Professional Documents
Culture Documents
Hierarchical
Sliding Mode
Control for Under-
actuated Cranes
Design, Analysis and Simulation
Hierarchical Sliding Mode Control
for Under-actuated Cranes
Dianwei Qian Jianqiang Yi
Hierarchical Sliding
Mode Control
for Under-actuated Cranes
Design, Analysis and Simulation
123
Dianwei Qian Jianqiang Yi
School of Control and Computer Institute of Automation
Engineering Chinese Academy of Sciences
North China Electric Power University Beijing
Beijing China
China
Dianwei Qian
Jianqiang Yi
Preface
Overhead cranes are widely used in many places, such as warehouses, disaster sites,
nuclear plants, shipyards and construction sites. Overhead crane control has been
paid more and more attention in recent years. Concerning the control problem,
numerous theoretical studies and practical implementations have been carried out.
Accordingly, various control methods have been presented. Among the diversity of
control methods, sliding mode control is recognized as one of the most efcient
design tools.
Many control designs and applications in the elds of sliding mode control for
overhead cranes have been published in various journals and conference proceed-
ings. In spite of these remarkable advances in this eld, most of the current
researches only focus on some special control applications, rather than a systematic
methodology.
The methodology of hierarchical sliding mode control lls the gap between
sliding mode control and its applications in overhead cranes. Hierarchical sliding
mode control presents several control structures for the sliding mode control
applications in overhead cranes. It is a systematic and effective design tool, which
has both theoretical and practical signicances.
This book provides readers with a comprehensive overview of sliding mode
control for overhead crane systems with detailed proofs of the fundamental results.
Capturing the structure characteristic of overhead cranes, this is possible to achieve
novel control structures by the method of sliding mode control. With the under-
standing that the physical structure of overhead crane systems is dismantled, some
appropriate control structures are constructed and hierarchical sliding mode control
is developed.
The key feature of hierarchical sliding mode control architectures is the hier-
archical sliding surfaces with the guaranteed stability. By explicitly building the
structure specication into the problem formulation, it is possible to construct and
analyze novel sliding surface structures. With hierarchical sliding mode control
architectures, sliding motion can be depicted by phase plane and this appears to be
benecial both for scientic researches and studies.
vii
viii Preface
The book has six chapters. Each chapter concludes with appendices about
simulation programs.
Chapter 1 starts with a brief introduction of overhead crane systems. It proceeds
with a brief historical overview of sliding mode control. A review about overhead
crane control is considered next. This chapter proceeds with analysis of some
typical control problems associated with sliding mode control for overhead cranes.
Chapter 2 investigates modeling of overhead crane systems. First, equations of
motion for single-pendulum-type overhead cranes are presented. Then, dynamics of
double-pendulum-type overhead cranes are described. Uncertainties of the two
models are considered next. The chapter proceeds with analysis of oscillations for
pendulum-type motions according to the linearized models of the two types of
overhead cranes.
Chapter 3 introduces several typical design methods of sliding mode control.
The chapter proceeds with applications of these design methods for overhead
cranes. Some simulation results are demonstrated. MATLAB codes about the
simulations are also available in the appendix part.
Chapter 4 presents hierarchical sliding mode control for overhead cranes. Three
hierarchical structures are designed for single-pendulum-type overhead cranes and
one hierarchical structure is considered for double-pendulum-type overhead cranes.
For each hierarchical structure, it is proved that both the hierarchical sliding sur-
faces and the whole control system are of asymptotically stability in the sense of
Lyapunov. Numerical simulations illustrate the feasibility of these designed hier-
archical structures. MATLAB codes about the simulations are also attached in the
appendix part.
Chapter 5 extends the method of hierarchical sliding mode control to accom-
modate unmatched uncertainties. It starts with compensator design for the unmat-
ched uncertainties of overhead cranes. The compensator and the controller work
together to realize the robust overhead crane control. Since the compensator design
needs a strict assumption, this chapter proceeds with the design of intelligent
compensator based on the hierarchical structure. Some simulations are conducted to
verify the effectiveness of the presented control scheme. MATLAB codes about the
simulations are also presented.
Chapter 6 summarizes some of the further extensions not captured within this
book, states the open problems, and the challenges for future thinking.
The book can be used for teaching a graduate-level special-topics course in
sliding mode control.
In this book, all the control algorithms and their programs are described sepa-
rately and classied by the chapter name, which can be run successfully in
MATLAB 7.5.0.342 version or in other more advanced versions. If you have
questions about algorithms and simulation programs, please feel free to contact
Dianwei Qian by E-mail: dianwei.qian@gmail.com.
ix
Contents
1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Crane Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Overhead Cranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Boom Cranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.3 Tower Cranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Review of Sliding Mode Control . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Variable Structure Control . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Sliding Mode Control. . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.3 Equivalent-Control-Based Sliding Mode Control . . . . . . . 11
1.2.4 Chattering Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.5 Sliding Mode Control Design for State Space Model . . . . 15
1.2.6 Robustness Against Uncertainties. . . . . . . . . . . . . . . . . . 17
1.2.7 Sliding Order and Sliding Sets . . . . . . . . . . . . . . . . . . . 22
1.3 A Review of Crane Control. . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.1 Open-Loop Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.2 Closed-Loop Control . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4 Challenges of Sliding Mode-Based Crane Control . . . . . . . . . . . 33
1.4.1 Theoretical Challenges . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.4.2 Practical Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
A Matlab Codes to Plot Fig. 1.5a. . . . . . . . . . . . . . . . . . . . . . . 35
B Matlab Codes to Plot Fig. 1.5b. . . . . . . . . . . . . . . . . . . . . . . 35
C Matlab Codes to Plot Fig. 1.7 . . . . . . . . . . . . . . . . . . . . . . . 35
D Simulink Model to Plot Figs. 1.8 and 1.9 . . . . . . . . . . . . . . . 36
E Simulink Model to Plot Figs. 1.10 and 1.11 . . . . . . . . . . . . . . 37
F Matlab Codes to Plot Fig. 1.12 . . . . . . . . . . . . . . . . . . . . . . . 38
G Simulink Model to Plot Figs. 1.13 and 1.14. . . . . . . . . . . . . . 38
H Simulink Model to Plot Figs. 1.15 and 1.16. . . . . . . . . . . . . . 39
I Simulink Model to Plot Figs. 1.17, 1.18, 1.19 and 1.20 . . . . . . 41
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
xi
xii Contents
A vast number of cranes are used worldwide. Cranes are employed to transport
heavy loads at shipyards, construction sites, warehouses, and factories throughout
the world because of their heavy payload capabilities.
In the late sixth century BC, the rst construction cranes in Fig. 1.1 were
invented by the ancient Greeks and were powered by men or beasts of burden [1].
These cranes were used for the construction of buildings. Larger cranes were later
developed, employing the use of human tread-wheels, permitting the lifting of
heavier weights. In the High Middle Ages, harbor cranes were introduced to
load/unload ships and assist with their construction. The earliest cranes were
constructed from wood, but cast iron, iron, and steel took over with the coming of
the Industrial Revolution [2].
Cranes exist in an enormous variety of forms. According to their primary dynamic
properties and the coordinate system that most naturally describes the location of
the suspension cable connection point, cranes can roughly be categorized into three
types, i.e., overhead cranes, boom cranes, and tower cranes [3, 4].
The rst type is overhead cranes, also known as entitled bridge cranes. This type of
cranes, illustrated in Fig. 1.2a, b, often operate in Cartesian space, where the trolley
moves along a bridge and the motion of the bridge is perpendicular to that of the
trolley. Sometime, overhead cranes travel on a mobile base. Under this condition,
mobile overhead cranes are often called gantry cranes. Overhead cranes are com-
monly utilized in factories, warehouses, and shipyards.
The second major type of cranes is named as boom cranes. The schematic repre-
sentation of boom cranes is shown in Fig. 1.3a, b. Boom cranes are usually
described by spherical coordinates. In the coordinates, a boom rotates around axes
both perpendicular and parallel to the ground. Displayed in Fig. 1.3b, is the
rotation around the vertical Z axis and is the rotation around the horizontal Y axis.
The payload is supported by a suspension cable at the end of the boom. Different
from the other two types of cranes, the boom cranes can support loads in com-
pression. Boom cranes are typically compact than the other two types of cranes
under the condition of similar load carrying capacities. Such a compact advantage
also lends well to being mounted on a mobile base such as trucks, tracked vehicles,
and ships. Boom cranes are often located at building construction sites, harbors, and
shipyards.
1.1 Crane Types 3
The third major type of cranes is called tower cranes, like the ones displayed in
Fig. 1.4a, b. For convenience, tower cranes can be described by cylindrical coor-
dinates. From the schematic drawing in Fig. 1.4b, the horizontal jib arm can rotate
around a vertical tower. The payload is supported by a cable from the trolley and
the trolley moves along the jib arm in the radial direction. Tower cranes often give
the best combination of height and lifting capacity. They are commonly found in
the construction of tall buildings because, this type of cranes has the merit of having
a small footprint-to-workspace ratio.
Whatever type the cranes are, a common characteristic among all cranes is that
the payload is supported by a suspension cable. The structural characteristic pro-
vides the basic functionality of cranes, i.e., hoisting and lowering. Meanwhile, it
also presents some challenges. Motion of a crane will denitely result in its payload
oscillation, which is one of the most challenging issues. Inherently, the oscillations
determined by the crane structure are pendulum-type. The oscillations have many
4 1 Introduction
detrimental effects and a series of drawbacks may be induced, including but not
limited to safety hazards, time delays, and degradation of positioning accuracy.
The ubiquity of cranes versus the propensity to payload oscillations result in
cranes to be the bottleneck in many practical applications. As a result, improving
the control and efciency of cranes will benet a lot because of having enormous
economic impact. Signicant research attention has been paid to resist payload
oscillations.
Compared to overhead cranes, tower cranes and boom cranes are more chal-
lenging in the aspect of control design because their nonlinear dynamics originate
from the rotational nature of the cranes. This book concentrates on novel structures
of sliding mode control for cranes and adopts crane systems as a research platform.
For the purpose of simplication, only overhead cranes are considered in the rest of
the book.
1.2 Review of Sliding Mode Control 5
Sliding mode control is also named as sliding mode variable structure control or
variable structure control with sliding mode [5]. From the two aliases, sliding mode
6 1 Introduction
control is often associated with variable structure control and it has a close rela-
tionship with variable structure control. To understand sliding mode control, some
background information about variable structure control is presented.
Variable structure control has a long history. Although the idea evolved from the
pioneering work in Russia in the early 1960s, variable structure control appeared
worldwide only when it was reported by Utikin [5].
Suggested by its name, a variable structure control system has a control law that
deliberately changed during the control process according to some predened rules.
The rules depend on the state of the system. For the purpose of illustration, consider
a second-order system (1.1).
xt ut 1:1
ut kxt; 1:2
k1 xt if x_x\0
ut 1:3
k2 xt otherwise
Vt x2 x_ 2 1:4
From (1.4) and (1.5), V > 0 and V\0. _ In the sense of Lyapunov, the
second-order system with the compound control law is of asymptotic stability. This
fact indicates that the distance from the origin is always decreasing. The theoretical
deduction coincides with the previous intuitive analysis. As a result, a stable
closed-loop system has been built up by a rule combining two control laws that
cannot independently stabilize the system. The two control laws result in two
different system dynamics (structures) in the phase plane such that the design
method is entitled variable structure control.
As mentioned, sliding mode control has close relevance to variable structure control.
Briefly, it is just a special case of variable structure control [6]. A characteristic that
variable structure control is different from other control methods is the design of
switching rules.
In a large part, the switching rule (1.3) originates from intuition and analysis. To
systematize the design tool of variable structure control, another variable structure
control law can be given as
1 if sx; x_ [ 0
ut 1:6
1 if sx; x_ \0
sx; x_ mx x_ 1:7
ut sgns 1:8
1.2 Review of Sliding Mode Control 9
Apply the control law (1.8) to the second-order system (1.1). To consider the
stability of this control system in the sense of Lyapunov, the Lyapunov function
candidate is dened as
1
Vx; x_ s2 1:11
2
Differentiating V with respect to time t and substituting (1.1) and (1.8) into the
derivative of V yields
V_ s_s sm_x x
sm_x u sm_x sgns 1:12
\jsjmjx_ j 1
Equation (1.12) indicates the system trajectories on either side of the following line
L that will point toward the line when mjx_ j\1.
Ls f x; x_ : sx; x_ 0 g 1:13
when m = 1, Fig. 1.8 illustrates the trajectory slide along the line to the origin.
Illustrated in Fig. 1.8, the system trajectory repeatedly crosses the line L. Each cross
indicates that the control system is switching between the two different control
structures. Intuitively, such repeated crosses have a direct effect on the system
performance. In fact, such switching makes the control input choppy. Displayed in
Fig. 1.9, the effect is the high-frequency switching of the control signal and the
fluctuation of the system output. Figures 1.8 and 1.9 are obtained by a simulink
model of MATLAB, which is shown in Appendix D.
Provided that the switching frequency is innite, the system trajectory can be
constrained to remain on the line L. Under this assumption, the motion conned to
the line L satises the differential equation obtained from (1.13). That is,
x_ mx 1:14
Equation (1.14) means that a straight line crosses the origin in the phase plane,
where m is the slope of the line.
10 1 Introduction
As a result, some terms to describe sliding mode control can be drawn. The line
L is entitled as the sliding surface, the trajectory sliding along the surface is termed
the ideal sliding mode, the trajectory moving toward the surface is called the
reaching mode, the high-frequency switching of the control signal is described by
chattering and the condition in (1.12) from the Lyapunov direct method is referred
to as the reachability condition, which are usually written as
s_s\0 1:15
or equivalently
Consider the system (1.1), design the sliding surface (1.7), and adopt the sliding
mode control law (1.8). Then, the existence of the sliding mode is guaranteed.
Suppose that the system trajectory reaches the sliding surface at time ts and that the
sliding mode exists thereafter. These linguistic descriptions can be formulated as
s 0 and s_ 0 1:17
s_ m_x ut 1:18
Since s_ 0 for all t > ts, the control law (1.19) mains the sliding motion of the
second-order system on the line L.
ut m_x t ts 1:19
The control law (1.19) is named equivalent control. Usually, the equivalent
control is marked by ueq. Note that the equivalent control action is not the real
control signal applied to the system but may be treated as the control signal applied
on average. According to the equivalent control law, the original control signal can
be divided into two parts, i.e., equivalent control and switching control.
In such a design, both of the controls in (1.20) cooperate with each other to
generate the real control signal in the reaching mode stage but the visual equivalent
control is solely applied to the system in the sliding mode stage. To illustrate the
superior and design of the equivalent-control-based sliding mode control, the fol-
lowing case study is conducted.
Consider the system (1.1), design the sliding surface (1.7) and adopt the sliding
mode control law (1.20). The switching control law usw can be deduced from the
reachability condition (1.15). Substituting (1.1), (1.7), (1.19), and (1.20) into the
left of (1.15) yields
s_s sm_x x
1:21
sm_x ut sm_x ueq usw
s_s\0 1:24
Compared with the sliding mode control law (1.8), the equivalent-control-based
sliding mode control law (1.25) consists of two parts, where the visual equivalent
control law is continuous.
To demonstrate the feasibility of the equivalent-control-based sliding mode
control, some simulation results are shown in Figs. 1.10 and 1.11. For the purpose
of comparison with the precious control law (1.8), the value of k in (1.25) is kept
unchanged from k = 1 and the parameter of the sliding surface is also designed by
m = 1 in (1.7).
Such an equivalent-control-based design can dramatically reduce the chattering.
Compared with Figs. 1.8 and 1.10, the phase trajectory in Fig. 1.10 is much smooth
and the system performance is apparently improved. Further, the system perfor-
mance under the equivalent-control-based sliding mode control law is displayed in
Fig. 1.11. Compared with the results in Fig. 1.9, the curves are smoother and have
less fluctuation. Figures 1.10 and 1.11 are obtained by a simulink model of
MATLAB, which is shown in Appendix E.
As mentioned, the purpose of the control law (1.8) is to ensure the phase tra-
jectory moves toward and is forced to remain on the sliding surface. The purpose
Fig. 1.10 Phase portrait of a sliding motion of the system when m = 1 and k = 1
1.2 Review of Sliding Mode Control 13
Fig. 1.11 System performance when m = 1 and k = 1. a System output x; b control signal u
results in a dramatic switching of the control signal generated by (1.8). The idea of
equivalent-control-based sliding mode control divides the whole control action into
two parts, where the control action on the sliding surface is continuous, which can
reduce the chattering phenomenon and improve the system performance.
Note that the chattering is an inherent drawback of sliding mode control, which
is also a gap between sliding mode control and real applications. The equivalent-
control-based sliding mode control design lls the gap and such a control design
makes the sliding mode control theory applicable.
where s denotes the sliding surface variable and is a small positive constant.
14 1 Introduction
Numerical solutions of (1.26) are shown in Fig. 1.12. It can be visualized that the
function tends point-wise to the signum function as 0. The variable can be
used to trade-off the requirement of maintaining ideal performance with that of
ensuring a smooth control action. From Fig. 1.12, bigger the value of is, smoother
the curve is. The replacement can improve the performance of control signal, but it
decreases the robustness of sliding mode control, degenerates sliding mode into
quasi-sliding mode, and loses guaranteed system stability. The MATLAB programs
of the example are given in Appendix F.
As far as the equivalent-control-based sliding mode control is concerned, such an
alternative will only effect the second term in the control law given by Eq. (1.25).
Concerning the alternative, the new control law can be rewritten as
s
ut m_x k 1:27
j sj d
In (1.27), (1.26) replaces sgn(s) in (1.25). Thus, in the region around the sliding
surface determined by the value of , the system is effectively a high gain feedback
system and possesses the properties of certain classes of disturbance rejection and
parameter invariance. Adopt the control (1.27) for the system (1.1), where the
values of m and k are kept unchanged from m = 1 and k = 1 and the value of is 0.1.
The system phase trajectory and the control signal are illustrated in Figs. 1.13 and
1.14, which are obtained by a simulink model of MATLAB in Appendix G.
Compare Fig. 1.13 with Fig. 1.10. The two curves of phase trajectories almost
make no difference in the phase plane. But the control signals in Figs. 1.11 and 1.14
illustrate the effects of the smoothing function method. In Fig. 1.11, the control
signal severely jumps back and forth at about time t = 1 s because the signum
function is triggered by the sliding motion in the phase plane. On the other hand, the
control signal becomes smooth in the dynamic process by introducing the
smoothing function in Fig. 1.14.
1.2 Review of Sliding Mode Control 15
Fig. 1.13 Phase portrait of a sliding motion of the system when when m = 1, k = 1 and = 0.1
Fig. 1.14 System performance when m = 1, k = 1 and = 0.1. a System output x; b control signal u
In Fig. 1.14, the performance of the control signal can be improved by increasing
the value of . But a series of drawbacks may be induced by the alternative
smoothing function method, such as performance, stability, accuracy, and so on.
The value of must be determined by a trade-off among the performance indexes.
Although it is an intuitive and effective tool, phase plane analysis is hardly utilized
to analyze multivariable systems and high-order systems. A more general frame
will need to be established. The state space approach provides a possible choice for
accomplishing this task.
16 1 Introduction
x , x x_ T 1:28
Then, the system (1.1) can be rearranged in the form of state space, whose
expression is as follows:
0 1 0
x_ x u 1:29
0 0 1
s cT x 1:30
here c m 1 T .
Concerning (1.28) and (1.30), the aforementioned design methods, i.e., sliding
mode control, equivalent-control-based sliding mode control and chattering
reduction, can be described in the form of matrix as well. The details are exhibited
in Chaps. 35.
