Licenta Mecanica
Licenta Mecanica
Licenta Mecanica
Junfeng Zhao
in
(Mechanical Engineering)
Air pollution, global warming, and rising gasoline prices have lead govern-
ments, environmental organizations, and consumers to pressure the auto-
motive industry to improve the fuel efficiency of cars.
Since alternative fuels such as hydrogen are still quite far from being
commercially viable, improving the existing internal combustion engine is
still an important priority. Traditional internal combustion engines use a
camshaft to control valve timing. Since the camshaft is rigidly linked to
the crankshaft, engineers can optimize the camshaft only for one particular
speed torque combination. All other engine operating points will suffer from
a suboptimal compromise of torque output, fuel efficiency, and emissions. In
an engine with a camless valve actuation system, valve events are controlled
independently of crankshaft rotation. As a result, fuel consumption and
emissions may be reduced by 15% ∼ 20% and torque output is enhanced in
a wide range of engine speeds.
The Fully Flexible Valve Actuation (FFVA) system is our approach to
construct a camless valve actuation system. Within the limits of the dynamic
bandwidth of the system, it allows for fully user definable valve trajectories
that can be adapted to any need of the combustion process. The system
is able to achieve 8mm valve lift in 3.4ms, which is suitable for an engine
operating at 6000RPM. The valve seating velocity is similar to conventional
valve trains that achieve 0.2m/s at high engine speeds and 0.05m/s at engine
idle conditions. Finally, the energy consumption measured in an experimen-
tal test bed matches the friction losses of conventional valve trains and it
can further be improved by using an optimized motor.
This thesis describes the progress that has been made towards designing
this technology. A design methodology is derived and important operation
features of the mechanism are explained. Modeling and simulation results
show significant advantages of the FFVA over previously designed electro-
magnetic engine valve drives.
ii
Table of Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . xii
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background and Motivation . . . . . . . . . . . . . . . . . . 2
1.2 Principles of Camless System . . . . . . . . . . . . . . . . . . 4
1.3 Current State of Research . . . . . . . . . . . . . . . . . . . . 7
iii
Table of Contents
4 Motor Optimization . . . . . . . . . . . . . . . . . . . . . . . . 30
4.1 Stator Saturation Constraint . . . . . . . . . . . . . . . . . . 30
4.2 Demagnetization Constraint . . . . . . . . . . . . . . . . . . 32
4.3 Linearity Constraint . . . . . . . . . . . . . . . . . . . . . . . 35
4.4 Energy Constraint . . . . . . . . . . . . . . . . . . . . . . . . 36
4.5 Motor Design Example . . . . . . . . . . . . . . . . . . . . . 38
6 Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.1 3 Phase Inverter . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2 Position (or PD) Controller . . . . . . . . . . . . . . . . . . . 54
6.3 States Observer . . . . . . . . . . . . . . . . . . . . . . . . . 54
iv
Table of Contents
8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
8.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
8.2 Suggestions for Future Works . . . . . . . . . . . . . . . . . . 80
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
v
List of Tables
vi
List of Figures
3.1 Trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Jerk limited smooth trajectory . . . . . . . . . . . . . . . . . 23
3.3 Winding variation for QB02302 . . . . . . . . . . . . . . . . . 29
vii
List of Figures
5.1 QB02302 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2 Optimized motor . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.3 Mesh result . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.4 Flux density distribution of the optimized motor . . . . . . . 51
viii
Acronyms
ix
Nomenclature
r Rotor Radius
hs Slot Height
hrs Thickness of Stator Back
rsi Stator Inner Radius
rso Stator Outer Radius
l Motor Length
lg Airgap
lge Effective Airgap
lm Magnet Thickness
β Magnet Coverage Span
wt Tooth Width
ws Slot Width
KT Motor Torque Constant
KB Back EMF Constant
R Motor Terminal Resistance
L Motor Terminal Inductance
Z Total Number of Conductors
I Current in each Conductor
Itotal Total Current Itotal = ZI
T Motor Torque
ra Length of Excenter Arm
Jm Motor Inertia
m Valve Mass
J Total Inertia J = Jm + mra2
α Angular Acceleration
ω Angular Velocity
θ Rotation Angle
a Linear Acceleration
v Linear Velocity
s Linear Displacement
j Jerk(Derivative of a )
x
Nomenclature
E Energy Loss
ρ Density of Core and Magnet
Br Magnet Residual Flux Density
Bc Flux Density in Core
Bs Saturation Flux Density
Bgo Airgap Flux Density Due to Magnet Alone
BI Airgap Flux Density Due to Current Alone
Bg Airgap Flux Density Bg =Bgo + BI
BD Demagnetization Flux Density
wp Width of Each Pole
wm Width of Each Magnet
Ks Linear Current Density
γ Angular Displacement Between the Fields Produced by the Magnet and
the Stator Current
kw Stator Length Factor
Aca Copper Area per Slot
Aslot Slot Area
dw Wire Diameter
nw Number of Wires per Slot
kf ill Filling Factor
p Number of Poles
q Number of Slot per Pole per Phase
Q Total Number of Slot
→
−
H Vector of Magnetizing Field
→
−
D Vector of Electric Displacement
→
−
J Vector of Free Current Density
→
−
B Vector of Magnetic Field
→
−
A Potential Vector
ε Electric Permeability
µ Magnetic Permeability
σ Electric Conductivity
→
−n Normal Unit Vector
Pi Proportional Gain in PI Controller
Ki Integral Gain in PI Controller
Pd Proportional Gain in PD Controller
Kd Derivative Gain in PD Controller
xi
Acknowledgements
I would like to thank all people who have helped and inspired me during my
master study.
I especially want to thank my advisor, Prof. Rudolf Seethaler, for his
guidance during my research and study at University of British Columbia
Okanagan. His perpetual energy and enthusiasm in research had motivated
me. In addition, he was always accessible and willing to help me with my
research. As a result, research life became smooth and rewarding for me. I
am indebted to him more than he knows.
Prof. Holzman, Prof. Najjaran, and Prof. Koch deserve a special thanks
as my thesis committee members and advisors.
Finally, I thank my family for supporting me throughout all my studies
at University of British Columbia Okanagan.
xii
Chapter 1
Introduction
1
1.1. Background and Motivation
2
1.1. Background and Motivation
tants and greenhouse gas from motor vehicles must be reduced [7]. Fig-
ure 1.1 shows the EU standard of powertrain evolution in the future for
both gasoline engines and diesel engines, which are represented by red and
black spots respectively in the picture. These objectives are challenging and
conflict with each other. It can be seen that for both gasoline and diesel
engines there is still a long way to go before they can meet the proposed
standard. In practice, if N Ox emissions are reduced, the CO2 emissions
and fuel consumption inevitably rises. With the more stringent regulations
expected, new engine technologies are urgently required.
In Figure 1.2, key technologies for improving engine’s performance are
shown. The circled camless system, is on a very promising place on the map.
