume of ∼ 100 cm3 . The
volume expansion due to plunge movement in the regulating valve has there-
fore been neglected, and the dead volume has been considered constant. If a
much smaller dead volume is used, the expansion could have effects on the
system’s dynamics and should be considered to be included in the model.
The regulating valve can be modeled using Newtons Second Law, where
the sum of all the forces working on the system equals zero.
where ½
Fsf if ẋplunge = 0
Ff riction =
Fdf if ẋplunge 6= 0
Aair is the affected plunge area on the air side of the valve, and Aoil is the
affected plunge area on the oil side of the valve. From these equations the
plunge’s position can be derived. Note that the friction includes two cases,
static friction, Fsf , which is present when the plunge is at rest, and dynamic
friction, Fdf , when the plunge moves.
34
3.5 Model Summary
In this section the equations for the complete model have been summarized
in state-space form. First the input signals, output signals and the states are
defined, which is followed by the expressions for all the states.
Input signals:
u1 = uP W M,sup
u2 = uP W M,vent
Output signal:
y = Pch
States:
x1 = isupply x2 = xp,sup x3 = ẋp,sup
x4 = Pch
x5 = ivent x6 = xp,vent x7 = ẋp,vent
35
State-Space Equations:
u1 − Rx1
ẋ1 = Lof f −Lon
(3.26)
Lof f + x
lg,on −lg,of f 2
ẋ2 = x3 (3.27)
µ ¶2
µ0 Ap N 2 x21 π d0 ks b Fpld
ẋ3 = 2 + (Psup − x4 ) −
x2 − x3 −
2mp (lg,of f − x2 ) mp 2 mp mp mp
(3.28)
p µ µ ¶ µ ¶¶
Rgas Tair d0 π x4 Patm
ẋ4 = Cd,f ill Psup x2 Ψ − Cd,vent x4 x6 Ψ
Vch Psup x4
(3.29)
where
q
¡ 2 ¢ k−1
k+1
µ ¶
k k+1 if x4
≤ 0.528
x4 s µ ¶
Psup
Ψ = ³ ´k ³
2 ´ k
k+1
Psup
2k x4
− x4
if x4
> 0.528
k−1 Psup Psup Psup
q
¡ 2 ¢ k+1
µ ¶
sk k+1 k−1 if Patm
≤ 0.528
Patm µ³ ¶
x4
Ψ = ´ k2 ³ ´ k+1
x4
2k Patm
− Pxatm
k
if Patm
> 0.528
k−1 x4 4 x4
u2 − Rx5
ẋ5 = Lof f −Lon
(3.30)
Lof f + x
lg,on −lg,of f 6
ẋ6 = x7 (3.31)
µ ¶2
µ0 Ap N 2 x25 π d0 ks b Fpld
ẋ7 = 2 + (x4 − Patm ) − x6 − x7 −
2mp (xof f − x6 ) mp 2 mp mp mp
(3.32)
36
Chapter 4
Model Validation
When the real system has been modeled, the model has to be verified. This
is done by comparing the model’s behavior with the behavior of the real sys-
tem. Since the real system is not yet available, the two prototypes described
in Section 2.4.2 have been used for verification. The prototypes are physical
models of the system to be implemented in the future, and will work approx-
imately as the real system. They will give a good indication on how well
the model represents the true system. Experiments have been performed on
Prototype 1 and Prototype 2. Data describing the system’s behavior has
been acquired and will be presented here.
Some of the questions concerning the system’s dynamics that are examined
can be summarized in a list:
• How fast can the dead volume be filled to its maximum, Psup , and to
0.26Psup ?
• How fast can the dead volume be ventilated from maximum pressure,
Psup , to atmospheric pressure, and from 0.88Psup to 0.71Psup ?
• Does the flow behave different for pressures close to Psup than for pres-
sures close to atmospheric pressure, Patm ?
• Is friction present?
37
• How does the frequency affect the valves behavior?
From experiments, it has been observed that lower duty cycles are needed to
move the armature if the pressure on the valves’ inlet port are higher than
the pressure on the outlet port. This simply shows that the pressure helps
the armature to move rather than preventing it from opening, for both the
filling valve and the ventilation valve.
If the pressure on the inlet port is higher than on the outlet port, a move-
ment in the armature can be detected if a change in the chamber pressure
is detected. It has been observed that the duty cycle where this occurs is
different for the fill valve and the ventilation valve, and is probably due to
variations in the friction and preload parameters. Different duty cycles were
applied to the valves, while the pressure was measured. When the inlet port
of the valves is exposed to a pressure of Psup , the results can be summarized
in Table 4.1.
DCmin is the lowest allowed duty cycle to start moving the armature, while
DCmax is the minimum duty cycle that is needed to keep the valve fully
open. From Table 4.1 it can be seen that the valves open for different duty
38
Fill Valve
DCmin,f ill 28.5 %
DCmax,f ill 82 %
Ventilation Valve
DCmin,vent 25.5 %
DCmax,vent 80 %
Table 4.1: Duty cycle limits for valves to start to open and for valves to keep
fully open when a pressure force equal the system pressure helps to open the
valves
4.2 Prototype 1
As described in Section 2.4.2, this prototype consists of a fixed dead volume
of 100 cm3 and valves with an orifice of 1.9 mm. It cannot be mounted on
the truck, and is only used for early testing. In this section important system
properties for Prototype 1 are examined. A verification of the model is also
39
Fill Valve
DCmin,f ill 54.5 %
DCmax,f ill 82 %
Ventilation Valve
DCmin,vent 51.5 %
DCmax,vent 80 %
Table 4.2: Duty cycle limits for valves to start to open and for valves to keep
fully open when no pressure force is acting on the valves
The time to build up a pressure of Psup in the dead volume takes longer time.
The pressure in the dead volume for a duty cycle of 100 % can be seen in
Figure 4.1 (right), and shows that a pressure of Psup cannot be reached within
Treq seconds, which was stated as a requirement in Section 1.4. Actually it
takes about 1.4Treq to obtain a dead volume pressure the same as the system
pressure, including the time delay.
40
Pch/Psup
1
PWM Duty Cycle
0.3 1
Pch/Psup [Normalized]
0.8
Td = 0.04 Treq Tf = 0.10 Treq 0.8
0.2 0.6
0.6
0.4
0.4
0.1
0.2 0.2
0 0 0
420 422 424 426 428 430 432 420 430 440 450 460 470 480 490
Time [sample] Time [sample]
Figure 4.1: Filling the dead volume with a duty cycle of 100 %, i.e. the
valves are fully open and there is a maximum flow through the valves.
fast enough to fulfill what is required, i.e. the volume has to be small enough
and the orifice big enough to empty the volume in a certain amount of time.
A pressure drop from Psup to Patm is shown in Figure 4.2 (right).
0.9
1 Pch/Psup
PWM Duty Cycle
1
Pch/Psup [Normalized]
0.8
0.8
0.8
0.6
0.6
0.75
0.4
0.4
Tv = 0.08Treq
0.7
0.2 0.2
0.65
0 0
784 785 786 787 788 789 790 780 800 820 840 860
Time [sample] Time [sample]
One can see from the figure that a time of 1.4Treq is needed to completely
empty the dead volume, given that the fill valve is deactivated. The time to
ventilate from 0.88Psup to 0.71Psup takes approximately 0.08Treq as seen in
Figure 4.2 (left), and does not fulfill the requirement of 0.05Treq .