From the viewpoint of control theory, (1.28) depicts the state space model of a
continuous time-invariant system. Further, the more general state space represen-
tation of a linear continuous time-invariant system with p inputs, q outputs, and
n state variables can be written in the following form
x_ Ax Bu
1:31
y Cx Du
Here x 2 <n is called state vector, A 2 <nn is termed input matrix, B 2 <np is
called input matrix, u 2 <p is termed control vector, y 2 <q is named output vector,
C 2 <qn is called output matrix, D 2 <qp is termed feed-forward matrix, x_
Ax Bu is named state equations, and y Cx Du is called output equations.
Since this book considers control design by the state-feedback-based sliding
mode control, only the state equations of the general state space representation are
employed in the rest chapters. Thus, the general expression of (1.29) is written as
x_ Ax Bu 1:32
1.2 Review of Sliding Mode Control 17
Differentiating (1.30) with respect to time t and substituting (1.32) into the
derivative of s yield
s_ cT x_ cT Ax Bu 1:33
In (1.35), dx; t is the distance term and the original Eq. (1.29) are called
nominal systems. To demonstrate the effects of matched uncertainty on the system
performance, dene
0
dx; t 0:1 rand; 1:36
1
0 1 0
x_ x u 0:1 rand 1:37
0 0 1
From (1.37), the disturbance term enters the system (1.1) by the control channel.
Consequently, it undoubtedly belongs to matched uncertainty. To verify the system
stability, we resort to the reachability condition of sliding mode (1.15). Designing
the sliding surface (1.30), differentiating (1.30) with respect to time t and substi-
tuting the derivative of s into (1.15) yield
supk0:1 randk 0:1 in (1.41), indicating we can choose any k [ 0:1 as the
parameter of the switching control law. k is a design parameter, which is only
related to the boundary of uncertainty rather than any physical parameters. This
property is termed invariance of sliding mode. For the purpose of illustration, adopt
the equivalent-control-based sliding mode control design (1.20) and keep the
parameters of the sliding surface and the switching control law unchanged from the
precious example. Some numerical simulation results from the initial conditions
x(0) = 1 and x_ 0 0 are displayed in Figs. 1.15 and 1.16, which are obtained by a
simulink model of MATLAB in Appendix H.
Compared with Fig. 1.10, the phase portrait in the phase plane in Fig. 1.15 has
no ideal sliding mode. The trajectory frequently crosses the sliding surface because
of the existence of matched uncertainty. Once the trajectory deviates from the
sliding surface, the switching control law and the equivalent control law will draw
the trajectory toward the surface together. The switching in Fig. 1.16b reflects this
process. Illustrated in Fig. 1.16a, the system output can eventually tend to be
1.2 Review of Sliding Mode Control 19
Fig. 1.15 Phase portrait of a sliding motion of the system with matched uncertainty when m = 1
and k = 1
Fig. 1.16 Performance of the system with matched uncertainty when m = 1 and k = 1. a System
output x; b sliding surface s; c control signal u
asymptotically stable. Compared with Fig. 1.11b, the chattering phenomenon of the
control signal is severe in Fig. 1.16c, to resist the uncertainty.
To demonstrate the effects of unmatched uncertainty, dene
1
dx; t 0:1 rand 1:42
0
From (1.43), the disturbance term can hardly enter the system (1.1) by the control
channel. Consequently, the uncertain term is unmatched. The equivalent-control-
based sliding mode control design (1.20) is still adopted. Since the unmatched
20 1 Introduction
Fig. 1.17 Phase portrait of a sliding motion of the system with unmatched uncertainty when m = 1
and k = 1
Fig. 1.18 Performance of the system with unmatched uncertainty when m = 1 and k = 1. a System
output x; b sliding surface s; c control signal u
Fig. 1.19 Phase portrait of a sliding motion of the system with unmatched uncertainty at t = 40 s
22 1 Introduction
Fig. 1.20 Control performance of the system with unmatched uncertainty at t = 40 s. a System
output x; b sliding surface s; c control signal u
From Figs. 1.17, 1.18, 1.19 and 1.20, the sliding mode control system is of
asymptotically stability under both matched and unmatched uncertainties. Under
either kind of uncertainty, the sliding mode will deviate from the sliding surface.
The sliding mode control law tends to make the system trajectory move toward the
sliding surface by the sliding mode control law. Consequently, both kinds of
uncertainties affect the sliding modes and deteriorate the chattering phenomena.
The sliding mode control system can realize the control task quickly against the
matched certainty because the matched certainty has no relationship with the
physical parameters and can directly affect the system stability. On the other hand,
the unmatched certainty affects the system stability by coupling the physical
parameters, which delays the system response.
A function is of class Ck, i.e., f is continuous with its derivatives up to the order
k. Correspondingly, if the sliding surface variable s is of class Ck, then the sliding
mode on the surface s(x, t) = 0 is referred as (k + 1)th order sliding mode, when
x denotes the system states. Note that the sliding mode of order one corresponds to a
sliding variable of class C0, with discontinuous derivative. The order of a sliding
mode [6] represents the smoothness degree associated to the motion constrained on
the sliding manifold, dened as follows.
Denition 1.1 The sliding order r is the number of continuous total derivatives
(including the zero one) of the function s = s(x, t) whose vanishing denes the
equations of the sliding mode.
1.2 Review of Sliding Mode Control 23
Denition 1.2 The sliding set of order r associated to the manifold s(x, t) = 0 is
dened by the equalities.
s s_ sr1 0 1:48
From Denitions 1.1 and 1.2, the sliding set in (1.13) is just the rst-order
sliding mode. In Chap. 3, some typical sliding mode control methods will be
introduced one by one though control problems of overhead cranes.
Optimal trajectory planning methods, also named motion planning, seek to elimi-
nate vibration by using preplanned trajectories to move the crane through the
workspace. These methods can avoid the sensing problems of closed-loop control
methods.
The control problem is typically formulated as a minimum time optimization
problem, subject to vibration constraints. Optimal design of the transportation
trajectory not only improves the control performance in terms of efciency or
safety, but also releases some pressure on control for the sequential strategy.
The rst to prose an optimal control strategy for automatic crane operation was
Field [9]. The similar control problems were also investigated by the Pontryagins
maximum principle [10] and some evolutionary algorithms [11]. Lee and his
research division solved several motion planning-based methods to realize the
high-performance anti-swing control of overhead cranes, i.e., solving the motion
planning problem through a kinematic problem [12], designing the motion planning
method by means of the model-following anti-swing control law [13], and fusing
the motion planning method on the sliding mode control [14]. Ross and Fahroo [15]
proved that the optimal motion planning problem was equivalent to a classic Bolza
problem of the calculus of variations and their results were exhibited by the 2-D
crane model.
24 1 Introduction
To deal with the crane control problems in cluttered work environment, Blajer
and Kolodziejczyk [16] rst sketched a trajectory by a series of points in the work
space and then approximated the trajectory by spline functions. To coordinate
multiple construction cranes [17], the incremental coordination method is employed
to plan the motion of two or more cranes, where the kinematics and the geometrical
constraints of cranes are taken into consideration. Programming is an effective
optimization tool. Several programming methods have been applied to plan crane
trajectories, i.e., dynamic programming [18] and linear programming [19].
Generally speaking, trajectory planning is offline, which makes the method non
robust. To deal with the issue, AlBahnassi and Hammad [20] developed a frame-
work for near real-time motion planning of cranes. The framework can assist crane
operators to replan safe paths in near real time. Fang and his colleagues [21]
investigated a kinematic coupling-based offline trajectory planning method for
overhead cranes, where the trajectory was tuned by an iterative learning strategy to
guarantee accurate trolley positioning. Further, they [22] considered an offline
trolley trajectory planning method for underactuated overhead cranes, where some
rigorous geometric analysis were utilized to address the coupling behavior between
the actuated trolley motion and the underactuated payload swing.
The primary restriction of the optimal trajectory planning method is that the
desired motion of cranes must be known in advance, in addition to the initial
conditions of the maneuver. Another major drawback is that generating the optimal
proles can become computationally expensive because of the kinematic coupling
behavior between the underactuated payloads swing and the actuated trolleys
motion.
Input shaping is another open-loop control technology for cranes, which can dra-
matically reduce motion-induced payload oscillations by intelligently shaping the
reference commands. Unlike the optimal trajectory planning, input shaping is an
online method and it can be applied in real time.
To employ the input shaping method, an input shaper should be designed, which
is composed of a series of impulses. These impulses can cancel vibration. That is,
vibration caused by a part of the impulses is canceled by vibration caused by other
parts of the impulses. The amplitude and time locations of the impulses are obtained
from the systems natural frequencies and damping ratios.
The input shaping technology is widely used. So far, the method has been
implemented on several large bridge cranes at nuclear facilities [23], a 10-ton crane
at Georgia Tech [24], tower cranes [25] as well as coordinate measurement
machines [26]. Input shaping not only achieves crane control directly, but also it
can improve crane operator performance [27].
Crane control research in the input shaping eld involves developing shapers to
minimize the sway of crane payloads. The use of input shaping dramatically
reduces the time to the completion of move operations and greatly increases the
1.3 A Review of Crane Control 25
safety of the crane by allowing for a smaller danger zone around the crane. Alsop
et al. [28] rst proposed a shaping-type control strategy to control cranes in 1965,
which could be treated as the rudiment of the current input shaping method.
Thereafter, a large research effort has been directed at crane control applications via
the input shaping method owing to the effectiveness of input shaping.
Yamada et al. [29] proposed an optimal input shaping method, where the control
command was generated by Pontryagins maximum principle to achieve minimum
transfer time. Lin [30] designed an input shaping method by a pole/zero placement
approach, where the pole/zero placement method could improve system perfor-
mance and simplify design procedures. Singhose and his colleagues also suggested
several shaping methods, i.e., convolved and simultaneous two-mode input shapers
[31], expert-system-based shaper [32], and negative-impulse-included shaper [33].
Recently, some novel shaping methods are often reported, i.e., nonlinear input
shaping [34], graphical-approach-based input shaping design [35], shaping design
considering limited-state constraints [36], frequency-modulation input shaping [37].
Publications concerning crane control by input shaping have resulted in dozens
of papers. However, input shapers are located at the feed-forward channel, meaning
that they are hard to resist any error without special design. Especially, the draw-
back becomes deteriorative for some types of double-pendulum cranes because
there are severe residual vibrations even for a small modeling error.
For the purpose of illustration, the survey of open-loop control for cranes is
demonstrated in Fig. 1.21.
Closed-loop crane control aims to nd a feedback control law that stabilizes the
system in the presence of various uncertainties and external disturbances, which
uses information about the current states of the system (e.g., payload swing angle,
trolley position, etc.) to generate commands that drive the system toward the
desired state. Concerning crane control, the control input is the force or torque
applied to the trolley in order to suppress oscillations due to the acceleration and
deceleration of the trolley. The complexity of the crane control eld originates from
the crane structure that the number of actuators is less than the degree-of-freedom to
be controlled. Direct applications of many traditional nonlinear control methods
also suffer from such structure.
Recently, many control methods based on feedback linearization, energy,
backstepping, sliding mode, and fuzzy logic have been developed for crane control.
This section provides a view to highlight some of these popular control methods.
For the purpose of illustration, the survey of closed-loop control for cranes is
demonstrated in Fig. 1.22.
26 1 Introduction
Linear control calls for linearized crane model. Based on the linearized crane
model, Hazlerigg [38] was one of the rst to propose this method in 1972. Since
then, a variety of linear control methods has been applied to crane control practice.
These methods include some classic control design, i.e., linear quadratic regulator
(LQR) control by Grassin et al. [39], proportional-integral and proportional
derivative controllers by Lee [40], state feedback control by Nguyen [41], com-
bination of input shaping and linear control by Yoon et al. [42], and parameter
optimization of linear controller by Hamalainen et al. [43].
However, the crane model is inherently nonlinear and the linearized method
cannot provide the sufcient accuracy of information about position error and load
swing. In particular, linear control methods are sensitive to changes in the cable
length because they are invariably tuned to counter the effects of the natural
1.3 A Review of Crane Control 27
A useful technique for crane control is called partial feedback linearization, which is
a method providing a natural global change of coordinates that transforms the
system into a strict feedback form, and the control method can be easily applied to
the new form of the system. There are two partial feedback linearization techniques
presented as collocated and non-collocated partial feedback linearization methods,
respectively.
Olfati-Saber [44] introduced a global change of coordinates that can decouple
these two subsystems but leave the linear subsystem invariant. The method has
been extensively applied to several types of cranes. Zhang et al. [45] investigated
the feasibility of partial feedback linearization for crane control problems; Tuan
[46, 47] extended the method to overhead cranes with varying cable length and
3-Doverhead cranes; Sun and Fang [48] reported their results about the applications
of partial feedback linearization with saturation constraint; Wu et al. [49] designed a
novel control method on the basis of partial feedback linearization, where the
overall crane system is divided into two subsystems.
28 1 Introduction
Uncertainty is a common yet intractable issue for crane control. For any control
design, there are typical discrepancies between a practical system and its theoretical
model because of unmodelled dynamics, parameter uncertainty, and external dis-
turbance. In the past few decades, a lot of effort has been directed in adaptability
and robustness to these factors. As two main branches, adaptive control and robust
control are developed to solve the adaptability and robustness issues of uncertainty
compensation, respectively.
Adaptive control [65] is the control method used by a controller which must
adapt to a controlled system with parameters which vary, or are initially uncertain.
An adaptive control system utilizes online identication of either system parameter
or controller parameter, which does not need a priori information about the bounds
on these uncertain or time-varying parameters.
Hurteau and Desantis [66] suggested a simplied adaptive control law for
cranes, where their adaptive controller was made of a state regulator block and a
gain tuning module. Butler et al. [67] investigated the model reference adaptive
control method for overhead cranes. dAndrea-Novel and Boustany [68, 69]
reported two indirect adaptive control methods for cranes. Yang and Yang [70]
developed an adaptive control method for 3-D cranes, where all the control system
parameters were not needed for the controller design in a priori. Recently, some
adaptive design approaches against cable length changes [71], external distance
[72], and unknown parameters [73] are also reported.
These adaptive control approaches consider their control design in the sense of
Lyapunov, according to the traditional adaptive control approaches. Besides, there
are still some research results by combining adaptive control and other control
methods, i.e., adaptive sliding mode control [74], motion planning-based adaptive
control [75], adaptive sliding mode fuzzy control [76], adaptive coupling control
[77], and adaptive control based on fuzzy cerebellar model articulation controller
[78], to name but a few.
Robust control is a branch of control theory. The theory of robust control began
in the late 1970s and early 1980s and soon developed a number of techniques for
dealing with bounded system uncertainty [79]. Such a technique is developed under
the assumption that uncertain parameters or disturbances are found within some
compact set, which aims to achieve robust performance and stability in the presence
of bounded modeling errors. Ackermann [80] investigated the parameter space
30 1 Introduction
design of robust control and illustrated the use of the design tool by a crane control
system.
Another important example of a robust control technique is H loop-shaping.
The technique [79], developed by Duncan McFarlane and Keith Glover of
Cambridge University, minimizes the sensitivity of a system over its frequency
spectrum. This kind of design can guarantee that the system will not greatly deviate
from expected trajectories when disturbances enter the system. Adopting H
loop-shaping, Kar et al. [81] developed a multimode vibration control method to
suppress crane vibrations.
Besides the classic H loop-shaping robust control, other robust control
methods also are developed against the changes of cable length and payload mass,
i.e., wave-based robust control by Yang and OConnor [82], pole-placement-based
robust control by Uchiyama et al. [83], hybrid control on the basis of proportional-
differential control and H scheme [84], partial-state-feedback-based robust control
by Uchiyama [85], operator-based robust right coprime factorization design
methods by Wen et al. [8688], linear matrix inequality-based optimal robust
control [89], and gain scheduling-based robust control [90], to name but a few.
Although both the control approaches aim to resist the uncertainties of crane
systems, such as cable length and payload mass, the motivations behind the two
approaches are different from each other. The idea of robust control [79] can
guarantee a designed robust control system to be insensitive to all uncertainties by
using a xed control structure, indicating that the robust control law need not be
changed if the changes are within given bounds. On the other hand, the design of an
adaptive control system [65] is concerned with the adaptive control law changing
itself to suppress all uncertainties. Owing to the nature of these two control
structures, robust control is only suitable for dealing with small uncertainty, while
adaptive control is suitable for a wide range of parameter variation, but it is sen-
sitive to unstructured uncertainty.
Model predictive control is an advanced control method that has been in use in the
process industries, in chemical plants and oil reneries, since the 1980s. In recent
years, it has also been used in crane control problems [91].
Kimiaghalam et al. [92] rst proposed a model predictive control method for
cranes, where an optimizer was employed to nd an open-loop solution at each
sampling interval for a given horizon, and model predictive controller was located
at feedback channel to compensate for the optimizer. Subsequently, Deng and
Becerra [93] explored the constrained predictive control for cranes. Arnold et al.
[94] investigated the application of model predictive control of boom cranes.
Since model predictive controllers rely on dynamic models of the process, most
often, linear empirical models obtained by system identication but nding a
closed-loop control law for cranes is usually quite complex because of their highly
nonlinear equations of motion. As a result, there are noreports on the aspect of
1.3 A Review of Crane Control 31
model predictive control of cranes until Van den Broeck et al. [95] presented a
time-optimal formulation within the model predictive control framework.
Thereafter, model predictive control of the crane becomes active again. Some
novel approaches on different sides of model predictive control for cranes are
reported, i.e., solving constraint problems of crane control [96, 97], designing state
observers to approximate unmeasurable states [98], and nonlinear model predictive
control against crane nonlinearities [99102], to name but a few.
Suggested by its name, the model predictive control algorithm has the ability to
anticipate future events and it can take control actions accordingly. The advantage
of model predictive control [91] is the fact that it allows the current timeslot to be
optimized, while keeping future timeslots in account. This is achieved by opti-
mizing a nite time-horizon, but only implementing the current timeslot. However,
main challenges of model predictive control are that the control algorithm highly
depends on system model and that the model predictive control algorithm has no
guaranteed stability.
design of fuzzy rules [106108], analysis of the crane dynamics [109, 110],
inference mechanism of fuzzy systems [111], optimization of fuzzy systems [112],
and control practice of fuzzy crane systems [113], to name but a few.
Thereafter, most reports have focused on combination of fuzzy logic and other
control techniques. One type can be categorized as fuzzy modeling [114117],
where fuzzy logic is employed to approximate the nonlinear crane model and the
nal control command is generated by other control methods. The other type can be
categorized as fusion of multiple control methods, where one of the control method
is fuzzy logic. Further, the second type can also be subcategorized as fusion of
fuzzy logic and evolutionary algorithms [118, 119], combination of fuzzy control
and LQR design [120123], utilization of fuzzy control and proportional-integral-
differential control [124127], and a blend of fuzzy control and adaptive control
[128131]. More recent reports about fusion of fuzzy control of other control
methods, i.e., gain scheduling [132], Lyapunov redesign [133], and iterative and
evolutionary optimization [134] can be found.
Owing to the nature of strong robustness, the versatile sliding mode control method
has received considerable devotion from researchers. Pieper and Surgenor [135]
rst explored the sliding mode control method for gantry cranes. Since then, some
investigations about crane control problems by sliding mode have been reported in
recent years.
These recent contributions can be divided into two types. One type is to inves-
tigate various siding mode design methods for cranes. The other type is to fuse other
control methods to improve performance of sliding mode-based control systems for
cranes.