Also, it can be seen from Figure 1.3 that the camless technology is the most
effective, since it can improve the fuel economy by 15% ∼ 20%.
But a combination of different technologies needs to be developed, since
no single one can fully satisfy the future standard. The large number of mu-
tually interactive variables and sub-systems is a major engineering challenge,
but it also provides the flexibility that is essential for converging to an op-
timal solution. After all, a properly matched combination of fuel, injection
system, air management, and exhaust treatment should make it possible to
3
1.2. Principles of Camless System
4
1.2. Principles of Camless System
during which both of the intake and exhaust valves are closed. Then the
mixture in the cylinder is compressed when the piston moves upwards. Ig-
nition is started when the piston is close to the end of the stroke. Then the
pressure in cylinder increases rapidly.
The third stroke from TDC to BDC is defined as a power stroke. When
combustion is completed, the pressure in the cylinder reaches a maximum
value. Then the cylinder starts expanding, which pushes the piston to move
until the exhaust valve opens, which is called exhaust valve opening (EVO).
Generally before BDC, the first portion of the burned gases is allowed to be
discharged.
The fourth stroke from BDC to TDC is defined as an exhaust stroke, as
the piston pushes the remaining burned gases out. Near the TDC position,
the intake valve opens again and the exhaust valve closes, which is called
exhaust valve closing (EVC). The timing is usually slightly after the TDC
position.
Figure 1.4 shows schematic p-V diagrams of a four-stroke engine with
a conventional mechanical valvetrain (plot 1) and a camless systems with
early intake valve closing (EIVC, plot 2) and late intake valve closing (LIVC,
plot 3). The shadowed area corresponds to the work needed to induct and
expel gases. These losses are referred to as pumping losses.
The pumping losses due to the throttling operation in a conventional
engine with a mechanical valvetrain are quite significant [10]. A challenge in
improving engine efficiency is to reduce the pumping losses while the friction
loss is not excessively increased. Two strategies are available for controlling
the amount of air drawn into the combustion chamber with variable valve
timing systems: Early intake valve closing (EIVC) and late intake valve
closing (LIVC). With EIVC, the engine load is controlled by closing the
intake valve early to trap the desired amount of charge, instead of throttling
the incoming charge conventionally by means of a throttle plate. Likewise,
LIVC also controls engine load with reduced pumping loss, but in this case
by returning unwanted charge to the intake manifold.
The comparison between pressure cycles clearly shows the advantages
obtainable by adopting a camless valve actuation system in the same engine
due to a significant reduction in pumping losses.
No matter whether EIVC or LIVC is implemented in the combustion
process, the full controllability of each valve must be obtained by replacing
the conventional valve system with a camless system, which must have some
5
1.2. Principles of Camless System
6
1.3. Current State of Research
In recent years, some relatively basic variable valve actuation systems have
been presented in publications or even introduced into production engines.
These forms mainly use mechanical, electro-hydraulic or electromagnetic
valve actuation technologies.
The mechanical systems could be very simple in structure as in Fig-
ure 1.5. The simplest ones, the cam phasers, change only valve timing.
Other types, slightly more complex, can change the valve lift. These mech-
anisms are quite simple and effective, and some of them are already on the
market. But the mechanical mechanisms can only provide limited flexibility
of the valve’s motion, and their dynamic response is too slow for guaran-
teeing optimum valve timing for transient engine operations. The Honda
VTEC mechanism [17] and the Toyota VVT-i system [18] are examples of
mechanical variable valve timing systems.
Electro-hydraulic systems [21] [22] are conceptually quite simple. The
electro-hydraulic camless systems proposed so far usually offer a contin-
uously variable and independent control of all aspects of valve motion.
Electro-hydraulic systems typically use piezo-actuated valves to control the
hydraulic fluid flow that is used to displace the valve. Unfortunately, hy-
draulic systems suffer from viscosity changes across the required temperature
range, since engine oils are typically used as the hydraulic liquid. Thus, the
performance deteriorates at low temperatures. In addition, it is very dif-
ficult to achieve good energy efficiency with hydraulic systems, since there
is no simple way to recover the kinetic energy of the valves when they are
slowed down [23].
Electromagnetic systems are characterized by a spring system that is
used to accelerate and decelerate the valve. Magnets or motors are used to
hold the valves in the end positions and to compensate for friction loss as
well as for combustion forces. Devices using solenoids such as in the one
shown Figure 1.7, are able to generate flexible valve phase and duration,
however, high valve seating velocity is difficult to control and the valve lift
is limited by the structure. Nevertheless, this is a very popular technology
and needs to be discussed in more detail.
7
1.3. Current State of Research
8
1.3. Current State of Research
9
1.3. Current State of Research
10
1.3. Current State of Research
11
Chapter 2
A schematic of the FFVA system is shown in Figure 2.1. There are three
main parts: the actuation part, the valve control unit and the engine control
unit.
The engine control unit controls the engine operation and provides the
required valve timing information to the valve control unit. The engine
control unit is designed separately, and not part of this research.
12
2.1. Actuator
The valve control unit receives the desired valve timing and lift from the
engine control unit and uses this information together with measured valve
position and measured current in order to regulate the amount of voltage
applied to the actuator.
The actuator consists of BLDC motor, a valve and a linkage structure
connecting the two. A state observer, including sensors and digital filters,
is connected to the actuator and used to measure the valve motion, which
is then provided to the valve control unit.
2.1 Actuator
Figure 2.3 shows a detailed picture of the linkage between motor and valve.
An excenter arm of length, ra , is attached to the motor’s shaft, which in turn
is connected to the valve through a small bracket. This structure transfers
the motor’s angular movement to the valve’s linear movement:
a = αra (2.1)
If the rotor angle is small, the relationship between lift and rotor angle
can be assumed linear. The excenter arm is made of aluminium and its
inertia can be neglected when compared to the rotor inertia and the valve
mass. The motor torque T is used to accelerate the rotor with an inertia,
Jm , and the valve with a reflected inertia mra2 , where m is the mass of the
valve:
T = (Jm + mra2 )α (2.2)
13
2.2. Principles of Brushless DC Motor
The figure also shows several parameters, which play an important role
in determining the system’s performance. These parameters, especially the
rotor angle θ and excenter arm length ra , will be discussed in the chapter of
optimization.
As shown above, a brushless DC motor is the key part of the FFVA system.
The basic structure of a permanent magnet brushless DC motors has three
elements: a stator with windings, a rotor with permanent magnets attached
to it, and a sensor to measure the rotor position. A picture of QB02302 [27]
and its 3D model is given in Figure 2.2.
The motor’s torque constant and other properties are all related to the
structure and material of the motor. Figure 2.4 provides the dimensioning
details of an interior rotor BLDC motor (2D geometry of motor QB02302).
The rotor radius r, the slot height hs and the length of the motor l are treated
as variables for an optimized design goal in subsequent chapters. The airgap
lg and magnet thickness lm are selected to give sufficient flux density in the
airgap. Magnets are glued on the rotor surface. The coverage of the magnets
on the rotor is called span, represented by β. In our application, the span is
about 0.8. Inside the stator, the tooth width wt and the slot width ws are
almost the same.
