Semiactive Cab Suspension Control for Semitruck Applications Florin M. Marcu Dissertation submitted to the Faculty of the Virginia Polytechnic Institute and State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Mechanical Engineering Mehdi Ahmadian, Chair Steve C. Southward, Co-Chair John B. Ferris Stefan B. Jansson Corina Sandu April 3, 2009 Blacksburg, Virginia Keywords: Truck cab suspension, Magneto-Rheological, Skyhook, Semiactive, Hierarchical control Copyright 2009, Florin M. Marcu
169
Embed
Semiactive Cab Suspension Control for Volvo Truck Applications
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Semiactive Cab Suspension Control for Semitruck Applications
Florin M. Marcu
Dissertation submitted to the Faculty of the
Virginia Polytechnic Institute and State University
in partial fulfillment of the requirements for the degree of
Semiactive Cab Suspension Control for Semitruck Applications
Florin M. Marcu
ABSTRACT
Truck drivers are exposed to vibrations all day as a part of their work. In addition torepetitive motion injuries the constant vibrations add to the fatigue of the driver whichin turn can have safety implications. The goal of this research is to lower the vibrationsan occupant of a class 8 semitruck cab sleeper is exposed to by improving the ride quality.Unlike prior research in the area of ride comfort that target the chassis or seat suspension, thiswork focuses on the cab suspension. The current standard in cab suspensions is comprisedof some type of spring and passive damper mechanism. Ride improvements can most easilybe accomplished by replacing the stock passive dampers with some type of controllabledamper; in this case Magneto-Rheological (MR) dampers. MR dampers can change dampingcharacteristics in real time, while behaving like a passive damper in their OFF state. Thismeans that in case of a failure to the power supply, the dampers still retain their functionalityand can provide some level of damping. Additionally, MR dampers can be packaged suchthat they do not require any redesign of mounting bracketry on the cab or the frame, theiruse as a retrofitable device. The damper controller is based on the skyhook control policypioneered by Karnopp et al. in the 1970s. A variation on skyhook control is chosen calledno-jerk skyhook control. A controller called Hierarchical SemiActive Control (HSAC) isdesigned and implemented to allow the no-jerk skyhook controller to adapt to the roadconditions. It also incorporates an endstop controller to better handle the limited rattlespace of the cab suspension. The development and initial testing of the controller prototypeis done in simulation using a model of the cab and its suspension. The model is derivedfrom first principles using bond graph modeling. The controller is implemented in Simulinkto ease the transition to hardware testing. The realtime prototype controller is tested on aclass 8 semitruck in a lab environment using dSPACE and road input at the rear axles. Thelaboratory results are verified on the road in a series of road tests on a test truck. The roadtests showed a need for HSAC controller. The HSAC is implemented on the test truck ina final prototype system. The test results with this system show significant improvementsover the stock passive suspension, especially when dealing with transient excitations. Theoverall research results presented show that significant ride improvements can be achievedfrom a semiactive cab suspension.
Acknowledgments
I would like to dedicate this work to my parents, Georgeta and Mircea Marcu, without whose
support, encouragement, and personal sacrifice I would not be where I am today. They gave
up a comfortable life at the peak of their careers, and left their homeland to provide me with
the freedom and the opportunities they never had.
I would like to thank the love of my life, Amber, who came into my life when I least expected
it and filled it with meaning. She put things in perspective and provides the balance in my
life.
I would also like to thank all the faculty, staff, and students at the Center for Vehicle Systems
& Safety who have provided both their technical expertise to help me complete such a big
project, and a fun work environment which made my time there seem much shorter than it
This document will provide a detailed description of the project “Semiactive Cab Suspension
Control for Semitruck Applications”. This chapter will provide a short narrative summary
of the work and a list of contributions to the body of knowledge.
The purpose of this study is to improve the ride quality of semitrucks through the use of
semiactive suspensions. The primary focus of the investigation is the truck cab suspension.
The current cab suspension setup employed by Volvo Trucks North America (VTNA) is
studied at Center For Vehicle Systems & Safety (CVeSS) and a prototype system is designed,
implemented, and evaluated through simulations and dynamic testing in both a controlled
laboratory environment and on the road. The significance of this work lies in the “system
level” approach to the problem where modeling and simulation is not considered sufficient.
Not only did this work develop a model of a controllable cab suspension but, unlike similar
simulation studies of both primary and secondary suspensions in other literature, it also
1
CHAPTER 1. INTRODUCTION
provides a prototype cab suspension and controller, complete with road tested validation
and analysis.
To successfully achieve the goal of this study, a dynamic model of the cab and its suspension
is developed. The model is used to perform simulation studies to aid with the control
development. To validate the dynamic model a truck is instrumented with sensors and a
series of laboratory tests are conducted using a Volvo VN770 semitruck available at CVeSS.
The bond graph modeling approach is used for developing the dynamic model, resulting
in state space equations. A bond graph is a graphical representation of a dynamic system
that shows the energy flow through the system [36]. The energy flow is described in terms
of two generic power variables, effort and flow. These generic variables have more specific
interpretations depending on the system type. For example, in the mechanical domain
“effort” is equivalent to force or torque, and flow is equivalent to linear or angular velocity.
Similarly, in the electrical domain, effort is voltage and flow is current. Because of the generic
nature of bond graphs, they can easily be used to model complex systems spanning multiple
energy domains. In addition to its interdisciplinary advantages, the bond graph approach
provides an algorithmic and relatively “fool proof” method for deriving state space equations
for multi-domain systems. Bond graph modeling is suitable for modeling large systems
with many states, including Ordinary Differential Equation (ODE)s, Partial Differential
Equation (PDE)s, and combinations of both. It yields a complete state space mathematical
model with a minimal number of states [36]. The bond graph modeling approach also allows
for easy addition and removal of components from the system model. The bond graphs will
prove useful because in the need to transform the passive suspension modeled early in the
study into a semiactive model later in the study.
Due to the large number of components in a semitruck suspension, many of which include
non-linear characteristics, a number of simplifying assumptions are necessary. A parameter
2
CHAPTER 1. INTRODUCTION
optimization algorithm is used to compensate for these assumptions and to bring the dynamic
model closer to the test measurements in the lab. This optimization uses a cost function to
indicate how close the simulated response is to the measured response of the test rig. The
inputs to both systems are the same and after optimizing eleven parameters the output of
the simulated system is found to closely match the measured output during lab testing.
Magneto-Rheological (MR) dampers are installed and tested with constant (non-varying)
current, to ensure that the MR dampers in passive mode can perform as well as the stock
dampers. MR dampers are chosen to replace the stock passive dampers due to their con-
trollability, potential for improved performance and their robustness [40]. MR technology
works on the idea that by suspending iron particles in a carrier fluid, one can change the
damping characteristics of the damper by applying a magnetic field to the fluid. In the
presence of a magnetic field, the yield stress of the fluid increases, allowing the ability to
adjust the damping force between a minimum and maximum amount in a nearly continuous
manner [7]. The tests are conducted with the MR dampers in their “off” and “on” state
and it is found that the MR dampers outperform the stock dampers by providing higher
damping force in the full on state and lower damping force in the off state.
After the cab model is completed and validated, the controller development begins. A con-
tinuous skyhook policy is developed and tested. Skyhook control is selected due to its proven
superior performance over other common control strategies [51]. A no-jerk skyhook policy is
also implemented and tested. It is found that the no-jerk skyhook control outperforms the
skyhook controller in terms of ride comfort. No-jerk control provides a smoother ride due to
its built in attenuation function that smooths out the transition between the high and low
damping forces [7].
The advantage of using skyhook, or a variation thereof, is that the algorithm is computa-
tionally efficient. The algorithm simply compares two signals, decides if the damper should
3
CHAPTER 1. INTRODUCTION1.1. CONTRIBUTIONS
be turned on or off. If the decision is made to turn the damper on, the amount of damping
generated is proportional to the of the absolute velocity signal.
Finally, the controllers are tested both in the laboratory using sinusoidal and random inputs,
and on the road using a predetermined route around Blacksburg, VA consisting of highway,
interstate and city driving situations with a number of common driving conditions such as
left turn, right turn, road bumps, exit ramps, gear shifting, stopping and idling.
The initial road tests with different variants of skyhook control indicated situations where
no-jerk skyhook control proved insufficient for significantly improving the ride as compared
to the stock suspension. Therefore a Hierarchical SemiActive Control (HSAC) is developded
so that it can adjust the no-jerk controller in real time. HSAC consists of three control
hierarchies. The top level is a type of endstop control that is designed to keep the suspension
from crashing into the mechanical endstops. The middle level is the algorithm that selects
and adjusts the skyhook gain in the lowest level. The lowest level is comprised of a no-jerk
skyhook controller. The HSAC controller which is implemented on the test truck provides
a better ride in the sleeper portion of the cab than other suspension configurations that are
tested during road tests.
1.1 Contributions
The primary contributions of this research are:
• A modular cab dynamic model that includes a controllable suspension and can be used
for cab suspension development.
• A novel Hierarchical Semiactive Control method that can be readily used for cab
suspensions and possibly seat suspensions.
4
CHAPTER 1. INTRODUCTION1.1. CONTRIBUTIONS
• A comprehensive implementation of semiactive cab suspension for a two-point sus-
pended cab of the type that is commonly used in North America.
• A complete set of test data on the effect of semiactive cab suspensions that extends
the analytical and numerical results that are available in the open literature.
• An easily retrofitable turn-key prototype semiactive cab suspension system for semitrucks.
5
Chapter 2
Background and Literature Review
This chapter provides the background information related to the topics of this research. The
topics discussed in more detail are cab suspensions, bond graph modeling, MR technology,
skyhook control, and hierarchical control.
2.1 Cab Suspensions
The cab suspension is what connects the truck cab to the truck frame. Cab suspensions
emerged from the need for vibration isolation between the cab and the rest of the truck. In
the early 1970s, Crosby noted that due to the high location of the driver inside the truck cab,
high fore-aft motion can occur despite relatively small pitch angles of the truck itself [22].
In a study conducted in 1973, Van Deusen noted that significant improvements to the ride
quality of heavy trucks can be achieved by softening the primary suspension of the truck [53].
6
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.1. CAB SUSPENSIONS
On the other hand, doing that can negatively impact the roll stiffness of the vehicle and
generate large variations in handling due to loading conditions. Poor handling manifests
itself especially in tractor-trailer combinations as forces from the trailer are transfered to the
tractor through the fifth wheel inducing a pitching motion of the truck [22]. Implementing
a cab suspension will allow a stiff primary suspension yet still keep the vibration levels in
the cab at a comfortable level.
2.1.1 Cab Isolation
The first types of cab isolators were simple rubber mounts. Although they do a good job
of dealing with high frequency, low amplitude vibrations coming from the engine, they do
a poor job of handling low frequency, high amplitude inputs. In essence, the cab is still
susceptible to the undesirable low-frequency (less than 6Hz.) fore-aft motions induced by
the truck pitching [33]. Further improvements to cab suspension lead first to the introduction
of steel followed later by air springs with shock absorbers. This new configuration allowed
for significant relative motion between the cab and the truck frame that greatly improved
the ride quality by lowering the accelerations in the cab.
2.1.2 Truck Cab Types
This section will describe the different truck cab configurations and explain why a large
portion of the related literature is primarily focused on the Cab Over Engine (COE) truck
configuration. This work is primarily targeted at conventional heavy trucks which are the
most common trucks on the market at this time. Portions of the work could easily be applied
to COE trucks but will not be discussed at length at this time. It is, however, important to
have an understanding of the evolution of semitruck cabs to better grasp the needs of the
7
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.1. CAB SUSPENSIONS
Figure 2.1: Illustration of federal truck size regulation as of 2004. Note that the federalregulation merely imposes a minimum trailer length that all states must allow without con-sidering the length of the tractor itself. [5]
trucking industry today.
There are a number of different cab suspension designs depending on the type of cab and the
market it was designed for. The two prevailing cab types are the conventional cab and the
COE cab. The COE cab style was pioneered by Mack Trucks in 1905 [2] but became very
popular for heavy truck applications in the United States in the 1970s primarily due to length
limitations on heavy truck sets which imposed a maximum overall length on tractor-trailer
sets of 55 ft. [43, 44]. The shorter cab allowed for the cargo area to be longer while staying
within the legislated maximum length. These laws have since been relaxed to not include the
tow vehicle. The most recent Federal Highway Administration regulations state “A State
may not impose an overall length limit on a truck tractor pulling a single semitrailer or a
limit on the distance between the axles of such a truck tractor. A truck tractor is defined as a
non-cargo-carrying power unit used in combination with a semitrailer.” Since the truck itself
no longer counts toward the overall length of the vehicle (see Figure 2.1), the conventional
truck cabs have yet again taken over the long-haul heavy truck market [5]. The COE trucks
still have some significant advantages such as ease of maintenance due to unrestricted access
to the engine and transmission when the cab is tilted forward, better maneuverability due
8
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.1. CAB SUSPENSIONS
to generally shorter wheel bases, and greater driver visibility. Thus the COE cabs are still
popular for medium sized trucks designed for delivery and city use. For long-haul, highway
use the conventional cab is preferred due to its higher comfort, lower noise (because the cab
is not directly over the engine) and better crash worthiness (COE trucks have very limited
front crumple zone).