The rst type includes discrete sliding mode control by Pieper and Surgenor
[136], second-order sliding mode control by Bartolini et al. [137, 138] and Vazquez
et al. [139141], high-order sliding mode control by Chen and Saif [142], incre-
mental sliding mode control by Dong et al. [143], terminal sliding mode control by
Cao et al. [144], integral sliding mode control by Defoort et al. [145] and Xi and
Hesketh [146], and hierarchical sliding mode control by Tuan and Lee [147] to
name but a few.
The second type is characterized by a blend of sliding mode control and other
control methods, which is abundant because of the diversity of control methods.
Some representative methods are adaptive sliding mode fuzzy control by Liu et al.
[76] and Chang et al. [148], fuzzy sliding mode control by Liu et al. [149, 150] and
Sun et al. [151], suboptimal integral sliding mode control by Liu et al. [152, 153],
neural network sliding mode control by Wang et al. [154] and Tsai et al. [155], and
adaptive sliding mode control by some researchers [74, 156160].
1.3 A Review of Crane Control 33
The sliding mode-based crane control has grown exponentially in the past two
decades. Concerning the details of the aforementioned literature, their contributions
mainly focus on the following viewpoints:
Apply the method of sliding mode control. The aspect is exhibited by applica-
tions of mature sliding mode design for cranes to verify the feasibility. Most of the
early investigations are in the aspect of, for example, discrete sliding mode control
[136] and second-order sliding mode control [137, 138].
Design some novel structures of sliding surfaces. Crane systems inherently have
highly nonlinear characteristics, which may impede direct applications of mature
sliding mode design. Some reports on this aspect have been reported, for instance,
hierarchical sliding mode control [147, 159] and incremental sliding mode control
[143].
Suppress uncertainties in crane dynamics. There are many uncertain factors in
crane systems. For the purpose of illustration, the factors cover friction nonlinearity,
flexible cables, payload mass fluctuations, wind disturbances, etc. All the practical
factors are ideal during the crane modeling. To reject the factors, a blend of sliding
mode control and other control methods are reported, i.e., fuzzy sliding mode
control [148151], neural network sliding mode control [154, 155], adaptive sliding
mode control [74, 156160], and observer-based sliding mode control [161].
Optimize the control command generated by sliding mode. To improve the
control performance of sliding mode control systems, sliding mode control is
sometime compound with optimization methods, i.e., optimal discrete sliding mode
control [136], optimal integral sliding mode control [152], and suboptimal sliding
mode control [153].
Owing to inherent nonlinearities, not all of these control methods for crane systems
are practically applicable because of system constraints (e.g., actuator power lim-
itation). This section reviews these issues from the points of view of theoretical and
practical challenges.
Unmatched uncertainty. Although the dynamics of cranes has been well under-
stood, practical cranes cover uncertainties, for example, parameter changes, un-
modelling errors, and external disturbances. Although the methodology of sliding
mode control is of invariance against matched uncertainties, not all the uncertainties
are as ideal as matched uncertainties. How to suppress the effects of unmatched
uncertainties on the stability of sliding mode control systems becomes crucial and
important. Vazquez et al. [139141] and Xi and Hesketh [146] explored this eld.
34 1 Introduction
They analyzed the effects of unmatched uncertainties on the stabilities of the dis-
crete sliding mode and the second-order sliding mode closed-loop control systems,
respectively. So far, other sliding mode design methods in the eld are kept
untouched and remain problematic.
Physical constraints. First, cranes are driven by actuators that are subject to their
rated powers, indicating that crane control systems have a control input saturation
constraint. Second, cranes move on rails that are subject to limited lengths and
payload angles should have their limits, meaning that variables of crane systems
have their boundary constraints. Third, not all variables of crane systems are
measurable. Usually, trolley position and its velocity are easy to measure but
payload angle and angular velocity are hard to measure, this fact hints the mea-
surability constraints of crane systems.
Guaranteed Stability. A sole sliding mode control system has guaranteed sta-
bility. However, a blend of sliding mode control and optimization methods can
complicate the system stability analysis. Especially, the stability analysis is rather
difcult under those ideas that directly optimize or plan the control command
generated by sliding mode control [152, 153].
Industrial needs. Crane systems are widely used in industry. Although crane sys-
tems have received many well-established successes in well-structured environ-
ment, their control techniques for complex uncertain environment are still
immature. However, there is always a desiderative expectation of new applications
of crane systems that requires autonomous operation in an unstructured and pos-
sibly dynamically changing environment.
Fault tolerant detection and control. Sensor failure results in delayed or missing
feedback signal which affects overall tracking performance of crane control sys-
tems. Therefore, the issue is to be considered by scientists and engineers in the
design of crane control system and a fault detection mechanism is highly recom-
mended [161].
Networked crane control systems. Another practical concern is networked
control. Once network is introduced, communication package delay or loss may
lead to poor control performance, since the communication between actuator and
sensor can be interrupted [58]. Crane control systems should be robust to network
delay, and a network control technique (e.g., predictive control) is required.
Appendices 35
Appendices
Plot program:
figure(1);
sol=dsolve('Dx1=x2','Dx2=-0.5*x1','x1(0)=1','x2(0)=0','t');
tval=(0:0.1:10)'; % column vector with t-values
yval=double(subs([sol.x1,sol.x2],'t',tval)); % 2 columns with y1,y2
plot(yval(:,1),yval(:,2),'k') % plot col.2 of yval vs. col.1 of yval
axis([-1,1,-1 1]); xlabel('x');
latexStr=['$$\dot x$$'] ;
lgh=ylabel(latexStr);
set(lgh,'interpreter','latex');
Plot program:
figure(1);
sol=dsolve('Dx1=x2','Dx2=-5.5*x1','x1(0)=1','x2(0)=0','t');
tval=(0:0.1:10)'; % column vector with t-values
yval=double(subs([sol.x1,sol.x2],'t',tval)); % 2 columns with y1,y2
plot(yval(:,1),yval(:,2),'k') % plot col.2 of yval vs. col.1 of yval
axis([-1,1,-1 1]);xlabel('x');
latexStr=['$$\dot x$$'] ;
lgh=ylabel(latexStr);
set(lgh,'interpreter','latex');
Plot program:
figure(1)
y1(1)=1;y2(1)=0;
dt=0.001;
for i=1:1:30000
if((y1(i)*y2(i))<0)
k(i)=0.5;
else
k(i)=5.5;
end
y1(i+1)=y2(i)*dt+y1(i);
y2(i+1)=y2(i)-dt*k(i)*y1(i);
end
plot(y1,y2,'k');
xlabel('x');
latexStr=['$$\dot x$$'] ;
lgh=ylabel(latexStr);
set(lgh,'interpreter','latex');
36 1 Introduction
x=-100:1:100;
y1=sign(x);
y2=x./(abs(x)+0.2);
y3=x./(abs(x)+0.5);
y4=x./(abs(x)+1);
plot(x,y1,'k-')
hold on;
plot(x,y2,'k-.')
hold on;
plot(x,y3,'k:')
hold on;
plot(x,y4,'k--');
axis([-110,110,-1.2,1.2])
legend('no deta','deta=0.2','deta=0.5','deta=1')
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 2;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 2;
sizes.NumInputs = 1;
sizes.DirFeedthrough = 0;
sizes.NumSampleTimes = 0;
sys = simsizes(sizes);
x0 = [1 0];
str = [];
ts = [];
function sys=mdlDerivatives(t,x,u)
sys(1)=x(2);
sys(2)=u(1)+0.1*rand();
function sys=mdlOutputs(t,x,u)
sys(1)=x(1);
sys(2)=x(2);
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 2;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 2;
sizes.NumInputs = 1;
sizes.DirFeedthrough = 0;
sizes.NumSampleTimes = 0;
sys = simsizes(sizes);
x0 = [1 0];
str = [];
ts = [];
42 1 Introduction
function sys=mdlDerivatives(t,x,u)
sys(1)=x(2)+0.1*rand();
sys(2)=u(1);
function sys=mdlOutputs(t,x,u)
sys(1)=x(1);sys(2)=x(2);
subplot(1,3,2),plot(t,simout(:,4))
xlabel('t');ylabel('s')
subplot(1,3,3),plot(t,simout(:,3),'k');
xlabel('t');ylabel('u');
References
1. Kurrer KE (2008) The history of the theory of structures: from arch analysis to computational
mechanics. Ernst & Sohn, Berlin
2. Lancaster L (1999) Building Trajans column. Am J Archaeol 103(3):419439
3. Vaughan J (2008) Dynamics and control of mobile cranes. Dissertation, Georgia Institute of
Technology
4. Vaughan J, Kim D, Singhose W (2010) Control of tower cranes with double-pendulum
payload dynamics. IEEE Trans Control Syst Technol 18(6):13451358
5. Utikin VI (1992) Sliding modes in control and optimization, 2nd edn. Springer, Berlin
6. Khalil HK, Grizzle JW (1996) Nonlinear systems. Prentice Hall, New Jersey
7. Edwards C, Spurgeon S (1998) Sliding mode control: theory and applications. CRC Press,
Padstow
8. Bequette BW (2003) Process control: modeling, design, and simulation. Prentice Hall, New
Jersey
9. Field JA (1961) The optimization of the performance of an ore bridge. Trans Eng Inst Canada
5(3):163169
10. Manson GA (1982) Time-optimal control of an overhead crane model. Optim Control Appl
Meth 3(2):115120
11. Kiniaghalam B, Homaifar A, Bikdash M, Dozier G (1999) Genetic algorithms solution for
unconstrained optimal crane control. In: Proceedings of the congress on evolutionary
computation, Washington, DC, USA, pp 21242130
12. Lee HH (2004) A new motion-planning scheme for overhead cranes with high-speed
hoisting. J Dyn Syst Meas Control Trans ASME 126(2):359364
References 43
13. Lee HH (2005) Motion planning for three-dimensional overhead cranes with high-speed load
hoisting. Int J Control 78(12):875886
14. Lee HH, Liang Y, Segura D (2006) A sliding-mode antiswing trajectory control for overhead
cranes with high-speed load hoisting. J Dyn Syst Meas Control Trans ASME 128(4):842
845
15. Ross IM, Fahroo F (2004) Pseudospectral methods for optimal motion planning of
differentially flat systems. IEEE Trans Autom Control 49(8):14101413
16. Blajer W, Kolodziejczyk K (2007) Motion planning and control of gantry cranes in cluttered
work environment. IET Control Theory Appl 1(5):13701379
17. Kang SC, Miranda E (2008) Computational methods for coordinating multiple construction
cranes. J Comput Civ Eng 22(4):252263
18. Zameroski D, Starr G, Wood J, Lumia R (2008) Rapid swing-free transport of non-linear
payloads using dynamic programming. J Dyn Syst Meas Control Trans ASME 130(4).
doi:10.1115/1.2936384
19. Da Cruz JJ, Leonardi F (2013) Minimum-time anti-swing motion planning of cranes using
linear programming. Optim Control Appl Meth 34(2):191201
20. AlBahnassi H, Hammad A (2012) Near real-time motion planning and simulation of cranes
in construction: framework and system architecture. J Comput Civ Eng 26(1):5463
21. Sun N, Fang YC, Zhang YD, Ma BJ (2012) A novel kinematic coupling-based trajectory
planning method for overhead cranes. IEEE ASME Trans Mechatron 17(1):166173
22. Sun N, Fang Y, Zhang X, Yuan Y (2012) Transportation task-oriented trajectory planning for
underactuated overhead cranes using geometric analysis. IET Control Theory Appl 6
(10):14101423
23. Singer N, Singhose W, Kriikku E (1997) An input shaping controller enabling cranes to
move without sway. In: Proceedings of 7th topical meeting on robotics and remote systems,
Augusta, GA, pp 225231
24. Sorensen K, Singhose W, Dickerson S (2007) A controller enabling precise positioning and
sway reduction in bridge and gantry cranes. Control Eng Pract 15(7):825837
25. Parker GG, Petterson B, Dohrmann C, Robinett RD (1995) Command shaping for residual
vibration free crane maneuvers. In: Proceedings of American control conference, Seattle,
USA, pp 934938
26. Jones S, Ulsoy AG (1999) An approach to control input shaping with application to
coordinate measuring machines. J Dyn Syst Meas Control Trans ASME 121(2):242247
27. Khalid A, Huey J, Singhose W, Lawrence J, Frakes D (2006) Human operator performance
testing using an input-shaped bridge crane. J Dyn Syst Meas Control Trans ASME 128
(4):835841
28. Alsop CF, Forster GA, Holmes ER (1965) Ore unloader automation-A feasibility study. In:
Proceedings of IFAC workshop on systems engineering for control systems, Tokyo, Japan,
pp 295305
29. Yamada S, Fujikawa H, Matsumoto K (1983) Suboptimal control of the roof crane by using
the microcomputer. In: Proceedings of the annual conference on industrial electronics, San
Francisco, USA, pp 323328
30. Lin TC (1993) Design an input shaper to reduce operation-induced vibration. In: Proceedings
of American control conference, San Francisco, USA, pp 25022506
31. Singhose W, Crain E, Seering W (1997) Convolved and simultaneous two-mode input
shapers. IEE Proc Control Theory Appl 144(6):515520
32. French L, Singhouse W, Seering W (1999) An expert system for the design of input shapers.
In: Proceedings of the 1999 IEEE international conference on control applications, Kohala
Coast-Island, USA, pp 713718
33. Singhose W, Mills B, Seering W (1999) Vibration reduction with specied-swing input
shapers. In: Proceedings of the 1999 IEEE international conference on control applications,
Kohala Coast-Island, USA, pp 533538
34. Daqaq MF, Masoud ZN (2006) Nonlinear input-shaping controller for quay-side container
cranes. Nonlinear Dyn 45(12):149170
44 1 Introduction
35. Masoud ZN, Daqaq MF (2006) A graphical approach to input-shaping control design for
container cranes with hoist. IEEE Trans Control Syst Technol 14(6):10701077
36. Sung YG, Singhose W (2009) Limited-state commands for systems with two flexible modes.
Mechatronics 19(5):780787
37. Masoud Z, Alhazza K (2014) Frequency-modulation input shaping control of
double-pendulum overhead cranes. J Dyn Syst Meas Control Trans ASME 136(2). doi:10.
1115/1.4025796
38. Hazlerigg A (1972) Automatic control of crane operations. In: Abstracts of the IFAC 5th
world congress, Paris, France, pp 1113
39. Grassin N, Retz T, Caron B, Bourles H et al (1991) Robust control of a travelling crane. In:
Proceedings of the 1st European control conference, Grenoble, France, pp 21962201
40. Lee HH (1997) Modelling and control of 2-dimensional overhead crane. In: Proceedings of
the ASME dynamic systems and control division, Dallas, USA, pp 535542
41. Nguyen HT (1994) State-variable feedback controller for an overhead crane. J Electr Electron
Eng Aust 12:7584
42. Yoon JS, Park BS, Lee JS, Park HS (1995) Various control schemes for implementation of
the anti-swing crane. In: Proceedings of the ANS 6th topical meeting on robotics and remote
systems, Monterey, USA, pp 472479
43. Hamalainen JJ, Marttinen A, Baharova L, Virkkunen J (1995) Optimal path planning for a
trolley crane: fast and smooth transfer of load. IEE Proc Control Theory Appl 142(1):5157
44. Olfati-Saber R (2001) Nonlinear control of underactuated mechanical systems with
application to robotics and aerospace vehicles. Dissertation, Massachusetts Institute of
Technology
45. Zhang X, Gao B, Chen H (2005) Nonlinear controller for a gantry crane based on partial
feedback linearization. In: Proceedings of international conference on control and
automation, Budapest, Hungary, pp 10741078
46. Tuan L, Janchiv A, Kim GH, Lee SG (2011) Feedback linearization control of overhead
cranes with varying cable length. In: Proceedings of international conference on control,
automation and systems, Gyeonggi-do, Korea, pp 906911
47. Tuan L, Kim GH, Lee SG (2012) Partial feedback linearization control of the three
dimensional overhead crane. In: Proceedings of IEEE international conference on automation
science and engineering, Seoul, Korea, pp 11981203
48. Sun N, Fang YC (2013) A partially saturated nonlinear controller for overhead cranes with
experimental implementation. In: Proceedings of IEEE international conference on robotics
and automation, Karlsruhe, Germany, pp 44734478
49. Wu XQ, He XX, Sun N, Fang YC (2014) A novel anti-swing control method for 3-D
overhead cranes. In: Proceedings of American control conference, Portland OR, USA,
pp 28212826
50. Maschke B, Ortega R, Van der Schaft AJ (2000) Energy-based Lyapunov functions for
forced Hamiltonian systems with dissipation. IEEE Trans Autom Control 45(8):14981502
51. Karkoub MA, Zribi M (2002) Modelling and energy based nonlinear control of crane lifters.
IEE Proc Control Theory Appl 149(3):209216
52. Sun N, Fang YC (2012) New energy analytical results for the regulation of underactuated
overhead cranes: an end-effector motion-based approach. IEEE Trans Ind Electron 59
(12):47234734
53. Alli H, Singh I (1999) Passive control of overhead cranes. J Vib Control 5(3):443459
54. Guo W, Liu D, Yi J, Zhao D (2004) Passivity-based-control for double-pendulum-type
overhead cranes. In: Proceedings of IEEE region 10 annual international conference, Chiang
Mai, Thailand, pp 546549
55. Collado J, Lozano R, Fantoni I (2000) Control of convey-crane based on passivity. In:
Proceedings of American control conference, Chicago IL, USA, pp 12601264
56. Thull D, Wild D, Kugi A (2006) Application of a combined flatness- and passivity-based
control concept to a crane with heavy chains and payload. In: Proceedings of IEEE
international conference on control applications, Munich, Germany, pp 656661
References 45
57. Fang Y, Dixon WE, Dawson DM, Zergeroglu E (2003) Nonlinear coupling control laws for
an underactuated overhead crane system. IEEE ASME Trans Mechatron 8(3):418423
58. Liu Y, Yu HN (2013) A survey of underactuated mechanical systems. IET Control Theory
Appl 7(7):921935
59. Kokotovic PV (1992) The joy of feedback: nonlinear and adaptive. IEEE Control Syst Mag
12(3):717
60. Kokotovic PV, Krstic M, Kanellakopoulos I (1992) Backstepping to passivity: recursive
design of adaptive systems. In: Proceedings of the 31st IEEE conference on decision and
control, Tucson AZ, USA, pp 32763280
61. d Andra-Novel B, Coron JM (2000) Exponential stabilization of an overhead crane with
flexible cable via a back-stepping approach. Automatica 36(4):587593
62. Cao LZ, Wang HW, Niu C, Wei SB (2007) Adaptive backstepping control of crane hoisting
system. In: Proceedings of IEEE international conference on automation and logistics,
Qingdao, China, pp 245248
63. Tsai CC, Wu HL, Chuang KH (2012) Backstepping aggregated sliding-mode motion control
for automatic 3D overhead cranes. In: Proceedings of IEEE/ASME international conference
on advanced intelligent mechatronics, Kaohsiung, Taiwan, pp 849854
64. Tsai CC, Wu HL, Chuang KH (2012) Intelligent sliding-mode motion control using fuzzy
wavelet networks for automatic 3D overhead cranes. In: Proceedings of 51st annual
conference of the society of instrument and control engineers of Japan, Akita, Japan,
pp 12561261
65. strm KJ, Wittenmark B (2013) Adaptive control. Courier Corporation, Chelmsford
66. Hurteau R, Desantis RM (1983) Microprocessor-based adaptive control of a crane system. In:
Proceedings of the 22nd IEEE conference on decision and control, San Antonio TX, USA,
pp 944947
67. Butler H, Honderd G, van Amerongen J (1991) Model reference adaptive control of a gantry
crane scale model. IEEE Control Syst Mag 11(1):5762
68. dAndrea-Novel B, Boustany F (1991) Adaptive control of a class of mechanical systems
using linearization and Lyapunov methods: a comparative study on the overhead crane
example. In: Proceedings of the 30th IEEE conference on decision and control, Brighton,
England, pp 120125
69. Boustany F, dAndrea-Novel B (1992) Adaptive control of an overhead crane using dynamic
feedback linearization and estimation design. In: Proceedings of 1992 IEEE international
conference on robotics and automation, Nice, France, pp 19631968
70. Yang JH, Yang KS (2006) Adaptive control for 3-D overhead crane systems. In: Proceedings
of American control conference, Minneapolis MN, USA, pp 18321837
71. Messineo S, Serrani A (2008) Offshore crane control based on adaptive external models. In:
Proceedings of American control conference, Seattle WA, USA, pp 24982503
72. Sun N, Fang Y, Chen H (2014) Adaptive control of underactuated crane systems subject to
bridge length limitation and parametric uncertainties. In: Proceedings of the 33rd Chinese