To understand how these elements work as a motor, consider some ele-
mentary magnetics [28]. When a current carrying wire is placed in a mag-
netic field so that the current flow is perpendicular to the direction of the
field, a force is exerted between the field producing element and the wire.
The electromagnetic model of the actuator is approximated by an equiv-
alent linear single phase DC-motor model. The mathematical model of the
BLDC motor is based on the following assumptions: 1) stator resistances
of all the three phases are equal and the self and mutual inductances are
constant; 2) the motor is operated within the rated condition and hence the
saturation effect due to current level is neglected; 3) iron losses are negligible.
This approach is followed, since most motor specifications are based on
this model and we want to be able to select an optimum motor from the
manufacturer’s motor specifications. The motor current I, is defined by
a first order differential equation in terms of the applied voltage U , and
14
2.3. Valve Control Unit
the back EMF voltage KB ω, the winding resistance R, and the winding
inductance L:
dI 1
= (U − KB ω − IR) (2.3)
dt L
The main function of the valve control unit is to move the valve from the
closed to the opened position (and vice-versa) avoiding noise, which is caused
by nonzero seating velocity. This is achieved using a cascaded tracking
controller with feed forward.
In Chapter 3, valve trajectories for minimizing energy and maximizing
acceleration will be derived, and in Chapter 6 the design of the tracking
controller will be described.
15
2.3. Valve Control Unit
16
2.3. Valve Control Unit
17
Chapter 3
18
3.1. Mechanical Optimization
valve:
T = (Jm + mra2 )α (3.1)
By setting the derivative equal to zero, one finds the excenter arm length
that provides maximum acceleration:
r
Jm
ra = (3.4)
m
It will now be shown that the same excenter arm length also provides
minimum energy consumption, given a desired valve acceleration profile.
When the motor’s rotation angle, denoted as θ, is small enough, the valve
motion can be regarded as linear. Then the relationship between valve’s
linear displacement s and motor’s rotational displacement θ can be expressed
as:
s = ra θ (3.5)
Thus,
d2 s
= ra · α (3.6)
dt2
where α is angular acceleration. The torque that the motor provides is:
J d2 s
T = Jα = (3.7)
ra dt2
where J is the total inertia including motor inertia Jm and load inertia mra2 :
J = Jm + mra2 (3.8)
19
3.2. Motor Selection
Now consider the motor’s copper loss, which depends on current in the
form of:
d2 s 2 Jm + mra2 2 d2 s
Z Z Z
2 J 2
E = I Rdt = ( ) R ( 2 ) dt = ( ) R ( 2 )2 dt
K T ra dt K T ra dt
(3.10)
Take the derivatives on both sides:
Jm Jm + mra2 2 d2 s 2
Z
∂E 2R
= 2 (m − 2 )( ) R ( ) dt (3.11)
∂ra KT ra K T ra dt2
Few off-the-shelf motors are designed for the highly dynamic operating con-
ditions found in the FFVA system. However, it is very important for a suc-
cessful implementation that the motor operates with high efficiency. Thus,
using the findings from the mechanical optimizations, criteria for selecting
a motor that provides maximum valve acceleration and minimum energy
consumption are derived in the following section.
The mechanical optimization assumed that the rotor motion would be
small enough in order to ensure a linear relationship between valve and
motor motion. Using this assumption, the identical Equations 3.4 and 3.12
describe the optimal excenter arm length. One can now go backwards and
describe the minimum motor inertia required in order to achieve a rotor
angle smaller than θmax , for a maximum valve motion, smax , and a valve
mass, m:
smax 2
Jm > m( ) (3.13)
θmax
20
3.2. Motor Selection
Since minimizing the size of the motor reduces cost and facilitates pack-
aging the motor in the cylinder head, one would usually choose motors with
inertia close to the linearity constraint.
In addition to the linearity requirement, the motor is also required to
provide large acceleration. Substituting Equation 3.4 back into Equation 3.2
provides an equation for the maximum valve acceleration in terms of motor
and valve parameters:
Tmax ra Tmax
amax = = √ (3.14)
Jm + mra2 2 Jm m
d2 s 2
Z
Jm R
E = 4m × × ( ) dt (3.15)
KT2 dt2
This equation indicates that there are three parts of the actuator that
need to be optimized in order to minimize copper losses.
• The energy cost term of the motor consists of three parameters, the
motor inertia multiplied by the electrical resistance divided by the
square of the motor torque constant. The energy cost term provides
a convenient guideline to compare and select stock motors using their
specification data sheets. Chapter 4 will derive an optimum design
procedure for the motor.
• The last term in Equation 3.15 is the integral of the acceleration tra-
jectory of the valve. The next section will show how to design energy
optimal trajectories.
21
3.3. Trajectory Generation
The usual requirement for a valve trajectory is that the valve needs to travel
the desired lift within a prescribed transition time t4 , which is given by the
engine control unit. Both lift, and transition time will vary with changing
engine speeds and engine torque requirements. Thus, a simple online al-
gorithm that generates energy optimal valve trajectories is required. The
problem can be solved by calculus of variations [30], which leads to the
following valve acceleration [31]:
2t
a(t) = amax (1 − ) (3.16)
t4
where amax is the maximum acceleration required to achieve the desired
valve lift and transition time with minimum energy consumption. The wave
form is shown as the plot of ”Optimal” in Figure 3.1 and the resistive copper
22
3.3. Trajectory Generation
where smax is the maximum valve travel, and t2 = t4 /2 is half the valve
travel time.
In practice, the optimal trajectory suggested above is not feasible due to
two physical constraints.
First, maximum current constrains the available acceleration for a motor
23
3.3. Trajectory Generation
If the motor parameters are fixed, then the voltage becomes the limiting
factor for jerk. The optimal trajectory can be followed only when the motor
can provide infinite jerk, which requires infinite voltage.
The remaining section then outlines a procedure for imposing a jerk lim-
ited smooth trajectory. The kinematic profiles used in trajectory generation
are illustrated in Figure 3.2. The suboptimal acceleration has a triangular
profile with maximum slopes (i.e. jerk values) at the start and at the end
of the valve motion. It requires a slightly higher amax than the optimal
trajectory, which can be seen from Figure 3.1.
Given a time t1 during which the initial maximum jerk j1 is applied, the
value of this maximum jerk is defined as:
3smax
j1 = t 1 t2 (3.20)
2t2 − t1
The energy lost in the copper windings during a single transition from
closed to open or from open to closed, with the sub-optimal trajectory is
given by:
6s2max J 2 R
Z
Esubopt = I 2 Rdt = (3.21)
2t2 (t1 − t2 )2 ra2 KT2
We now aim our attention towards finding the minimum value for t1 .