2.1.3 Importance of Cab Frame Dynamics
The reason it is important to clearly specify the type of cab being studied is due to the
influence of the truck frame dynamics on the ride characteristics of the cab. As the cab
mounting systems are different for different types of cabs, it is important to know what type
of cab one is dealing with.
All current heavy trucks are built on a truck frame. This frame is the backbone of the truck
and is the one component that connects all other parts of the vehicle [31]. The truck frame,
which is a long steel c-channel spanning the entire length of the vehicle, has its own dynamics
mainly caused by its first bending and torsional flexural modes. As periodic loads are applied
at various points on the frame (such as at the suspension mounting points), the frame can
begin to oscillate. Empirically, it has been found that the first “beaming” mode lies in the
range of 6–9 Hz. for a loaded truck [33]. The first beaming mode of the frame has nodes near
the front and rear end of the truck and large vertical displacement near the middle of the
truck. Because of the geometry and mounting locations of COE and conventional cabs, the
location of the front node can be used to improve the ride inside the cab while simplifying
the cab suspension.
Ideally, one would use an independent suspension at each corner of the cab, but Flower [32]
showed that by strategically placing either the rear mounting point of the COE cabs or
9
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.1. CAB SUSPENSIONS
the front mounts of the conventional cab on this node, one can simplify the cab suspension
by replacing a full-fledged suspension with simple rubber mounts leading to only a small
increase in cab vibration. This was confirmed by the work of Gillespie [33]. The combination
of rubber mounts and complete suspension is particularly common in the US market where
the customer desire for high cab suspension roll stiffness and increased road feel exceeds the
desire for high comfort. In other parts of the world (Europe, Japan) drivers are willing to
accept more cab motion in exchange for lower vibrations [33,34,41].
2.1.4 Controllable Truck Cab Suspensions
The notion of a controllable suspension is relatively new in the truck cab suspension field.
Although air spring suspensions with load leveling valves provide adjustability to varying
load, they are not designed to provide real-time control of the cab dynamics [33]. The latter
requires much faster response time than the few seconds that it takes for a load leveling
system to react to the cab dynamics. All production trucks currently use a passive cab
suspension to provide isolation from the remainder of the truck. There are two researchers
that have started looking at novel ways of improving the ride of the cabs through using more
modern damper designs and various control algorithms.
One of the major contributors to this field is Mohamed M. ElMadany. He has done exten-
sive simulation work describing both fully active and semiactive cab suspension systems and
comparing their performance with passive systems [26–30]. ElMadany performed a simu-
lation study in 1988 where he tested a fully active cab suspension with a linear stochastic
optimal controller with great success [29]. In one of his papers on this topic, ElMadany
established that semiactive suspensions can yield superior vibration isolation compared to
passive suspensions with the only penalty being a slight increase in cab displacement [30].
10
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.1. CAB SUSPENSIONS
To reach these results, ElMadany used the Hooke and Jeeves Pattern search algorithm to
solve the non-linear system problem and to find the elements of the controller gain matrix.
Most of his work on the subject took place during the 1980s and 90s, just before the advent
the first truly practical semiactive solution, the MR dampers. To this day, there still are
no practical, fully active solutions available for vehicle suspension applications. This may
explain why most of ElMadany’s work remained in the simulation world.
Around the same time ElMadany was working with controllable truck cabs, Chew performed
an interesting simulation study of a variety of cab mounts which included semiactive mounts
using skyhook control [20]. He found that continuous skyhook control can be successfully
used to improve the ride of both 4-point and 2-point cab suspensions. In his simulations,
he discovered that a continuous skyhook controller can perform comparably to a fully active
system.
As noted above, however, neither ElMadany nor Chew have ventured beyond the simulation
stage, into real world implementation and road testing.
Tsujiuchi et al. eveloped a semiactive suspension for an agricultural tractor that they tested
with great success in simulation, but yet again not in practice. [52].
A number of other studies have been performed using fully active control using relatively
complicated actuator systems. These have been implemented with good results. None have
gone into production due to the inherent reliability issues and failure modes related to fully
active suspensions.
Hiromatsu et al. developed a fully active suspension that used an electric motor to control the
motion of the cab [35]. The results were promising with relatively low power requirements
(<100W). Nakano et al. took this work one step further and developed a self-powered electric
suspension [42]. It uses a capacitor and an algorithm that controls a number of relays which
11
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.1. CAB SUSPENSIONS
change the flow of current depending on whether the suspension can produce energy or needs
to consume energy. They described a method to strike a balance between consumed and
generated power and found that there is a clear trade-off inherent to this balance.
The closest thing to a real world implementation of a controllable cab suspension is mentioned
in a patent by Catanzarite that describes a system very similar to this work [17]. Catanzarite
proposes using MR dampers and a host of sensors measuring everything from throttle, brake
and steering input to cab accelerations, displacements and roll. These measurements are
combined in one controller that weighs everything and calculates a control signal to be
sent to the dampers. The major difference between the work presented in this document
and Catanazarite’s work is that the work presented is using two independent controllers,
one for each damper, that are far less complicated than what Catanzarite is proposing. In
addition, this work describes the entire process of developing and testing the controllable
cab suspension
2.1.5 Truck Cab Suspension State-of-the-Art
Based on the literature discussed in the previous sections, the current state-of-the-art in
truck cab suspensions for conventional on-highway trucks on the US. market is a set of
rubber mounts at the front of the cab at or near the front frame beaming node combined
with an air spring and damper suspension near the back of the cab. The air spring suspension
usually incorporates a load leveling system to keep the suspension natural frequency constant
despite changes in loading conditions.
12
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.2. BOND GRAPH MODELING
2.2 Bond Graph Modeling
Bond graph modeling is a straight-forward way of describing a system by graphically captur-
ing the power flow through the system. The founding work was introduced by Henry Paynter
in 1959 [47] and has since been developed further by Dean Karnopp, Ronald Rosenberg, and
Donald Margolis into a more powerful technique [36–38,49]. Bond graph modeling gives the
user a more easy way of finding the equations of motion of a dynamic system. The beauty
of the bond graph modeling procedure is that it guarantees a set of equations of motion
that contain the minimum number of states necessary to describe a particular system. Bond
graph modeling also provides a systematic “turn the crank” procedure for generating the
equations of motion. It uses a universal graphical notation that allows it to be cross disci-
plinary. Thus, it is an excellent tool for modeling mechatronic systems and other systems
involving components from multiple energy domains.
The structure of a bond graph is composed of bonds and nodes. The bonds describe how flow
and effort travels through the system. The nodes contain information on the energy sinks,
sources and storage devices of the system in addition to operations that can be performed
on the flow of power. There are a number of good summary papers [9] and books [36]
on the topic that go into great detail with examples on how to use bond graph modeling.
Another useful resource and repository of bond-graph-related information is the website
http://www.bondgraph.info/.
2.3 Magneto-Rheological Technology
Magneto-Rheological technology came about in the late 1940s when it was developed by
Jacob Rabinow [48]. He was granted a patent in 1954 on a “Magnetic Fluid Shock Absorber.”
13
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.3. MAGNETO-RHEOLOGICAL TECHNOLOGY
Although MR technology has been around for so long, it has been largely unknown, and
especially Rabinow’s contributions have been overlooked partly due to Willis Winslow’s
work on Electro-Rheological (ER) fluids which included a discussion on MR fluids [55,56].
There is little mention of MR technology in the literature from the early days until the
1980s and 1990s when Lord Corporation took a new interest in the technology and started
to commercially develop and produce MR fluid and devices under the leadership of David
Carlson. Lord Corporation has a number of patents related to MR technology [11–14] and
has been able to market MR devices for numerous applications such as seat suspensions,
motor mounts, and devices to provide resistance in exercise machines. It can safely be
said that Lord Corporation is currently the leader in MR research and production and
their contribution to the field has reached many global markets, most notably the passenger
transportation industyr.
Another important contributor to the development and implementation of MR technology
is Mehdi Ahmadian who has performed and supervised a number of projects of significance
to MR technology in general and to heavy truck applications in particular. In one of his
most interesting papers Ahmadian gives a detailed description of the isolation properties
of MR dampers [6]. In the late 1990s, Ahmadian and his student Angela Carter worked
to successfully improve roll stability of heavy vehicles by using MR suspensions and fuzzy
logic control [15]. A few years later Ahmadian and his student David Simon studied the
effects of MR dampers on the primary suspension of semitrucks. They found that the
benefits are greatest from equipping the front axle with MR dampers. Yet again, the list
of contributions is too long to mention, but perhaps the greatest contribution Ahmadian
made to the field of MR was by never being satisfied with just theoretical evaluations and
simulation results. Most of his work was extended into the real world with actual product
development and testing. Additionally, not being affiliated with a particular corporation
14
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.4. SKYHOOK CONTROL
allowed him to publish his results extensively, which has been of great benefit to the overall
body of knowledge.
2.4 Skyhook Control
Skyhook control is a control policy that tries to emulate the behavior of a dynamic system
where the sprung mass is somehow connected to an inertial reference frame in the sky through
a damper called a skyhook damper. In theory, this is a great idea since the purpose of a
suspension connected to a fixed point of reference is to minimize the absolute vibrations.
Unfortunately, it is very difficult to connect mobile devices such as vehicles to an inertial
reference. Instead, an active or a semiactive device can be inserted between the sprung mass
and the unsprung mass to try to emulate the forces generated by the imagined skyhook
damper.
The idea of skyhook damping was pioneered by Karnopp et al. in the early 1970s [40]. Since
then, a number of variations on the original skyhook control have appeared.
One major trend was to drift away from skyhook control into fuzzy logic. Numerous works
in the 1990s applied fuzzy logic based on lessons learned from skyhook to control semiactive
suspensions and showed that it was a good alternative [15,21] both in theory and in practice.
Others chose to model the behavior of a system with skyhook dampers and then try to use
other control techniques to follow that behavior. Sammier et al. compared skyhook with a
nonlinear H∞ controller and were able to get better results from H∞. As they admitted in
their paper, it was a rather complex solution to the problem [50].
One highly effective, yet simple, variation was proposed by Ahmadian, Southward et al.
and is called no-jerk skyhook control [7]. It uses an attenuation function to smooth out the
15
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.5. HIERARCHICAL CONTROL
transition between damper high and low states which alleviates the problem of jerk entering
the system.
Literature shows that numerous approaches have been taken to controlling semiactive dampers.
Nearly all have been shown to work better than the passive damper, but none have stood
the test of time like semiactive skyhook. Due to its simplicity and elegance, skyhook control
has essentially become the benchmark for all other semiactive control methods.
2.5 Hierarchical Control
This section will describe a few ways in which hierarchical control has been used in past
suspension designs. As the works cited below show, most of the hierarchical control work
relates to some type of higher level controller that coordinates the efforts of controlled actu-
ators acting at various parts of the vehicle [23, 54]. This enables a hierarchical controller to
achieve a better performance than local controllers that act independently.
In the late 1990s der Hagopian et al. proposed a two level hierarchical controller for a
fully active suspension for off-road military applications. The top level decides on a global
control strategy based on the overall pitch and ground clearance of the vehicle and passes
the decisions on to the local controllers that are responsible for each bogie assembly [23].
Around 2005 Dong et al. proposed a Human-Simulation Intelligent Control (HSIC) with
three levels to deal with the non-linearity and time delay characteristics of MR suspension
systems. This was only evaluated in simulation. The lowest level is the control strategy
selected to control the MR dampers. The second level makes adjustments to the parameters
16
CHAPTER 2. BACKGROUND AND LITERATURE REVIEW2.5. HIERARCHICAL CONTROL
in the lowest level, and the third level, labeled the task adjustment level, linearizes the
nonlinear behavior of the MR damper and compensates for any controller delay that could
cause instability in a system [24].
17
Chapter 3
Modeling
In this section the modeling approach and the resulting state space model will be described.
Insight will be provided on how the parameter optimization was performed and how the
model was validated.
The modeling task can be accomplished using two main approaches. One is the unstructured
model approach. This approach uses mathematical tools to look at a known input and a
measured output and derives a relationship between the two that can be used as a model for
the system to be modeled. This method does not use any physical parameters and is great
for use with systems that do not change their composition. If, however, a change is made
to the system (such as changing a spring or damper) the previously derived model must be
rederived. The unstructured method generates a “black box” model that is not suitable for
this application.