control conference, Nanjing, China, pp 35683573
73. He W, Zhang S, Ge SS (2014) Adaptive control of a flexible crane system with the boundary
output constraint. IEEE Trans Ind Electron 61(8):41264133
74. Ngo QH, Hong KS (2012) Adaptive sliding mode control of container cranes. IET Control
Theory Appl 6(5):662668
75. Fang Y, Ma B, Wang P, Zhang X (2012) A motion planning-based adaptive control method
for an underactuated crane system. IEEE Trans Control Syst Technol 20(1):241248
76. Liu D, Yi J, Zhao D, Wang W (2005) Adaptive sliding mode fuzzy control for a
two-dimensional overhead crane. Mechatronics 15(5):505522
77. Yang JH, Yang KS (2007) Adaptive coupling control for overhead crane systems.
Mechatronics 17(2):143152
78. Yu W, Moreno-Armendariz MA, Rodriguez FO (2011) Stable adaptive compensation with
fuzzy CMAC for an overhead crane. Inf Sci 181(21):48954907
79. Zhou KM, Doyle CJ (1999) Essentials of robust control. Prentice Hall, New Jersey
46 1 Introduction
80. Ackermann J (1980) Parameter space design of robust control systems. IEEE Trans Autom
Control 25(6):10581072
81. Kar IN, Seto K, Doi F (2000) Multimode vibration control of a flexible structure using H-
based robust control. IEEE ASME Trans Mechatron 5(1):2331
82. Yang TW, OConnor WJ (2006) Wave based robust control of a crane system. In:
Proceedings of 2006 IEEE/RSJ international conference on intelligent robots and systems,
Beijing, China, pp 27242729
83. Uchiyama N, Takagi S, Sano S (2006) Robust control of rotary cranes based on pole
placement approach. In: Proceedings of the 9th IEEE international workshop on advanced
motion control, Istanbul, Turkey, pp 647652
84. Toda M (2007) A unied approach to robust control of flexible mechanical systems. In:
Proceedings of the 46th IEEE conference on decision and control, New Orleans LA, USA,
pp 57875793
85. Uchiyama N (2009) Robust control of rotary crane by partial-state feedback with integrator.
Mechatronics 19(8):12941302
86. Wen SJ, Deng MC, Ohno Y, Wang DY (2011) Operator-based robust right coprime
factorization design for planar gantry crane. In: Proceedings of 2010 international conference
on mechatronics and automation, Xian, China, pp 15
87. Wen SJ, Deng MC, Wang DY (2011) Operator-based robust nonlinear control for a crane
system with constraint inputs. In: Proceedings of the 30th Chinese control conference,
Yantai, China, pp 61096114
88. Wen SJ, Deng MD, Inoue A (2012) Operator-based robust non-linear control for gantry crane
system with soft measurement of swing angle. Int J Model Ident Control 16(1):8696
89. Sano S, Ouyang H, Uchiyama N (2012) Robust control of rotary cranes under rope length
variance via LMI optimization. In: Proceedings of 2012 IEEE international conference on
industrial technology, Busan, Korea, pp 846851
90. Ushida Y, Narita M, Ushiro T, Chen G, Takami I (2014) Robust control for crane
considering all varying parameters in the dynamics. In: Proceedings of the 11th IEEE
international conference on control and automation, Taichung, Taiwan, pp 785790
91. Camacho EF, Bordons C (2004) Model predictive control. Springer, Berlin
92. Kimiaghalam B, Homaifar A, Sayarrodsari, B (2001) An application of model predictive
control for a shipboard crane. In: Proceedings of American control conference,
Arlington VA, USA, pp 929934
93. Deng JM, Becerra VM (2004) Application of constrained predictive control on a 3D crane
system. In: Proceedings of IEEE conference on robotics, automation and mechatronics,
Singapore, pp 583587
94. Arnold E, Sawodny O, Neupert J, Schneider K (2005) Anti-sway system for boom cranes
based on a model predictive control approach. In: Proceedings of IEEE international
conference on mechatronics and automation, Niagara Falls, Canada, pp 15331538
95. Van den Broeck L, Diehl M, Swevers J (2011) A model predictive control approach for time
optimal point-to-point motion control. Mechatronics 21(7):12031212
96. Kapernick B, Graichen K (2013) Model predictive control of an overhead crane using
constraint substitution. In: Proceedings of American control conference, Washington DC,
USA, pp 39733978
97. Schaper U, Arnold E, Sawodny O, Schneider K (2013) Constrained real-time
model-predictive reference trajectory planning for rotary cranes. In: Proceedings of
IEEE/ASME international conference on advanced intelligent mechatronics, Wollongong,
Australia, pp 680685
98. Bauer D, Schaper U, Schneider K, Sawodny O (2014) Observer design and flat-ness-based
feedforward control with model predictive trajectory planning of a crane rotator. In:
Proceedings of American control conference, Portland OR, USA, pp 40204025
99. Barisa T, Bartulovic M, Zuzic G, Iles S, Matusko J, Kolonic F (2014) Nonlinear predictive
control of a tower crane using reference shaping approach. In: Proceedings of the 16th
References 47
international power electronics and motion control conference and exposition, Antalya,
Turkey, pp 872876
100. Iles S, Matusko J, Kolonic F (2014) Real-time predictive control of 3D tower crane. In:
Proceedings of IEEE 23rd international symposium on industrial electronics, Istanbul,
Turkey, pp 224230
101. Khatamianfar A, Savkin AV (2014) A new tracking control approach for 3D over-head crane
systems using model predictive control. In: Proceedings of European control conference,
Strasbourg, France, pp 796801
102. Bock M, Kugi A (2014) Real-time nonlinear model predictive path-following control of a
laboratory tower crane. IEEE Trans Control Syst Technol 22(4):14611473
103. Stengel RF (1991) Intelligent failure tolerant control. IEEE Control Syst Mag 11(4):1423
104. Antsaklis PJ, Passino KM (1993) An introduction to intelligent and autonomous control.
Kluwer Academic Publishers, Dordrecht
105. Michels K, Klawonn F, Kruse R, Numberger A (2006) Fuzzy control: fundamentals, stability
and design of fuzzy controllers. Springer, New York
106. Takeuchi S, Fujikawa H, Yamada S (1988) The application of fuzzy theory for a rotary crane
control. In: Proceedings of the 14th annual conference of IEEE industrial electronics society,
Singapore, pp 415420
107. Yamada S, Fujikawa H, Takeuchi O, Wakasugi Y (1989) Fuzzy control of the roof crane. In:
Proceedings of the 15th annual conference of IEEE industrial electronics society,
Philadelphia PA, USA, pp 709714
108. Suzuki Y, Yamada S, Fujikawa H (1993) Anti-swing control of the container crane by fuzzy
control. In: Proceedings of international conference on industrial electronics, control, and
instrumentation, Maui HI, USA, pp 230235
109. Itoh O, Migita H, Itoh J, Irie Y (1993) Application of fuzzy control to automatic crane
operation. In: Proceedings of international conference on industrial electronics, control, and
instrumentation, Maui HI, USA, pp 161164
110. Nowacki Z, Owczarz D, Wozniak P (1996) On the robustness of fuzzy control of an overhead
crane. In: Proceedings of the IEEE international symposium on industrial electronics,
Warsaw, Japan, pp 433437
111. Yi J, Yubazaki N, Hirota K (2002) Anti-swing fuzzy control of overhead traveling crane. In:
Proceedings of the 2002 IEEE international conference on fuzzy systems, Honolulu HI,
USA, pp 12981303
112. Kijima Y, Ohtsubo R, Yamada S, Fujikawa H (1995) An optimization of fuzzy controller and
its application to overhead crane. In: Proceedings of international conference on industrial
electronics, control, and instrumentation, Orlando FL, USA, pp 15081513
113. Benhidjeb A, Gissinger GL (1995) Fuzzy control of an overhead crane performance
comparison with classic control. Control Eng Pract 3(12):16871696
114. Al-Humaidi HM, Hadipriono Tan F (2009) Mobile crane safe operation approach to prevent
electrocution using fuzzy-set logic models. Adv Eng Softw 40(8):686696
115. Chen YJ, Wang WJ, Chang CL (2009) Guaranteed cost control for an overhead crane with
practical constraints: Fuzzy descriptor system approach. Eng Appl Artif Intell 22:639645
116. Smoczek J, Szpytko J (2014) Evolutionary algorithm-based design of a fuzzy TBF predictive
model and TSK fuzzy anti-sway crane control system. Eng Appl Artif Intell 28:190200
117. Precup RE, Filip HI, Rdac MB et al (2014) Online identication of evolving Takagi
SugenoKang fuzzy models for crane systems. Appl Soft Comput J 24:11551163
118. Liu D, Yi J, Tan M (2002) Proposal of GA-based two-stage fuzzy control of over-head crane.
In: Proceedings of 2002 IEEE region 10 conference on computer, communications, control
and power engineering, Beijing, China, pp 17211724
119. Petrenko YN, Alavi SE (2010) Fuzzy logic and genetic algorithm technique for non-liner
system of overhead crane. In: Proceedings of 2010 IEEE region 8 international conference on
computational technologies in electrical and electronics engineering, Irkutsk Listvyanka,
Russia, pp 848851
48 1 Introduction
120. Adeli M, Zarabadi SH, Zarabadipour H, Aliyari Shoorehdeli M (2011) Design of a parallel
distributed fuzzy LQR controller for double-pendulum-type overhead cranes. In: Proceedings
of IEEE international conference on control system, computing and engineering, Penang,
Malaysia, pp 6267
121. Adeli M, Zarabadipour H, Zarabadi SH, Aliyari Shoorehdeli M (2011) Anti-swing control
for a double-pendulum-type overhead crane via parallel distributed fuzzy LQR controller
combined with genetic fuzzy rule set selection. In: Proceedings of IEEE international
conference on control system, computing and engineering, Penang, Malaysia, pp 306311
122. Adeli M, Zarabadipour H, Shoorehdeli MA (2011) Crane control via parallel distributed
fuzzy LQR controller using genetic fuzzy rule selection. In: Proceedings of international
conference on control, instrumentation and automation, Shiraz, Iran, pp 390395
123. Adeli M, Zarabadipour H, Shoorehdeli MA (2011) Anti-swing control of a
double-pendulum-type overhead crane using parallel distributed fuzzy LQR controller. In:
Proceeding of international conference on control, instrumentation and automation, Shiraz,
Iran, pp 401406
124. Solihin MI, Wahyudi (2007) Fuzzy-tuned PID control design for automatic gantry crane. In:
Proceedings of 2007 international conference on intelligent and advanced systems, Kuala
Lumpur, Malaysia, pp 10921097
125. Yang Y (2008) Fuzzy-PI damping control for hydraulic crane tip. In: Proceedings of 5th
international conference on fuzzy systems and knowledge discovery, Jinan, China, pp 7579
126. Lin F, Zhang X, Zhai X, Huang H (2011) The application study of heave compensation
control based on motion prediction and fuzzy-PID for intelligence crane. In: Proceedings of
international conference on mechatronics and automation, Beijing, China, pp 23702374
127. Liu C, Zhao H, Cui Y (2014) Research on application of fuzzy adaptive PID controller in
bridge crane control system. In: Proceedings of IEEE international conference on software
engineering and service science, Beijing, China, pp 971974
128. Chang CY (2007) Adaptive fuzzy controller of the overhead cranes with nonlinear
disturbance. IEEE Trans Ind Inf 3(2):164172
129. Park MS, Chwa D, Hong SK (2007) Adaptive fuzzy nonlinear anti-sway trajectory tracking
control of uncertain overhead cranes with high-speed load hoisting motion. In: Proceedings
of international conference on control, automation and systems, Seoul, South Korea,
pp 28862891
130. Pal AK, Mudi RK (2012) An adaptive fuzzy controller for overhead crane. In: Proceedings of
IEEE international conference on advanced communication control and computing
technologies, Ramanathapuram, India, pp 300304
131. Smoczek J, Szpytko J (2013) Fuzzy logic-based adaptive control system prototypying for
laboratory scaled overhead crane. In: Proceedings of 18th international conference on
methods and models in automation and robotics, Midzyzdroje, Poland, pp 9297
132. Smoczek J, Szpytko J (2012) Design of gain scheduling anti-sway crane controller using
genetic fuzzy system. In: Proceedings of 17th international conference on methods and
models in automation and robotics, pp 573578
133. Sharkawy AB, Moustafa K, El-Awady H et al (2014) Control of overhead crane based on
Lyapunov approach to fuzzy controller synthesis. In: Proceedings of 23rd international
conference on robotics in Alpe-Adria-Danube region, Smolenice Castle, Slovakia, pp 16
134. Smoczek J, Szpytko J (2014) Iterative and evolutionary optimization for interval
analysis-based designing a fuzzy controller for a planar crane model. In: Proceedings of
19th international conference on methods and models in automation and robotics,
Midzyzdroje, Poland, pp 258263
135. Pieper JK, Surgenor BW (1994) Discrete time sliding mode control applied to a gantry crane.
In: Proceedings of 33rd IEEE conference on decision and control, Lake Buena Vista FL,
USA, pp 829834
136. Pieper JK, Surgenor BW (1994) Discrete time sliding mode control applied to a gantry crane.
In: Proceedings of 33rd IEEE conference on decision and control, Lake Buena Vista FL,
USA, pp 829834
References 49
137. Bartolini G, Orani N, Pisano A, Usai E (2000) Load swing damping in overhead cranes by
sliding mode technique. In: Proceedings of 39th IEEE conference on decision and control,
Sydney, Australia, pp 16971702
138. Bartolini G, Pisano A, Usai E (2002) Second-order sliding-mode control of container cranes.
Automatica 38(10):17831790
139. Vazquez C, Collado J, Fridman L (2012) On the second order sliding mode control of a
parametrically excited overhead-crane. In: Proceedings of American control conference,
Montral, Canada, pp 62886293
140. Vazquez C, Fridman L, Collado J (2013) Second order sliding mode control of an
overhead-crane in the presence of external perturbations. In: Proceedings of IEEE 52nd
annual conference on decision and control, Florence, Italy, pp 28762880
141. Vazquez C, Aranovskiy S, Freidovich L, Fridman L (2014) Second order sliding mode
control of a mobile hydraulic crane. In: Proceedings of IEEE 53rd annual conference on
decision and control, Los Angeles CA, USA, pp 55305535
142. Chen WT, Saif M (2008) Output feedback controller design for a class of MIMO nonlinear
systems using high-order sliding-mode differentiators with application to a laboratory 3-D
crane. IEEE Trans Ind Electron 55(11):39853997
143. Dong Y, Wang Z, Feng Z, Cheng J (2008) Incremental sliding mode control for
double-pendulum-type overhead crane system. In: Proceedings of 27th Chinese control
conference, Kunming, China, pp 368371
144. Cao L, Li XB, Lou FP, et al (2010) Controller design and simulation of the crane based on
non-singular terminal sliding mode method. In: Proceedings of international conference on
logistics systems and intelligent management, Harbin, China, pp 484487
145. Defoort M, Maneeratanaporn J, Murakami T (2012) Integral sliding mode anti-sway control
of an underactuated overhead crane system. In: Proceedings of 9th France-Japan and 7th
Europe-Asia congress on mechatronics and research and education in mechatronics, Paris,
France, pp 7177
146. Xi Z, Hesketh T (2010) Discrete time integral sliding mode control for overhead crane with
uncertainties. IET Control Theory Appl 4(10):20712081
147. Tuan LA, Lee SG (2013) Sliding mode controls of double-pendulum crane systems. J Mech
Sci Technol 27(6):18631873
148. Chang CY, Hsu KC, Chiang KH, Huang GE (2006) An enhanced adaptive sliding mode
fuzzy control for positioning and anti-swing control of the overhead crane system. In:
Proceedings of IEEE international conference on systems, man and cybernetics, Taipei,
Taiwan, pp 992997
149. Liu DT, Yi JQ, Zhao DB (2003) Fuzzy tuning sliding mode control of transporting for an
overhead crane. In: Proceedings of international conference on machine learning and
cybernetics, Xian, China, pp 25412546
150. Liu DT, Yi JQ, Zhao DB, Wang W (2004) Swing-free transporting of two-dimensional
overhead crane using sliding mode fuzzy control. In: Proceedings of American control
conference, Boston MA, USA, pp 17641769
151. Sun H, Chen ZM, Meng WJ (2012) Fuzzy sliding mode anti-swing control for tower crane
base on time-delayed lter. In: Proceedings of 24th Chinese control and decision conference,
Taiyuan, China, pp 22052210
152. Liu R, Li S, Chen X (2013) An optimal integral sliding mode control design based on
pseudospectral method for overhead crane systems: In: Proceedings of 32nd Chinese control
conference, Xian, China, pp 21952200
153. Liu R, Li S (2014) Suboptimal integral sliding mode controller design for a class of afne
systems. J Optim Theory Appl 161(3):877904
154. Wang W, Yi JQ, Zhao DB, Liu XJ (2004) Incremental neural network sliding mode
controller for an overhead crane. In: Proceedings of international conference on intelligent
mechatronics and automation, Chengdu, China, pp 166171
50 1 Introduction
155. Tsai CC, Wu HL, Chuang KH (2012) Intelligent sliding-mode motion control using fuzzy
wavelet networks for automatic 3D overhead cranes. In: Proceedings of 51st annual
conference of the society of instrument and control engineers of Japan, Akita, Japan,
pp 12561261
156. Tuan LA, Moon SC, Kim DH, Lee SG (2012) Adaptive sliding mode control of three
dimensional overhead cranes. In: Proceedings of IEEE international conference on cyber
technology in automation, control, and intelligent systems, Bangkok, Thailand, pp 354359
157. Tuan LA, Moon SC, Lee WG, Lee SG (2013) Adaptive sliding mode control of overhead
cranes with varying cable length. J Mech Sci Technol 27(3):885893
158. Qian DW, Tong SW, Yi JQ (2013) Adaptive control based on incremental hierarchical
sliding mode for overhead crane systems. Appl Math Inform Sci 7(4):13591364
159. Tuan LA, Lee SG, Nho LC, Kim DH (2013) Model reference adaptive sliding mode control
for three dimensional overhead cranes. Int J Precis Eng Manuf 14(8):13291338
160. Park MS, Chwa D, Eom M (2014) Adaptive sliding-mode antisway control of uncertain
overhead cranes with high-speed hoisting motion. IEEE Trans Fuzzy Syst 22(5):12621271
161. Dhahri S, Hmida FB, Sellami A, Gossa M (2009) Actuator fault reconstruction for linear
uncertain systems using sliding mode observer. In: Proceedings of 3rd international
conference on signals, circuits and systems, Medenine, Tunisia, pp 16
Chapter 2
Crane Mathematic Model
2.1.1 Modeling
Figure 2.1 shows the coordinate system of an overhead crane system with its
payload. Apparently, the crane system consists of two subsystems, i.e., trolley and
payload [1]. The former is driven by a force. The latter is suspended from the trolley
by a rope.