A conclusion is that a valve with 1.9 mm orifice and a volume of 100 cm3
cannot fulfill the requirements of filling and emptying times.
41
4.2.3 Friction
In most systems there is friction due to contact with e.g. walls or ground. The
dynamic friction coefficient bs has been provided by the valve manufacturer.
To identify whether static friction is present or not, one can disconnect the
pressure supply, let the dead volume have atmospheric pressure, and let the
ventilation valve be deactivated. Then no pressure force affects the fill valve.
It should be sufficient to investigate the fill valve since the valves are of the
same kind. When the armature does not move, there is no spring force or
dynamic friction counteracting the valve. The only forces that are present
are a magnetic force due to applied current to the fill valve, preload force
and possible static friction. A duty cycle just beneath the limit identified
earlier, is used as input. For this duty cycle the armature just about does
not move, i.e. that the magnetic force has to overcome preload and static
friction to move. Calculating the magnetic force by reading the current and
subtracting with the preload force the static friction is identified as
Unfortunately, as was experienced in tests, the provided preload did not seem
to be reasonable for a usable model, and had to be adjusted which will be
discussed in the next chapter. Therefore the static friction has not been
identified with an exact result, but rather been tuned in a reasonable way to
get the desired model.
4.2.4 Validation
In this section the model outputs and the outputs from prototype experi-
ments will be compared to each other. One of the factors that is vital in
getting a good model is the current used as input to the valves. The ECU
contains a lot of electronics that can be very complex, and as seen in the
modeling part, the ECU was simplified significantly. The current is one of
the outputs that can be validated, and will together with the pressure give
an indication of the model’s quality. It has also been seen that the system
behaves different for various duty cycles, so a scheme of different DC’s is in
the last part of the validation section used as input to the system where the
response has been observed and compared to the model.
Current
Recalling Chapter 3, the inductance in the coil changes according to the
armature’s position, and will for this reason affect the current in the coil.
42
The armature’s position affects the inductance in the solenoid, and thereby
the current. Since the armature starts to move for different PWM’s whether
pressure supply is connected or not, the current with pressure supply will
differ from the current without pressure supply for the same duty cycles.
The current in the coil when no pressure supply is connected and a duty
cycle of 75 % is applied to one of the valves can be seen in Figure 4.3.
250
duty cycle [%] and current [mA]
200
150
100
From Section 3.3, it was depicted that the current in a solenoid valve charges
and discharges according to (3.2) and (3.5). These equations can be simplified
by neglecting the freewheeling diode. The voltage balance in the circuit is
then expressed as
di dL
UP W M − Ri − L dt −i =0
dt
|{z}
≈0
It is important to know that the freewheel diode has been used in the sim-
ulations, and that this simplification just has been done in the manner to
portray the solenoid’s nature and time constant. The last term, which is the
electro motive force, is zero when the armature is at rest. This is the case
when the coil is fully charged, i.e. for infinite charging time, or when the
valves are deactivated. Solving the differential equation, the current can now
be expressed as
UP W M ³ −R t
´
i (t) = 1−e L (4.2)
R
L
where R = τ is the time constant. If a step is applied to the solenoid, when
time elapses to infinity, the solenoid will be fully energized and the current
43
reaches the steady state
The current of 292 mA is the maximum current that ever will be present in
the solenoid as long as the parameters provided from manufacturers are used,
and corresponds to the maximum magnetic force that can lift the armature.
Depending on the duty cycle used for the PWM-signal, the steady current
varies between zero and 292 mA.
200
180
160
duty cycle [%] and current [mA]
140
120
100
80
60
40
Duty Cycle [%]
20 Measured Current [mA]
Modeled Current [mA]
0
5.76 5.78 5.8 5.82 5.84 5.86
time [s]
Comparing the modeled current and the measured current in Figure 4.3, it
can be seen that the time constant and the frequency of the steady oscilla-
tions are pretty much the same. In the model, a constant time delay of 15
milliseconds has been used that seen from the figure does not exactly cor-
respond to the real delay. This is probably the main reason to the current
displacement. The average steady state differs with about 15 mA, and can
probably be due to the coil resistance variation. The top to bottom steady
oscillations are nearly the same, but differs a bit. This gets worse for lower
duty cycles, as illustrated in Figure 4.4, where a duty cycle of 40 % has been
used.
Still the frequency and the average steady state are the same for the mod-
eled and the measured current, but the maximum and minimum current in
steady state differs significantly, and is apparently due to missing dynam-
ics in the electrical model. In Figure 4.4, the peak in the time interval
5.77 ms − 5.78 ms, is due to a movement in the armature, resulting in an
44
induced electro motoric force (emf). When the iron armature moves into the
solenoid, the induced emf counteracts the current already present in the coil,
and results in a longer charging time and a slower pressure increase in the
dead volume.
Pressure
The pressure in the dead volume determines the braking torque in the re-
tarder and is for this reason of high interest to investigate. Both filling and
emptying the volume should be studied, using different duty cycles to the
valves. First the filling and venting are studied separately with a duty cycle
that keeps the valves fully open. This will give the fastest filling and venting
time that is possible in the system and can give an indication of how well
the pneumatic model coincides with the real flow characteristic. Figure 4.5
shows the modeled pressure and the measured pressure when an 82 % duty
cycle has been applied to the fill and empty valve separately.
1
Measured Pressure
Modeled Pressure
Pch/Psup [Normalized]
0.8
0.6
0.4
0.2
0
410 420 430 440 450 460 470 480
Time [Sample]
1
Measured Pressure
Modeled Pressure
Pch/Psup [Normalized]
0.8
0.6
0.4
0.2
0
1260 1270 1280 1290 1300 1310 1320
Time [Sample]
Figure 4.5: The modeled pressure and the measured pressure when 82 % duty
cycle has been applied to the fill (top) and empty (bottom) valve separately
It can be seen that for this duty cycle the modeled pressure follows very
much the measured pressure, and is a result of tuning pneumatic parameters
such as discharge coefficients and critical pressure ratio. The model fills the
dead volume a bit faster than what happens in reality. When the pressure
in the chamber is close to the system pressure, the air flows slower through
the valve and is due to different flow patterns, such as choked and unchoked
flow, as seen in the pneumatic model in Chapter 3. In the real system,
45
a leak can also be observed, which has not been considered in the model.
Figure 4.6 shows the pressure in the dead volume when both valves are
activated simultaneously, using a DC of 82 %. Also for these inputs, the
model corresponds well to the real system. This experiment can be used to
give an indication of stationary points suitable for linearization if a model
based controller is to be designed.