This study seeks to generate a physics-based model that can be used even after changes are
18
CHAPTER 3. MODELING3.1. MODELING STRATEGY AND SIMPLIFYING ASSUMPTIONS
performed to the system (for example if a component is exchanged for a different component).
Because of this, the structured model approach was chosen. The structured model approach
derives a model from first principles and is based on real-life parameters.
To derive the model, the bond graph approach was used. A bond graph is a graphical
representation of a dynamic system that shows how power flows through the system. It uses
a universal nomenclature and the generic power variables flow and effort [1, 36].
3.1 Modeling Strategy and Simplifying Assumptions
Before the modeling could begin, it was necessary to decide on what exactly needed to
be modeled to accurately describe the cab and its suspension. The model itself is needed
to speed up the controller design process by facilitating controller design in a controlled
simulation environment prior to real-world implementation. This allows for quick testing of
many scenarios without the complications of lab experiment design and setup. It has been
shown that relatively simple truck models can yield reasonably good results [53]. Therefore,
the decision was made to simplify the model as much as possible without compromising the
usefulness of the model. Once the model is developed and validated, controller design and
testing can proceed at a rapid pace before final implementation and testing on an actual
truck.
A semitruck cab is isolated from the frame through a rear suspension consisting of springs and
dampers and two front bushings, as depicted in Figure 3.1. The front acts much like a hinge
that allows the cab to pivot about the horizontal axis in pitch. The front bushings provide
a limited amount of vibration isolation, although they are mainly designed to maintain the
connection between the cab and frame, and to some extent provide a limited amount of
controlled motion between the two.
19
CHAPTER 3. MODELING3.1. MODELING STRATEGY AND SIMPLIFYING ASSUMPTIONS
Figure 3.1: Schematic of cab with suspension and inputs.
In the rear, the truck has a set of two air springs and two hydraulic dampers that work to
restrict the vertical pivoting motion of the cab. In addition to these springs and dampers, a
panhard rod connected to a torsional spring is used to limit the lateral motion of the cab. The
panhard rod provides the lateral strength needed for crash worthiness. Early in the model
development it was decided to neglect the influence of the torsional spring as it has little
effect on the ride quality of the truck. This assumption was validated by Volvo engineers
who confirmed that ride quality is mainly influenced by the vertical and pitch motion of the
cab [41]. Thus the model does not include the dynamics of the torsional spring.
In the rear, the model includes the two air springs and two dampers, as shown in Figure 3.1.
The dampers on the truck are not vertical, but for the sake of the model the dampers are
20
CHAPTER 3. MODELING3.1. MODELING STRATEGY AND SIMPLIFYING ASSUMPTIONS
Figure 3.2: Schematic of the simplifying assumptions used when modeling the front of thecab.
assumed to be vertical, as the lateral force contributions by the damper does not play a role
in vertical or pitch motion of the cab.
In the front, the bushings are modeled as a relatively stiff vertical spring and damper on each
side. The location of the springs coincides with the location of the dampers, since in reality
the front bushings exhibit both stiffness and damping. In addition, the front is modeled as
connected to the ground through the previously mentioned springs and dampers, as depicted
in Figure 3.2. This is a simplifying assumption based on the fact that the front mounts are
located on or near the truck frame beaming node, and the interest is in isolating the relative
motion between the frame and the cab. There are also limitations on the lab equipment
which do not allow accurate measurement of the inputs from the truck frame to the front
of the cab. As will be seen later on, the parameter optimization algorithm takes this into
21
CHAPTER 3. MODELING3.2. BOND GRAPH MODEL
account and the response of the system is not significantly affected.
The cab itself is modeled as a rigid box with three Degrees of Freedom (DOFs); a rotational
Degree of Freedom (DOF) around X (roll), a rotational DOF around Y (pitch) and a vertical
displacement DOF in the Z direction (heave).
In order to model the rear inputs to the system, the rigid cross beam is modeled to transfer
the motion from the Linear Voltage Differential Transformer (LVDT) inputs at the truck
frame to the cab suspension. The beam is assumed to be massless, which is a reasonably
valid assumption since the beam itself is much smaller than the rest of the truck (the beam
only weighs around 20 lb, which is negligible compared to the cab’s weight of 3000 lb).
This beam is modeled to receive vibration inputs from the truck frame at two points and to
transmit them onto the cab suspension. The beam has both a heave and a roll component.
3.2 Bond Graph Model
Now that the system and its simplifying assumptions have been described, the mathematical
model can be derived. In order to reduce the possibility of errors in the model, the cab
and its suspension were divided into three subsystems: the cab subsystem, the suspension
subsystem, and the cross beam subsystem. The beauty of the bond graph approach is
that multiple subsystems from different physical domains can easily be connected together
which greatly simplifies the troubleshooting of the model and the extraction of the equations
later on. The bond graphs for the three subsystems can be seen in Figures 3.3 and 3.4.
The element labels in the bond graphs can be cross referenced with Figure 3.1. For all the
transformer elements, there are coefficients that correspond to distances to the various spring
and damper components. All the distances are measured from the CG with “l” indicating
a length along the length of the truck (x-axis) and “w” indicating a width along the width
22
CHAPTER 3. MODELING3.2. BOND GRAPH MODEL
Figure 3.3: Cab subsystem bond graph.
of the truck (y axis). The smaller illustrations on the left indicate the sign conventions used
when deriving the equations of motion.
The derivation of the equations of motion is relatively straightforward and involves starting
from an energy storage/dissipation device (spring or damper) and following the various
branches of the bond graph until the origin of the energy is completely traced. The end
result is a set of first order differential equations that comprise the state space model for the
various subsystems that can be combined into a global state space system [36].
The schematic representation of the state space system can be seen in Figure 3.5. The
disturbances are the input velocities from the road, transmitted through the truck frame to
the cross beam and onto the cab. The input velocities are easily measured during lab testing
on the actual truck by placing two LVDTs on the floor directly under the cross beam and
23
CHAPTER 3. MODELING3.2. BOND GRAPH MODEL
Figure 3.4: Suspension and beam subsystem bondgraph.
24
CHAPTER 3. MODELING3.2. BOND GRAPH MODEL
Figure 3.5: Schematic representation of the state space system.
attaching them to the truck frame on both the right and the left frame member. The control
inputs are the currents supplied to the two MR dampers to provide the most suitable damping
force for isolating the cab from the road disturbances. The control inputs act directly on
the cab suspension. The outputs of the system are displacement and acceleration outputs
at various locations on the cab. A detailed schematic of where the inputs and outputs are
located on the truck are shown in Figure 3.6.
3.2.1 Kinematic Equations
This section contains the kinematic equations at all the points of interest on the cab. These
points are the mounting locations of the front cab bushings and the rear springs and dampers.
The equations are summarized in Table 3.1 and represent the velocities (x, y, and z compo-
nent) at the specified locations with respect to the velocity at the Center of Gravity (CG)
of the truck cab and the roll (θx), pitch (θy) and yaw (θz) angular velocity. The governing
assumption for these equations is the small angle approximation. This is a valid assumption
25
CHAPTER 3. MODELING3.2. BOND GRAPH MODEL
Figure 3.6: Locations of all the sensors on the truck
The parameter optimization code is built around the Matlab “fmincon” function that uses
an iterative approach to minimize a cost function by altering a set of system parameters. The
user inputs the system model, the cost function, and the parameters including a valid range
for the parameters. The function iterates until it finds a minimum for the cost function, as
schematically represented in Figure 3.7. It is important to note that the cost function may
not be as smooth as depicted and the outcome of the optimization does depend on the initial
values of the parameters.
The parameters optimized include the cab mass, the moments of inertia of the cab around x
34
CHAPTER 3. MODELING3.3. PARAMETER OPTIMIZATION
Figure 3.7: Illustration of a cost function. It is noteworthy that the actual cost function maynot be as smooth as depicted.
and y (Jx and Jy), the cab’s front spring and damper coefficients, and the cab’s rear spring
and damper coefficients. For this particular cab, the nominal values along with a set of
best-fit values resulting from the parameter optimization process are included in Table 3.2.
Table 3.2: Nominal and Best Fit values of optimization parameters
Parameter Nominal Value Best Fit Value
Cab Mass 1534kg. 1702kg.Jx 4560kg −m2 4560kg −m2
Jy 4080kg −m2 4600kg −m2
Left Front Spring 39100N/m 39100N/mLeft Front Damping 5000N/(m/s) 4993N/(m/s)Right Front Spring 39100N/m 39100N/mRight Front Damping 5000N/(m/s) 4993N/(m/s)
Left Rear Spring 33000N/m 32967N/mLeft Rear Damping 8000N/(m/s) 5972N/(m/s)Right Rear Spring 33000N/m 32957N/mRight Rear Damping 8000N/(m/s) 5630N/(m/s)
After the best fit values of the uncertain parameters are found, the model output using the
optimized parameters is compared with the experimental test results from the lab. Figure 3.8
shows a comparison of one sensor output (in this case, LLV) from the optimized model with
35
CHAPTER 3. MODELING3.3. PARAMETER OPTIMIZATION
Figure 3.8: Comparison between optimized model output and lab measured output of onesensor.
the equivalent sensor output as measured in the lab.
The optimized model yields a reasonably good approximation of the real system response
and is sufficient for use in developing control algorithms. The response of the system model
closely resembles the response of the truck cab as tested in the lab and the conclusion is
drawn that the model is suitable for controller development in simulation. It is worth noting
that this model will only be used for initial controller development and troubleshooting in
a controlled simulation environment. The controllers developed in simulation will then be
implemented on the truck and throughly tuned and tested both in the lab and on the road.
The road testing is what will be used to evaluate the performance of the various controllers
and therefore the fidelity of the cab suspension model is not as critical as it may be if the
controller evaluation was done in simulation.
The validation of the model against the actual cab completes the cab model. Next, a set of
controllers will be designed.
36
Chapter 4
Initial Vehicle Preparation and
Testing
This chapter will discuss the details of the test setup and the preliminary testing performed
on a semitruck (a Class 8 Volvo VN 770) using the dynamic test rig at CVeSS.
4.1 Truck Modifications
Although the primary truck suspension is not tested in this study, it is still important to
the test setup because the actuation of the truck occurs through the primary suspension.
Initially, the suspension was configured as seen in Figure 4.1 where the aft drive axle was
raised off the ground and the fore drive axle was being actuated using two hydraulic actuators
connected to both ends of the axle.
37
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.1. TRUCK MODIFICATIONS
(a) Rear axle on bogie. (b) Front axle on bogie.
Figure 4.1: Original truck suspension setup before modifications.
This setup was used in previous projects and testing performed for VTNA [46], attempting
to use this method caused several drawbacks that lead to the necessity to modify the test
setup. When actuating the truck in roll, the truck would start swaying in a yaw motion.
This phenomenon introduced undesired dynamics into the truck and was attributed to the
fact that the truck was essentially unconstrained in the lateral direction. Another problem
was that the measured outputs on the cab were showing very low outputs, due to isolation
resulting from the primary suspension at higher frequencies. Since the system of interest
for this study is the cab suspension, it is important to get clean and ample vertical motion
transmitted from the truck frame to the cab in order to achieve good measurable signals at
the sensors on the cab. It was determined that by transmitting the motion from the actuators
through the primary suspension, the dynamics in the range between 4 and 8 Hz is filtered
out, resulting in actuator input not efficiently reaching the cab suspension.Thus the primary
suspension was modified to enable more vibration energy to reach the cab suspension. The
modifications are shown in Figure 4.2.
The suspension modifications included re-enabling the rear drive axle and installing wheels
on both sides. Now, the static weight is carried by the tires and the front drive axle is used
38
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.1. TRUCK MODIFICATIONS
(a) Before modification. (b) After modification.
Figure 4.2: Truck suspension setup before and after modification. Green color indicates partof the actuation system and red color indicates immobile components.
solely for actuating the truck frame and all components attached to it. The use of tires in
a loaded configuration provides the lateral stiffness needed to prevent the truck rear end
from swaying. Additionally, the air springs on the front drive axle are replaced with a rigid
member that enables the hydraulic actuators to directly shake the frame.
Several other changes had to be made to the test vehicle before dynamic testing could begin.
In order to have greater control over the stiffness of the suspension and the ride height of the
vehicle, the load-leveling system and accompanying plumbing had to be modified. In their
stock configuration, the two driver-side and passenger-side air springs were linked together.
The plumbing for the air suspension was rerouted so that the front drive axle was on an
independent air supply and the rear drive axle was on another. This allowed the ride height
of each drive axle to be modified independent of the other, and allowed for controlling the
weight ratio that is carried by the tires on the rear drive axle and the actuators on the front
drive axle. By removing the connection between the truck’s air supply and the drive axles,
changes to the ride height could be made without affecting other truck components on the
air supply, such as the cab suspension air springs and the vehicle brake system.
39
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.1. TRUCK MODIFICATIONS
Figure 4.3: Air dryer inlet bypass hose.