Other symbols in Fig. 2.1 are described as the trolley mass M, the payload mass
m, the rope length L, the swing angle of the payload with respect to the vertical line
, the trolley position with respect to the origin x, and the driven force applied to the
trolley f.
Consider that the crane in Fig. 2.1 is static and the payload is in its downward
position. If the trolley moves toward the right direction by a positive driven force,
then the payload will rotate clockwise. Apparently, the payload angle is inherently
1 1
K M x_ 2 mv2 2:2
2 2
where vx x_ Lh_ cos h and vy Lh_ sin h. Note that the payload is assumed to be
a particle such that its moment of inertia is not considered in (2.2). When it is failed
2.1 Modeling of Single-Pendulum-Type Cranes 53
to assume that the payload is a particle, its moment of inertia has to be taken into
considerations.
From Fig. 2.1, the potential energy of the trolley subsystem is kept unchanged.
Owing to this fact, the system potential energy in Fig. 2.1 is only exhibited by the
potential energy of the payload subsystem, dened as
Here, g is the gravitational acceleration. From (2.2) and (2.4), La has the form
1 1
La K P M x_ 2 mv2 mgL1 cos h 2:5
2 2
@La
0 2:6
@x
@La
M x_ m_x Lh_ cos h 2:7
@ x_
@La
m_x Lh_ cos hLh_ sin h
@h
Lh_ sin hLh_ cos h mgL sin h 2:10
@La
m_x Lh_ cos hL cos h Lh_ sin hLh_ sin h
@ h_ 2:11
mL_x cos h mL2 h_
54 2 Crane Mathematic Model
From (2.9) and (2.13), the dynamic model [4] of this overhead crane system with
respect to x and can be obtained by means of the Lagrangian method.
::
m M x mLh cos h h_ 2 sin h f 2:14
::
x cos h Lh g sin h 0 2:15
Further, the above dynamic model composed of (2.14) and (2.15) can be
transformed to the following state space model [5], formulated as
x_ 1 x2
x_ 2 f1 x b1 xu
2:16
x_ 3 x4
x_ 4 f2 x b2 xu
1
b1 x
M m sin2 x3
Note that the model (2.16) is ideal and it contains no uncertainties. Due to imperfect
modeling and effects of environment, it is impossible to avoid uncertainties and external
disturbance in real dynamical systems. In reality, overhead crane systems often are
operated under uncertainty conditions such as parameter variations, unmodeled
dynamics, skidding and slipping, etc. Considering the possible effects of these uncer-
tainties, the dynamic model of the overhead crane in Fig. 2.1 can have the form
x_ 1 x2
x_ 2 f10 x b10 xu
2:17
x_ 3 x4
x_ 4 f20 x b20 xu
Dfi0 x bi xD~fi x
Dbi0 x bi xD~bi x
Apparently, all the uncertainties in (2.19) enter the dynamic model (2.17) by the
control tunnel, indicating that they are matched. Such an entering tunnel makes this
kind of uncertainties resistible by suitable control methods.
In the case that there are unmatched uncertainties, it is challenging to suppress
them because it is hard to formulate these kinds of uncertainties. A common
56 2 Crane Mathematic Model
x_ 1 x2
x_ 2 f1 x b1 xu n1 x; u
2:20
x_ 3 x4
x_ 4 f2 x b2 xu n2 x; u
The nonlinear single-pendulum-type overhead crane model has been discussed and
it is composed of (2.14) and (2.15). Because = 0 is the sole stable equilibrium of
the overhead crane system, both of the equations can be linearized around the point.
The linearized equations can be written as
:: ::
m M x mL h f 2:21
:: ::
x L h gh 0 2:22
::
According to Newtons second law, f = (m + M) x can be obtained. Substituting
::
x f =m M into (2.21) yields
f ::
L h gh 0 2:23
Mm
Equation (2.24) reveals the system frequency depending on the rope length and the
mass ratio. It is of interest to investigate how the frequency changes as a function of
the system parameters. Such information can be used for physical insights of the
overhead crane system. Figure 2.2 demonstrates the function revealed in (2.24).
The MATLAB programs of the example are given in Appendix A.
From Fig. 2.2, the frequency changes very little with respect to the mass ratio
when the rope length is more than 4 m. On the other hand, the frequency value has a
strong dependence on the mass ratio when the rope length is short.
2.1.4.1 Modeling
d @Lad @Lad
Ti 2:25
dt @ q_ i @qi
1 1 1
Kd m0 x_ 2 m1 v21 m2 v22 2:26
2 2 2
here, the vectors v1 and v2 denote the hook and payload velocities, respectively.
They are dened as
Lad Kd Pd
1 1 1
m0 x_ 2 m1 v21 m2 v22 2:29
2 2 2
m1 gl1 1 cos h1 m2 gl1 1 cos h1 l2 1 cos h2
@Lad
0 2:30
@x
@Lad
m0 m1 m2 _x m1 m2 l1 h_ 1 cos h1 m2 l2 h_ 2 cos h2 2:31
@ x_
m2 l2 h_ 22 sin h2
@Lad
m1 m2 l1 x_ h_ 1 sin h1
@h1 2:34
m2 l1 l2 h_ 1 h_ 2 sinh1 h2 m1 m2 gl1 sin h1
@Lad
m1 m2 l1 x_ cos h1 l21 h_ 1 m2 l1 l2 h_ 2 cosh1 h2 2:35
_
@ h1
60 2 Crane Mathematic Model
d @Lad @Lad
m1 m2 l1x cos h1 m1 m2 l21 h1
dt @ h_ 1 @h1
m2 l1 l2 h2 cosh1 h2 m2 l1 l2 h_ 2 sinh1 h2
2 2:37
m1 m2 gl1 sin h1
=0
@Lad
m2 l1 x_ h_ 2 sin h2 m2 l1 l2 h_ 1 h_ 2 sinh1 h2 m2 gl2 sin h2 2:38
@h2
@Lad
m2 l1 x_ cos h1 m2 l22 h_ 2 m2 l1 l2 h_ 1 cosh1 h2 2:39
@ h_ 2
From (2.33), (2.37), and (2.41), the dynamic model of the double-pendulum-type
overhead crane system with respect to x, 1, and 2 [8] can be obtained by means of
the Lagrangian method.
m0 m1 m2 x m1 m2 l1 h1 cos h1 m2 l2 h2 cos h2
2:42
m1 m2 l1 h_ 21 sin h1 m2 l2 h_ 22 sin h2 F
Rearrange (2.42), (2.43), and (2.44) in the form of a matrix. The three equations
can be rewritten as
::
Mq q Cq; q_ q_ Gq s 2:45
Further, (2.45) can be transformed to its state space expression. The expression
[9, 10] has the form
62 2 Crane Mathematic Model
x_ 1 x2
x_ 2 f1 x b1 xu
x_ 3 x4
2:46
x_ 4 f2 x b2 xu
x_ 5 x6
x_ 6 f3 x b3 xu
In (2.46), the vector x is dened by [x1, x2, x3, x4, x5, x6]T; x1 = x; x3 = 1; x5 = 2;
x2 is the trolley velocity; x4 is the angular velocity of the hook; x6 is the angular
velocity of the payload; u = F is the control input; and fi(x) and bi(x) (i = 1, 2, 3) are
nonlinear functions of the vector x, formulated by fi (x) = i/ and bi(x) = i/.
Here i, i, and are determined by
D = m1 m2 m2 l21 l22 m0 m1 m2 m1 m2 cos2 x3
m22 l21 l22 m1 m2 cos2 x5 m0 m1 m2 cos2 x3 x5
2m1 m2 cosx3 cosx5 cosx3 x5
C1 m1 m2 m2 l21 l22 m22 l21 l22 cos2 x3 x5 m1 m2 l1 x24 sinx3
m2 l2 x26 sinx5 m1 m2 m2 l1 l22 cosx3
m22 l1 l22 cosx5 cosx3 x5 m2 l1 l2 x26 sinx3 x5
m1 m2 gl1 sinx3 m1 m2 m2 l21 l2 cosx5
m2 l21 l2 cosx3 cosx3 x5 m2 l1 l2 x24 sinx3 x5 m2 gl2 sinx5
Note that the model (2.45) has two important assumptions that are mass-point
hook and payload. Usually, the mass-point assumption can be satised for the hook
subsystem. However, the mass-point assumption for the payload subsystem can be
satised under some operating conditions. Concerning these extreme operating
conditions, the moment of inertia of the payload subsystem cannot be ignored and it
has to be taken into consideration.
x_ 1 x2
x_ 2 f1 x b1 xu n1 x; u
x_ 3 x4
2:47
x_ 4 f2 x b2 xu n2 x; u
x_ 5 x6
x_ 6 f3 x b3 xu n3 x; u
q:: Kq 0
M 2:48
From (2.49) and (2.50), the two natural frequencies not only depend on the
length of the cables but also depend on the masses of payload and hook. It is
interesting to investigate how the frequencies change as a function of the system
physical parameters.
To simplify this problem, R = m2/m1 is dened as the payload-to-hook mass
ratio, the cable length between hook and payload l2 is considered as a variable when
the total length l determined by l1 plus l2 is held constant at 6 m. Figure 2.4
illustrates the two oscillation frequencies as a function of R and l2.
In Fig. 2.4, 1 changes very little for a constant l1 + l2. It corresponds closely to
the frequency of a single pendulum with the length of l1 + l2. On the other hand, the
value of 1 is maximized for a constant l1 + l2 when the two cable lengths are equal
to l1 l
2 ; but it can be dramatically changed by the hoisting operation [12].
2
2.1 Modeling of Single-Pendulum-Type Cranes 65
Appendices
References
1. Abdel-Rahman EM, Nayfeh AH, Masoud ZN (2003) Dynamics and control of cranes: a
review. J Vib Control 9(7):863908
2. Lee HH (1998) Modeling and control of a three-dimensional overhead crane. J Dyn Syst Meas
Control Trans ASME 120(4):471476
3. Spong MW, Hutchinson S, Vidyasagar M (2006) Robot modeling and control. Wiley, New
York
4. Oguamanam D, Hansen JS, Heppler G (2001) Dynamics of a three-dimensional overhead
crane system. J Sound Vib 242(3):411426
5. Liu D, Yi J, Zhao D, Wang W (2005) Adaptive sliding mode fuzzy control for a
two-dimensional overhead crane. Mechatronics 15(5):505522
6. Cheng C, Chen CY (1996) Controller design for an overhead crane system with uncretainty.
Control Eng Pract 4(5):645653
7. Park MS, Chwa D, Hong SK (2008) Antisway tracking control of overhead cranes with
system uncertainty and actuator nonlinearity using an adaptive fuzzy sliding-mode control.
IEEE Trans Ind Electron 55(11):39723984
8. Tuan LA, Lee SG (2013) Sliding mode controls of double-pendulum crane systems. J Mech
Sci Technol 27(6):18631873
9. Liu D, Guo W, Yi J (2008) Dynamics and GA-based stable control for a class of underactuated
mechanical systems. Int J Control Autom Syst 6(1):3543
10. OConnor W, Habibi H (2013) Gantry crane control of a double-pendulum, distributed-mass
load using mechanical wave concepts. Mech Sci 4:251261
11. Vaughan J, Kim D, Singhose W (2010) Control of tower cranes with double-pendulum
payload dynamics. IEEE Trans Control Syst Technol 18(6):13451358
12. Singhose W, Kim D, Kenison M (2008) Input shaping control of double-pendulum bridge
crane oscillations. J Dyn Syst Meas Control Trans ASME 130(3): doi:10.1115/1.2907363
Chapter 3
Overhead Crane Control by Sliding Mode
Methods
Overhead, crane control includes several control problems [1], i.e., trolley posi-
tioning, payload swing reduction, and crane scheduling, where the crane scheduling
problem should resort to optimization of crane operations and the other two control
problems can be dealt with control algorithms.
Concerning transport control of a crane system, its control task [2] is to transport
the payload to a desired position as accurately and quickly as possible. Meanwhile,
its control system should suppress payload oscillations no matter what kind of
pendulum-type they are. The control task has two aspects, i.e., fast crane posi-
tioning and small payload swing. Inherently, the two aspects contradict each other.
The physical properties of cranes cannot simultaneously realize the two aspects. To
solve the inherent contradiction and achieve the transport task of cranes, a control
mechanism or algorithm has to be applied to crane systems.
As mentioned, the task of the transport control problem fuses both the trolley
positioning and the payload swing reduction. As a result, the transport control
problem of crane systems is representative. In this book, overhead crane control just
means the transport control problem without independent claim.
Since the sliding mode control (SMC) methodology is a feedback control design,
there are two ways to design a SMC system [3], i.e., output feedback design and
state feedback design, where the SMC design based on state feedback is adopted in
Chap. 2. Provided that all the crane states are measurable, the state feedback-based
SMC design is adopted in the rest of the book without independent claim.
Among a diversity of sliding mode design methods, the rst-order SMC method is a
basic one, but it is important and representative. To illustrate its design, the
rst-order SMC method in Sect. 3.2 is applied to two overhead cranes. The cranes
cover both single-pendulum-type and double-pendulum-type crane systems. Recall
Chap. 2. The mathematic models of the two types of cranes have been established.
In this chapter, they are directly adopted for control design without more
explanations.
s cT x 3:1
where s is the sliding surface variable; x in (2.16) is the state vector; and c is the
parameter vector of this sliding surface and it should be predened.
Recall the equivalent-control-based SMC law (1.20) in Chap. 1. Owing to its
advantages, the design of the equivalent-control-based SMC is adopted in the rest
of the book. Design the control law of the crane system (1.20), described by the
equivalent control law ueq plus the switching control law usw.
To obtain its equivalent control law, differentiating s in (3.1) with respect to time
t and substituting the crane model (2.16) into the derivative of s yield
s_ cT x_ c1 x_ 1 c2 x_ 2 c3 x_ 3 c4 x_ 4
c1 x2 c2 f1 x b1 xu c3 x4 c4 f2 x b2 xu 3:2
c1 x2 c3 x4 c2 f1 x c4 f2 x c2 b1 xu c4 b2 xu
3.2 First-Order Sliding Mode Control 69
When the system states keep sliding along the sliding surface (3.1), the sole
equivalent control law is applied to the crane control system and s_ 0 exists in
(3.2). These linguistic descriptions are formulated by
s_ c1 x2 c3 x4 c2 f1 x c4 f2 x c2 b1 xu c4 b2 xueq 0 3:3
c1 x2 c3 x4 c2 f1 x c4 f2 x
ueq 3:4
c2 b1 xu c4 b2 x
Compared with (1.20), only the equivalent control law ueq is obtained. The
switching control law usw is still kept unknown. To obtain the whole control law u,
the switching control law will be deduced from the system stability analysis.
Concerning the crane control system based on the rst-order SMC, a Lyapunov
function candidate in the Lyapunovs stability scheme can be dened by
1
Vt s2 3:5
2
Differentiating V with respect to time t and substituting the crane model (1.16),
the sliding surface (3.1), and the equivalent-control-based SMC law (1.20) into the
derivative of V yield
_
Vt s_s sc1 x_ 1 c2 x_ 2 c3 x_ 3 c4 x_ 4
sfc1 x_ 1 c2 f1 x b1 xu c3 x_ 3 c4 f2 x b2 xug
sfc1 x2 c3 x4 c2 f1 x c4 f2 x c2 b1 x c4 b2 xug
sfc1 x2 c3 x4 c2 f1 x c4 f2 x c2 b1 x c4 b2 xueq usw g 3:6
Consider the expression of the equivalent control law (3.4). Substituting it into
(3.6) yields
_
Vt susw c2 b1 x c4 b2 x 3:7
_
From (3.5), Vt 0. In the sense of Lyapunov, Vt\0 should exist to make
the crane control system asymptotically stable. As a result, dene
70 3 Overhead Crane Control by Sliding Mode Methods
js g sgns
usw 3:8
c2 b1 x c4 b2 x
u ueq usw
c1 x2 c3 x4 c2 f1 x c4 f2 x js g sgns 3:9
c2 b1 xu c4 b2 x c2 b1 x c4 b2 x
_
Vt s_s sc1 x_ 1 c2 x_ 2 c3 x_ 3 c4 x_ 4
sfc1 x_ 1 c2 f1 x b1 xu n1 x; u
c3 x_ 3 c4 f2 x b2 xu n2 x; ug
3:10
sfc1 x2 c3 x4 c2 f1 x c4 f2 x
c2 b1 x c4 b2 xu n1 x; u n2 x; ug
sfc2 b1 x c4 b2 xusw n1 x; u n2 x; ug
_
Vt 0 in (3.5). In the sense of Lyapunov, Vt\0 should exist to make the
uncertain crane control system asymptotically stable. As a result, dene
j s g sgns
usw 3:11
c2 b1 x c4 b2 x
Finally, the rst-order SMC law of the crane system with unmatched certainties
can be deduced from (3.4) and (3.11):
3.2 First-Order Sliding Mode Control 71
u ueq usw
c1 x2 c3 x4 c2 f1 x c4 f2 x j s g sgns 3:12
c2 b1 xu c4 b2 x c2 b1 x c4 b2 x
Compared with (3.8), the control law for crane systems with unmatched
uncertainties (3.11) needs a larger coefcient of the signum function to resist the
adverse effects of unmatched uncertainties on the system stability.
To demonstrate the control performance of the rst-order SMC law for nominal
overhead crane systems, some simulation results will be displayed and some
analyses will be presented. Physical parameters of the overhead crane system are
shown in Table 3.1 [4], so are the initial and desired state vectors. The
equivalent-control-based SMC law (3.9) is adopted, where the controller parameters
are determined by = 4, = 0.05, and c = [3 3 10.3 1]T after trial and error. Note
that the physical parameters of the crane system in Table 3.1 are very small
compared with a real crane. Here, they are just employed for the purpose of
illustration. In practice, they can be scaled up.
From Figs. 3.1 and 3.2, the crane system spends about 3 s arriving at the desired
position in Fig. 3.1a. The rst-order SMC law can effectively suppress the
single-pendulum-type motions of the payload in Fig. 3.1b when the control task is
achieved. From Fig. 3.2a, the rst-order SMC law is composed of the equivalent
control law in Fig. 3.2c and the switching control law in Fig. 3.2d. The three curves
show that the equivalent-control-based SMC method can effectively decrease the
chattering. Figure 3.2b displays the curve of the sliding surface s. In Fig. 3.2b, the
sliding motion begins at about 1 s. Thereafter, the switching control law is kept
zeroth in Fig. 3.2d and the sole equivalent control law forces the sliding mode along
the sliding surface. Figures 3.1 and 3.2 can be obtained by a Simulink model of
MATLAB, which is shown in Appendix A.
Fig. 3.1 System performance by the rst-order SMC method for the overhead crane. a Trolley
position; b Trolley velocity; c Payload angle; d Payload angular velocity
Fig. 3.2 Control performance of the rst-order SMC method. a Control input u; b Sliding surface
s; c Equivalent control ueq; d Switching control usw
3.2 First-Order Sliding Mode Control 73
Recall the crane model with unmatched uncertainties (2.19). Assume that both the
uncertain terms in the model are determined by 0.1 [12 rand()], where rand() is
a MATLAB command to generate a uniformly distributed pseudorandom number
on the open interval (0, 1).