1
Measured Pressure
0.95 Modeled Pressure
Pch/Psup [Normalized]
0.9
0.85
0.8
0.75
0.7
0.65
1000 1010 1020 1030 1040 1050 1060 1070 1080 1090
Time [Sample]
Figure 4.6: Pressure in the dead volume when filling valve and venting valve
are both applied a duty cycle of 82 %
As shown in Figure 4.7 the model pressure for low duty cycles does not
correspond as well to the real pressure as for high duty cycles. The error is
at its maximum 50 % and is not satisfying enough. This is believed to be due
to a bad model of the current in the solenoid, which affect the armature, and
can be explained as follows. For lower duty cycles it was seen in Section 4.2.4
that the model current was oscillating much more than the real current. In
the case where the duty cycle was 40 %, the minimum current was around
60 mA, and is not enough to lift the armature. The same happens with a
duty cycle of 50 %, so the armature is in the model closed for a longer time
than in the real process and can thereby not fill as fast as the valve does in
reality as shown in Figure 4.7. As will be seen in the next sections, this will
cause problems when operating with lower duty cycles.
46
Pressure Validation
1.2
Measured Pressure
Pch/Psup [Normalized]
1 Modeled Pressure
0.8
0.6
0.4
0.2
0
300 400 500 600 700 800 900 1000 1100 1200
Time [Sample]
Error
60
Error
40
Error [%]
20
−20
300 400 500 600 700 800 900 1000 1100 1200
Time [Sample]
Figure 4.7: Pressure in the dead volume when filling valve and venting valve
are both applied a duty cycle of 50 %
than in reality for low duty cycles, causing a longer fill or vent time. Since the
armature’s position is not possible to measure, it is very difficult to have an
exact understanding of how the armature moves. What is known, is that the
inductance changes as the armature moves in and out of the solenoid. This
can be seen in the measured current data, and may be of help for improving
the model. Due to the thesis time restrictions, this has not been investigated.
4.3 Prototype 2
Prototype 2, as described in previous sections, is a physical model that can
be mounted directly on the retarder. The air pressure affects the regulating
valve that determines the oil pressure in the retarder, which corresponds to
a braking torque in the truck. According to calculations done by various
valve manufacturers, Prototype 2 has been derived with a nozzle of 1.0 mm
and a fixed volume of 75 cm3 including the volume in the housing and the
regulating valve, to compete with the specification requirements. The volume
can easily be extended by adding extra volume to the chamber mounted on
the prototype, as well as orifices with another dimension can be inserted on
the valve’s inlet port. Friction, open and close duty cycle limits, and current
properties should all be the same as for Prototype 1, this because the same
valves are used, and will not be investigated for Prototype 2. On the other
47
8 Measured pressure
Model pressure
Pressure [Bar]
0
2 4 6 8 10 12 14
time [s]
Duty Cycle [Normalized]
1
Duty cycle fill valve
Duty cycle vent valve
0.5
0
2 4 6 8 10 12 14
time [s]
Figure 4.8: Modeled and real pressure when a scheme of different duty cycles
have been used as input to the fill and vent valve) valve separately
hand, the flow behavior is different and will be handled in this section.
48
0.3 1.2 0.9
Pch/Psup
PWM Duty Cycle
0.25 1
0.85
Td = 0.02 Treq Tf = 0.19 Treq
Pch/Psup [Normalized]
Pch/Psup [Normalized]
0.2 0.8
0.8
0.15 0.6
0.75
0.1 0.4
Tv = 0.15 Treq
0.7
0.05 0.2
0 0 0.65
364 366 368 370 372 374 376 378 870 872 874 876 878
Time [sample] Time [sample]
Figure 4.9: Filling (left) and ventilating (right) the dead volume with and
orifice of 1.0 mm
small, it was increased to be 1.3 mm. The results can be seen in Figure 4.10.
Also for this orifice size, the filling and ventilation are too slow. Changing
the orifice to 1.9 mm, the fastest fill time is of 0.066Treq for a duty cycle of
100 %, and is shown in Figure 4.11 (left). If the valve response time and the
ECU time delay is to be considered, which is 0.04Treq , then the filling time
is 0.106Treq , and is very close to fulfill the time constraints.
0.9
Pch/Psup
0.3 PWM Duty Cycle
1
PWM dutycycle [Normalized]
0.85
Pch/Psup [Normalized]
Pch/Psup [Normalized]
0.25
0.8
0.8
0.2
0.6
Td = 0.06 Treq Tf = 0.12 Treq
0.15
0.75
0.4
0.1 Tv = 0.092 Treq
0.7
0.05 0.2
0 0 0.65
480 485 490 382 384 386 388 390
Time [sample] Time [sample]
Figure 4.10: Filling (Left) and ventilating (Right) the dead volume with an
orifice of 1.3 mm
49
(Right). Examining the figures, one can conclude that also for venting the
dead volume, the orifices of 1.0 mm and 1.3 mm are too small to empty the
dead volume fast enough. The prototype having an orifice of 1.9 mm and a
volume of 75 cm3 is the one that closest fulfills the requirement of ventilation
of 0.05Treq .
0.9
Pch/Psup
PWM Duty Cycle
0.25 1
0.85
Pch/Psup [Normalized]
0.2 0.8
Tf = 0.066Treq
0.8
Td = 0.04Treq
0.15 0.6
0.75
0.1 0.4
Tv = 0.052Treq
0.7
0.05 0.2
0 0 0.65
486 488 490 492 1178 1179 1180 1181 1182 1183
Time [sample] Time [sample]
Figure 4.11: Filling (Left) and ventilating (Right) the dead volume with and
orifice of 1.9 mm
Summarizing the results from filling and ventilation experiments for Proto-
type 1 and 2 in Table 4.3, one can conclude that a volume of 75 % and an
orifice of 1.9 mm is the only available prototype that is close to fulfill the
time constraints on filling and ventilation of the dead volume.
Table 4.3: Filling from Patm to 0.26Psup and ventilation from 0.88Psup to
0.71Psup
The results can be used to give an indication on how big the volume and
the orifice should be. To fulfill the requirements, the dead volume should be
a bit less than 75 cm3 and an orifice of 1.9 mm, or the orifice a bit bigger
50
Table 4.4: Filling from Patm to Psup and ventilation from Psup to Patm
with a volume of 75 cm3 . Having a bigger orifice than 1.9 mm would imply
that a bigger valve should be needed, which could result in a slower valve
response and more time delays in the system. A smaller volume is therefore
to be preferred.
4.3.3 Validation
It should be sufficient to verify the model for one orifice since this is the only
parameter that differs through the experiments on Prototype 2. The orifice
of 1.0 mm has been chosen in the verification. In order to validate the model
and the prototype, the model has to be modified by changing the orifice and
the volume. Since the orifice has been inserted in the valve housing, the
affected armature area by the pressure can still be thought to be the area
with diameter 1.9 mm, since this is the fixed orifice in the valve, and can be
seen in Figure 4.12.
Housing Valve
As was done for Prototype 1, the modeled pressure and the measured pressure
will be compared, both for filling and venting the dead volume.
51
Pressure
A duty cycle of 82 % is first applied to both the filling and ventilation valve
separately to fill from Patm to Psup , respective ventilate from Psup to Patm .
For this duty cycle, the armature is fully open in both model and reality, so
that the pressure does not depend on the armature’s position. In this way
the pneumatic part can be verified. The simulation and the measured data
for the pressure when 82 % duty cycle was used are shown in Figure 4.13.