Figure 4.4: External air hookup.
40
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.1. TRUCK MODIFICATIONS
The main air supply hose from the engine-mounted air compressor was disconnected from the
truck’s air supply system. The supply hose was originally connected to the air dryer located
under the cab as shown in Figure 4.3. The hose was replaced with new tubing connected to
an external valve and fitting, as shown in Figure 4.4, for providing air through the shop air
supply.
Figure 4.5: Weight stack simulating the trailer load.
To simulate the weight of a loaded truck, the vehicle is loaded by placing steel weights on
the back of the truck where the fifth wheel normally resides. A one-inch thick steel plate
(called the fifth wheel adapter plate) was mounted to the frame so that the truck can accept
the weight plates for dynamic testing. To simulate the trailer vertical load, a stack of plates
was added to the vehicle frame on top of the fifth wheel adapter plate. This additional
weight consisted of a stack of 34 metal plates, 350 pounds (159 kg) each, which was placed
on locator pins and strapped to the fifth wheel plate as shown in Figure 4.5. With these
modifications, the weight on the rear axle was approximately 15,000 pounds (6800 kg). Most
41
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.2. VEHICLE ACTUATION HARDWARE
Figure 4.6: Sketch of dynamic actuation setup before air spring removal.
of the vertical static load is held up by the rear axle and transferred to the ground through
the truck tires.
4.2 Vehicle Actuation Hardware
The attachment of the actuators to the truck is achieved through the front drive axle as
illustrated in Figures 4.6 and 4.7. One of the two actuators and its mounting system is also
shown in Figure 4.7. By adjusting the air pressure in the rear drive axle air springs, each ac-
tuator supported approximately 3000 lb (1361 kg) at rest and thus they were not overloaded
during testing (the maximum weight each actuator can support is 5000lb (2270kg)).
4.3 Actuation and Data Acquisition
Hydraulic actuators are used to excite the truck at different modes. During this excitation,
accelerometers and LVDTs are used to record the response of the system to various inputs.
42
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.3. ACTUATION AND DATA ACQUISITION
Figure 4.7: Picture of dynamic actuation setup before air spring removal.
LVDTs are used to measure the displacement of the truck chassis at various locations and
the accelerometers are used to measure the accelerations of the truck components at multiple
key locations such as the frame, rear cab and B-post. The following sections will describe
the test instrumentation used to excite the system, the instrumentation used to record the
dynamic response of the system, and a discussion of the instrument locations.
4.3.1 Truck Actuation
A computer is used to control the actuation of the suspension during dynamic testing. For
each test, the input, a band limited random noise signal, is generated in Simulink and then
downloaded into dSPACE Control Desk. dSPACE provides the user interface for controlling
the tests and recording the data. The dSPACE output is used as an external input to the
MTS 458.20 hydraulic controller, shown in Figure 4.8. The controller regulates the motion
43
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.3. ACTUATION AND DATA ACQUISITION
of two MTS Model 248.03 hydraulic actuators, which are mounted at each end of the axle.
Each actuator is controlled independently, exciting the axle in both heave and roll. During
the heave tests, the actuators move in phase, whereas in roll tests they move 180 degrees out
of phase with each other. The physical setup for actuation of the suspension during dynamic
testing, including the hydraulic actuator and attachments, is shown in Figure 4.6.
4.3.2 Data Acquisition
As mentioned in the previous subsection, the data collection and recording of all measure-
ments is performed with dSPACE. A dSPACE AutoBox DS 2201 data acquisition unit
records data from all the measurement devices onto a laptop computer. A sampling rate of
1000 Hz is selected to ensure a high enough sampling frequency to lower the risk of aliasing.
The highest test frequency in any of the input signals is 15 Hz. After the analog signals
are converted to digital signals at the higher sampling frequency, they are passed through
a second order low-pass Butterworth filter with a break frequency of 15 Hz prior to being
down-sampled at a rate of 200 samples per second. The measurements performed in dynamic
testing include accelerations at various points on the truck, relative displacements between
the cab and frame, relative displacement between the frame and ground, and the actuator
Figure 4.8: MTS 458.20 hydraulic controller
44
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.3. ACTUATION AND DATA ACQUISITION
Figure 4.9: PCB Model U352C65 accelerometer
load and displacement. Accelerations are measured by PCB Model U352C65 accelerometers,
such as the one shown in Figure 4.9. These accelerometers have a sensitivity of 100mV/g
and are capable of measuring accelerations up to ±50g with a frequency range of 0.5Hz–
10kHz. These are good, rugged accelerometers for automotive use, but their sensitivity is
lower than is desired for use inside the cab. A PCB ICP 16 channel signal conditioner is used
with a 100x gain to power the accelerometers and increase the resolution of their output.
The accelerometers plug into the signal conditioner, which has BNC outputs that go to the
AutoBox.
Displacements and velocities are measured with the Unimeasure VP510-10 LVDTs as shown
in Figure 4.13. The VP510-10 is both a displacement and velocity transducer capable of
measuring displacement up to 10 in, with a maximum wire acceleration of 50 g. For these
tests it is used to measure both, depending on the location of the sensor.
4.3.3 Instrument Locations
On the test vehicles, acceleration, velocity and position are recorded at several points, includ-
ing on the frame, at the back of the cab, and at the driver-side B-post. This instrumentation
45
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.3. ACTUATION AND DATA ACQUISITION
arrangement is used throughout the tests. The location of each sensor is shown can be seen
in Figure 4.10.
Figure 4.10: Locations of all the sensors on the truck
LVDTs are used to measure vertical displacement and velocity. In addition, accelerometers
are used to measure acceleration in three different directions. The accelerometers can be used
in two configurations, uni-axial and tri-axial. The tri-axial accelerometers are configured
using three unidirectional accelerometers mounted inside of an enclosed box, oriented along
the vertical, lateral and fore-aft directions. The tri-axial accelerometers are used in two
locations on the truck as seen in Figure 4.10. The Accelerometer Rear Lower (ARL) and
Accelerometer Inside (AI) are both tri-axial accelerometers. ARL is mounted on the outside
of the cab on the center line of the truck at the bottom of the cab (see Figure 4.11). AI is
mounted on the bulk head near the B-pillar inside the cab at head level for the driver (see
46
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.4. MR DAMPER IMPLEMENTATION
Figure 4.12).
In addition to the tri-axial accelerometers, one unidirectional accelerometer Accelerometer
Rear Upper (y-direction) (ARU) is placed high up on the cab oriented in the lateral direction
for the sole purpose of studying the roll response of the cab. Two LVDTs are used on the
frame and two LVDTs are used across the cab suspension. The cab LVDTs LLV and RLV
are mounted on the frame rails under the cab just to the front of the cab suspension as seen
in Figure 4.13. They measure the cab suspension’s vertical displacement. The frame LVDTs
Left Rear Input Linear Voltage Transformer (LRin) and Right Rear Input Linear Voltage
Transformer (RRin), mounted directly under the cab suspension cross member, measure
velocity inputs to the truck frame relative to the floor as shown in Figure 4.14. These
measurements are used as inputs to the model of the cab for validation purposes.
4.4 MR Damper Implementation
Before designing the control system, it is important to determine the type of controllable
device to be used. At the beginning of the study it was relatively clear that some type
of MR device was to be used in place of the stock cab dampers. However, until it was
proved that the MR dampers would perform at least as well as the stock passive dampers,
the study could not proceed. Thus, as soon as the modeling and optimization tasks were
completed, the focus of the study shifted to finding a suitable MR damper. Two aspects are
important in selecting a suitable damper, packaging and force/velocity performance. The
packaging aspect is important because ideally a damper would be found that fits in the
stock damper location with minimal modification. The highest control effect due to MR
dampers occurs when the off state force is smaller and the on state force is larger than the
that of the stock dampers. This guarantees that the range of damping force of the device
47
CHAPTER 4. INITIAL VEHICLE PREPARATION AND TESTING4.4. MR DAMPER IMPLEMENTATION
Figure 5.11: Simulation results using semiactive skyhook control, 1 second snapshot, 4Hzsine input
Figure 5.12: Simulation results using semiactive skyhook control, 1 second snapshot, randominput
67
CHAPTER 5. CONTROLLER DEVELOPMENT5.5. IMPLEMENTATION OF CONTROL POLICIES FOR LAB TESTING
discontinuity visible in both the sine and the random response. For this case where the
control forces are low it’s not a big problem, but when the forces get higher, these peaks can
become an issue that needs to be resolved.
Finally, as can be seen in Figures 5.13 and 5.14, the semiactive no-jerk skyhook policy
completely removes the discontinuities without any detrimental effect on the cab response.
This is great news and clearly shows that the project can move on from the simulation stage
to lab testing on the actual truck.
5.5 Implementation of Control Policies for Lab Testing
Once the simulation confirmed that there was benefit to continuing this research, certain
changes had to be made to the code to convert the simulation setup for dSPACE use.
Figure 5.15 displays the modified high-level Simulink controller. The same color scheme has
been used here as in Figure 5.5 to facilitate the transition between the two.
The red block illustrates the source of excitation. For lab testing, it is necessary to create
independent signals that are sent to each actuator that will excite the truck. The actuators
have the pet names “Ethel” and “Fred” and shake the left and right side of the truck
respectively.
The details of the lab setup were discussed in Chapter 4 and will not be discussed at length
here. Figure 5.16 is included to show a schematic view of the actuator configuration (Fred
showed). Note how the bulk of the truck weight is supported by the rear axle and the
dynamic actuation is transferred directly to the truck frame through the disabled front drive
axle.
From the red block in Figure 5.15, the actuation signal is sent directly to the output of the
68
CHAPTER 5. CONTROLLER DEVELOPMENT5.5. IMPLEMENTATION OF CONTROL POLICIES FOR LAB TESTING
Figure 5.13: Simulation results using no-jerk semiactive skyhook control, 1 second snapshot,4Hz sine input
Figure 5.14: Simulation results using no-jerk semiactive skyhook control, 1 second snapshot,random input
69
CHAPTER 5. CONTROLLER DEVELOPMENT5.5. IMPLEMENTATION OF CONTROL POLICIES FOR LAB TESTING
Figure 5.15: Simulink diagram of the lab testing controller, high level view
dSPACE system, here illustrated by the yellow block. As before, the green block represents
the sensor measurements on the cab. The outputs from that block remain two absolute
accelerations and two relative velocities. These go straight into the light blue controller
block. This block is unchanged from the simulation as far as its control logic is concerned.
A few features are added to allow for on-the-fly changing between the control policies once
in the lab, but no changes whatsoever were introduced to the controllers. These features are
not relevant for the operation of the controllers and are thus omitted from this document.
70
CHAPTER 5. CONTROLLER DEVELOPMENT5.5. IMPLEMENTATION OF CONTROL POLICIES FOR LAB TESTING
Figure 5.16: Truck actuation system. Green color indicates actuator attachment link andred color indicates rigid component that transfers the input excitation to the truck frame.
71
Chapter 6
Laboratory Testing
As with the simulation results presented earlier, a series of plots will be presented for the
response of the truck in response to various excitations from laboratory testing. The data
represents test results for four MR damper settings: damper off state, damper on state with
1A current, semi-active skyhook, and semi-active no-jerk skyhook. For the damper off state,
the dampers provide a maximum force of approximately 100N. In their on state, the dampers
yield a maximum force of roughly 1200N. The MR damper test using the on state was
included because it resembles the stock dampers force. The MR damper on state was selected
so that it can mimic the stock damper as closely as possible. It is important to note that
a direct A-B force comparison cannot be done between passive and MR dampers. Passive
dampers use mechanical valving that restricts fluid flow according to the relative velocity
across the damper, thereby providing a nonlinearly tuned force-velocity characteristic. The
MR dampers in their on state behave almost symmetrically in extension and compression.
72
CHAPTER 6. LABORATORY TESTING
Figure 6.1: Time trace of laboratory truck testing results with 3Hz sine excitation. Top:Cab acceleration. Center: Cab suspension relative displacement. Bottom: Cab suspensionrelative velocity.
In contrast to the simulation data, the lab testing results are presented in an overlapped
manner such that the measured data of all the control policies are superimposed to illustrate
the differences between them. Since each line in the graph indicates a different test run, it
was impossible to exactly time the data to superimpose nicely. A best effort has been made
to manually match the timing of the data. Also, the displayed graphs were selected to be at
or around the natural frequency to show how effective the skyhook control policy can be at
lowering the cab acceleration. Figure 6.1 shows the response of the cab when excited below
its natural frequency of 3.5Hz. As is well known, at frequencies below the natural frequency
of the suspended body (in this case, the truck cab), the two ends of the dampers move in
phase with each other resulting in little relative motion (displacement and velocity) across
the damper. As such, no significant difference is seen in Figure 6.1 between the damper
configurations tested.