For the purpose of comparison, the parameter vector of the sliding surface c is
kept unchanged. The physical parameters of the crane, the initial state vector, and
the desired state vector are also kept unchanged from Table 3.1. To make the crane
control system asymptotically stable, * is set to 0.25 because sup jjn1 x; u
n2 x; ujj\ 0:2 and the other parameter * = = 4.
The Simulink model of MATLAB to plot Figs. 3.3 and 3.4 is very similar to that
of plotting Figs. 3.1 and 3.2. Figure 3.3 shows the crane performance though the
state variable curves of the uncertain crane system. Compared with Fig. 3.1a, the
trolley in Fig. 3.3a can likewise achieve its positioning at about 3 s. In contrast, the
control law (3.9) applied to the nominal crane model can perfectly suppress the
single-pendulum oscillations of the payload in Fig. 3.2c. However, the payload has
some slightly residual oscillations in the presence of uncertainties in Fig. 3.3c. This
fact indicates that some lightly residual oscillations in a real crane control system
are evitable because of unmatched uncertainties. Figure 3.4 shows the control
performance though the control input and sliding surface curves. Compared with
Fig. 3.3 System performance by the rst-order SMC method for the crane with unmatched
uncertainties. a Trolley position; b Trolley velocity; c Payload angle; d Payload angular velocity
74 3 Overhead Crane Control by Sliding Mode Methods
Fig. 3.4 Control performance of the rst-order SMC method for the crane with unmatched
uncertainties. a Control input u; b Sliding surface s; c Equivalent control ueq; d Switching control
usw
Fig. 3.2d, the switching control works all the time to resist the unmatched uncer-
tainties in the system dynamic process because of the existence of unmatched
uncertainties. In Fig. 3.4b, some small jumps exist, indicating that the system
trajectory deviates from the sliding surface due to the adverse effects of uncer-
tainties. Such jumps also make the control input in Fig. 3.4a jump back and forth.
To exhibit the development of SMC, a novel design is introduced. Tuan and Lee
[5] designed a novel sliding surface for such a double-pendulum-type overhead
crane. Compared with the standard equivalent-control-based SMC law, their idea
can be treated as an order-reduction sliding mode, where the sliding mode exists in
a subspace of the standard sliding mode. For the purpose of illustration, their
design, analysis, and simulation are investigated step by step.
Consider the crane dynamics (2.46). The novel sliding surface is dened by
In (3.13), , , and are the sliding surface parameters, e = x1 x1d and the
derivative of e is equal to x2. Compared with the standard sliding surface, two
variables x4 and x6 in (3.13) are ignored. This fact means that the sliding surface
does not need the angular velocities of the payload and hook. Such a denition can
simplify the parameter tuning because only three parameters exist in (3.13) other
than ve parameters in the standard sliding surface.
To obtain its equivalent control law, differentiating s in (3.13) with respect to
time t and substituting the crane model (2.46) into the derivative of s yield
When the system states sliding along the sliding surface (3.13), s_ 0 exists and
the sole equivalent control law is applied to the crane control system. From (3.14),
these linguistic descriptions are formulated by
The equivalent control (3.15) can guarantee all state trajectories slide on the
sliding surface (3.13) when the sliding mode is reached. However, to keep these
system states on the sliding manifold, the switching control law should be com-
plemented such that the nal control law can be written by
The standard sliding surface has six state variables such that the derivative of
s covers the hook and payload controls are included in the nal control law. But
there is only the trolley control in (3.14). The method presented by Tuan and Lee
[5] can be treated as mandatory decoupling. Consequently, the system stability has
76 3 Overhead Crane Control by Sliding Mode Methods
1
V s2 3:17
2
Differentiating V with respect to time t and substituting the Eq. (2.42) and the
equivalent control law (3.15) into the derivative of V yield
K
V_ s_s s sgns 3:18
m0 m1 m2
Substitute (3.19) into (2.42) and (2.43). Then, rearranging (2.42) and (2.43) yield
When the state trajectories slide on the surface and convergent to the origin, the
system dynamics are composed of (3.19), (3.20), and (3.21). Apparently, all the
equations are nonlinear. According to the Lyapunovs indirect method, the system
stability can be touched by the linearization technique.
Linearize (3.19), (3.20), and (3.21). Then the linearized dynamics are written by
x_ Ax 3:22
3.2 First-Order Sliding Mode Control 77
here
2 3
0 1 0 0 0 0
6 0 k 0 a 0 b 7
6 7
60 0 0 1 0 0 7
A6
6 0 lk
7
6 lg1 la1 0 lb1 7
7
40 0 1 5
1
0 0 0
0 lk2 0 la2 lg2 lb2
To illustrate the feasibility of the rst-order SMC law, the control law will be
carried out to a double-pendulum-type overhead crane, where the crane physical
parameters [5] are determined by Table 3.2. The controller parameters of the
rst-order SMC law [5] are selected by K = 70; = 0.5; = 17; = 11. Displayed
in (3.18) and (3.23), the signs of the four parameters coincide with the proven
results. Some numerical simulation results are demonstrated in Figs. 3.5 and 3.6,
obtained by a Simulink model of MATLAB, which is shown in Appendix B.
In Fig. 3.5, the transport task of the double-pendulum-type crane is set as the
initial state vector x0 = [0 m 0 m s1 0 rad 0 rad s1 0 rad 0 rad s1] and the desired
state vector xd = [4 m 0 m s1 0 rad 0 rad s1 0 rad 0 rad s1]. Figure 3.5 shows the
curves of the crane state variables during the transport dynamics, where the trolley
position, the trolley velocity, the hook angle, the hook angular velocity, the payload
angle, and the payload angular velocity are located in Fig. 3.5af, respectively.
From Fig. 3.5a, the crane can arrive at the desired position at about 7.5 s.
Meanwhile, the controller can effectively resist the double-pendulum oscillations in
Fig. 3.5c and e. On the other hand, the curve in Fig. 3.5a has no overshooting,
indicating that the trolley can directly arrive at the desired position instead of
moving back and forth around the desired position.
The control performance by the rst-order SMC law is shown in Fig. 3.6. In
Fig. 3.6a, the control input, i.e., the driven force applied to the trolley, jumps back
and forth at the outset to suppress the double-pendulum oscillations. In addition, the
chattering phenomenon is greatly reduced because of the equivalent-control-based
control design. Figure 3.6b shows the curve of the novel sliding surface dened by
Fig. 3.5 System performance by the rst-order SMC method for the double-pendulum-type
overhead crane system. a Trolley position; b Trolley velocity; c Hook angle; d Hook angular
velocity; e Payload angle; f Payload angular velocity
Fig. 3.6 Control performance by the rst-order SMC method for the double-pendulum-type
overhead crane system. a Control input; b Sliding surface
a part of the crane state variables. From Fig. 3.6b, the sliding mode is reached at
about 3 s. Thereafter, the state trajectories of the closed-loop crane control system
by the sliding mode keep sliding along the sliding surface (3.13) and converge to
the origin.
3.3 Integral Sliding Mode Control 79
In 1996, Utkin and Shi [6] proposed an improved sliding control method named
integral sliding mode control (ISMC). The characteristic of the sliding surface of an
ISMC system is that there is an integral term in the surface. Concerning other SMC
methods, the system motion under sliding mode has a dimension short of that of the
state space. However, the integral term in the surface breaks out of this point and it
promotes the dimension of the sliding mode. Under integral sliding mode, the
system motion has a dimension equal to that of the state space. This characteristic
makes the ISMC method attractive because the system trajectory in an ISMC
system always starts from the sliding surface. Accordingly, the robustness of the
ISMC system against matched uncertainties is promised in the whole state space.
For the purpose of illustrating the ISMC design, only single-pendulum-type over-
head crane systems are considered. To develop such an integral sliding mode
controller, recall the dynamic model of a nominal crane system (2.16). The model to
can be rewritten by
x_ Ax Bu F 3:24
Zt
s Cx x0 CA BKxsds 3:25
0
Adopt the equivalent-control-based SMC design [6]. Dene the ISMC law as
where ueq is the equivalent control law and usw is the switching control law.
To obtain the expression of ueq, differentiating s in (3.25) with respect to time
t and substituting the crane model (3.24) into the derivative of s yield
When the crane states slide along the integral sliding surface (3.25), s_ 0 exists
and the sole equivalent control law ueq is applied to the crane control system. From
(3.27), these linguistic descriptions are formulated by
The equivalent control law has been obtained. In (3.26), the switching control
law is still kept unknown. To obtain the whole ISMC law, the switching control law
will be deduced from the stability analysis of the crane ISMC system.
1
Vt s2 3:29
2
Differentiating V in (3.29) with respect to time t and substituting the crane model
(3.24), the integral sliding surface (3.25), and the equivalent-control-based ISMC
law (3.26) into the derivative of V yield
V_ sCBusw 3:31
_
Vt 0 in (3.29). In the sense of Lyapunov, Vt\0 in (3.31) should exist to
make the crane control system asymptotically stable. As a result, dene
3.3 Integral Sliding Mode Control 81
where both and are the positive constants. Finally, the ISMC law of the crane
system can be deduced from (3.28) and (3.32), determined by
The ISMC law for nominal single-pendulum-type overhead cranes has been
drawn in (3.33). However, uncertainties exist everywhere. As mentioned, the
invariance property of SMC can resist matched uncertainties, indicating that the
stability of the ISMC system is kept unchanged when the control law (3.33) is
carried out on crane systems with matched uncertainties (2.19).
To test the effects of unmatched uncertainties on the system stability, the crane
model with unmatched uncertainties (2.20) is taken into considerations. Rearrange
the crane model (2.20). Equation (3.34) can be drawn.
x_ Ax Bu F dx; u 3:34
dx; u 0 n1 x; u 0 n2 x; uT
Differentiating V in (3.29) with respect to time t and substituting the crane model
with unmatched uncertainties (3.34), the integral sliding surface (3.25), and the
equivalent-control-based ISMC law (3.26) and the equivalent control law (3.28)
into the derivative of V yield
V_ sCBusw Cd 3:36
_
Vt 0 is dened in (3.29). Vt\0 exists as well. In the sense of Lyapunov,
the ISMC system for single-pendulum-type overhead cranes with unmatched
uncertainties is of asymptotic stability. At last, the ISMC law can be deduced from
(3.28) and (3.37), written by
So far, the ISMC laws for nominal and uncertain cranes have been completed.
Compare the two control laws (3.33) and (3.39). Both the expressions almost look
the same. The sole difference is the amplitude of the signum function. In (3.33), is
employed to generate two kinds of dynamic equations so that it can be arbitrarily
small. However, * in (3.39) not only has the ability to generate two kinds of
dynamic equation, but also helps the control system resist unmatched certainties. *
in (3.39) should coincide with the inequality (3.38). Otherwise, the ISMC system
for overhead cranes has no guaranteed stability against unmatched certainties.
Fig. 3.7 System performance by the ISMC method for nominal single-pendulum-type overhead
crane systems. a Trolley position; b Trolley velocity; c Payload angle; d Payload angular velocity
motions very well if the crane model is accurate enough. However, an accurate
dynamic model is hard to obtain. Many uncertainties exist in a real crane system. To
make the designed ISMC law practical, it is meaningful to test the system per-
formance in the presence of uncertainties.
The control performance of the closed-loop system is demonstrated in Fig. 3.8,
where the control input and the integral sliding surface are located in Fig. 3.8a, b,
respectively. To accelerate the trolley as fast as possible, the driven force f in
Fig. 3.8a is applied to the trolley as largely as possible at the outset. The trolley
moves toward the desired position but the payload has an angular deviation because
Fig. 3.8 Control performance by the ISMC method for nominal single-pendulum-type overhead
crane systems. a Control input; b Integral sliding surface
84 3 Overhead Crane Control by Sliding Mode Methods
of its inertia. Then, the driven force f decreases and reverses its direction against the
angular deviation. Meanwhile, the trolley decreases its velocity till the control task
is achieved.
From Fig. 3.8b, the ISMC system for the nominal crane enters its integral sliding
mode at t = 0 s. Thereafter, the state trajectories keep sliding on the integral sliding
surface and converge to the origin.
Recall the crane model with unmatched uncertainties (2.19). In the model, both the
uncertain terms 1(x, u) and 2(x, u) are determined by 0.1 [12 rand()], where
rand() is a MATLAB command to generate a uniformly distributed pseudorandom
number on the open interval (0, 1). For the purpose of comparison, the parameter
vectors C and K of the ISMC sliding surface are kept unchanged from the ISMC
system for the nominal overhead crane. * = = 10. From (3.38), * should
coincide with the inequality (3.38). According to the parameter tuning, ||C|| = 3.
Further, ||1(x, u) || = ||2(x, u) || = 0.1. Consequently, * = 0.6 is set to guarantee the
asymptotic stability of the ISMC system for the crane in the presence of uncer-
tainties. Simulation results are shown in Figs. 3.9 and 3.10.
The Simulink model of MATLAB to plot Figs. 3.9 and 3.10 is very similar to
that of plotting Figs. 3.7 and 3.8. Figure 3.9 shows the system performance by the
ISMC method where the trolley position, the trolley velocity, the payload angle,
and the payload angular velocity are shown in Fig. 3.9ad, respectively. From
Fig. 3.9, the payload has some residual oscillations because of the existence of
unmatched uncertainties although the crane can arrive at the desired position at
about 4 s. The maximum angular deviation is about 0.12 rad. Such a deviation is
small enough in practice. Compared with the zeroth residual oscillations in Fig. 3.7,
some slight oscillations in Fig. 3.9 exist in the crane system because of the effects of
unmatched uncertainties. Since it is hard to accurately model the uncertainties,
Fig. 3.9 indicates that some slight oscillations of the payload are inevitable in a real
crane system.
Figure 3.10 illustrates the control performance by the ISMC method, where the
control input and the sliding surface are shown in Fig. 3.10a, b, respectively. From
Fig. 3.10a, the control input slightly jumps back and forth to resist the adverse
effects of unmatched uncertainties. In Fig. 3.10b, although the sliding mode begins
at t = 0 s, the state trajectories deviate from the integral sliding surface because of
the unmatched uncertainties. Compared with the ideal integral sliding mode in
Fig. 3.8b, the integral sliding mode in Fig. 3.10b is not ideal because of the
existence of uncertainties. The uncertainties always tend to make the state trajec-
tories deviate from the sliding surface. Once the trajectories are out of the sliding
3.3 Integral Sliding Mode Control 85
Fig. 3.9 System performance by the ISMC method for the single-pendulum-type overhead crane
system with unmatched uncertainties. a Trolley position; b Trolley velocity; c Payload angle;
d Payload angular velocity
Fig. 3.10 Control performance by the ISMC method for the single-pendulum-type overhead crane
system with unmatched uncertainties. a Control input; b Sliding surface
mode, the ISMC law draws the trajectories back to the surface. The process con-
tinues till the end of the simulation such that the curve of the control input in
Fig. 3.10a does not smooth.
86 3 Overhead Crane Control by Sliding Mode Methods
In 1993, a new type of SMC, named terminal sliding mode (TSM), was invented by
Venkataraman and Gulati [7]. Evolved out of seminal work on terminal attractors,
the main idea of TSM control is evoked by the concept of terminal attractors which
guarantees nite-time convergence of the states. In contrast, the asymptotic stability
of the rst-order SMC is promised which leads to the convergence of the states to
the origin. But this convergence can just be guaranteed within innite time.
In a TSM control system, a nonlinear term is introduced in the sliding surface
design so that the manifold is formulated as an attractor. After the sliding surface is
intercepted, the trajectory is attracted within the manifold and converges to the
origin following a power rule. The TSM control has been widely applied to non-
linear process control, for example, rigid robot control [8], nonlinear spacecraft
attitude [9], continuous stirred tank reactor [10], etc. Several open questions still
remain on the mathematical treatment of the systems behavior at the origin since it
is non-Lipschitz. In Sect. 3.4, the TSM control method will be employed to deal
with the crane control problem.
Bu F
X 3:40
In (3.40), u is the control input, i.e., driven force applied to the trolley;
X = [x ]T, B = [b1(x), b2(x)]T and F = [f1(x), f2(x)]T, where x, b1(x), b2(x), f1(x),
and f2(x) are dened in (2.16).
To design the TSM control law for the type of cranes, a terminal sliding surface
should be dened at rst. The surface s has the form of
s cT1 Xe Pt cT2 X_ e Pt
_ 3:41
In (3.41), the vectors c1 and c2 2 <21 are constant; the error vector Xe is dened
by X Xd where X is the state vector and Xd is the desired state vector, and the
error vector can be written by Xe = [e1 e2]T; the vector P(t) is dened by P(t) = [p1
p2]T where p1 and p2 are assumed to satisfy p1 and p2: R+R, p1 and p2 2 C2[0,),
p1, p_ 1 , p2 and p_ 2 2 L. For a positive constant T > 0, pi is bounded as 0 t T and
pi 0 ei 0, p_ i 0 e_ i 0 and pi 0 ei 0 here i = 1, 2.
3.4 Terminal Sliding Mode Control 87
Select pi as
8
>
> 1 :: c02 ::
>
> e 0 e_ 0t e 0t 2
c00
e 0 c01
_
e 0 e 0 t3
>
>
i
i 2 i T 3
i T 2 i T i
>
>
>
> ::
< cT104 ei 0 cT113 e_ i 0 cT122 e0 t4
pi t i
>
> if 0tT
>
> c22 ::
>
> c20 c21
_
5 ei 0 T 4 e 0 3 e0 t ;
5
>
> T i T
>
>
i
: if t[T
0;
3:42
Solve (3.43), (3.44), and (3.45) by the Gaussian elimination algorithm. These
parameters are obtained and they are listed by
8 8 8
< c00 10 < c01 6 < c00 1:5
c 15 c 8 c 1:5 3:46
: 10 : 11 : 10
c20 6 c21 3 c20 0:5
Substitute (3.46) into (3.41). Then, the terminal sliding surface can be drawn.
Adopt the equivalent-control-based SMC method. Dene the TSM control law as
where ueq is the equivalent control law and usw is the switching control law.
To obtain the expression of ueq, differentiating s in (3.41) with respect to time
t and substituting the crane model (3.40) into the derivative of s yield
s_ t c1 X_ e Pt
_ c2 X e Pt
v
c1 X_ e Pt
_ c2 X X d Pt
v 3:48
c1 X_ e Pt
_ c2 Bu F X d Pt
v
88 3 Overhead Crane Control by Sliding Mode Methods
When the crane states slide along the terminal sliding surface (3.41), s_ 0 exists
and the sole equivalent control law ueq is applied to the crane control system. From
(3.48), these linguistic descriptions are formulated by
c2 X
ueq c2 B1 c2 F c2 Pt d c1 X_ e Pt
_ 3:49
From the TSM control law (3.47), the switching control law is still kept
unknown. To obtain the whole TSM control law, the switching control law will be
deduced from the system stability analysis.
1
Vt s2 3:50
2
Differentiating V in (3.50) with respect to time t and substituting the crane model
(3.40), the terminal sliding surface (3.41), and the equivalent-control-based TSM
control law (3.49) into the derivative of V yield
V_ st_st
3:51
st c1 X_ e Pt
_ d Pt
c2 Bueq usw F X
In (3.49), c2 X
c2 Bueq c2 F c2 Pt d c1 X_ e Pt
_ 0 exists.