Equation 3.19 indicates that during the initial period where j = j1 , the
24
3.3. Trajectory Generation
voltage must be continuously increasing, since the jerk is constant and both
velocity and acceleration are increasing. Thereafter, the voltage decreases,
since jerk is reversed. Thus, it is expected that the maximum voltage will
take place at t = t1 . Equations 3.19 and 3.20 for t = t1 lead to a quadratic
equation in terms of t1 :
(3KT2 smax +2KT t2 ra U )t21 +6smax LJ +(6smax RJ −4KT t22 ra U )t1 = 0 (3.23)
25
3.4. Design Example
The same idea can be applied to the velocity and position profiles. For
constant jerk, acceleration profiles are linear, velocity profiles are parabolic
and displacement profiles are cubic. The advantage of using a jerk limited
profile is that triangular acceleration profiled trajectories have smoother
velocity, acceleration and jerk characteristics compared to other profiled
trajectories. The control signals resulting from the utilization of such a ref-
erence trajectory will also be smoother, hence reducing the risk of impacting
the drives, or exciting the machine’s structural dynamics.
In the simulation of FFVA system, a Waveform Generator is programmed
to generate the reference trajectory. The transition timing and the displace-
ment of the valve can be set.
From the sections above, two important criteria for motor selection were
derived which are summarized as follows:
Tmax
• amax = √
2 Jm m
Jm R
• E∝ KT2
26
3.4. Design Example
In Table 3.1, three candidates with similar size are listed. The size is
chosen in order to provide the minimum inertia that fulfills the linearity con-
straint between valve movement and motor movement. Since the QB02302
series motor outperforms the other candidates in terms of energy consump-
tion and acceleration, it is used in the experimental test bed in Chapter
7.
It should be pointed out that the QB02302 motor can be ordered with
different windings.
To make a first approximation, changing the windings from N to N ∗
affects the motor parameters as follows:
N∗ N∗ 2 N∗ 2
KT∗ = KT , R ∗ = ( ) R, L∗ = ( ) L (3.26)
N N N
Note that the energy cost term is not affected by changing the number
of windings. Also, the maximum torque and thus, the available acceleration
does not change, even though the current required to achieve this torque will
be altered, since the torque constant KT is linearly related to the number
of windings:
N
Imax = Tmax (3.27)
KT N ∗
27
3.4. Design Example
Another interesting relationship for the FFVA system is the jerk available
for a given maximum supply voltage Umax :
K T ra N KT v aJ
j= (Umax ∗ − −R ) (3.28)
JL N ra K T ra
Equation 3.28 indicates that the available jerk scales linearly with volt-
age and the inverse of the winding ratio. In addition, the jerk is reduced
once the valve starts moving. To summarize, decreasing the number of wind-
ings increases the available maximum jerk. However, the current required to
achieve the maximum acceleration is increased at the same time. Selecting
the optimum number of windings then depends on the available voltage, cur-
rent, and the type of trajectory chosen to reach the required valve dynamics.
For our trajectories, a higher initial jerk leads to lower energy consumption.
It is expected then, that a lower winding number improves efficiency at the
expense of higher currents.
For the FFVA system with a QB02302 motor, Figure 3.3 shows a plot of
maximum current, maximum torque and energy consumption vs. winding
ratio: This plot shows that fewer windings require moderately less energy
and a slightly smaller maximum torque. However, these advantages are
offset by a large increase in required current. Note that there is a minimum
value for maximum current and the winding ratio chosen for the FFVA
system is close to that minimum value.
Before presenting an implementation of a system with the stock QB02302
motor, the following chapter derives how to design a custom motor for this
application.
28
3.4. Design Example
29
Chapter 4
Motor Optimization
The last chapter showed that motor parameters play an important role in
the energy consumption and the acceleration capability of the FFVA sys-
tem. Motor selection criteria were presented that can be used to choose
the best off-the-shelf motor. However, in the FFVA system, the motor is
used for a transient application, while the majority of motors on the market
are designed for continuous application. Thus, off-the-shelf motors usually
do not provide an optimal design for this application. This chapter then
attempts to provide an analytical method to design an optimum motor.
Several critical constraints should be considered when such a BLDC motor
is designed. First it will be demonstrated that the stator saturation deter-
mines the thickness of the permanent magnets used on the rotor. Then a
constraint on magnet demagnetization will provide a relationship between
the acceleration of the actuator and the motor length. Thereafter, the need
for a linear relationship between rotor movement and valve movement will be
used to define radius of the rotor. Finally, it will be shown, that a relation-
ship between rotor and the stator radius governs the energy consumption of
the actuator.
The iron core of the stator consists of teeth and slots. The slots are filled
with windings that create a stator field. The teeth are used to guide the
stator field due to the current in the stator windings and the rotor field due
to permanent magnets on the rotor. In order to ensure efficient use of the
stator material, the stator core should not be saturated. This section shows
that stator saturation can be used to determine an optimal thickness of the
permanent magnets on the rotor.
A variety of magnet materials are available to provide the required
rotating magnetic field in a brushless dc motor. The most popular are
30
4.1. Stator Saturation Constraint
where lge is the effective length of the airgap including the distance from
the stator teeth to the rotor iron core. Usually the effect of slotting is also
included. Approximately, lge = lg + lm .
Ideally speaking, reducing the airgap lg could obtain the same Bgo by
using less permanent magnet material. But in practice, the minimum value
of the airgap is set by mechanical considerations. The usual range of lg is
above 0.5mm.
Choosing the operating point of the magnet to be at the maximum energy
product of the magnet material will result in minimum magnet volume and
magnet cost [35]. However, the resultant average air gap flux density will be
low (about one half of the magnet Br ), and therefore the armature winding
will take a larger portion of the stator volume to provide the larger total
current in order to generate the same torque.
Using a thicker magnet results in a more expensive magnet, but increases
the airgap flux density, reduces the armature current loading, and results
in a better balance of stator lamination iron and stator copper. But the
flux density in the iron core Bc should also be considered before deciding
Bgo . The maximum flux density in the iron core is usually limited in order
to avoid saturation. For the QB02302 motor, the material of the stator is
31
4.2. Demagnetization Constraint
32
4.2. Demagnetization Constraint
where Ks is the rms linear current density along the stator periphery. No-
tation p represents the number of poles. So the demagnetization constraint
is given by [36]:
plge (Bgo − BD )
Ks ≤ √ (4.5)
2 2rµ0 sinβ
33
4.2. Demagnetization Constraint
This equation does not include saturation of the iron core. Especially for
small motors, this can lead to substantial errors. Substituting Equation 4.9
34
4.3. Linearity Constraint
and 3.12 into 3.14 provides an expression for maximum valve acceleration
in terms of geometric dimensions and material properties (the expression of
Jm can be found in next section) of the motor [38]:
√ √
Tmax ra 2plm Br (Bgo − BD )kw l
amax = = √ (4.10)
2Jm µ0 πρm r
35
4.4. Energy Constraint
assumption requires that the rotor rotation is small. In practice, one would
usually constrain the rotor rotation to a maximum value of θmax = 30o , thus
the length of the excenter arm is also constrained:
r
Jm s
ra = > (4.13)
m θmax
In practice, the density of the rotor core and the permanent magnets
attached to it, are very similar. Thus, the rotor can be assumed to be a
solid cylinder with density, ρ, motor length, l, and rotor radius, r. The
rotor’s inertia Jm can be expressed as:
1 1
Jm = πr2 lρr2 = πlρr4 (4.15)
2 2
So,
2s2 m2
r4 l > 2
(4.17)
θmax πρ
Since the length of the rotor has been determined by the acceleration
requirement of the application, Equation 4.17 provides a constraint for the
minimum rotor radius. At this point the rotor dimensions are fully defined
by the valve acceleration requirements, the constraints on stator saturation,
and linearity between motor and valve motion. The following section shows
how the stator diameter determines the energy consumption of the actuation
device.