73
CHAPTER 6. LABORATORY TESTING
Figure 6.2: Time trace of laboratory truck testing results with 3.5Hz sine excitation. Top:Cab acceleration. Center: Cab suspension relative displacement. Bottom: Cab suspensionrelative velocity.
Figure 6.3: Time trace of laboratory truck testing results with 4Hz sine excitation. Top:Cab acceleration. Center: Cab suspension relative displacement. Bottom: Cab suspensionrelative velocity.
74
CHAPTER 6. LABORATORY TESTING
Figure 6.4: Time trace of laboratory truck testing results with 4.5Hz sine excitation. Top:Cab acceleration. Center: Cab suspension relative displacement. Bottom: Cab suspensionrelative velocity.
Figure 6.2, Figure 6.3 and Figure 6.4 show the test results nearest the cab natural frequency.
It can be seen that the skyhook test runs show a 40-50% decrease in acceleration with only a
moderate increase in relative displacement. Figure 6.5 is included to show that at frequencies
above and away from the cab natural frequency the controlled cab behaves much like the
uncontrolled cab.
Figure 6.6 illustrates the behavior of the cab with a 15Hz band limited random excitation.
Since random signals were used, not every test used the exact same signal. The signals were
generated in the same way, and the statistical content of the signals was the same, but the
generated signals were not identical. As such, it was not possible to overlap the results and
a direct comparison is not possible. To account for this, a 10 second snapshot is displayed to
better show the general trend of each configuration. As the acceleration plots show, there is
little difference between test scenarios for the most part, because the random input includes
75
CHAPTER 6. LABORATORY TESTING
Figure 6.5: Time trace of laboratory truck testing results with 7Hz sine excitation. Top:Cab acceleration. Center: Cab suspension relative displacement. Bottom: Cab suspensionrelative velocity.
Figure 6.6: Time trace of laboratory truck testing results with bandlimited white noiseexcitation. Top: Cab acceleration. Center: Cab suspension relative displacement. Bottom:Cab suspension relative velocity.
76
CHAPTER 6. LABORATORY TESTING
a broad spectrum of energy up to 15 Hz. (the cut-off frequency). As the previous graphs have
illustrated, a significant improvement can only be noted near the natural frequency. Thus,
the skyhook controllers will result in similar performance to the stock dampers. However,
near the natural frequency the skyhook controllers will greatly improve the ride in the cab.
Such a scenario can be seen in the green line near 4, 7–8 and 9–10 seconds. These correspond
to large spikes in cab response to which the stiff damper (on state) results in large vibration
transmission. Neither the semi-active nor the no-jerk configuration shows spikes of that
magnitude in the 10-second snapshot illustrated in Figure 6.6. This can be confirmed by
the subjective observations done while performing the tests. Sitting in the cab while it was
shaking allowed the author to actually feel the difference between the control policies. As ride
quality is both subjective and objective, the way the ride felt cannot be discounted. Based on
personal observation, the difference between either of the skyhook policies and the on state
is significant enough to be felt by a casual rider in the cab. The skyhook policies provide a
much smoother ride. When comparing the skyhook policies with one another, the no-jerk
policy is noticeably smoother than the semi-active policy. The smoothest ride is delivered
by the OFF state but it comes at the price of large swaying motion. This large motion is
disturbing to the point that after a few minutes it becomes physically uncomfortable and
motion sickness sets in. Based on these observations, the conclusion is drawn that there is
merit to using the skyhook controller for this application and it is time to move the project
to actual road testing.
77
Chapter 7
Building Block Controller Road
Testing
This chapter will describe the road testing that has been performed on the building block
controllers descirbed in Chapter 6. The goal of this testing was to ensure the test platform
(the truck, sensors and all data acquisition systems) is ready for road testing and to establish
a baseline with the stock truck cab suspension as well as with the controllable MR damper
suspension being controlled by the building block controllers.
7.1 Signal Conditioning Box
Upon the decision to commence road testing, a new data acquisition system was purchased
for this project. The new system was needed to replace the dSPACE AutoBox system
78
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.2. DESIGN OF EXPERIMENT
previously used. The old system was being used for other projects and in-lab experiments in
addition to this project. It would not be feasible to constantly interrupt the other projects
sharing the system. To allow road testing without interrupting other projects, a brand
new MicroAutobox was acquired. Although the MicroAutobox is in may ways similar and
equivalent to the AutoBox, there is one major difference between them; the input voltage
range of the Analog to Digital (AD) channels is different. For the Autobox the AD voltage
range is −10V to +10V; for the MicroAutobox it is 0V to 5V. Because all the sensors used are
bipolar, a signal conditioning box was necessary. This signal conditioning box takes the ±10
V signal coming from the sensors and shrinks it by a factor of 4 prior to offsetting it by 2.5 V
This allows the signal to fit in the 0-5 V range with 0 V at the sensors corresponding to 2.5 V
just past the signal conditioning box. All this can be accounted for in the controller software
by simply applying the correct gains and offsets. Figure 7.1 shows the electrical diagram of
one of the circuits inside the signal conditioning box. Sixteen identical circuits were built to
allow for sixteen channels of data to be recorded simultaneously. Figure 7.2 shows the input
and output signals of the signal conditioning circuit in Figure 7.1. The circuit was tested
to ensure that it will perform well over the range of frequencies of interest to this work.
Signals with frequencies as high as 50 Hz were run through the signal conditioning box and
it successfully processed the signal without distortion or lag.
7.2 Design of Experiment
Prior to commencing the road testing, a design of experiment is performed in order to
minimize the time with the driver while maximizing the number of tests performed. This
involves breaking up the necessary testing into two distinct categories: Functionality tests
and cab suspension evaluation tests. The functionality tests are tests that establish the
79
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.2. DESIGN OF EXPERIMENT
Figure 7.1: Electical diagram of the signal conditioning box circuit. Only one circuit shown,but sixteen identical circuits are inside the box to allow for sixteen channels of data.
Figure 7.2: Illustration of input and output voltages from the signal conditioning circuit
80
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.2. DESIGN OF EXPERIMENT
in Table 7.4, a higher current tends to yield lower suspension displacements than what is
allowed by the stock suspension. In addition, the left damper appears to provide less damping
than right damper. This is most likely caused by the kinematics of the cab suspension and
possibly by uneven occupant distribution within the cab.
91
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
7.4.2 Tests With bsky = 90000
Based on the lab testing performed it was found that bsky = 90000 and Vo = 100 yield a good
ride. Thus the first controller tests were performed using these values to establish if the real
world corresponds to the lab testing. The complete test run can be seen in Figures 7.9 and
7.10. The figures show the accelerations in three directions at the B-pillar and as can be
observed in Figure 7.9, the general trend is that the stock cab suspension performs better
than the MR suspension with both the semi-active skyhook and the no-jerk control (labeled
in the figures “sa” and “nj” respectively). This was noticed both subjectively when riding
in the truck and objectively when observing the control current output of the controller, as
shown at the bottom of Figure 7.11. The current stayed on for most of the time, indicating
that the controller is trying to generate too high of a force even when it is not needed.
Due to these observations, the bsky = 90000 tests were limited to this Vo value and a lower
bsky = 50000 was studied more closely.
92
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.9: Time trace of B-post acceleration for stock damper and controlled MR damperwith bsky = 90000; Top: fore-aft; Center: lateral; Bottom: vertical.
Figure 7.10: PSD plot of B-post acceleration for stock damper and controlled MR damperwith bsky = 90000; Top: fore-aft; Center: lateral; Bottom: vertical.
93
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.11: Time trace of vertical displacement at the back of the cab for stock damperand controlled MR damper with bsky = 90000; Top: left side; Center: right side; Bottom:control current.
94
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
7.4.3 Tests With bsky = 50000
As for the bsky = 90000 test, a plot of the complete run with Vo = 100 was studied first.
This showed that the stock suspension causes lower accelerations than the semiactive skyhook
suspension, but the no-jerk suspension is comparable to the stock one, and often outperforms
it. This can be observed easiest in the PSD plot in Figure 7.13.
These observations warranted a closer look at this bsky level. Two more runs were made
where the parameter Vo was changed to 10 and 1000. This was done to study the effects of
Vo on overall comfort.
As shown in Figures 7.14-7.17, higher Vo values appear to result in a harsher ride both at
lower frequencies (where it is most uncomfortable) and at higher frequencies, for instance
near 9 Hz, where the natural frequency of the truck’s exhaust stacks can be found. This
makes sense because a higher Vo value will react slower to the input excitation and thus
there will be less damping whenever it is needed and conversely too much damping where
it is not needed. Another explanation could be that due to the extremely gradual slope of
the attenuation function at high values of Vo, the amount of damping commanded by the
no-jerk controller is too slow to react to the road conditions and may be hurting more than
it is helping. The results from the figures are summarized in Table 7.5. The table clearly
shows that a higher Vo will yield higher acceleration and that the semiactive controller does
minimize the peaks but it can yield a higher RMS acceleration than the no-jerk controller
with low Vo.
For these tests, just like for the tests described in the previous sections, the accelerations
at the B-post contain significant amounts of vibration propagating from the truck frame
into the cab through the front cab mounts. Because the front of the cab is mounted using
bushings, the vibration isolation properties of the front cab mounting points are noticeably
95
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.12: Time trace of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 and Vo = 100; Top: fore-aft; Center: lateral; Bottom: vertical.
Figure 7.13: PSD plot of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 and Vo = 100; Top: fore-aft; Center: lateral; Bottom: vertical.
96
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.14: Time trace of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 and Vo = 10; Top: fore-aft; Center: lateral; Bottom: vertical.
Figure 7.15: PSD plot of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 and Vo = 10; Top: fore-aft; Center: lateral; Bottom: vertical.
97
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.16: Time trace of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 and Vo = 1000; Top: fore-aft; Center: lateral; Bottom: vertical.
Figure 7.17: PSD plot of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 and Vo = 1000; Top: fore-aft; Center: lateral; Bottom: vertical.
98
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Table 7.5: RMS and peak acceleration (in m/s2) at the B-post for bsky = 50000.
RMS Peak+ Peak-
X
Stock 0.18 2.24 -1.60Semiactive 0.20 1.63 -2.04NJ Vo = 10 0.19 1.74 -1.58NJ Vo = 100 0.19 1.80 -1.59NJ Vo = 1000 0.20 2.16 -1.73
Y
Stock 0.22 1.73 -2.23Semiactive 0.24 1.43 -1.84NJ Vo = 10 0.22 2.23 -2.22NJ Vo = 100 0.22 2.58 -2.69NJ Vo = 1000 0.22 1.41 -1.56
Z
Stock 0.29 2.84 -3.51Semiactive 0.30 2.48 -3.06NJ Vo = 10 0.29 2.66 -2.87NJ Vo = 100 0.29 2.74 -2.87NJ Vo = 1000 0.33 3.54 -4.37
worse than what the rear suspension can provide. Since only the rear suspension is controlled
and the goal is to improve the ride at the back of the cab, a closer look at the accelerations
in the rear of the cab are warranted. The vertical accelerations at the back of the cab are
illustrated in Figures 7.18-7.23 and summarized in Table 7.6. When studying the response
at the back of the cab, the influence of Vo is even clearer than at the B-post. The PSD plots
are especially useful and when comparing Figure 7.23 to Figure 7.19 it is easy to see the
benefits of a lower Vo.
As noted in Table 7.6, the no-jerk control with Vo = 10 provides the lowest overall accelera-
tion. It, however, appears to allow larger spikes than semiactive control without significant
increase in damper stroke as illustrated in Figures 7.24-7.26. The three figures are summa-
rized in Table 7.7. The results show that Vo has little influence on the relative displacement.
Although the stroke of the MotionMaster damper is observed to be comparable to the stock
dampers, the testing shows that the suspension is moving throughout the entire range of mo-
99
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.18: Time trace of vertical acceleration at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 and Vo = 10; Top: left side; Bottom: right side.
Figure 7.19: PSD plot of vertical acceleration at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 and Vo = 10; Top: left side; Bottom: right side.
100
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.20: Time trace of vertical acceleration at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 and Vo = 100; Top: left side; Bottom: right side.
Figure 7.21: PSD plot of vertical acceleration at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 and Vo = 100; Top: left side; Bottom: right side.
101
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.22: Time trace of vertical acceleration at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 and Vo = 1000; Top: left side; Bottom: right side.
Figure 7.23: PSD plot of vertical acceleration at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 and Vo = 1000; Top: left side; Bottom: right side.
102
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Table 7.6: RMS and peak acceleration (in m/s2) at the back of the cab for bsky = 50000.
RMS Peak+ Peak-
Left
Stock 0.50 4.66 -4.95Semiactive 0.50 3.72 -3.89NJ Vo = 10 0.45 4.66 -3.73NJ Vo = 100 0.48 3.98 -4.69NJ Vo = 1000 0.54 5.73 -5.72
Right
Stock 0.44 4.11 -4.28Semiactive 0.49 3.13 -4.75NJ Vo = 10 0.44 4.14 -3.83NJ Vo = 100 0.47 3.32 -4.91NJ Vo = 1000 0.54 4.42 -4.54
Figure 7.24: Time trace of vertical displacement at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 and Vo = 10; Top: left side; Center: right side;Bottom: control current.