Substituting it into (3.51) yields
where both and are the positive constants. Then, the derivative of V in (3.52) can
be written as
V_ js gjsj\0
c2 X
u c2 B1 c2 F c2 Pt d c1 X_ e Pt
_ js g sgns 3:54
Bu F F
X 3:55
K n1 x; u n2 x; uT
Differentiating V in (3.50) with respect to time t and substituting the crane model
with unmatched uncertainties (3.55), the terminal sliding surface (3.41), and the
equivalent-control-based TSM control law (3.47) and the equivalent control law
(3.49) into the derivative of V yield
Substitute (3.57) and (3.58) into (3.56). Then, the derivative of V in (3.56)
becomes
_
Vt js2 gjsj sc2 K\ js2 gjsj jjc2 Kjj1 s
\ js2 gjsj jjc2 jj1 jjKjj1 s\0
Since Vt 0 in (3.50), the crane control system on basis of the TSM control
law is asymptotically stable in the sense of Lyapunov. Finally, the TSM control law
for the crane system with unmatched certainties can be deduced from (3.49) and
(3.57), written by
c2 X
u c2 B1 c2 F c2 Pt d c1 X_ e Pt
_ j s g sgns 3:59
90 3 Overhead Crane Control by Sliding Mode Methods
Fig. 3.11 System performance by the TSM control method for nominal single-pendulum-type
cranes. a Trolley position; b Trolley velocity; c Payload angle; d Payload angular velocity
3.4 Terminal Sliding Mode Control 91
Fig. 3.12 Control performance by the TSM control method for nominal single-pendulum-type
cranes. a Control input; b Sliding surface
around the terminal sliding surface because the nonlinear term P(t) exists in the
surface. Moreover, the sliding surface variable s jumps back and forth in the
dynamic process. Concerning the crane control based on TSM, the terminal
attractor in the sliding surface may deteriorate the chattering of the control input.
Recall the crane model with unmatched uncertainties (2.19). In the model, both the
uncertain terms 1(x, u) and 2(x, u) are determined by 0.1 [12 rand()], where
rand() is a MATLAB command to generate a uniformly distributed pseudorandom
number on the open interval (0, 1). Simulation results are shown in Figs. 3.13 and
3.14. The Simulink model of MATLAB to plot Figs. 3.13 and 3.14 is very similar
to that of plotting Figs. 3.11 and 3.12.
For the purpose of comparison, the parameter vectors c1 and c2 of the terminal
sliding surface in Figs. 3.13 and 3.14 are kept unchanged from the TSM control
system for the nominal overhead crane. * = = 1. From (3.38), * = 2 to guarantee
the asymptotic stability of the control system for the uncertain crane because
jjc2 jj1 jjKjj1 15 0:1\ 2 in (3.58).
The trolley position, the trolley velocity, the payload angle, and the payload
angular velocity are located in Fig. 3.13ad, respectively. The curves in Fig. 3.13a,
c seem smooth. However, the curves in Fig. 3.13b, d slightly jump back and forth,
indicating that the chattering phenomenon indeed exists in the closed-loop control
system. The payload has slight residual oscillations because of the existence of
unmatched uncertainties. The maximum angular deviation is about 0.25 rad (15).
Such a deviation is small enough in practice.
The control input and the sliding surface are shown in Fig. 3.14a, b, respectively.
From Fig. 3.14a, the control input severely jumps back and forth to resist the
adverse effects of unmatched uncertainties. To overcome the chattering of the
92 3 Overhead Crane Control by Sliding Mode Methods
Fig. 3.13 System performance by the TSM control method for uncertain single-pendulum-type
cranes. a Trolley position; b Trolley velocity; c Payload angle; d Payload angular velocity
Fig. 3.14 Control performance by the TSM control method for uncertain single-pendulum-type
cranes. a Control input; b Sliding surface
control input, one possible method is to employ the saturation function instead of
the signum function in (3.59). The sliding surface curve is neither smooth nor
gradual in Fig. 3.14b. It looks random but bounded because of both the uncertain
terms. Compared with Fig. 3.12b, the system trajectories severely deviate from the
sliding mode in Fig. 3.14b.
3.5 Second-Order Sliding Mode Control 93
Mentioned in Sect. 3.1, the rst-order SMC is usually called the SMC. In spite of its
robustness properties, such a resulting controller has a specic disadvantage, i.e.,
chattering phenomenon. The drawback indicates high-frequency vibrations of the
controlled system, which degrades the system performance and may lead to
inherent instability.
Preserving the main advantages of the rst-order SMC, an approach, called
higher order SMC, has been proposed in order to reduce the chattering phenome-
non. Instead of influencing the rst sliding variable time derivative, the signum
function acts on its higher order time derivative. Compared to the rst-order SMC,
the chattering effect is removed to higher derivatives so that the higher order SMC
method can also achieve a better accuracy with respect to discrete sampling time.
The problem of higher order SMC is formulated through the differentiation of
the sliding variable and is equivalent to the nite-time stabilization of higher order
integrator chains. Theoretically, arbitrary order sliding mode controllers can be
developed. In Sect. 3.5, the second-order SMC method is taken into consideration
because of a trade-off between control accuracy and design complexity.
For the purpose of illustration, recall the crane mode (3.40). To develop the
second-order SMC law for such a single-pendulum-type overhead crane system, a
second-order sliding surface for the control design [11] should be dened by
Zt
s a X_ e bT Xe cT
T
Xe sds 3:60
0
e bT X_ e cT Xe aT Bu F X
s_ aT X d bT X_ e cT Xe 3:62
Let
where the scalar parameters , , and are the constants and they are positive.
Substitute (3.65) into (3.64). Then, (3.66) can be drawn.
_ aT Bu_ aT F_
ks j_s gsgns aT Bu
3:66
d bT Bu F X
aT X d cT X_ e
To verify the stability of the crane control system under the second-order SMC law
(3.67), some theoretical analyses are presented in the sense of Lyapunov. Since the
second derivative of the sliding surface variable exists in the second-order SMC
method, its Lyapunov candidate cannot only have the sole sliding surface variable.
Dene a novel Lyapunov candidate function as
1 1
V ks2 s_ 2 gjsj 3:68
2 2
3.5 Second-Order Sliding Mode Control 95
V_ j_s2 3:70
From (3.68), V > 0 exists. Since the derivative of V is negative in (3.70), the
crane system (3.40) by the second-order SMC law (3.67) is of asymptotic stability.
The control law (3.67) can asymptotically stabilize nominal crane systems
(3.40). To test the effects of unmatched uncertainties on the system stability, the
crane model with unmatched uncertainties (3.55) is taken into considerations.
Concerning the second-order SMC law for crane systems in the presence of
unmatched uncertainties, dene the sliding surface as (3.60) and differentiating s in
(3.60) with respect to time t. Then, (3.61) can be drawn. Substituting (3.55) into
(3.61) yields
e bT X_ e cT Xe aT Bu F K X
s_ aT X d bT X_ e cT Xe 3:71
Let
where the scalar parameters , , , and are the constants and they are positive.
Rearrange (3.73) and (3.74). The second-order SMC law for
single-pendulum-type overhead cranes in the presence of unmatched uncertainties
has a form of
1 h
T : :
u_ aT B a B bT B u bT F aT F
3:75
cT X_ e ks j_s gsgns qsgn_s bT X
d aT X
d
96 3 Overhead Crane Control by Sliding Mode Methods
where is subject to
_ bT Kjj
q [ jjaT K 3:76
1
To verify the stability of the crane control system by the second-order SMC law
(3.67), dene a Lyapunov candidate function as (3.68), differentiate V with respect
to time t, and substitute (3.73) and (3.75) into the derivative of V. Then, (3.77) can
be drawn.
_ bT K
V_ s_ ks s gsgns s_ j_s qsgn_s aT K 3:77
_
Since Vt 0 in (3.68) and V\0 in (3.78), the second-order SMC law (3.75)
can guarantee that the crane system subject to unmatched uncertainties is asymp-
totically stable in the sense of Lyapunov.
Fig. 3.15 System performance by the second-order SMC method for nominal
single-pendulum-type cranes. a Trolley position; b Trolley velocity; c Payload angle; d Payload
angular velocity
Figure 3.16 shows the control performance of the closed-loop crane control
system, where the control input and the sliding surface are located in Fig. 3.16a, b,
respectively.
As mentioned, the purpose of the second-order SMC law is to reduce the
chattering phenomenon by covering the signum function in the twice derivative of
the sliding surface s. From Fig. 3.16a, the curve of the driven force f applied to the
Fig. 3.16 Control performance by the second-order SMC method for nominal single-
pendulum-typecranes. a Control input; b Sliding surface
98 3 Overhead Crane Control by Sliding Mode Methods
trolley almost has no chattering phenomenon and the design purpose has been
achieved. On the other hand, the maximum control input is just 0.6 N, which is also
the smallest one among the four SMC methods. The fact means that only a small
rated power driver can be employed to achieve the control task in practice. From
Fig. 3.12b, the state trajectories arrive at the second-order sliding surface at 7 s, and
then the system trajectories keep sliding on the surface till the end of the dynamic
process.
Recall the crane model with unmatched uncertainties (2.19). In the model, both the
uncertain terms 1(x, u) and 2(x, u) are determined by 0.1 [12 rand()], where
rand() is a MATLAB command to generate a uniformly distributed pseudorandom
number on the open interval (0, 1). Due to the existence of uncertainties, the control
law again unmatched uncertainties (3.75) is adopted
Compared with (3.67), one more term exists in (3.75) and the term is very
important because it is related to the system stability. For the purpose of compar-
ison, the controller parameters , , , , , and are kept unchanged from the
second-order SMC law (3.67) for the nominal overhead crane. According to (3.76),
= 2.5 are selected to guarantee the asymptotic stability of the control system for
the uncertain crane because of jjaT K _ bT Kjj \2:5. Some numerical simulation
1
results are shown in Figs. 3.17 and 3.18. The Simulink model of MATLAB to plot
Figs. 3.17 and 3.18 is very similar to that of plotting Figs. 3.15 and 3.16.
The trolley position, the trolley velocity, the payload angle, and the payload
angular velocity are located in Fig. 3.17ad, respectively. The curve in Fig. 3.17a
seems smooth. However, the curves in Fig. 3.13bd slightly jump back and forth,
indicating that the chattering phenomenon indeed exists in the closed-loop control
system. The payload has slight residual oscillations because of the existence of
unmatched uncertainties. The maximum angular deviation is about 0.02 rad (1.15).
Such a deviation is small enough in practice.
From Fig. 3.17a, the trolley arrives at the desired position at about 20 s. Such a
slow transport denitely reduces the throughput of the crane system in reality.
However, the payload endures the minimum angular deviation. The contradiction
between the trolley positioning and the payload swing is completely illustrated by
the second-order SMC method. Such a method is a good technique to resist the
payload oscillations, but it fails to a highly effective transport.
The control input and the sliding surface are displayed in Fig. 3.18a, b,
respectively. From Fig. 3.18a, the control input jumps back and forth to resist the
adverse effects of unmatched uncertainties. But the chattering is not very severe
because only a small control input is needed to make the trolley slowly move.
3.5 Second-Order Sliding Mode Control 99
Fig. 3.17 System performance by the second-order SMC method for uncertain
single-pendulum-type cranes. a Trolley position; b Trolley velocity; c Payload angle; d Payload
angular velocity
Fig. 3.18 Control performance by the second-order SMC method for uncertain
single-pendulum-type cranes. a Control input; b Sliding surface
100 3 Overhead Crane Control by Sliding Mode Methods
Appendices
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 4;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 4;
sizes.NumInputs = 4;
sizes.DirFeedthrough = 0;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
x0 = [0 0 0 0];
str = [];ts = [];
function sys=mdlDerivatives(t,x,u)
Appendices 101
%parameter
g=9.81;M=1;m=0.8;l=0.305;
A=(M+m);
B=(m*l*cos(x(3)));
C=(cos(x(3)));
sys(1)=x(2);
sys(2)=((u(1)+m*l*x(4)^2*sin(x(3)))*l+g*B*sin(x(3)))*inv(A*l-B*C);
sys(3)=x(4);
sys(4)=((u(1)+m*l*x(4)^2*sin(x(3)))*C+g*A*sin(x(3)))*inv(B*C-A*l);
function sys=mdlOutputs(t,x,u)
sys=x;
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 0;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 4;
sizes.NumInputs = 4;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlOutputs(t,x,u)
% parameter of system
g=9.81;M=1;m=0.8;l=0.305;
A=(M+m);
B=(m*l*cos(u(3)));
C=(cos(u(3)));
D=(m*l*sin(u(3)));
102 3 Overhead Crane Control by Sliding Mode Methods
%state error;
e1=(u(1))-1;
e2=u(2);
e3=u(3);
e4=u(4);
s=c(1,1)*e1+c(1,2)*e2+c(1,3)*e3+c(1,4)*e4;
ds=ita*sign(s)+k*s;
f1=(D*(u(4))^2+C*D*g/l)/(A-B*C/l);
b1=1/(A-B*C/l);
f2=(D*C*(u(4))^2+A*g*sin(u(3)))/(B*C-A*l);
b2=C/(B*C-A*l);
con_law=-(c(1,1)*u(2)+c(1,2)*f1+c(1,3)*u(4)
+c(1,4)*f2+ds)/(c(1,2)*b1+c(1,4)*b2);
sys(1)=con_law;
sys(2)=s;
sys(3)=-(c(1,1)*u(2)+c(1,2)*f1+c(1,3)*u(4)+c(1,4)*f2)/(c(1,2)*b1+c(1,4)*b2);
sys(4)=-ds/(c(1,2)*b1+c(1,4)*b2);
figure(1)
subplot(2,2,1); %displacement of trolley
plot(tout,simout(:,1));hold on;
xlabel({'Time(s)';'(a)'});
ylabel('x(m)');
subplot(2,2,2);%velocity of trolley
plot(tout,simout(:,2));hold on;
xlabel({'Time(s)';'(b)'});
ylabel('v(m/s)');
subplot(2,2,3);%theta
plot(tout,simout(:,3));hold on;
xlabel({'Time(s)';'(c)'});
ylabel('\theta(rad)');
subplot(2,2,4);%angular velocity
plot(tout,simout(:,4));hold on;
xlabel({'Time(s)';'(d)'});
ylabel('w(rad/s)');
Appendices 103
figure(2)
subplot(2,2,1)
plot(tout,simout1(:,1));hold on;
xlabel({'Time(s)';'(a)'});
ylabel('f(N)');
subplot(2,2,2);%velocity of trolley
plot(tout,simout1(:,2));hold on;
xlabel({'Time(s)';'(b)'});
ylabel('sliding surface s');
subplot(2,2,3);%theta
plot(tout,simout1(:,3));hold on;
xlabel({'Time(s)';'(c)'});
ylabel('Equivalent control');
subplot(2,2,4);%angular velocity
plot(tout,simout1(:,4));hold on;
xlabel({'Time(s)';'(d)'});
ylabel('Switching control');
function[sys,x0,str,ts]=plant(t,x,u,flag)
switch flag,
case 0,
[sys,x0,str,ts]=mdlInitializeSizes;
case 1,
sys=mdlDerivatives(t,x,u);
case 3,
sys=mdlOutputs(t,x,u);
case {2,4,9},
sys=[];
otherwise
error(['Unhandled flag=',num2str(flag)]);
end
function[sys,x0,str,ts]=mdlInitializeSizes
sizes=simsizes;
sizes.NumContStates=6;
sizes.NumDiscStates=0;
sizes.NumOutputs=8;
sizes.NumInputs=1;
sizes.DirFeedthrough=1;
sizes.NumSampleTimes=0;
sys=simsizes(sizes);
x0=[0,0,0,0,0,0];
str=[];ts=[];
function sys=mdlDerivatives(t,x,u)
mt=50;mc=2;mh=10;l1=3;l2=0.3;b=0;g=9.81;
x_=x(1);
dx=x(2);
phi=x(3);
dphi=x(4);
theta=x(5);
dtheta=x(6);
tol=u(1);
q=[x_;phi;theta];
dq=[dx;dphi;dtheta];
U=[tol;0;0];
M=[mt+mh+mc -(mh+mc)*l1*cos(phi) -mc*l2*cos(theta);...
-(mh+mc)*l1*cos(phi) (mh+mc)*l1^2 mc*l1*l2*cos(phi-theta);...
-mc*l2*cos(theta) mc*l1*l2*cos(phi-theta) mc*l2^2];
C=[b (mh+mc)*l1*sin(phi)*dphi mc*l2*sin(theta)*dtheta;...
0 0 mc*l1*l2*sin(phi-theta)*dtheta;...
0 -mc*l1*l2*sin(phi-theta)*dphi 0];
G=[0;(mh+mc)*g*l1*sin(phi);mc*g*l2*sin(theta)];
ddq=inv(M)*(U-G-C*dq);
ddx=ddq(1,1);
ddphi=ddq(2,1);
ddtheta=ddq(3,1);
sys(1)=x(2);
sys(2)=ddq(1,1);
Appendices 105
sys(3)=x(4);
sys(4)=ddq(2,1);
sys(5)=x(6);
sys(6)=ddq(3,1);
function sys=mdlOutputs(t,x,u)
mt=50;mc=2;mh=10;l1=3;l2=0.3;b=0;g=9.81;
x_=x(1);
dx=x(2);
phi=x(3);
dphi=x(4);
theta=x(5);
dtheta=x(6);
tol=u(1);
q=[x_;phi;theta];
dq=[dx;dphi;dtheta];
U=[tol;0;0];
M=[mt+mh+mc -(mh+mc)*l1*cos(phi) -mc*l2*cos(theta);...
-(mh+mc)*l1*cos(phi) (mh+mc)*l1^2 mc*l1*l2*cos(phi-theta);...
-mc*l2*cos(theta) mc*l1*l2*cos(phi-theta) mc*l2^2];
C=[b (mh+mc)*l1*sin(phi)*dphi mc*l2*sin(theta)*dtheta;...
0 0 mc*l1*l2*sin(phi-theta)*dtheta;...
0 -mc*l1*l2*sin(phi-theta)*dphi 0];
G=[0;(mh+mc)*g*l1*sin(phi);mc*g*l2*sin(theta)];
ddq=inv(M)*(U-G-C*dq);
ddx=ddq(1,1);
ddphi=ddq(2,1);
ddtheta=ddq(3,1);
sys(1)=x_;
sys(2)=dx;
sys(3)=phi;
sys(4)=dphi;
sys(5)=ddphi;
sys(6)=theta;
sys(7)=dtheta;
sys(8)=ddtheta;
function[sys,x0,str,ts]=Controller(t,x,u,flag)
switch flag,
case 0,
[sys,x0,str,ts]=mdlInitializeSizes;
case {1,2,4,9},
sys=[];
case 3,
sys=mdlOutputs(t,x,u);
otherwise
error(['Unhandled flag=',num2str(flag)]);
end
function[sys,x0,str,ts]=mdlInitializeSizes
sizes=simsizes;
106 3 Overhead Crane Control by Sliding Mode Methods
sizes.NumContStates=0;
sizes.NumDiscStates=0;
sizes.NumOutputs=2;
sizes.NumInputs=8;
sizes.DirFeedthrough=1;
sizes.NumSampleTimes=0;
sys=simsizes(sizes);
x0=[];str=[];ts=[];
function sys=mdlOutputs(t,x,u)
x_=u(1); dx=u(2); phi=u(3); dphi=u(4);
ddphi=u(5); theta=u(6); dtheta=u(7); ddtheta=u(8);
mt=50;mc=2;mh=10;l1=3;l2=0.3;b=0;xd=4;g=9.81;
K=70;lambda=0.5;alpha=17;beta=-11;
xd=4;
e=x_-xd;
de=dx;
s=de+lambda*e+alpha*phi+beta*theta;
tol=-(mh+mc)*l1*cos(phi)*ddphi-mc*l2*cos(theta)*ddtheta+b*dx+...