1
Absolute Pressure [Bar]
Measured Pressure
Modeled Pressure
0.8
0.6
0.4
0.2
0
360 370 380 390 400 410 420 430 440
time [s]
Venting: Orifice 1.0 mm and Volume 75 cm3
1 Measured Pressure
Pch/Psup [Normalized]
Modeled Pressure
0.8
0.6
0.4
0.2
0
860 880 900 920 940 960 980 1000 1020
Time [Sample]
Figure 4.13: Filling and Ventilation Verification with 1.0 mm orifice, 75 cm3
volume, and 82 % applied duty cycle to fill and vent valve separately
Filling and ventilation of the dead volume takes approximately the same
time in the model as in reality, about 1.1Treq respective 2.4Treq . On the
other hand, the curves do not correspond that well in the transreal phase.
As was the case for Prototype 1 for low duty cycles, that problem will
not disappear using another prototype. The same will happen here. The
results for a duty cycle of 60 % is shown in Figure 4.14 and as can be seen,
the valves are too weak to fill and vent to the real levels in the same time.
With this result it can again be concluded that in reality the valves switch
less between the on and off position and are open for a longer time during
the period the fill or vent valve is activated, which makes the fill- and vent
response much faster.
For duty cycles lower than 60 %, the results are even worse, and can in worst
case deteriorate the performance of a model based controller.
52
Filling: Orifice 1.0 mm and Volume 75 cm3
1.4
Measured Pressure
Modeled Pressure
1.2
Pch/Psup [Normalized]
0.8
0.6
0.4
0.2
0
0 200 400 600 800 1000 1200 1400 1600 1800
Time [sample]
Figure 4.14: Filling and Ventilation Verification with 1.0 mm orifice, 75 cm3
volume, and 60 % applied duty cycle to fill and vent valve as seen in the
figure
Chapter 5
Model Refinements
55
data, a model was obtained that far from the real data matched the real
system. In the armature on-position, the provided air gap corresponded to
a so strong magnetic force causing the valves to keep fully open already for
duty cycles below 60 %, while in fact the limits are 82 % and 80 %.
According to the valve manufacturer it was clear that the air gap inside
the solenoid changed when the armature moved, between lg,on and lg,of f ,
which is the length of the air gap in the on and off position. However, the total
air gap that results in a magnetic field inside the solenoid does not necessarily
change as much as the provided air gap, which can be explained in a simple
way using Figure 2.1 in Chapter 2 as reference: When the armature starts
to move, the air gap between the armature and the inlet port (outside the
solenoid) increases, while the air gap on the back side of the armature (inside
the solenoid) decreases. In the on-position, the inside air gap has reached its
minimum, lon , while the air gap outside the solenoid is maximum. Taking
into consideration the air gap outside the solenoid, the effective air gap can
be considered almost constant. The magnetic field is hard to measure, and
the magnetic field lines are therefore hard to have an understanding of. Using
a nearly constant air gap has therefore been seen as a reasonable solution,
and a model more similar to the real system was obtained.
56
near unity for Reynolds number between 20000 and 100000. However, other
values of the discharge coefficients could be real for lower Reynolds numbers
or for orifice geometries where the length-to-diameter ratios are greater than
unity. For these cases, the discharge coefficient was identified to be between
0.6 and 1. The former results have been used as a reference in finding the
discharge coefficient for the valves used in this Master’s Thesis.
Figure 5.1 shows how the pressure in the dead volume changes when a step
is first applied to the fill valve, then to both valves simultaneously, using dif-
ferent discharge coefficients for the fill valve. Here the discharge coefficient
for the ventilation valve is set to a constant value 0.97. One can see that a
Cd closer to one makes the fill valve fill faster.
10 10
8 8
pressure [bar]
pressure [bar]
6 6
4 4
Cd,sup=0.60 Cd,sup=0.60
Cd,sup=0.75 Cd,sup=0.75
2 2
Cd,sup=0.90 Cd,sup=0.90
Figure 5.1: Pressure Change in the Dead Volume as Different Cd’s are used
for the Fill Valve when Fill Valve is first activated, then both valves si-
multaneously (Left). Zoomed plot of the Filling Characteristics (Right).
Cd,vent=0.97.
The discharge coefficient is varied until the measure data and the simulated
data for the pressure corresponds to each other. A discharge coefficient close
to unity as seen in the figure is used to get the same fast response in the
model as have been observed in the true system.
57
in this case also prevent the valves from opening. When the system model
first was derived, data for preload force, valve orifice diameter, spring coef-
ficient, air gap, and damping coefficient provided by the manufacturer, were
used in simulations. Experiments have shown that the output in pressure
and current for the model does not match the outputs from the real system;
the modeled valves start to open and keep open for duty cycles even lower
than the limits identified in experiments. Either the preload force or the
static friction have been assumed too weak, or the inlet orifice too big. In
order to let the modeled valves open for the same DC limits as identified,
the parameters have been considered necessary to tune.
Increasing the preload force or the static friction and keeping the other pa-
rameters constant makes the valves open closer to the DC limits. Tuning it
so that the lower limit is correct makes the upper limit, i.e. the DC where
the valve keeps open, too high. To still keep the lower limit correct, Fsf or
Fpld could be decreased while the orifice, corresponding to a pressure force,
is adjusted. In this way, correct values for both the lower and the upper
limit can be obtained. The orifice only affects the valve as long as there is a
pressure difference between the upper and downer side of the valve. In other
words, changing the orifice will not affect the upper limit, since Pu − Pd = 0,
and thereby Fprs = 0, when the valves stay fully open. In order to tune the
model so that the limits are correct, it has been a matter of adjusting the
preload force, the friction and the orifice.
Recalling (3.13), one can see that another effect of having a big orifice is
that less magnetic force has to be applied to the solenoid to open the valve,
since the pressure force affects a bigger area helping the armature to open.
Seen from experiments is that the fill valve and the ventilation valve starts
to open for 28.5 % duty cycle respective 25.5 % duty cycle when they are
exposed to a pressure of Psup , presuming that there is atmospheric pressure
on the outlet port of the valves. As seen in the previous section, in order
to let the valves keep open at 82 % / 80 % duty cycle, the preload force
was increased. As depicted earlier, having a correct upper limit causing an
error in the lower limit, so the orifice area was increased. This was done with
58
the motivation that the geometry of the valve is pretty complex and hard
to have an opinion of. It is believed and presumed from valve descriptions
and sketches that a bigger area affects the mechanics in the valve, with the
same principle as a bigger orifice affected the mechanics in Prototype 2 with
an orifice of 1.0 mm in Section 4.3.3. Remember that the orifice also affects
the flow in the pneumatic part. Increasing or decreasing the area that affects
the armature to move in the mechanical part, does not mean that air flows
through a smaller or bigger orifice. This area should therefore always be kept
constant so that the flow model is kept correct.