In the FFVA system, the energy loss should be less than 2.5J/cycle. When
the motor is designed, the index function JKm2R should be taken into account
T
in order to minimize the energy loss.
36
4.4. Energy Constraint
T 1
KT = = · (Z · 2r · l · Bgo )kw . (4.19)
I 3
Z 2l
R = f( ) (4.20)
Aca
which shows that the cost term is proportional to r2 and A1ca , but is inde-
pendent of the stator length l or number of conductors Z. If we want to
reduce the energy loss, the radius of the rotor should be decreased or the
copper area in each slot increased. The minimum radius is constrained by
the linearity requirement from the previous section. The copper area is con-
strained by the slot area and the ability to fill the slot area with copper. The
remainder of this section is then concerned with the relationship between
stator geometry and copper area in each slot.
The copper area in each slot is related to the wire diameter, dw , and the
number of wires in each slot, nw :
dw 2
Aca = nw π( ) (4.22)
2
The stator slot cannot be completely filled with conductors, since there is
always space left around the wires. The ratio of slot area, Aslot , to conductor
area, Aca , is defined as the filling factor, kf ill :
Aca
kf ill = (4.23)
Aslot
37
4.5. Motor Design Example
The slot area is related to the number of slots, Q, the stator inner radius,
rsi , the slot height, hs , and the tooth width, wt :
π
Aslot ≈ [(rsi + hs )2 − rsi
2
] − wt hs (4.24)
Q
where hrs is the required thickness of the stator back. In Figure 2.4, it can
be seen that the stator back must be larger than half the tooth thickness:
1
hrs ≥ wt (4.26)
2
The DC motor (QB02302) chosen for the FFVA system has the parameters
shown in Table 7.1. This is a very good motor, but it was not specifically
designed for our application. This section then shows how to optimize the
motor for the FFVA system.
2. Rotor Radius r: In the last step, the length of the motor was decreased,
thus the rotor inertia Jm becomes smaller. The linearity constraint
needs to be verified. Equation 4.17 can be rewritten as:
2s2 m2
r4 > 2
θmax πρl
38
4.5. Motor Design Example
For lift s = 8mm, valve’s mass m ≈ 40g, motor length l = 28mm the
density of steel ρ = 7800kg/m3 , and θmax = 30o , the limit of r is given
by:
r > 12.8mm
The rotor radius is kept at 14mm, still fulfills the linearity constraint.
3. Slot Area Aslot : The two steps above are used to determine the rotor
size. Energy consumption determines the stator size. It was shown
in Section 3.4 that the energy consumption of the stock system is
expected to be at least 7% larger than required. Thus, the goal of this
step is to reduce the cost term JKm2R of the optimized motor to half of
T
the value of the original QB02302 motor. The energy cost term has not
been affected by the change in rotor length and the parameters left for
us to manipulate are mainly the slot height hs and wire diameter dw .
The number of conductors will be kept. Thus, since the rotor length
was cut in half, the inductance of the motor should also be halved.
This should allow for more energy efficient trajectories.
(a) First, the slot area and related parameters for the original mo-
tor QB02302 are determined. The measured parameters are slot
width ws = 5.1mm, tooth width wt = 5.0mm, number of wires
per slot nw = 72, stator inner radius rsi = 14.5mm, number
of slots Q = 9, airgap lg = 0.5mm, thickness of stator back
hrs = 4.65mm. wire diameter dw = 0.723mm [40] (Gauge 21:
constant maximum current is about 9A, transient maximum cur-
rent is usually 6 ∼ 7 times of that, which is around 50A; the
resistance is 42.0Ohm/km) and slot height hs = 8.5mm.
According to Equation 4.22 the copper area of one slot is given
by:
dw
Aca = nw π( )2 = 29.6mm2
2
Thus, Equation 4.24 provides the expression for the slot area:
π
Aslot = [(rsi + hs )2 − rsi
2
] − wt hs = 68.8mm2
Q
The fill factor can be calculated by using Equation 4.23:
Aca
kf ill = = 43%
Aslot
39
4.5. Motor Design Example
Aca = 74.8mm2
Aslot = 140mm2
kf ill = 53%
rso = 35mm
Thus, the copper cross section area and slot area have both in-
creased. The filling factor becomes a little higher, but it is still in
a reasonable range. The outer radius has increased by 7.35mm.
Most importantly, the cost term JKm2R has decreased. From Equa-
T
tion 4.21, we know the copper loss of the system for an FFVA
system with constant rotor radius is inversely proportional to the
copper area:
Jm R 1
E∝ ∝ (4.28)
KT2 Aca
29.6mm2
Which means theoretically, the copper loss now is 74.8mm2
=
39.6% of the value that QB02302 consumes.
40
4.5. Motor Design Example
41
Chapter 5
Because of the motor’s symmetry along the axis of the shaft, a 2D model
is sufficient to analyze the important properties, while the complexity of
the calculations are reduced. The transverse sections are given in Figure 5.1
and Figure 5.2. The structures of both motors are also symmetrical by every
120o , only one third of the layout is simulated, which is sufficient and concise
for the Finite Element Analysis. Comparing the two models, it can be seen
that the rotor structure is kept the same and the stator structure is changed
by varying the slot height and outer diameter of the stator.
42
5.2. Maxwell Equations
→
− →
−
where E is the vector of the electric field, B is the vector of the magnetic
field, ε is the electric permeability, µ is the magnetic permeability and σ is
the electric conductivity.
43
5.3. Governing Equations in 2D Electromagnetic Field
Or,
∂ 1 ∂Ax ∂ 1 ∂Ay
( )+ ( ) + Jz = 0 (5.3)
∂x µ ∂x ∂y µ ∂y
where Jt is the total surface current while Jz is the component of Jt in the
→
−
z-direction, and the flux density vector B is defined in terms of the potential
→
−
vector A using the auxiliary equation:
− −
→ → −→
∇×A =B (5.4)
44
5.3. Governing Equations in 2D Electromagnetic Field
In the laminated stator, the iron core and air gap are
Ω1 : ∂ ( 1 ∂Ax ) + ∂ ( 1 ∂Ay ) = 0
∂x µ ∂x ∂y µ ∂y (5.6)
Γ : −
→
A =0
u
The term σ ∂A
∂t exists in the solid rotor iron core where eddy currents
cannot be ignored.