103
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.25: Time trace of vertical displacement at the back of the cab for stock damperand controlled MR damper with bsky = 50000 and Vo = 100; Top: left side; Center: rightside; Bottom: control current.
104
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.26: Time trace of vertical displacement at the back of the cab for stock damperand controlled MR damper with bsky = 50000 and Vo = 1000; Top: left side; Center: rightside; Bottom: control current.
105
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
tion regardless of control policy. This observation clearly shows a need for a stroke-limiting
controller that will keep the damper from hitting the end stops.
In the following sections, the four events described in Section 7.4 will be studied in detail.
Table 7.7: RMS and peak relative displacement (in cm) over cab suspension for bsky = 50000.
RMS value Peak+ Peak-
Left
Stock 0.25 1.09 -1.75Semiactive 0.27 1.39 -1.91NJ Vo = 10 0.28 1.52 -1.93NJ Vo = 100 0.28 1.82 -2.02NJ Vo = 1000 0.28 1.90 -2.16
Right
Stock 0.36 2.31 -1.94Semiactive 0.38 2.57 -2.12NJ Vo = 10 0.39 2.86 -2.22NJ Vo = 100 0.38 2.75 -2.23NJ Vo = 1000 0.40 2.99 -2.26
106
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
7.4.4 Sharp Left Turn (25 mph)
The location of the turn described in this section is on Ramble Road when traveling from
Christiansburg toward the Virginia Tech airport. The road has a speed limit of 35 mph and
the turn is such that the driver has to slow down to about 25 mph to successfully negotiate
it.
Figure 7.27 shows the response of the cab at the B-post. The time trace indicates that the
acceleration amplitudes are smallest for the disturbance in all directions when using no-jerk
control. The lowest fore-aft and lateral accelerations are achieved when Vo = 1000 and
the largest when using the stock damper. For the vertical direction, the best performance
comes from the no-jerk controller with Vo = 10. It is interesting to note that the semiactive
controller is generally at least as good as the stock suspension.
The acceleration time trace in Figure 7.28 shows that the no-jerk controllers generally provide
a lower acceleration than both the stock suspension and the semiactive controller at the
disturbance. Everywhere else all the controllers appear to perform similarly.
The displacement plots in Figure 7.29 illustrate that the improvements shown in the accel-
eration plots mentioned earlier come without a large relative displacement penalty.
7.4.5 Sharp Right Turn (25 mph)
The location of the turn described in this section is on Ramble Road when traveling from
Christiansburg just past the Virginia Tech airport. The road has a speed limit of 35 mph and
the turn is such that the driver has to slow down to about 25 mph to successfully negotiate
it.
As for the left hand turn discussed earlier, the no-jerk controller outperforms the stock
107
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.27: Time trace of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 in response to sharp left hand turn at approximately 25 mph; Top: fore-aft;Center: lateral; Bottom: vertical.
Figure 7.28: Time trace of vertical acceleration at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 in response to sharp left hand turn at approximately25 mph; Top: left side; Center: right side; Bottom: control current.
108
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.29: Time trace of vertical displacement at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 in response to sharp left hand turn at approximately25 mph; Top: left side; Center: right side; Bottom: control current.
Figure 7.30: Time trace of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 in response to sharp right hand turn at approximately 25 mph.; Top:fore-aft; Center: lateral; Bottom: vertical.
109
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.31: Time trace of vertical acceleration at the back of the cab for stock damperand controlled MR damper with bsky = 50000 in response to sharp right hand turn atapproximately 25 mph; Top: left side; Center: right side; Bottom: control current.
Figure 7.32: Time trace of vertical displacement at the back of the cab for stock damperand controlled MR damper with bsky = 50000 in response to sharp right hand turn atapproximately 25 mph; Top: left side; Center: right side; Bottom: control current.
110
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
suspension in all directions. Figure 7.30 shows that a high value of Vo provides significantly
better roll stiffness with minimal impact on the other directions. This is similar to the result
for the left hand turn discussed earlier. It is noteworthy that the stock suspension had a
much higher lateral response than the controlled suspensions. The response of the no-jerk
controlled suspension with Vo = 100 in the fore-aft direction shows some unusual peaks in
the time series plot in Figure 7.30 that are not visible in the other test runs. These can
be attributed to late braking when entering the turn. The resulting forward pitching of the
cab can also be observed in Figure 7.32 as a vertical, equal and uniform displacement of the
cab with respect to the truck frame on both sides of the cab. Minor variations like this are
inevitable when performing road tests in traffic with interference from other motorist and
the inherent variations stemming from a human driver.
7.4.6 Road Bump (35 mph)
The location of the bump in the road studied in this section is on Ramble Road shortly after
the left hand turn described earlier. It is located in front of the entrance to the Virginia
Tech airport. The road has a speed limit of 35 mph and the bump is on a straight section
where the vehicle is traveling at the posted speed limit.
Figure 7.33 shows the response of the system when going over the bump in front of the
airport entrance. In this case it appears that there is no configuration that is clearly optimal
in terms of acceleration. Both the stock suspension and the no-jerk suspension appear to
yield similar results and the high Vo = 1000 value in particular appears to stand out. This
warrants a closer look at the accelerations at the back of the cab, shown in Figure 7.34.
From the time traces it is clear that the peak acceleration at the back of the cab is significantly
lower with all of the no-jerk controllers and Vo = 100 appears to yield the best results.
111
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.33: Time trace of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 in response to road bump at approximately 35 mph.; Top: fore-aft; Center:lateral; Bottom: vertical.
Figure 7.34: Time trace of vertical acceleration at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 in response to a bump at approximately 35 mph;Top: left side; Center: right side; Bottom: control current.
112
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.35: Time trace of vertical displacement at the back of the cab for stock damperand controlled MR damper with bsky = 50000 in response to a bump at approximately 35mph; Top: left side; Center: right side; Bottom: control current.
Figure 7.35 shows that this is accomplished with no more than 15 mm. of damper stroke.
113
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.36: Time trace of B-post acceleration for stock damper and controlled MR damperwith bsky = 50000 in response to road bump at approximately 55 mph; Top: fore-aft; Center:lateral; Bottom: vertical.
7.4.7 Road Bump (55 mph)
The location of the bump described in this section is on highway 460 in the eastbound
direction between Southgate Drive and the South Main Street exit. The road has a speed
limit of 55 mph and the bump is located where the vehicle is traveling in a straight line at
the posted speed limit.
In Figure 7.36 it is observed that the controllers perform very similar to the stock suspension.
The controllers appear to mimic the stock damper quite well. Unfortunately this means that
all the controllers allow the suspension to bottom out at the bump and yet again the need
for some type of stroke limiting control becomes apparent. It can also be observed that
the semiactive controller absorbs the bump better than the other controllers at highway
speeds. The explanation to this is that at highway speeds a controller without the no-
jerk attenuation function can respond faster to the road inputs and can thus provide more
114
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.4. TEST RESULTS
Figure 7.37: Time trace of vertical acceleration at the back of the cab for stock damper andcontrolled MR damper with bsky = 50000 in response to a bump at approximately 55 mph;Top: left side; Center: right side; Bottom: control current.
damping force quicker. In this case, this means that the cab suspensions impact with its
mechanical endstops is less violent.
Figure 7.37 shows that at highway speeds the quick response of the semiactive controller
can provide a better ride experience despite the potential for jerk. A moderate attenuation
such as that provided by Vo = 10 provides a comparable ride but at the expense of greater
relative displacement, as illustrated in Figure 7.38. There is concern that the low Vo allows
the suspension to get very close to the suspension bump stops. Endstop control may resolve
this.
115
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.5. SUMMARY OF RESULTS
Figure 7.38: Time trace of vertical displacement at the back of the cab for stock damperand controlled MR damper with bsky = 50000 in response to a bump at approximately 55mph; Top: left side; Center: right side; Bottom: control current.
7.5 Summary of Results
The observations made during the course of the road testing can be summarized in the
following list:
• bsky = 50000 is better suited for on-road driving than bsky = 90000.
• Vo = 100 provides the best performance most of the time although no one controller
configuration was identified as being superior in all situations.
• Pure semiactive control works best at higher speeds when dynamic jerks gets overshad-
owed by road noise.
• Low Vo generally provides lower cab accelerations at the cost of getting very close to
the bump stops.
116
CHAPTER 7. BUILDING BLOCK CONTROLLER ROAD TESTING7.5. SUMMARY OF RESULTS
• The stroke of the MotionMaster damper appears to be sufficient based on the tests
scenarios studied.
• The force capabilities of the MotionMaster damper appear to be sufficient and, with
proper control, can outperform the stock damper.
• All building block controllers that were considered in the study have at least one
strength over other configurations and the stock damper.
• Cab loading conditions can influence controller selection.
117
Chapter 8
Hierarchical Semiactive Control
Development
The work that has been completed has shown that no one controller, or no one controller
configuration provides the best performance. Much of the preliminary work was based on
the study performed by Y. Shen [51] that showed that it is very hard to improve on the
performance of the skyhook control policy in this application. Thus the various control
schemes selected were all variations on skyhook control. Several trends were discovered
and there was reason to be confident that combining or altering these controllers in an
intelligent manner would provide a higher level of comfort in the truck cab. To accomplish
this, a Hierarchical SemiActive Control (HSAC) scheme was developed that could provide a
structured approach to selecting the best possible controller configuration for the situation.
This section describes the development of the HSAC controller.
118
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.1. HIERARCHICAL CONTROL BACKGROUND
8.1 Hierarchical Control Background
The idea behind hierarchical control is not new. In fact, hierarchical control can be observed
in many complex systems in nature and in everyday life. The human body is a perfect
example from biology. A human being is composed of numerous body parts that all work
together to achieve a greater good than each part can achieve on its own. The highest level in
the hierarchy can be for example the wishes and desires of the individual, briefly summarized
in the functions of the brain. Lower levels can be the muscle groups, the sensory organs, the
digestive and pulmonary organs, etc.; all the way down to the cell level. Similarly, other man
made, complex systems make use of hierarchical controls. Governments, armies and large
businesses are all good examples [8]. Lately, hierarchical control has been studied extensively
in the area of unmanned systems where teams of vehicles must work together to complete a
task [18].
A good, in-depth example that helps illustrate some of the components of hierarchical control
is the study of a human being that wishes to stand up from being seated. The top level,
the brain, sends the signal to the rest of the body to stand up. This signal reaches the next
lower level in the hierarchy, the muscular system, which uses energy gathered by even lower
level systems such as the digestive system and the pulmonary system to commence the act
of standing up. Sensory systems, which gather the information about the surroundings and
processes it, return feedback signals to the brain which are being used to establish when
the goal has been achieved, and whether unexpected events are occurring that may require
a reaction. These tasks can be summarized in the following categories: decision making,
actuation, energy storage, energy production/conversion and sensing.
Teaching a system to perform all the necessary steps to complete all the tasks listed would be
highly complex if it were not for the decomposition of the global task into smaller subtasks
119
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.1. HIERARCHICAL CONTROL BACKGROUND
and the assigning of these subtasks to specialized subsystems. The specialized systems do
not have to know the big picture to complete their task and, in fact, can perform their
limited tasks more efficiently due to being able to specialize and focus on a simple, specific
task. The subsystems do require a higher level in the hierarchy that can observe a slightly
bigger picture and can make decisions on how to assign tasks based on what they observe
and the knowledge of the capabilities of the subsystems. Conversely, the systems higher up
in the hierarchy perform their functions better by not being bothered with the details of
each little task. Take, for example, the interaction between the muscles and the digestive
system. The muscles do not know how food is processed to create energy. All they know
is how to access the energy stored by the digestive system. Likewise, the digestive system
does not know how the muscles do their job of propelling the body. It does, however, know
how to prepare the food into usable energy. If the digestive system generates the energy in
a way accessible to the muscles, the muscles will be able to perform their tasks which helps
the digestive system gather more food. At the next higher level, the brain does not know
how the muscles propel the body. It does however know that in response to certain electrical
signals, the body will move [10].
This leads to the idea of calibration. The human body is not automatically capable of
performing all these tasks in an coordinated fashion. Therefore, children go through a
calibration phase in their infancy where they learn how to perform higher level tasks such as
walking, talking, riding a bicycle, etc. One could say that during this time, infants generate
a series of lookup tables that are stored locally in the subsystems. One such lookup table is
commonly known as “muscle memory” and is the reason why some skills, once learned, are
never forgotten. A few good examples are riding a bicycle, snow skiing, or tying shoe laces.