59
duty cycle [%] and current [mA]
Duty Cycle [%]
200 Measured Current [mA]
150
Time Delay
100
50
0
4.74 4.742 4.744 4.746 4.748 4.75 4.752 4.754 4.756 4.758
time [s]
duty cycle [%] and current [mA]
80
60 Time Delay
40
20
0
3.9 3.902 3.904 3.906 3.908 3.91 3.912 3.914 3.916 3.918 3.92
time [s]
∆Eint = Q + W (5.2)
nRgas ∆T Cp
∆Eint = k−1
where k = Cv
≈ 1.4 (5.3)
60
295
current [mA]
PWM dutycycle [%]
290
285
current [mA]
280
275
270
10 20 30 40 50 60
time [s]
Combining (5.3), (5.4), and (5.5) an expression for the change in the tem-
perature can be derived.
k−1
Ṫ = (Ri2 − C(T − Tsurround )) (5.6)
nRgas
R = R0 (1 + α (T − T0 )) (5.7)
| {z }
∆T =Ṫ
where T0 is the initial temperature at rest. From these relations, the re-
sistance can be modeled. As simple dynamics is desired, the effect of the
resistance has been neglected with the motivation that its dynamics is very
slow compared to the total period of time the valves are activated. This can
also be seen in Figure 5.3. The current decreases from about 292 mA to
273 mA in 60 seconds, which is slow compared to the time the valves are
activated. It can be concluded that this effect is not only due to a long valve
activation time, but is also present if the truck is used in extreme surrounding
temperatures.
61
5.4 System Identification
As seen from the validation part for the current, the model error is large
for low duty cycles. As stated in Section 4.2 this is due to dynamics in the
real process, more specific in the ECU, which has not been considered in the
model. An alternative is to model all the electrical drives included in the
ECU, which could be a very complex approach. Using system identification
to model the current could be an alternative approach to the mathematical
equations used for the electrical model for the valves. This can be done using
System Identification Toolbox in Matlab. Using the circular flow principle
presented in Figure 5.4, an ARX1 model of the electrical part of the valves
is created. As seen in the figure, the first part of the identification process is
to gather data.
Construct
experiment and
gather data
Data
Data
Model
Choose model Accommodate
structure model to data
Evaluate model
Data Model structure
Not OK Not OK
No Can model be
accepted?
Yes
Figure 5.4: System identification’s circular flow. The rectangles are the com-
puter’s main responsibilities, and the ovals are user’s main responsibilities. [3]
The input and the output are then defined in separate vectors, and as long
as filtering is not required a linear model describing the relationship between
the input and output is generated. Eventually the model is validated. For
best results, an altering input signal consisting of both high and low duty
cycles, should be used. If this is not done, i.e. the input signal is based on
1
ARX model is a type of linear parametic model, which is described by:
62
one specific duty cycle, the identification may give a poor result. Because of
the limitations on time for the project, the method has only been tested for
a constant input, and due to non-satisfying results system identification has
not been implemented in the model. However, better results are thought to
be obtained using a wider range of inputs.
63
Chapter 6
Control Design
6.1 Background
A common approach in control design is to start with a simple control design.
Based on experience from employees at Scania and results from former Mas-
ter’s Thesis on On/Off control of exhaust gas recirculation [12], PID control
using a pressure feedback principle has been used as the base in the control
design. Additional to the PID controller, intuitive rules, a boosting action
using non-linear control and prediction is introduced for improvements, and
has been examined closer in upcoming sections. Giving the modeling part a
meaning, a model based controller is of interest to examine. Due to model
65
errors and time restriction on the thesis, a real-time tuned PID controller
has been developed and evaluated instead. The differences between a model
based and a tuned controller will be discussed in the next section.
r + e u y
Gc Gp
-
The output feedback structure is shown in Figure 6.2, and is equipped with
a feed forward block, Gf f . With the output feedback structure more design
freedom can be achieved, due to the feed forward part that can be used
for pole cancellation and separate derivative action on the reference signal.
In this Master’s Thesis only the error feedback is considered, hence output
feedback is not described further.
66
r + u y
Gff Gp
-
Gc
Error Feedback
A control law for the PID-controller using the error feedback principle is
according to [10] in continuos time given by
Zt
de (t)
u (t) = P e (t) + D + I e (t) dt
dt
t0
Zt
de (t) 1
= K e (t) + Td + e (t) dt (6.1)
dt Ti
t0
where K is the gain, Td is the derivative action time constant, and T1i is the
integral action time constant. The error, which is the difference between the
reference pressure and the actual pressure, is manipulated by a proportional,
derivative and integral action, and used as the new input to the process,
which is called the control signal u.
67
Proportional action
When there still is an error between the reference and the actual output value,
the error should as fast as possible be compensated for so that the reference
can be reached. The proportional part, also called the gain, K, is used to
amplify the error to get a faster action on the control signal. Choosing a too
high gain can result in oscillations in the output, or even instability in the
system.
Derivative action
The derivative part differentiate the error and uses the result as a subsidy in
the control signal to compensate for error in the output signal. Its intention
is to decrease the oscillations on the output caused by a high gain. The
derivative action is highest when the change in the error is high, and is least
significant when the change in the error is low.
Integral action
The integral part has as intention to remove stationary error on the output.
When the error is positive, the control signal increases due to integral action,
and when the error is negative, the control signal decreases. The integral
action is slower than the derivative and proportional action.
68
approach cannot be used because of security reasons or because the system
does not exist.
An indication on how the parameters should be chosen can be obtained
through simulations where the parameters can be adjusted. A reference signal
is generated and applied to the discrete hard-ware implemented controller.
With a real-time software, data such as the reference, the output and the
control signal can be acquired in real-time. Adjusting the control parameters
while comparing the real time data, a desired set of parameters have been
found when the desired control behavior is achieved. Both Ziegler-Nichols
method and Åströms rules have been used as reference for the choice of
PID-parameters [17].
6.2.4 Implementation
For simulations in computer programs such as Simulink, a continuos time
controller is sufficient. This is not true if the controller is to be implemented
in a real system, which mostly is the case. Then a discrete time controller
has to be implemented and is given by
à n
!
en − en−1 Ts X
un = K en + Td + ek (6.2)
Ts Ti k=0
6.3 Approaches
The PID-controller generates a control signal based on the error. A positive
error means that the pressure in the dead volume is too low compared to the
reference pressure. Opposite, if the error is negative, the pressure is too high.
Basically, seen from (6.1), the generated control signal is positive for positive
error and negative for negative error. Since there are two valves and just one
control signal, the controller has to distribute the signal to the two valves.
A controller including a distributor can look like the setup in Figure 6.3.
There are several ways for the distributor to handle how the valves should
work for different control signals, e.g. if both valves should act simultaneously
or separately, if the whole control signal range should be used or not etc.
Based on earlier research mentioned in the introduction, two different ways
of doing this have been investigated, where one of the approaches will be
investigated in this Master’s Thesis and the other is left for the parallel
work [13].
69
ufill
1 error u u fcn Gp 1
r uvent y
Controller Plant
y Distributor
Figure 6.3: A setup with the controller and a distributor distributing the
control signal to either of the two valves
One of the drawbacks with this approach is that for control signals in the
interval [−DCmin,vent , DCmin,f ill ], the valves will not react, meaning that the
control signal contains a dead band. From the validation in Chapter 4 it
was seen that a duty cycle of DCmax was sufficient to fully open the valves.
Using a higher duty cycle as in the traditional linear approach, the valves
will be over magnetized, causing the coil to discharge even slower, resulting
in a longer time to close the valves when needed. To prevent these two phe-
nomenas, Pulsing Scheme 1 has been developed, where the output from the
distributor can be seen in Figure 6.4 (right).