There are two models which are commonly used to represent perma-
nent magnets: the magnetization vector method and the equivalent current
sheet method. The magnetizing vector method is used in this model for the
permanent magnets:
∂ 1 ∂Ax ∂ 1 ∂Ay 1
Ω3 : ∂x ( µ ∂x ) + ∂y ( µ ∂y ) = ∇ × ( µ Br )
−−−→ −−→ (5.8)
Γt2 : 1 ∂ A− PM 1 ∂ Aair
+ = Js2
µ →
n µ →
−n
PM air
Here, the free current density is not 0 anymore, because there is induced
current in the permanent magnet when the magnetic field is changing. The
linear current density between air and the permanent magnet gives the nat-
ural boundary condition.
45
5.4. Finite Element Simulation for BLDC Motor
Since there is free current going through the windings in the positive
or the negative direction, the current density can be expressed as the total
current over area. The third equation is the coupling of the field and the
external drive circuit. The boundary conditions were already given in the
previous three parts.
These are governing equations for different parts of motors. They are not
directly solved by the FEA software, because they are all partial differential
equations. They have to be transferred to a weak form, which then can be
solved numerically by the software.
The previous sections described the motor geometry, the governing equa-
tions, and the boundary conditions of the motor. These models are imple-
mented in Maxwell SV using Galerkin’s method for 3-node elements. This
section describes the meshing process, the Galerkin’s formulations, and typ-
ical plots of flux density and current density that are obtained with Maxwell
SV.
Mesh generation for the FEM should be simple and robust, and the rotor
mesh should be allowed to rotate easily. In this approach, the FEM mesh of
the cross section of the BLDC motor is divided into three parts: the stator,
the rotor and the magnet, with each including a part of the air gap. When
the rotor is rotated according to the time step, the shape of the mesh for
both the stator and rotor can be kept constant and only the coordinates of
the rotor mesh and the periodic boundary condition on the interface need
to change. Therefore, in this approach the stator mesh and the rotor mesh
are required only need to be generated once. This can greatly reduce the
computing time required to generate the FEM mesh at each time step. A
46
5.4. Finite Element Simulation for BLDC Motor
typical mesh for the stock QB02302 motor is shown in Figure 5.3. It is
automatically generated by the software, but a finer mesh along the air gap
was specified manually, because this is the most important part of the flux
distribution. Additionally, the software checks for convergence and suggests
finer meshes, if the convergence criteria are not met.
Galerkin’s method is usually employed for the finite element formula-
tion. This method uses particular weighted residuals for both the weighting
functions and the shape functions.
According to the Galerkin’s method, for a 3-node triangular element, the
magnetic vector potential can be expressed as [42]
3
X
A= N i Ai (5.10)
i=1
47
5.4. Finite Element Simulation for BLDC Motor
function and the Ai is the approximation of the vector potential at the nodes
of the element.
The Galerkins formulation of the laminated stator iron core and air gap
ZZ 3 3
∂Ni ∂ 1 X ∂Ni ∂ 1 X
( N j Aj + Nj Aj )dxdy = 0 (5.11)
∂x ∂x µ ∂y ∂y µ
j=1 j=1
In matrix form
[G]A = Brx [ci ] − Bry [bi ] (5.16)
48
5.5. Evaluation of Optimized Motor
In matrix form:
L ∂A ∂A di
V = [({Q}{ })n − ({Q}{ })p ] + Ri + L (5.20)
S ∂t ∂t dt
where, ZZ ∆e
6 if i = j
Tij = Ni Nj dxdy = ∆e
12 6 j
if i =
ZZ
∆e
Q= Ni dxdy =
3
ZZ
∂Ni ∂Nj ∂Ni ∂Nj bi bj + ci cj
G= ( + )dxdy =
∂x ∂x ∂y ∂y 4∆e
and ∆e is the triangular area of the element.
The field equations above and the circuit equation have to be solved
simultaneously. Rotor movement has to be coupled. The coefficient matrix
is symmetric.
The computed steady-state flux distributions when maximum current
100A is applied in the stator winding is shown in Figure 5.4 (QB02302),
which indicates that the iron core is not saturated even maximum current
is applied. Forces and torques are calculated by integrating the Maxwell’s
stress tensors along a closed path in the air gap.
49
5.5. Evaluation of Optimized Motor
term, which means that it consumes less energy for completing the same
transition. Also, there is good agreement between the FEM simulation and
the analytical expressions. The parameters of size are inherently the same
since the models are identical. The difference of energy cost term between
FEM and theory is mainly caused by two parameters: motor resistance and
torque constant. The motor resistance given by the software includes the
extra windings stretching out of the stator, which is not estimated in our
theocratical calculation. The torque constant in the analytical calculation
is an estimated value, with some assumptions, such as the flux density in
the iron core is uniform and the magnet thickness is even. The FEM model
is likely closer to the true motor parameters.
The value of the energy cost term in the optimized motor is almost half
of that of the stock QB02302. Thus, the energy loss for the same operation is
cut down to half in theory. After optimizing the geometry of the stator, the
quantity of material used in the laminated core is increased. Even though
the core loss has actually increased, it is relatively small comparing to the
copper loss. It can be seen that the overall performance of the optimized
motor is far better than the stock motor QB02302.
This chapter used a finite element simulation in order to validate the
theoretical predictions of the optimized motor. This was achieved by sim-
ulating the original QB02302 motor and the new optimized motor. The
QB02302 motor compared well with the specification sheet and the opti-
mized motor compares well with the theoretical predictions from Chapter
4. The parameters obtained with the FEM model for the optimized motor
will be used in Chapter 7 in order to simulate the overall actuation system.
50
5.5. Evaluation of Optimized Motor
51
Chapter 6
Control Strategy
In this chapter, the technology for controlling the FFVA system is discussed.
The main function of the control system is to move the valve from the closed
to the opened position (and vice-versa) avoiding noise, which is caused by
nonzero seating velocity. Similar to typical modern servo control systems, a
cascaded control architecture is employed (see Figure 6.1). A conventional 3
phase inverter provides PI current control and commutation using measured
values of position and current as well as reference current. The reference
current is provided by a PD+Feedforward position controller that are im-
plemented in a dSpace rapid prototyping system. The position controller
receives its reference position from an energy minimized trajectory genera-
tor that is also implemented in dSpace. The algorithms for the trajectory
generator have been provided in Chapter 3. Positions measurements are
provided by an encoder. In order to minimize noise when deriving velocity
from the position signal, a state observer is employed. A detailed description
of the different parts of the controller follows in this chapter.