All are complicated tasks composed of many subsystems working in coordinated unison. It
is nearly impossible to describe the task completely to the point that an uninitiated person
120
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.2. HSAC INTRODUCTION
could master it. Yet with the help of calibration (commonly called practice), these tasks can
be learned. A similar method will be utilized to attempt to teach the truck suspension how
to provide a better ride.
8.2 HSAC Introduction
During the preliminary testing a number of events and driving situations were studied and
conclusions were drawn regarding which control scheme was observed to provide the best
results in each situation. By thoroughly studying these events, it is possible to gain a good
understanding of which controller performs the best for a variety of driving conditions. If
this information can be assembled into a decision process, all that remains to be done is to
create a higher-level control strategy that can identify the current conditions and select the
appropriate controller configuration.
The initial idea behind HSAC came from observing the behavior of the truck cab when
traveling at highway speed, for example as shown in Figure 7.37. It was observed that the
cab suspension would bottom out and it became apparent that some type of endstop control
would be necessary. HSAC was only going to incorporate endstop control as a higher priority
controller overlaid on top of one of the semiactive controllers discussed previously. A review
of the literature yielded the work of Dong et al. [24] which inspired the three level structure
of the proposed HSAC controller. Since the cab suspension system studied in this work relies
completely on a semiactive suspension, stability due to controller delay is not a problem.
Since MR dampers are dissipative control devices there is no risk of instability. Thus the idea
of a three level hierarchical controller described by Dong et al. can be modified to replace
the third level with endstop control.The three levels in the hierarchy of the HSAC controller
are illustrated in Figure 8.1.
121
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.3. ENDSTOP CONTROL
Figure 8.1: Conceptual sketch of HSAC.
The highest level is the endstop controller which will have a higher authority than all other
controllers. The mid level is comprised of an algorithm that configures the bsky parameter of
the controller in the lowest level in the hierarchy. The fundamental controller selected for the
lowest level was the no-jerk skyhook controller because of it showing the best performance
in the preliminary road tests described in Chapter 7.
8.3 Endstop Control
The endstop control designed is an algorithm that reads the relative displacement of the cab
suspension and outputs a control signal that ensures a smooth transition to the mechanical
endstops of the suspension. This removes the jolts that are measured as jerk and acceleration
spikes and provides a smoother ride while lowering the wear on the suspension components,
induced from endstop collisions.
Catanzarite et al. proposed in their patent a method for auto-calibration of controllable
122
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.3. ENDSTOP CONTROL
Figure 8.2: Sketch illustrating the endstop control range.
damper suspensions [16] that might at first glance seem like a feasible way of detecting the
cab suspension mechanical limits. Indeed, it would work if the suspension did not have the
load leveling sensor removed from the vicinity of the air springs. In this case it is better
to perform a static measurement and hard-code the allowable suspension rattle space into
the control algorithm. The mechanical endstops are not expected to vary over time, which
makes an automated endstop detection system an unnecessary complication.
The endstop algorithm is designed to activate when the suspension relative displacement is
15mm from the nominal ride height and damper control current saturation is achieved at
20mm of displacement. This is illustrated in Figure 8.2. When the endstop controller is
activated, the control signal takes the shape described by Equation 8.1 which is depicted in
Figure 8.3. It should be noted that the endstop controller is only activated when in Zone
2. Equation 8.1 does not provide adequate control values outside Zone 2. Therefore, the
endstop controller is deactivated in Zone 1.
123
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.3. ENDSTOP CONTROL
Figure 8.3: Plot of endstop control signal and the polynomial estimation.
Icontrol = Isat
(e
disp−endstophibufferhi + e
endstoplo−disp
bufferlo − 1
)or for this case
Icontrol = 2(e
disp−0.0150.005 + e
−0.015−disp0.005 − 1
) (8.1)
To try to generate the most effective code possible, a curve fit was performed to replace
the exponentials in the endstop control code with a polynomial function. This will make
the code much more efficient and allow it to run on less powerful systems while maintaining
the same functionality. The curve fit described by Equation 8.2 yields a similar response as
shown in Figure 8.3.
Icontrol = Isat
(8750 · disp2 − 2
)(8.2)
124
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
Figure 8.4: Simulink implementation of the endstop control algorithm.
The endstop controller implementation in Simulink is illustrated in Figure 8.4. A slight
modification has been implemented to ensure that the endstop control is only in effect when
the suspension is approaching the endstops. When the suspension is moving away from the
endstops the endstop controller shuts off to allow the suspension to go back to its nominal
ride height as quickly as possible. The saturation block ensures that the endstop controller
never commands more than 2 A of current, which is the limitation of the current generator.
A simulation of the endstop controller behavior can be seen in Figure 8.5
8.4 Controller Configuration Decision Process
The controller configuration decision process is what adjusts the nojerk low level controller
in response to road conditions. Based on the testing performed previously it was found
that a nojerk controller with Vo = 100 is consistently better than the alternatives. There
was, however, no consistently superior bsky value. Thus, it was decided to pursue a decision
125
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
Figure 8.5: Plot of endstop control simulation.
process that would adjust the bsky parameter in realtime. This can be done in a multitude
of ways, but since a significant amount of experimental data was available it was decided to
pursue a method that would take advantage of it. Thus it was decided to construct lookup
tables that can build on the experimental observations and would provide the appropriate
bsky value for each situation.
The next step was to develop a road condition detection algorithm that can make use of
empirical knowledge collected in the lookup tables. Because of the nature of on-road driving,
there are a number of situations that must be taken into account. Common situations
that are encountered are turning, negotiating bumps, and straight line driving. The road
condition detection algorithm must be able to detect all these situations. Since everything is
happening in real time, there is a need for a compromise between how quickly the algorithm
reacts and how much it recalls from the past. The algorithm looks at a period of time and
establishes how the cab has responded in the past few seconds and draws some conclusion
126
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
based on that. This is done by calculating the moving average over the past few seconds.
This allows for a statistical analysis of a period of time with the goal of attempting to predict
what will happen based on what just happened. This is suitable for driving on a road with
a consistent composition, but will not react well to sudden changes. One could argue that
the endstop controller is meant to handle any sudden changes. Indeed, that is the case.
But what if the sudden event is not large enough to trigger the endstop controller but large
enough to warrant a change in bsky?
To deal with this, a peak counter is implemented. The peak counter simply counts the
number of peaks over a set threshold within one second, which is selected because it allows
for easy correlation to the natural frequency calculation. For example, it is known that the
natural frequency of the cab and its suspension is approximately 4 Hz. Thus, four peaks
within a second correlates to an excitation at the natural frequency, which logically should
warrant a change in the damping to shift the damped natural frequency of the system. This
will avoid an undesirable large response.
8.4.1 Moving Average Calculation
Since a semitruck spends most of its time driving on a relatively smooth road in a straight
line it makes sense to primarily focus on providing the best possible ride in this situation.
Therefore, the most important part of the controller is the moving average portion. It selects
the damper current by looking back at the past five seconds and calculating two moving
averages; one for the positive and one for the negative relative displacement. The selection
of a five second window was not arbitrary. From observing initial laboratory and road test
results it was noticed that most transient effects are over within a few seconds. To ensure
that the moving average portion of the controller only reacts to changes in steady state
127
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
behavior, a time window of five seconds was selected. From the two moving averages, the
one yielding the highest control current is selected to control the dampers. This is unusual
because most often the RMS value is used [19,25,57].
There are several benefits to calculating two running means. Unlike the RMS value, this
method makes a better differentiation between a signal that has an offset and one that does
not. This is illustrated in Figure 8.6. The figure illustrates the difference between the 5
second moving RMS response and the 5 second maximum moving average response to what
could be a truck driving through an interstate interchange and then continuing on in a
straight line. Notice how the moving average calculation is generally lower, especially in the
straight line driving situation. Where there is a steady state offset, such as what would be
expected in a clover leaf interchange, the RMS and the maximum moving average are nearly
identical.
These differences best illustrate how the moving averages can be used to better distinguish
these two driving situations, allowing the controls designer to select the best possible con-
troller configuration for each situation.
The reason why this method is chosen over the RMS method is because the configuration of
the cab suspension and its load leveling system makes it prone to DC offsets in cab suspen-
sion relative displacement. This is especially prevalent in this cab suspension configuration
because the load leveling sensor is located at the center of the cab. This makes the load
leveling system insensitive to constant roll excitations where one side suspension is in com-
pression and the other side is in extension. Since the load leveling sensor is in the center of
the cab, it detects a ride level in between the two levels observed by the left and right side
suspensions. This manifests itself in a constant cab lean that could occur due to a sharp
turn, the crown of the road, uneven loading in the cab or even lateral wind loading which
all would be completely undetected by the load leveling system. This can cause each side
128
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
Figure 8.6: Plot illustrating the difference between moving RMS calculation and calculatingtwo moving averages and selecting the greater of the two.
of the cab to be dangerously close to the endstops with the load leveling system unable to
react. Therefore the damper control system needs to be able to identify and react to this
situation. This illustrates why it is important to avoid locating the relative displacement
sensor at the center of the cab. By locating the relative displacement sensors near the cab
damper locations one can collect much more accurate measurements for use in the various
control algorithms.
Let’s look at how a few scenarios may play out in Table 8.1. The table describes a few
common scenarios that a truck cab suspension may experience during normal operating
conditions. It is worth noting that since this portion of the controller has a relatively slow
response time, it can only successfully respond to changes in steady state behavior. There
are other controllers that can complement the moving average control scheme to provide fast
response to dynamic situations. When studying Table 8.1 it is useful to keep this in mind
129
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
Table 8.1: Different driving scenarios and their likelihood of endstop impact
ScenarioSuspension Motion Likelihood of
displacement Amplitude endstop impactLeft Right Left Right
1. Straight line driv-ing, smooth road
@ ride height @ ride height Low Low Low
2. Straight line driv-ing, rough road
@ ride height @ ride height High Medium Medium
3. Straight line driv-ing on crowned road
@ ride height < ride height Low Low Medium
4. Heavily loadedsleeper
<< ride height << ride height High High High
5. Lightly loadedsleeper
>> ride height >> ride height Low High High
6. Sharp right turn << ride height >> ride height Low High High
and to envision the moving average controller being paired up with the fast acting endstop
controller. This way the moving average controller can study the past few seconds of driving
in an attempt to anticipate the likelihood of the suspension impacting the endstops, and
adjust the damping accordingly. If the dynamic input is too large, the endstop controller
can take over and provide a smooth transition to the endstops instead of a sudden impact.
Table 8.2 shows how an RMS method and a moving average method may select the damping
for each situation based on observations made in Figure 8.6. Notice how the moving average
selection exactly matches the likelihood of endstop impact described in Table 8.1.
Figure 8.7 shows the Simulink implementation of the moving average algorithm. It should
be noted that the Simulink “Weighted Moving Average” block is used to calculate a simple
moving average over a 5 second time interval. The algorithm outputs an updated moving
average value every time step.
130
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
Table 8.2: Comparison between RMS and moving average damping selection in response todifferent driving scenarios based on observations made in Figure 8.6.
ScenarioDamping selection Damping selection based
based on RMS on moving averageLeft Right Left Right
1. Straight line driving, smooth road Medium Medium Low Low2. Straight line driving, rough road High High Medium Medium3. Straight line driving on crowned road Medium Medium Low Medium4. Heavily loaded sleeper High High High High5. Lightly loaded sleeper High High High High6. Sharp right turn High High High High
Figure 8.7: Simulink implementation of the moving average algorithm for calculating thepositive and negative moving averages.
131
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
Figure 8.8: Simulink implementation of the peak counter algorithm.
8.4.2 Peak Counter
The purpose of the peak counter is to fill the gap between the endstop controller and the
moving average algorithm. As described previously, the endstop controller takes precedent
and reacts instantly if a sudden event forces the cab suspension close to its endstops. The
moving average algorithm is, by comparison, a slow reacting algorithm that is mostly de-
signed to handle the monotonous excitation of straight ahead driving. This leaves a need
for something that can handle the situations in between. This is where the peak counting
algorithm shines. It counts the number of peaks in a one second time window and makes
adjustments to the bsky multiplier accordingly. The idea is to let the moving average algo-
rithm handle the smooth driving conditions and to prepare the suspension for a transition
to a rougher road by counting the peaks above a certain threshold that is lower than the
endstop controller threshold. For this application, the threshold is set at 10mm. As the peak
counter algorithm is relatively simple, it can be observed directly in Figure 8.8. The peaks
are counted as excursions above a certain threshold and the “Detect change” block ensures
that only the first time step above the threshold is counted. The counter is incremented
every time the signal is above the set threshold but as time passes the old values are flushed
out. Each peak is only remembered for one second.
132
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
8.4.3 Lookup Tables
The outputs from the moving average calculator and the peak counter are used as inputs for
two lookup tables which will produce multipliers to be multiplied by the nominal value which
was set to bsky = 50000. These tables were derived from studying the results in Section 7.