Now the duty cycle range [0, 100] is resolved and linearized into the active
range [DCmin , DCmax ] instead, so that the dead band is nearly eliminated.
70
Traditional Linear Scheme PWM Pulsing Scheme 1
80 80
Valve Duty Cycel [%]
40 40
20 20
0 0
Figure 6.4: Traditional Pulsing Scheme (left) and Pulsing Scheme 1 (right)
The reason why this only holds almost, is connected to the fact that Puls-
ing Scheme 1 does have a drawback letting both valves be DCmin when the
control signal is zero. Setting one of the valves to DCmin when the valve
is desired to be deactivated would give a faster response when the valve is
to be activated again. What is not so good is that when the valve should
deactivate, it would take much longer time to discharge the coil, and thereby
closing the valve. This will result in a delay that may give a poor control
performance. Therefore Pulsing Scheme 1 has been extended by including
an error tolerance. When the error is within the predetermined tolerance,
both valves should be deactivated. The tolerance has been chosen to be the
required control tolerance on ± Ptol as required in the specifications.
Letting the valves operate separately, i.e. using Scheme 1, a minimum use of
the valves could be possible, since one of the valves is always at rest. This is
an advantage considering the life time of the valves.
71
6.4 Results - Scheme 1
The controller is first implemented and simulated in Simulink. Doing this,
the regulator is tuned until a desired set of control parameters have been
obtained. The parameters identified from the simulations are then used as
reference when tuning the controller in the real system.
6.4.1 Simulations
Using a valve orifice of 1.3 mm and a volume of 75 cm3 , the simulation results
of a PID-controller can be seen in Figure 6.5.
Pch
Relative Pressure [bar]
6 Pref
0
2 4 6 8 10 12 14 16
Time [s]
Control Signal
100
DC fill
80 DC vent
Duty Cycle [%]
60
40
20
0
2 4 6 8 10 12 14 16
Time [s]
The control signal is better seen in Figure 6.6 where the controller’s behavior
clearly shows that close to the reference, the control signal decreases, meaning
that less action is needed. The reason that the output never exactly reach
the reference for any reference pressures is that the tolerance of ± Ptol has
been applied to the controller, i.e. when the error is less than this tolerance,
the valves will be deactivated. This has been done with respect to the real
system, where oscillations would be a result of not having the tolerance. It
is thereby not said that oscillations cannot arise, but the tolerance is rather
used to decrease eventual oscillations. Looking at Figure 6.5 again, one can
see from the control signal that only one valve acts at the moment, which
is the idea with scheme one. An example showing the integral part acting,
can be drawn by looking at the time period from t ≈ 4 s to t ≈ 5 s. The
72
fill valve is first activated, but has a too strong effect such that the pressure
rises over the reference and even outside the tolerance. The controller has
to ventilate the dead volume until the error is within the tolerance. The
control signal is in this area seen to be integrating until the valve reacts and
empty the dead volume. When the desired pressure is reached, the valve is
deactivated. The controller works fairly well, but with some overshoots. It
acts in these situations very slow, so a boosting action could be considered
to be included. The main reason that the controller is slow close to reference
pressures is that the minimum duty cycle value used in the distributor is
for all conditions set to 25.5 %. In fact this is not optimal because when
the pressure difference between the inlet port and outlet port of the valve
is very small, less pressure force helps the valve to open. Actually a duty
cycle between 25.5 % and 54.5 % is needed to lift the filling armature, where
the latter is needed when no pressure force is present. For pressures close to
reference pressure, a very low duty cycle is required. If the gain is too low,
the integrator will in the worst case need to integrate the control signal to be
54.5 % before the valve even will react and give a response. The integrator
speed is limited by Ti , and will in such cases cause a time delay in the output
response close to references.
Pch
Relative Pressure [bar]
7
Pref
4
15.6 15.65 15.7 15.75 15.8 15.85 15.9
Time [s]
Control Signal
100
DC fill
80 DC vent
Duty Cycle [%]
60
40
20
0
15.6 15.65 15.7 15.75 15.8 15.85 15.9
Time [s]
73
Prototype 1
The controller is discretized according to (6.2) and by using the toolbox
”Real-Time Workshop” in Matlab, C-code for the controller is generated for
use in hardware. Using the same pressure reference input as was used in the
model validation and simulations, a controller can be applied to Prototype 1.
The controller is implemented as a regular PID-controller with no prediction
or boosting action. Since the prototype did not fulfill the requirements on
filling and ventilation time, there is of little relevance to implement a con-
troller on this prototype. The validation of the controller has therefore been
left for Prototype 2 which is closer to the real system and can be used for
real tests in truck.
Prototype 2
To fulfill the time constraints on filling and ventilation, the size of the inlet
diameter was identified to be at least 1.9 mm, while the dimension of the
dead volume should be less than 75 cm3 . Using an orifice of 1.9 mm and a
dead volume of 75 cm3 for the control, which is the only prototype available
that is close to fulfill the requirements, would best indicate how the controller
would perform on the real system. However, the controller has only been ap-
plied to a physical model having an orifice of 1.3 mm and a volume of 75 cm3
because of lack of time to physically change the orifice in the prototype. It
is believed that this will not differ much from the optimal dimensions since
similar results have been obtained for a controller applied to Prototype 1,
where the orifice is actually bigger. For an orifice of 1.3 mm and volume
75 cm3 the resulting PID-controller is shown in Figure 6.7.
Seen from the figure is that the performance cannot be compared to the simu-
lation results, due to large deviations from the reference pressure. In the real
system, leakage in the dead volume has been observed, which can be seen as
a clearly visible pressure drop close to pressure references. This has not been
considered in the model, but is believed to be much less present in a future
implementation. The PID-controller should anyhow be able to compensate
for leakage if present. The reason that the leakage is not compensated for in
Figure 6.7, is due to the problem using 25.5 % duty cycle as the lower limit.
For higher pressures this duty cycle is not enough to open the fill valve, so it
takes some time for the integrator and the boosting action to reach a control
signal that affects the valve to fill.
Mentioned in the model verification chapter, was that a volume less than
75 cm3 and a valve orifice of 1.9 mm was an requisite for the controller to
74
9
Pch
Pref
8
6
Relative Pressure [Bar]
0
0 5 10 15 20 25 30
Time [s]
compete with the filling and ventilation times. This is not exactly true, since
the retarder has a operating pressure range between 0.18Psup and 0.71Psup .
This implies that a pressure change of Psup never has to be performed, so
that filling and ventilation of the dead volume will always be done in less
time than presented in the results.
75
6.5.1 Anti-Windup
In practical applications there are often physical limitations that prevent the
system to act further. An example of a system with such limitations is the
retarder valve system in this thesis, where the valve cannot open more than
it is restricted to due to walls preventing it from opening. In a case where
a controller using an integrator is used to control the valves, a phenomena
called windup effect can occur, and is best explained by the following exam-
ple.
If the valves have been set to be fully open, but the reference has still not
been reached, the integrator in the controller will try to compensate for the
error by continuing integrate the control signal, with the thought to open the
valves even more, which is not possible because of the physical restrictions.