52
6.1. 3 Phase Inverter
53
6.2. Position (or PD) Controller
Then,
Pd = ωp2 J; Kd = 2ζωp J
54
6.3. States Observer
resolution(θ)
resolution(ω) =
SamplingT ime
resolution(ω)
resolution(α) =
SamplingT ime
Since the typical sampling time is 0.025ms, the resolution for velocity
and acceleration using a simple backward difference operator deteriorates
drastically. The quantization noise on the measured velocity is small for
the encoder type chosen in the experimental setup. However, real sensors
in actual engine applications would usually provide a much lower resolution
that would then lead to significant quantization errors in the derived velocity
measurements. A low pass filter (LPF) can be introduced to reduce this noise
55
6.3. States Observer
If we set the sampling time at 0.25e−4 ms, and use the parameters given
above, the discrete state space can be calculated:
1 0.001529 0 0
A = 0 0.9794 0 B = 1.0e−4 0
0 121.9 0.9999 0.001529
C= 1 0 0 D=0
The predictor corrector type procedure of the Kalman filter can then be
written as
q(k) = q̄(k) + G(k)[y(k) − Cq̄(k))] (6.4)
q̄(k + 1) = Aq(k) + Bu(k) (6.5)
T T −1
G(k) = M(k)C [CM(k)C + Rv ] (6.6)
P(k) = M(k) − G(k)CM(k) (6.7)
M(k + 1) = AP(k)AT + Rw (6.8)
where q̄(k) is the predicted state estimate at the sampling instant k, and
q(k) is the actual state estimate. G(k) is the Kalman gain. M(k) is the
56
6.3. States Observer
For sensor quantization, the error probability for any one sample is uni-
form within the region −0.5 to 0.5, and zero elsewhere. The covariance of
the noise for one sample can be calculated as:
Z ∞Z ∞
Rv = xpθ (x) × xpθ (x)dxdx
−∞ −∞
Z 0.5 Z 0.5
= xpθ (x) × xpθ (x)dxdx (6.10)
−0.5 −0.5
2(0.5)3
= = 0.0833
3
where, pθ (x) is the function of probability.
The Kalman Filter is simulated in Simulink together with other system
components. Simulation results comparing the performance of different fil-
tering techniques are given in chapter 7. They show the advantage of the
Kalman Filter over direct differentiation and low pass filtering.
57
6.3. States Observer
58
Figure 6.2: Cascaded control structure
6.3. States Observer
59
6.3. States Observer
60
Chapter 7
In the previous chapters, two setups were introduced. The first setup uses
a stock QB02302 motor, and the second setup uses an optimized custom
motor that provides lower power consumption.
The first part of this chapter describes the components of an experimen-
tal test bed used in this study. The second part of this chapter compares
the transition performance of the two motors using a Simulink simulation
for typical engine operating scenarios. The third part of this chapter looks
at the robustness of the actuation system towards parameter variations of
the mechanical design and variations of the motor parameters. The fourth
part of this section focusses on the ability of the control system to reject
external disturbances due to combustion pressures or friction. Finally, the
fifth part of this chapter describes the experimental results obtained with
the stock QB02302 setup.
The experimental test bed drives a single exhaust valve of a Honda cylinder
head (see Figure 7.1). A block diagram of the FFVA system is shown in
Figure 6.1. It contains a dSpace1103 controller board that creates trajecto-
ries and performs position control. For position measurements, a 20000 line
encoder from Quantum devices is used. The control board performs its tasks
at 40kHz in order to match the PWM frequency of the subsequent 3 phase
inverter from Maccon GmbH. The inverter performs current control and
commutation using high speed PWM controlled power MOSFETs that can
regulate up to 100A per phase. The inverter internally uses LEM modules
to feed the current controllers. In addition, three external LEM HTP100-
61
7.2. Simulation of Transition Performance
P current probes are employed to measure and display currents with the
dSpace system. The inverter drives a QB02302 motor that is mounted to a
small cylinder head from Honda. The light weight excenter arm is fabricated
in aluminum and has a length of 23mm in order to match the 37g valve to
the inertia of the motor. The excenter arm and the valve are joined using a
connecting rod made from 2mm piano wire. Table 7.1 lists the parameters
(QB02302 with customized winding) of the FFVA system.
62
7.2. Simulation of Transition Performance
Two separate simulations are performed. First, the stock motor QB02302
system is simulated and its performance is estimated. Second, the optimized
motor is simulated in order to compare its possible performance improve-
ment with the predictions from the design procedure.
The transition performance of the closed loop controlled system can be
quantified by using the following three indices [45]:
Transition Time: The time it takes for the valve to move from 5% of the
maximum lift to 95% of the maximum lift.
Valve Seating Velocity: The velocity of the valve when it contacts the
valve seats.
Energy Loss: The copper loss per valve per transition.
The initial simulation of the actuation system with the stock QB02302
motor is based on the parameters listed in Table 7.1.
A number of different operating conditions are simulated to demonstrate
the flexibility and performance of the FFVA system. Figure 7.2 shows the
63
7.3. Simulation of the Robustness Towards Parameter Variations
simulated valve lift curves and Figure 7.3 depicts the corresponding energy
consumptions.
The dark blue curve shows a typical 8mm valve lift curve for 6000rpm,
which corresponds to t2 = 2.4ms. As predicted in Section 3.4, this tra-
jectory requires 1.35J/transition which is slightly more than the desired
1.25J/transiton. This problem can be alleviated by reducing the lift. A
trajectory with half the lift and the same transition time is shown in green
and it requires approximately 0.35J which is less than a third of the energy
consumed for twice the lift.
In practice, the required lift reduction would be much smaller in order to
achieve the desired energy consumption. The red curve shows an 8mm lift
curve for 3000rpm. Since the additional valve open time was achieved by
inserting extra time with the valve completely open, rather than reducing
the valve accelerations, the energy consumption is still approximately 1.35J
in every transition. In practice, one would likely reduce the acceleration in
order to find a compromise between engine performance and valve energy
loss. The light blue curve represents a 4mm valve lift curve at 8000rpm.
Compared to the case of 6000rpm with the same lift, this mode requires
almost three times the energy. This highlights the nonlinear relationship
between energy consumption and transition time.
Even though the setup with the stock QB02302 motor performs quite
well in the simulations, the energy consumption at high rpm is slightly
larger than desired. The optimized custom motor was designed to address
this problem in Chapter 4. Figure 7.4 shows the displacement curves and
energy consumption for simulating the optimized motor at 6000rpm en-
gine speed. The energy consumption for this setup has been reduced from
1.35J/transtion to 0.66J/transition. This corresponds well with the pre-
dictions shown in Table 5.1.
The system parameters such as valve mass m, excenter arm length ra , motor
torque constant KT and winding resistance R can shift from the nominal
values due to manufacture accuracy, abrasion, temperature or other factors.
In the simulation, the values of the listed four parameters are all increased
64
7.3. Simulation of the Robustness Towards Parameter Variations
expected that actual sensitivity values will differ from the nominal values,
because large changes are made to the parameters and Equations 7.1 to 7.4
provide only a linearized approximation. Table 7.2 shows a comparison of
the estimated and the simulated sensitivity values. It demonstrates that the
torque constant, the resistance, and the valve mass have a large influence on
65
7.4. Simulation of Disturbance Rejection
the energy consumption, whereas the excenter arm length is not very sen-
sitive. This lets us conclude, that manufacturing tolerances of the linkage
system do not need to be very high in order to guarantee low energy con-
sumption. On the other hand all the other parameters need to be designed
carefully.