It was observed that the a higher bsky value will yield higher control forces which is not
very surprising. The testing showed that maintaining a higher bsky value for a prolonged
period of time will also increase cab accelerations. Since bsky = 50000 was found to work
well in general, the lookup tables were set up to default to a multiplier of 1 under nomindal
circumstances and to increase the multiplier value as the combination of amplitude and offset
of the relative displacement signal moves closer to the endstops. This ensures that higher
bsky values are only utilized when getting closer to the endstops and that as soon as the
system goes back to its nominal ride height, the multiplier defaults back to 1. The multiplier
lookup tables are displayed in Tables 8.3-8.4.
After the mean and peak counter multipliers have been found, they are multiplied together
with the nominal skyhook gain which in this case is bsky = 50000. The Simulink implemen-
tation of the lookup tables is illustrated in Figure 8.9. The results of the implementation can
be studied in the simulation illustrated in Figure 8.10. The graph shows a simulated relative
displacement signal designed to illustrate a transition from a smooth road to a rougher road
with twice the excitation amplitude. This transition occurs at the 20 second mark. The
figure shows how the bsky multiplier is influenced by the mean and peak counter algorithms
in response to the input signal.
Putting all the components together is illustrated in Figure 8.11. This shows how the no-
jerk skyhook control algorithm is combined with the bsky selection algorithm. The block
diagram in Figure 8.11 includes a selector switch that allows for directly comparing the
133
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
Figure 8.9: Simulink implementation of lookup tables and the product of the mean and peakmultiplier.
Table 8.3: bsky multiplier derived from moving average.
Mean relative displacement bsky mean multiplier[mm]
-13 20-6.5 5-1.3 11.3 16.5 513 20
Table 8.4: bsky multiplier derived from peak counter.
Number of peaks bsky peak multiplier
0 11 1.52 23 2.310 3
134
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
Figure 8.10: Simulation illustrating the moving average and the peak counter algorithmsand how they influence the bsky multiplier.
HSAC controller with the non-adaptive algorithms discussed earlier.
135
CHAPTER 8. HIERARCHICAL SEMIACTIVE CONTROL DEVELOPMENT8.4. CONTROLLER CONFIGURATION DECISION PROCESS
Figure 8.11: Simulink implementation when all the components of the HSAC algorithm arecombined.
136
Chapter 9
HSAC Road Testing
This chapter will describe the road test results from the testing performed with the HSAC
controller. The testing was performed on the same route described in Section 7.2.2 for
consistency.
9.1 Sharp Left Turn (25 mph)
As Figures 9.1-9.2 show the acceleration performance of the HSAC controller is very similar
to the stock suspension and the nojerk semiactive controller. The peaks have approximately
the same amplitude for all the test runs. The main difference can be observed in the relative
displacement where the no-jerk and HSAC controllers keep the suspension more centered in
its range, ie., closer to zero than the stock suspension.
137
CHAPTER 9. HSAC ROAD TESTING9.1. SHARP LEFT TURN (25 MPH)
Figure 9.1: Time trace of vertical acceleration at the back of the cab for stock damper, no-jerk and HSAC controlled MR damper in response to sharp left hand turn at approximately25 mph; Top: left side; Center: right side; Bottom: control current.
Figure 9.2: Time trace of vertical displacement at the back of the cab for stock damper, no-jerk and HSAC controlled MR damper in response to sharp left hand turn at approximately25 mph; Top: left side; Center: right side; Bottom: control current.
Figure 9.3: Time trace of vertical acceleration at the back of the cab for stock damper,no-jerk and HSAC controlled MR damper in response to road bump at approximately 35mph; Top: left side; Center: right side; Bottom: control current.
Figure 9.4: Time trace of vertical displacement at the back of the cab for stock damper,no-jerk and HSAC controlled MR damper in response to road bump at approximately 35mph; Top: left side; Center: right side; Bottom: control current.
Figure 9.5: Time trace of vertical acceleration at the back of the cab for stock damper,no-jerk and HSAC controlled MR damper in response to road bump at approximately 55mph; Top: left side; Center: right side; Bottom: control current.
Figure 9.6: Time trace of vertical displacement at the back of the cab for stock damper,no-jerk and HSAC controlled MR damper in response to road bump at approximately 55mph; Top: left side; Center: right side; Bottom: control current.
[3] Aurora bearing company website, 2009. http://www.aurorabearing.com/.
[4] Magnetostriction; principle, technology, and how it works, 2009.http://www.mtssensors.com/technology/how-magnetostriction-works/index.html.
[5] Federal Highway Administration. Federal size regulations for commercial motor vehicles,2004.
[6] Mehdi Ahmadian. On the isolation properties of semiactive dampers. Journal of Vi-bration and Control, 5(2):217, 1999.
[7] Mehdi Ahmadian, Brian Reichert, Xubin Song, and Steve S. Southward. No-jerk semi-active skyhook control method and apparatus, 2000.
[8] James S. Albus, Anthony J. Barbera, and Roger N. Nagel. Theory and practice ofhierarchical control, 1980.
[9] Peter Breedveld. Bond graphs, 2003.
[10] Neil A. Campbell and Jane B. Reece. Biology. Benjamin Cummings, 7th edition, 2004.
[11] J. David Carlson. Magnetorheological fluid devices and process of controlling force inexcercise equipment utilizing same, October 6 1998.
[12] J. David Carlson and Michael J. Chrzan. Magnetorheological fluid dampers, January11 1994.
[13] J. David Carlson, Michael J. Chrzan, and Frank O. James. Magnetorheological fluiddevices, March 21 1995.
[14] J. David Carlson and Keith D. Weiss. Magnetorheological materials based on alloyparticles, January 1995.
149
REFERENCES
[15] Angela K. Carter. Transient motion control of passive and semiactive damping forvehicle suspensions. Masters thesis, Virginia Tech, 1998.
[16] David M. Catanzarite, David J. Hamo, and John C. Holzheimer. Method for auto-calibration of a controllable damper suspension, 1999.
[17] David M. Catanzarite, Kenneth A. St. Clair, and Robert H. Marjoram. Controllablecab suspension, 2000.
[18] P. R. Chandler and M. Pachter. Hierarchical control for autonomous teams, 2001.
[19] Hong Chen and Kong-Hui Guo. Constrained h-infinity control of active suspensions:an lmi approach. Control Systems Technology, IEEE Transactions on, 13(3):412–421,May 2005.
[20] Chee Chuan Chew. An analysis of passive, semi-active and active truck cab suspensionsystems. Masters thesis, Univeristy of California, 1992.
[21] Michael Jacob Craft. Design optimization of magneshock magnetorheological shockabsorbers and development of fuzzy logic control algorithms for semi-active vehiclesuspensions. Master’s thesis, North Carolina State University, 2003.
[22] Michael J. Crosby and Rush E. Allen. Cab isolation and ride quality. Technical report,Society of Automotive Engineers, 1974.
[23] J. der Hagopian, J. Gaudiller, and B. Maillard. Hierarchical control of hydraulic ac-tive suspensions of a fast all-terrain military vehicle. Journal of Sound and Vibration,222(5):723–752, 1999.
[24] X. M. Dong, Miao Yu, S. L. Huang, Zushu Li, and W. M. Chen. Half car magnetorhe-ological suspension system accounting for nonlinearity and time delay. InternationalJournal of Modern Physics, 19(7, 8 and 9):1381–1387, 2005.
[25] E. M. Elbeheiry and D. C. Karnopp. Optimal control of vehicle random vibration withconstrained suspension deflection. Journal of Sound and Vibration, 189(5):547 – 564,1996.
[26] M. M. ElMadany. An analytical investigation of isolation systems for cab ride. Com-puters & Structures, 27(5):679–688, 1987.
[27] M. M. ElMadany. Nonlinear ride analysis of heavy trucks. Computers & Structures,25(1):69–82, 1987.
[28] M. M. ElMadany. Design and optimization of truck suspensions using covariance anal-ysis. Computers & Structures, 28(2):241–246, 1988.
150
REFERENCES
[29] M. M. ElMadany. Design of an active suspension for a heavy duty truck using optimalcontrol theory. Computers & Structures, 31(3):385–393, 1989.
[30] M. M. ElMadany and Z. Abduljabbar. Design evaluation of advanced suspension sys-tems for truck ride comfort. Computers & Structures, 36(2):321–331, 1989.
[31] James William Fitch. Motor truck engineering handbook. Society of Automotive Engi-neers, Warrendale, PA, 4th edition, 1994. 93035469 James William Fitch. ill. ; 24 cm.Includes index.
[32] Wallace C. Flower. Analytical and subjective ride quality comparison of front and rearcab isolation systems on a coe tractor. SAE Document 780411, 1978.
[33] Thomas D. Gillespie. Heavy truck ride. Society of Automotive Engineers, Warrendale,Pa., 1985.
[34] Alexander Gross and Roy Van Wynsberghe. Development of a 4-point-air cab suspensionsystem for conventional heavy trucks. SAE Document 2001-01-2708, 2001.
[35] Takashi Hiromatsu, Takeshi Inaba, and Yoshiki Matsuo. Development of an electrome-chanical active-cab-suspension, November 15-19 1993.
[36] Dean Karnopp, Donald L. Margolis, and Ronald C. Rosenberg. System Dynamics :Modeling and Simulation of Mechatronic Systems. John Wiley & Sons, Hoboken, N.J.,4th edition, 2006.
[37] Dean Karnopp and Ronald C. Rosenberg. Analysis and simulation of multiport systems;the bond graph approach to physical system dynamics. M.I.T. Press, Cambridge, Mass.,,1968. 68025381.
[38] Dean Karnopp and Ronald C. Rosenberg. System dynamics : a unified approach. Wiley,New York, 1975. 74022466.
[39] Dean C. Karnopp and Michael J. Crosby. System for controlling the transmission ofenergy between spaced members, Sep. 19, 1972 1974.
[40] Dean C. Karnopp, Michael J. Crosby, and R. A. Harwood. Vibration control usingsemi-active force generators. Journal of Engineering for Industry, 96:619–626, 1974.
[41] Florin Marcu. Communication with Volvo Trucks North America engineers., February3, 2006.
[42] Kimihiko Nakano, Yshihiro Suda, and Shigeyuki Nakadai. Self-powered active vibrationcontrol using a single electric actuator. Journal of Sound and Vibration, 260:213–235,2002.
151
REFERENCES
[43] American Association of State Highway and Transportation Officials. Recommendedpolicy on maximum dimensions and weights of motor vehicles to be operated over thehighways of the United States. The American Association of State Highway and Trans-portation Officials, Washington, D.C., 1974.
[44] American Association of State Highway Officials. Committee on Highway Transporta-tion. Recommended policy on maximum dimensions and weights of motor vehicles to beoperated over the highways of the United States. American Association of State HighwayOfficials, Washington, D.C., 1968.
[45] Katsuhiko Ogata. Modern control engineering, volume xi, 964 p. :. Prentice Hall, UpperSaddle River, N.J. :, 4th ed. edition, 2002.
[46] Paul Stephen Patricio. Effects of frame design and cab suspension on the ride quality ofheavy trucks. Masters thesis, Virginia Polytechnic Institute and State University, 2002.
[47] Henry Martyn Paynter. Hydraulics by analog - an electronic model of a pumping plant.J. Boston Society of Civil Engineering, pages 197–219, 1959.
[48] Jacob Rabinow. Magnetic fluid shock absorber, 1954.
[49] Ronald C. Rosenberg and Dean Karnopp. Introduction to physical system dynamics.McGraw-Hill series in mechanical engineering. McGraw-Hill, New York, 1983.
[50] D. Sammier, O. Sename, and L. Dugard. Skyhook and h8 control of semi-active sus-pensions: Some practical aspects. Vehicle System Dynamics, 39:279–308, 2003.
[51] Y. Shen, M. F. Golnaraghi, and G. R. Heppler. Semi-active Vibration Control Schemesfor Suspension Systems Using Magnetorheological Dampers. Journal of Vibration andControl, 12(1):3–24, 2006.
[52] Nobutaka Tsujiuchi, Takayuki Kizumi, Tomoyuki Jinde, and Eiichi Ishida. Reduction ofvibration in tractor using semi-active suspension. SAE Document 2002-01-1469, 2002.
[53] B. D. Van Deusen. Truck suspension system optimization. Journal of Terramechanics,9(2):83–100, 1973.
[54] Klaus Webers and Erik Coelingh. Method and device for vehicle motion control, August2004.
[55] W. M. Winslow. Method and means for translating electrical impulses into mechanicalforce, 1947. Invention of ER fluid.
[56] W. M. Winslow. Induced fibration of suspensions. Journal of Applied Physics, 20:1137–1140, 1949.
[57] A. Zaremba, R. Hampo, and D. Hrovat. Optimal active suspension design using con-strained optimization. Journal of Sound and Vibration, 207(3):351 – 364, 1997.