The integration part will continue forever if not a change in the reference
is done, and is called the windup effect. When a change in the reference
is first done, the integration part has become so large that it will be very
hard to compensate for, and oscillations and instability may be effects of the
phenomena. To prevent the integrator from integrating when physical limi-
tations have been reached, an anti-windup function is included in the control
design.
76
lation result of introducing the prediction approach, where the error in the
next sample has been predicted, is shown in Figure 6.8.
0.1
0 tolerance
−0.1
Error [Bar]
−0.2
−0.3
−0.4
Error
−0.5 Predicted Error
−20
Duty Cycle [%]
−40
−60
−80
Figure 6.8: The actual error and the predicted error in the next sample (top).
Control Signal when prediction is introduced (bottom).
From the figure it can be seen that when the predicted error is within the
predefined tolerance range, the control signal is set to zero as mentioned ear-
lier. The actual error is observed to continue decreasing because of the time
delay for the valve to react and to perform the desired action, in this case
switching off the valve. A worst case time delay of 0.04Treq for the valve to
react has been observed in experiments. Since one sample corresponds to
0.02Treq , a prediction of the error in the two next samples has been consid-
ered reasonable.
There are many ways to perform prediction, e.g. Model Predictive Control
(MPC), Generalized Predictive Control (GPC) or even the Dynamic Matrix
Method (DMM) [18], which all are based on information about the system.
In this Master’s Thesis, a transfer function describing the non-linear system
has not been derived. A prediction approach based on feed-back and where
the actual pressure, past pressure and reference pressure are used to calcu-
late the future pressure has therefore been investigated and derived. The
prediction can be included in the control design as illustrated in Figure 6.9.
A very simple differential approximation for a signal u is given by Euler
77
Switch:
Turnoff valves
if |ek+1|<tolerance
0
Controller u
u*
ek 1 ek-1 + ek+1
z +
Backward [17] as
1
u̇(t) ≈ ∆u(t) = (u(t) − u(t − T )) (6.6)
T
where t is time and T is sample time.
In discrete time, using Figure 6.10 as reference, the predicted error in the
next sample, en+1 , can be calculated as
en+1 = en + ∆en
= en + (en − en−1 )
| {z }
BackwardEuler
= 2en − en−1 (6.8)
(6.9)
and is based on the differential approximation Backward Euler for calcula-
tions of ∆en , i.e. the change in the error, in the actual sample, n. n ∈ [0, N ]
78
where N is the total number of samples. The same principle is used to predict
the error in the two next samples, en+2 , and is given by
en+2 = en+1 + ∆en+1
= en+1 + en+1 − en
| {z }
BackwardEuler
= 2en+1 − en
= 3en − 2en−1 (6.10)
e
en+2
en+1
en ∆ en+1
en-1 ∆ en
n
n-1 n n+1 n+2
Figure 6.10: Error in previous, actual and next sample, based on the predic-
tion calculation.
79
6.6 Comparison Between Scheme 1 and Scheme
2
Because of lack of time, there has been a limited improvement of the two
control approaches. However, looking at results from the parallel work where
Scheme 2 was investigated, Scheme 1 seems to be preferrable. The results
shown in Figure 6.11 clearly shows drawbacks with Scheme 2.
Control Results − Scheme 2
9
Pch
8 Pref
7
Relative Pressure [Bar]
0
0 5 10 15 20 25
Time [s]
Since any control signal distributes an action to both the filling valve and
the ventilation valve, it is in certain occasions hard to obtain the reference
pressure. When the output is close to the reference, it is seen that oscillations
are very often present and is one of the main disadvantages observed with
Scheme 2. Using Scheme 2 requires a very good distribution of the control
signal to avoid such phenomenas. One of the main reasons for bad results
which is more visible in Scheme 2, is very much because of the use of a
constant lower duty cycle limit of 25.5 %. For pressures in the dead volume
close to Psup , the pressure force is very strong on the ventilation valve, while
it is much weaker on the fill valve. For pressures close to the atmospheric,
the opposite is true, where the pressure force affects the fill valve more than
the ventilation valve. Taking a reference pressure of 7 bar and dead volume
pressure of 6.8 bar as an example, the fill valve might requires a duty cycle
of 45 % to open, while the ventilation valve only needs 30 %. Still the error
is positive, meaning that more air is needed. The controller might fill with
40 % duty cycle and ventilate with 32 %. This should in the theory imply a
positive flow into the dead volume, but since the ventilation valve reacts for
lower duty cycles at higher pressures in the dead volume, there will actually
be a ventilation. Another drawback with Scheme 2 is believed to be the
practical disadvantage that both valves are used simultaneously, resulting in
80
a more frequently use of the valves, causing more wear of the valves. This
is believed to be better for Scheme 1 and is important to consider when
robustness of the proportional valve used in current retarder system and the
system using on/off valves is to be compared. Using Scheme 2, is however
believed to allow for larger valve orifices, making it possible to have a faster
filling and ventilation response, which is not possible having a smaller orifice.
This is explained by the fact that while one of the valves are acting, the other
valve can at the same time be activated preventing the response to be too
fast when the error is so small that a slower response is needed to reach the
reference.
81
Chapter 7
7.1 Conclusions
The Master’s Thesis aimed to investigate if today’s system that uses a pro-
portional valve to control the air pressure in the Retarder, could be replaced
by two on/off solenoid valves and a pressure chamber with a pressure sensor,
in order to reduce costs and improve the system robustness. The idea was to
let one filling valve and one ventilation valve together with a pressure supply,
build a pressure in the chamber, determining an oil pressure in the Retarder
and eventually the braking torque in the truck.
A model of the system was obtained, but validation of the model shows
that the quality of the model is not optimal. The model contains some
weaknesses that in future work could further be investigated and improved
by including the probably missing dynamics in the electronic control unit
and the magnetic properties. The modelled current and pressure correspond
well to the real current and pressure for high duty cycles as input signals.
For low duty cycles, the modelled current differs from the real current in the
amplitude and causes the valves to be closed for longer time than in reality,
and thereby a longer time to fill or ventilate air from the chamber.
To control the valves, two different controllers were investigated. A con-
trol approach using a pulse width modulated pulsing scheme where the fill
valve acts for positive errors and the ventilation valve acts for negative errors
(Scheme 1) seemed to give best results, and was designed in this Master’s
Thesis. Simulations prove the excellence of using a regular PID controller
on the system. Results from tests on the real system have shown a strong
potential in having two on/off valves controlling the retarder, with addi-
tional integral boosting action and a method for reducing overshoots using
error prediction. With further research, the method using Scheme 1 can be
83
considered to be a promising approach.
Another approach, Scheme 2, which was investigated in the parallel work [13]
let the filling valve and the ventilation valve act simultaneously. This was
thought to be a faster and more controlled way to control the air pressure,
with the intention that a bigger valve orifice could be used to get a faster
response and even have the opportunity to reduce the action of one valve by
letting the other valve counteract at the same time. A poor control perfor-
mance was achieved using Scheme 2 and showed at the same time a worse
robustness taking life-span into consideration.
Experiments have been done on physical models of the system with different
dimensions in order to determine proper parameters for volume of the cham-
ber and orifice of the valve. Results show that a proper total dead volume
(total volume in the system including volume in the valve housing, chamber
and regulating valve), in order to fulfill actual time constraints and control
performance, should be a bit less than 75 cm3 using an orifice of 1.9 mm.