The FFVA system is designed for intake valves where only moderate com-
bustion pressure is expected on the valves. The shape of the pressure curve
during the opening or the closing transition is not well defined. Other distur-
bances could include some degree of viscous damping or coulomb damping in
the motor or in the valve guides. In this investigation we attempt to combine
all of these causes using disturbance torques on the motor that are modeled
with a viscous friction coefficient of 0.015N ms/rad and a coulomb coeffi-
cient of approximately 1.15N m. Each of the two parameters corresponds to
an equivalent disturbance force of 50N on the valve.
It should be pointed out, that the original controller was not designed
to compensate for disturbances in the resting positions at either end of
the valve motion. In order to achieve low steady state errors under these
circumstances, the original PD position controller needs to be extended to
a PID controller. The I term is used to reduce steady state error and the
PD terms are defined in the same fashion as in the original design.
Figure 7.6 shows simulation results that indicate that the system is able
to cope with significant disturbance forces. The additional energy required
by the actuation system is very close to the energy dissipated in friction.
We conclude, that the efficiency of the actuation system in the presence of
disturbances is still close to optimal.
The experimental system used in this thesis uses a very high resolution op-
tical encoder. In a real engine, this is not a suitable sensor, because it is
too expensive and cannot withstand the harsh environmental conditions. A
hall sensor is a more likely candidate, and it would usually have a resolu-
66
7.6. Experimental Results
tion somewhere between 7 and 10 bits. At the high sampling rates used
for this system, the low resolution would lead to considerable quantization
error, that would consequently introduce large noise in the position signal.
For eliminating measurement noise we will now compare a low pass filter
and a Kalman filter. The low pass filter is a simple backward difference
implementation of a continuous first order lag with a break frequency that
is ten times higher than the position loop bandwidth. The Kalman filter
implementation and the selection of the covariances for the disturbances
and the measurement follow the guidelines shown in Chapter 6. Figure 7.7
shows the systems response using an 7 bit sensor with 1 bit measurement
noise and different feedback strategies. When the sensor signal and a di-
rect backwards differentiator are used, the system is unstable. When the
sensor and the backwards differentiator are combined with low pass filters
with bandwidths at 10 times the bandwidth of the position controller, the
oscillations are reduced but not eliminated. Only the Kalman filter, is able
to provide a stable response for the low resolution sensor.
The reason for the difference in control performance can be found by
comparing position and velocity errors for the different feedback signals.
Figure 7.8 then shows this comparison. As expected, the position error of
the Kalman filter is slightly smaller than the actual sensor signal. However,
the low pass filter demonstrates considerable phase lag which manifests itself
in a position error proportional to velocity. The velocity error of the direct
backward difference differentiator is the same size as the full scale velocity
measurement. This error is reduced considerably by both the low pass filter
and the Kalman filter. However, the latter shows less than half the noise
levels of the low pass filter.
In summary, for a low resolution sensor system, the Kalman filter is the
only approach investigated here that guarantees sufficient robustness.
67
7.6. Experimental Results
ing performance is quite good. However, due to the flexibility in the linkage
between valve and motor, mechanical vibrations are induced. This cause
was determined by comparing the standard operation to a setup with an
equivalent inertia mounted rigidly to the motor. With the rigid inertia,
no noticeable vibration was observed. This leads to the conclusion that
the rather rudimentary linkage implementation shown here needs to be re-
designed to provide less flexibility in future setups. The velocity plot shows
that the mechanical vibrations due to the flexible link are primarily excited
during the high jerk periods at the beginning and at the end of the motion.
Consequently, the seating velocity at the end of the valve travel is rather
high. It should be noted that the transition time chosen here would be used
during high rpm of the combustion engine. However, the low valve seating
velocity of 0.05m/s is only required during engine idle, when the transition
times are much longer and the valve lift can be reduced significantly. Under
such circumstances, the FFVA system will provide valve seating velocities
well below the desired value.
Phase current, total current, and energy loss are plotted in Figure 7.10.
The motor draws approximately 100A during acceleration and 60A during
deceleration. The unsymmetrical behavior and the large current oscilla-
tions during accelerations are again attributed to the flexible linkage sys-
tem, which has unsymmetrical damping elements for opening and closing.
However, the energy consumption of 1.35J/transition in the experiment
corresponds well to the analytical predictions in Chapter 4. Table 7.3 shows
68
7.6. Experimental Results
69
7.6. Experimental Results
70
7.6. Experimental Results
71
7.6. Experimental Results
Figure 7.7: Simulation of transition response with a 7 bit sensor and various
filtering strategies
75
7.6. Experimental Results
76
7.6. Experimental Results
77
7.6. Experimental Results
78
Chapter 8
Conclusion
8.1 Summary
In this thesis, a Fully Flexible Valve Actuation system for the intake valves
of automotive internal combustion engines is demonstrated.
After reviewing the need for electronic valve actuation and current imple-
mentations of such systems, a direct drive system is proposed that provides
six main advantages:
79
8.2. Suggestions for Future Works
netic portion, and the control algorithms of the actuation system. A design
methodology for an actuation system based on stock motors is proposed
first. It provides guidelines how to select a good motor, how to design the
mechanical layout, and how to design an energy efficient control strategy. In
the course of this work, analytical expressions were developed to predict the
transition performance and the energy efficiency of the actuation system. In
addition, a lumped parameter model was developed to support the analyt-
ical performance predictions. The lumped parameter model was also used
to study the influence of varying system parameters, the ability to reject
disturbances, and effects of sensors with low resolution. Finally a test bed
was constructed that was used to validate the transition performance. These
investigations show that the FFVA system based on stock motors perform
well compared to other systems found in the literature.
To further improve the performance of the FFVA system, an analytical
design methodology for modifying the stock motors is then proposed. An
FEM model is used to validate the motor optimization methodology and
the lumped parameter model is applied again in order to predict the system
performance. This study shows that the modified motor should provide
superior energy efficiency without compromising transition speed.
80
8.2. Suggestions for Future Works
81
Bibliography
82
Chapter 8. Bibliography
[9] M. Theobald and B. L. and.R. Henry, “Control of engine load via elec-
tromagnetic valve actuators,” SAE Tech., vol. 103, pp. 1323–1334, 1994.
[16] A. Nix, “The future of 42v vehicle systems,” Vehicle Systems Integra-
tion in the Wired World, pp. 167–176, 2001.
[17] H. T. and H. M., “Development of the variable valve timing and lift
(vtec) engine for the honda nsx,” SAE transactions, vol. 100, no. 3, pp.
1–7, 1991.
83
Chapter 8. Bibliography
84
Chapter 8. Bibliography
85
Chapter 8. Bibliography
86