In order to make a better system model, more complexity from the elec-
trical drives in the ECU could be included, so that the modelled current
behaves more like the real current. The magnetic valve properties could be
investigated in more detail, where the magnetic field in the iron core in the
valves could be included. An improved model can be used to make a model
based controller. A model based controller have the advantage that the sys-
tem properties may be included in the design, resulting in more freedom
choosing control strategies and may improve the control performance.
It has been observed that depending on the pressure in the dead volume,
different duty cycles are needed to start opening the valves. Which has been
seen, a common problem causing time delays in Scheme 1 and oscillations
in Scheme 2, is the constant lower duty cycle limit of 25.5 % used in the
distributor. In future work, this lower duty cycle limit should be implemented
as a function of the actual pressure in the dead volume where the lower duty
84
cycle limit varies between 25.5 % and 51.5 %. In this way eventual time
delays in the integral part and oscillations in the output pressure can be
reduced.
85
Bibliography
[8] K-E.Rydberg. ”Basic theory for Pneumatic System Design”. Linkö ping
University. IKP, Fluid and Mechanical Engineering Systems. 1997.
86
[13] M.J.Tehrani. Pressure Control of a Pneumatic Actuator Using On/Off
Solenoid Valves. The School of Electrical Engineering. The Royal Insti-
tute of Technology. 2008.
87
Appendix A
Appendix
A.1 Linearization
According to [10] a nonlinear system ẋ = f (x, u) can be linearized around
some operating point {x0 , u0 } by considering a neighborhood around the
operating point and approximating the nonlinear model with a truncated
Taylor series. This is done by setting x = x0 + ∆x, u = u0 + ∆u and
y = y 0 + ∆y, then
¯ ¯
ẋ = f (x, u) + ∂f ¯ 0 ∆x + ∂f ¯x=x0 ∆u
∂x x=x ∂u
u=u0 u=u0
¯
∂g ¯
¯
∂g ¯ (A.1)
y = g (x, u) + ∂x x=x0
0 ∆x + ∂u x=x0
0 ∆u
u=u u=u
The system has been thought to be in equilibrium either when both valves
are deactivated, when only one of the valves is fully open, or when both
valves are fully open. The equilibrium when both valves are fully activated
has been considered. In this case the PWM signal is set to 100 % DC on
both valves, and the pressure will reach a constant pressure unless there are
any disturbances disturbing the valves. The steady position will correspond
to the current when the coil is fully charged, and will probably be outside
the physical position limitation in the valve, i.e. outside the walls. In this
equilibrium, the velocity will be zero.
89
A.1.1 Fill valve and Ventilation valve are both acti-
vated
Equilibrium points:
x01 x02 x03 x04
x05 x06 x07
Linearized Model:
˙ = A∆x + B∆u
∆x
˙ = C∆x + D∆u
∆y
ẋ1 x1
ẋ2 x2
ẋ3 x3
∆x =
˙
ẋ4 and ∆x =
x4
(A.2)
ẋ5 x5
ẋ6 x6
ẋ7 x7
¯
∂f ¯
¯
∂f ¯
A= ∂x x=x0
0 B= ∂u x=x0
0
u=u u=u
¯ ¯ (A.3)
∂f ¯ ∂f ¯
C= ∂x x=x0
0 D= ∂x x=x0
0
u=u u=u
90
¯ ¯
∂f1 ¯ ∂f1 ¯
∂x1 ¯ ∂x2 ¯
0 0 0 0 0
x=x0 x=x0
1 0 0 0 0
¯0 ¯0 ¡ ¢2
∂f3 ¯ ∂f3 ¯
∂x1 ¯ ∂x2 ¯
− mbp − mπp d20 0 0 0
x=x0 ¯x=x0 ¯ ¯
∂f4 ¯ ∂f4 ¯ ∂f4 ¯
A = 0 ∂x2 ¯
0 ∂x4 ¯
0 ∂x6 ¯
0
0
x=x 0 x=x ¯ ¯x=x0
∂f5 ¯ ∂f5 ¯
0 0 0 0 ∂x5 ¯ ∂x6 ¯
0
x=x0 x=x0
0 0 0 0 0 1
¡ d ¢2 ¯ ¯0
0 ∂f7 ¯ ∂f7 ¯
0 0 0 − mπp 2 ∂x5 ¯ ∂x6 ¯
− mbp
x=x 0 x=x 0
µ 1
¶
Lof f −Lon
0
x02 x −x
−Lof f
on of f
91
0 0
0 0
B = 0 0
µ 1 ¶
0 Lof f −Lon
x06 xon −xof f
−Lof f
0 0
0 0
£ ¤
C = 0 0 0 1 0 0 0
where
¯
∂f1 ¯¯ R
¯ = − ³ ´
∂x1 x=x0 Lof f −Lon
x02 xon −xof f
− Lof f
³ ´
¯ 0 0 Lof f −Lon
∂f1 ¯¯ − (u1 − Rx1 ) xon −xof f
= ³ ³ ´ ´2
∂x2 ¯x=x0 L f −Lon
x02 xofon −xof f
− L of f
¯
∂f3 ¯¯ µ0 Ap N 2 x01
= 2
∂x1 ¯x=x0 mp (xof f − x02 )
¯ 2
∂f3 ¯¯ µ0 Ap N 2 (x01 ) ks
¯ = 0 3 −
92
∂x2 x=x0 mp (xof f − x2 ) mp
s
¯ p µ k+1
¶ k−1
∂f4 ¯¯ Rgas Tair 2
= d0 πCd,sup Psup k
∂x2 ¯x=x0 Vch k+1
à à ! !
q 2 ³
k k+1 ´
v 2k Patm
+ k+1 Patm k
¯ p u õ ¶ 2 µ ¶ k+1 ! k−1
− k2 2+k k x04
u 0 k
∂f4 ¯¯ Rgas Tair 0 t 2k Patm k Patm k (x4 ) 0
¯ = − d0 πCd,vent x6 0
− 0
+ r x 4
∂x4 x=x0 Vch k−1 x4 x4 ³ ´2 ³ ´ k+1
k
Patm k Patm
2 x0
− x0
4 4
v ÃÃ ! Ã !!
¯ p u 2 k+1
Rgas Tair u 2k P k P k
∂f4 ¯¯ atm atm
= − d0 πCd,vent x04 t −
∂x6 ¯x=x0 Vch k−1 x04 x04
¯
∂f5 ¯¯ R
¯ = − ³ ´
∂x5 x=x0 L f −Lon
x06 xofon −xof f
− Lof f
³ ´
¯ 0 0 Lof f −Lon
∂f5 ¯¯ − (u 1 − Rx 5 ) xon −xof f
¯ = ³ ³ ´ ´2
∂x6 x=x0 Lof f −Lon
x06 xon −xof f
− L of f
¯
∂f7 ¯¯ µ0 Ap N 2 x05
= 2
∂x5 ¯x=x0 mp (xof f − x06 )
¯ 2
∂f7 ¯¯ µ0 Ap N 2 (x05 ) ks
= 3 −
∂x6 ¯ 0
x=x m (x − x0 )
p of f 6
mp
93