(mh+mc)*l1*sin(phi)*(dphi)^2+mc*l2*sin(theta)*(dtheta)^2....
-(mt+mh+mc)*(lambda*dx+alpha*dphi+beta*dtheta)-K*sat(s);
sys(1)=tol;
sys(2)=s;
figure(1)
subplot(3,2,1); plot(x(:,1),x(:,2));hold on;
xlabel({'Time(s)';'(a)'}); ylabel('x(m)');
figure(2)
subplot(1,2,1),plot(u(:,1),u(:,2));hold on;
xlabel({'Time(s)';'(a)'}); ylabel('u (N)');
subplot(1,2,2),plot(s(:,1),s(:,2));hold on;
xlabel({'Time(s)';'(b)'}); ylabel('s');
Appendices 107
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 0;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 6;
sizes.NumInputs = 8;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
x0 = [];
str = [];
ts = [];
108 3 Overhead Crane Control by Sliding Mode Methods
function sys=mdlOutputs(t,x,u)
% parameter of system
g=9.81;
M=1;
m=0.8;
l=0.305;
a=(M+m);
b=(m*l*cos(u(3)));
c=(cos(u(3)));
d=(m*l*sin(u(3)));
%desired state
r=1;dr=0;ddr=0;
theta=0;dtheta=0;ddtheta=0;
R=[r;dr;theta;dtheta];
dR=[dr;ddr;dtheta;ddtheta];
%stste error
X=[u(1);u(2);u(3);u(4)];
e=X-R;
%parameter of controller
K=[1.2 1.7 -6 -2];
C=[0.5 1.7 -3 -1];
int=(A-B*K)*e;
E=[u(5);u(6);u(7);u(8)];
s=C*(e-E);
ita=0.01;k=10;
us=inv(C*B)*(-ita*sign(s)-k*s-C*B*K*e-C*F);
sys(1)=us;
sys(2)=int(1);
sys(3)=int(2);
sys(4)=int(3);
sys(5)=int(4);
sys(6)=s;
subplot(2,2,2);%velocity of trolley
plot(tout,simout(:,2));hold on;
xlabel({'Time(s)';'(b)'});ylabel('v(m/s)');
Appendices 109
subplot(2,2,3);%theta
plot(tout,simout(:,3));hold on;
xlabel({'Time(s)';'(c)'});ylabel('\theta(rad)');
subplot(2,2,4);%angular velocity
plot(tout,simout(:,4));hold on;
xlabel({'Time(s)';'(d)'});ylabel('\omega (rad/s)');
subplot(1,2,2),
plot(tout,simout2);hold on;
xlabel({'Time(s)';'(b)'});ylabel('s');
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 0;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 2;
sizes.NumInputs = 8;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlOutputs(t,x,u)
persistent e10 de10 dde10 e20 de20 dde20
% parameter of system
g=9.81;M=1;m=0.8;l=0.305;
A=(M+m);
B=(m*l*cos(u(3)));
C=(cos(u(3)));
D=l;
E=(m*l*sin(u(3)));
T=1.5;
% parameter of controller
c1=[-1 20];c2=[-15 -3.45];
K=0.04;
%e=x-xd;
e1=(u(1))-1;
de1=u(2);
dde1=u(6);
e2=u(3);
de2=u(4);
dde2=u(8);
e=[e1;e2];
if t==0
Appendices 111
e10=(u(1))-1;
de10=u(2);
dde10=u(6);
e20=(u(3));
de20=u(4);
dde20=u(8);
end
if t<=T
A0=-10/T^3*e10-6/T^2*de10-1.5/T*dde10;
A1=15/T^4*e10+8/T^3*de10+1.5/T^2*dde10;
A2=-6/T^5*e10-3/T^4*de10-0.5/T^3*dde10;
B0=-10/T^3*e20-6/T^2*de20-1.5/T*dde20;
B1=15/T^4*e20+8/T^3*de20+1.5/T^2*dde20;
B2=-6/T^5*e20-3/T^4*de20-0.5/T^3*dde20;
p1=e10+de10*t+1/2*dde10*t^2+A0*t^3+A1*t^4+A2*t^5;
dp1=de10+dde10*t+A0*t^2*3+A1*t^3*4+A2*t^4*5;
ddp1=dde10+A0*t*6+A1*t^2*12+A2*t^3*20;
p2=e20+de20*t+1/2*dde20*t^2+B0*t^3+B1*t^4+B2*t^5;
dp2=de20+dde20*t+B0*t^2*3+B1*t^3*4+B2*t^4*5;
ddp2=dde20+B0*t*6+B1*t^2*12+B2*t^3*20;
else
p1=0;dp1=0;ddp1=0;p2=0;dp2=0;ddp2=0;
end
s=c1*([e1;e2]-[p1;p2])+c2*([de1;de2]-[dp1;dp2]);
BX=[1/(A-B*C/D);C/(B*C-A*D)];
FX=[(E*(u(3))^2+m*g*sin(u(3))*C)/(A-
B*C/D);(E*(u(3))^2*C+A*g*sin(u(3)))/(B*C-A*D)];
con_law=-inv(c2*BX)*(c2*FX-c2*[ddp1;ddp2]+c1*[de1-dp1;de2-
dp2]+K*sign(s));
sys(1)=con_law;
sys(2)=s;
subplot(2,2,2);%velocity of trolley
plot(t,simout(:,2));hold on;
xlabel({'Time(s)';'(b)'});ylabel('v(m/s)');
subplot(2,2,3);%theta
plot(t,simout(:,3));hold on;
xlabel({'Time(s)';'(c)'});ylabel('\theta(rad)');
subplot(2,2,4);%angular velocity
plot(t,simout(:,4));hold on;
xlabel({'Time(s)';'(d)'});ylabel('w(rad/s)');
figure(2) %control force
subplot(1,2,1);plot(t,simout1(:,1));hold on;
xlabel({'Time(s)';'(a)'});ylabel('u(N)');
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 4;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 8;
sizes.NumInputs = 2;
sizes.DirFeedthrough = 0;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlDerivatives(t,x,u)
%parameter
g=9.81;M=1;m=0.8;l=0.305;
A=(M+m);B=(m*l*cos(x(3)));C=(cos(x(3)));
sys(1)=x(2);
sys(2)=((u(1)+m*l*x(4)^2*sin(x(3)))*l+g*B*sin(x(3)))*inv(A*l-B*C);
sys(3)=x(4);
sys(4)=((u(1)+m*l*x(4)^2*sin(x(3)))*C+g*A*sin(x(3)))*inv(B*C-A*l);
function sys=mdlOutputs(t,x,u)
g=9.81;M=1;m=0.8;l=0.305;
A=(M+m);B=(m*l*cos(x(3)));C=(cos(x(3)));D=(m*l*sin(x(3)));
f1=(D*(x(4))^2+C*D*g/l)/(A-B*C/l);
b1=1/(A-B*C/l);
f2=(D*C*(x(4))^2+A*g*sin(x(3)))/(B*C-A*l);
b2=C/(B*C-A*l);
sys(1)=x(1);sys(2)=x(2);sys(3)=x(3);sys(4)=x(4);sys(5)=b1;
sys(6)=b2;sys(7)=f1;sys(8)=f2;
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 3;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 2;
sizes.NumInputs = 10;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 1; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlDerivatives(t,x,u)
114 3 Overhead Crane Control by Sliding Mode Methods
%state error;
e1=u(1)-1;e2=u(2);
e3=u(3);e4=u(4);
e=[e1;e3];de=[e2;e4];
g=9.81;M=1;m=0.8;l=0.305;
A=(M+m);B=(m*l*cos(u(3)));C=(cos(u(3)));D=(m*l*sin(u(3)));
f1=(D*(u(4))^2+C*D*g/l)/(A-B*C/l);
b1=1/(A-B*C/l);
f2=(D*C*(u(4))^2+A*g*sin(u(3)))/(B*C-A*l);
b2=C/(B*C-A*l);
lamda=1;k=1.9;ita=0.1;
du=-inv(alpha*[b1;b2])*(lamda*s+(alpha*[u(5);u(6)]
+beta*[b1;b2])*x(3)+alpha*[u(7);u(8)]...
-alpha*[0;0]+beta*[f1;f2]-beta*[0;0]+gama*de+k*u(10)+ita*sign(s));
sys(1)=e1;
sys(2)=e3;
sys(3)=du;
function sys=mdlOutputs(t,x,u)
%state error;
e1=u(1)-1;e2=u(2);
e3=u(3);e4=u(4);
e=[e1;e3];de=[e2;e4];
sys(1)=con_law;
sys(2)=s;
figure(1)
subplot(2,2,1); %displacement of trolley
plot(t,simout(:,1));hold on;xlabel({'Time(s)';'(a)'});ylabel('x(m)');
subplot(2,2,2);%velocity of trolley
plot(t,simout(:,2));hold on;xlabel({'Time(s)';'(b)'});ylabel('v(m/s)');
subplot(2,2,3);%theta
plot(t,simout(:,3));hold on;xlabel({'Time(s)';'(c)'});ylabel('\theta(rad)');
subplot(2,2,4);%angular velocity
plot(t,simout(:,4));hold on;xlabel({'Time(s)';'(d)'});ylabel('w(rad/s)');
References
1. Abdel-Rahman EM, Nayfeh AH, Masoud ZN (2003) Dynamics and control of cranes: a
review. J Vib Control 9(7):863908
2. Liu D, Guo W (2013) Nonlinear controller design for the underactuated crane system. Int J
Control Autom 6(6):93104
3. Utkin VI (1992) Sliding modes in control and optimization. Springer, Berlin
4. Wang W, Yi J, Zhao D, Liu D (2004) Design of a stable sliding-mode controller for a class of
second-order underactuated systems. IEE Proc Control Theory Appl 151(6):683690
5. Tuan LA, Lee SG (2013) Sliding mode controls of double-pendulum crane systems. J Mech
Sci Technol 27(6):18631873
6. Utkin V, Shi J (1996) Integral sliding mode in systems operating under uncertainty conditions.
In: Proceedings of 35th IEEE conference on decision and control, Kobe, Japan, pp 45914596
7. Venkataraman S, Gulati S (1993) Control of nonlinear systems using terminal sliding modes.
J Dyn Syst Meas Control Trans ASME 115(3):554560
8. Man Z, Paplinski AP, Wu HR (1994) A robust MIMO terminal sliding mode control scheme
for rigid robotic manipulators. IEEE Trans Autom Control 39(12):24642469
9. Song ZK, Li HX, Sun KB (2014) Finite-time control for nonlinear spacecraft attitude based on
terminal sliding mode technique. ISA Trans 53(1):117124
10. Zhao DY, Zhu QM, Dubbeldam J (2015) Terminal sliding mode control for continuous stirred
tank reactor. Chem Eng Res Des 94:266274
11. Chiang HK, Fang CC, Lin WB, Chen GW (2011) Second-order sliding mode control for a
magnetic levitation system. In: Proceedings of 8th Asian control conference, Kaohsiung,
Taiwan, pp 602607
Chapter 4
Overhead Crane Control by Hierarchical
Sliding Mode
Keywords Overhead crane Hierarchical sliding surfaces Aggregated SMC
Incremental SMC
Combining SMC and adaptive control based hierarchical
sliding surfaces Transport control
What is the relationship between hierarchical sliding mode control (HSMC) and
hierarchical control system? Concerning the methodology of HSMC [2, 3], the
hierarchy means the hierarchical structure of the sliding surfaces instead of the
hierarchy of the SMC system. According to the hierarchical theory of control
systems [1], the HSMC method investigates the sliding mode controller design.
A hierarchical sliding mode controller is located at the lowest level, which does not
concern the two higher levels. Briefly, the HSMC method focuses on the hierar-
chical structure of the sliding surfaces and designs the control law on basis of the
hierarchical sliding surfaces.
Intuitively, a sliding surface covers several state variables. A rst-order sliding
surface is constructed by linear combination of the state variables. However, each
state variable plays a unique role in the system performance. These state variables
are not equal to each other and they have different degrees of importance. Some
may contribute more to the system performance than others. The rst-order sliding
surface weights each state variable by its coefcient. Another way to describe the
degree of importance of each state is by structure design of sliding surfaces. This is
our motivation to develop the methodology of HSMC.
To develop a diversity of the hierarchical sliding-surface structures for crane
control, a crane dynamic model should be taken into considerations. Recall the state
space model of single-pendulum-type overhead cranes (2.16). There are four state
variables in (2.16) and they consist of a state vector. In Chap. 3, the state vector is
directly employed for the control designs of several typical sliding mode methods.
These design methods treat the four state variables as a whole such that the sliding
surfaces have a single-layer structure.
On the other hand, the four state variables can arbitrarily combine to generate
sliding surfaces. Different sequences of combination can result in different types of
hierarchical structures. So far, there are three typical types of hierarchical structures
of sliding surfaces, i.e., aggregated hierarchical sliding surfaces, incremental hier-
archical sliding surfaces, and combining hierarchical sliding surfaces, which will be
introduced in Sects. 4.2, 4.3 and 4.4, respectively. Besides the typical hierarchical
structures, adaptive control design based on hierarchical sliding surfaces is also
investigated in this chapter.
Recall the crane structure in Fig. 2.1. The single-pendulum-type overhead crane is
composed of two subsystems, i.e., trolley and payload. Further, recall the system
state equations (2.16). There are four state variables in (2.16), where each sub-
system is described by two state variables. As a result, a novel type of sliding
surface structure can be constructed according to the mechanical structure of the
type of cranes.
The basic idea behind the novel structure is as follows [3]. Two sliding surfaces
can be designed for the two subsystems at rst. Then the two surfaces can be
4.2 Aggregated HSMC 119
Recall the crane model (2.16). Design the sliding surfaces of the trolley and payload
subsystems as
s 1 c1 x1 x2 4:1
s 2 c2 x3 x4 4:2
Here c1 and c2 are positive constants in order to have the stability of each
subsystem. According to the designed hierarchical structure, both the subsystem
sliding surfaces (4.1) and (4.2) are located at the rst layer. To construct the higher
lay sliding surface, aggregate the two sliding surfaces and design the second layer
sliding surface as
S as1 s2 4:3
120 4 Overhead Crane Control by Hierarchical Sliding Mode
where ueq1 and ueq2 are the equivalent control laws of the two subsystems on their
private rst-layer sliding surfaces; usw is the switching control law. According to
the hierarchical structure in Fig. 4.1, the switching control law should not only
make the closed-loop control system asymptotically stable, but also have to guar-
antee that the sliding mode of the second-layer sliding surface is reachable.
Concerning the stability analysis, the SMC law of the single-layer sliding surface
is drawn from the reachability condition of SMC such that the SMC law can
guarantee the system stability. However, there are two novel variables in the
aggregated HSMC system, i.e., the rst-layer sliding surface variables, s1 and s2. As
a result, the HSMC law has to stabilize the two variables by the design of the
switching control law usw.
Recall the crane model (2.16), where the two subsystems are described by
x_ 1 x2
4:5
x_ 2 f1 x b1 xu
4.2 Aggregated HSMC 121
and
x_ 3 x4
4:6
x_ 4 f2 x b2 xu
Adopt the equivalent-control-based SMC design. From (4.1) and (4.2), the
equivalent control laws of the two subsystems will be applied to the two subsystems
when the sliding modes of the two subsystems take place. To obtain the equivalent
control laws of the two subsystems, differentiating s1 and s2 with respect to time t,
substituting the subsystem model (4.5) into the derivatives of s1 and substituting the
subsystems mode (4.6) into the derivatives of s2 yield
c1 x2 f1 x
ueq1 4:7
b1 x
and
c2 x4 f2 x
ueq2 4:8
b2 x
Compared with the aggregated HSMC law (4.4), ueq1 and ueq2 have been drawn
in (4.7) and (4.8), respectively. The last term usw will be deduced from the system
analysis to make the closed-loop control system and the two rst-layer sliding
surface variables asymptotically stable.
Vt S2 =2 4:10
122 4 Overhead Crane Control by Hierarchical Sliding Mode
where S is the second-layer sliding surface variables. From (4.10), the motivation of
the HSMC law is to rst guarantee the stability the second-layer sliding surface.
Differentiating V with respect to time t in (4.10) and substituting the expression
of the second-layer sliding surface (4.3) into the derivative of V yield
_
Vt Sa_s1 s_ 2 4:11
_
Vt Sac1 x_ 1 x_ 2 c2 x_ 3 x_ 4 4:12
Substituting the subsystem models (4.5) and (4.6) into (4.12) yields
_
Vt Sac1 x2 f1 x b1 xu c2 x4 f2 x b2 xu 4:13
Consider the aggregated HSMC law (4.4) and the equivalent control laws at the
rst-layer sliding surfaces (4.7) and (4.8). Substituting (4.4), (4.7), and (4.8) into
(4.13) yields
_
Vt Sfac1 x2 f x1 b1 x ueq1 ueq2 usw
c2 x4 f2 x b2 x ueq1 ueq2 usw g 4:14
Sab1 x ueq2 usw b2 x ueq1 usw
_
From (4.10), Vt 0. In the sense of Lyapunov, Vt\0 should exist in (4.14)
to make the crane control system asymptotically stable. As a result, (4.9) should be
_
dened to achieve the condition Vt\0. From (4.7), (4.8), and (4.9), the aggre-
gated HSMC law has the form of
_
Since Vt 0 in (4.10) and Vt\0 in (4.14), the second-layer sliding surface
S is of asymptotic stability in the sense of Lyapunov.
Theorem 4.2 Consider the single-pendulum-type overhead crane system (2.16),
design the hierarchical sliding surfaces (4.1), (4.2), and (4.3), adopt the aggregated
HSMC law (4.4). Then, the rst-layer sliding surfaces s1 and s2 are asymptotically
stable.
_
Proof According to Theorem 4.1, Vt\0. _
This fact indicates that SS\0, i.e., the
reachability condition of SMC is satised on the second-layer sliding surface and
the sliding mode of the second-layer sliding surface is reachable in the nite time tf.
4.2 Aggregated HSMC 123
Consequently, the time domain can be divided into two parts by the time point tf.
On the time interval [0, tf], the system state trajectory moves toward the
second-layer sliding surface S. on the subsequent time interval (tf, ), the system
state trajectory keeps sliding on the second-layer sliding surface and converges to
the origin as proven in Theorem 4.1.
To check the stabilities of s1 and s2 in [0, tf], dene a truncated function as
1 t tf
ftf t 4:16
0 t [ tf
where i = 1, 2.
Equations (4.17) and (4.18) mean the rst-layer sliding surface variables s1 and
s2 are bounded on the closed interval [0, tf] and both of them can converge to the
second-layer sliding surface when the sliding mode is reached at tf. On the sub-
sequent open interval (tf, ), the system trajectory maintains on the second-layer
sliding surface S and the closed-loop crane control system becomes autonomous.
_ 0. Dene a positively invariant and compact set
From (4.14), Vt
Sc S 2 R2 jV_ 0 4:19
Further, dene
S S 2 SjV_ 0 4:20
rst-layer sliding surface and toward its subsystem coordinate origin constructed by
the subsystem state variables such that (4.23) can be drawn.
s1 s_ 1 0
4:23
s2 s_ 2 0
As a result, the largest invariant set inside S \ Sc is just the coordinate originate
origin constructed by the sliding surface variables s1 and s2 rather than other points.
In light of LaSalles principle, both the rst-layer sliding surfaces s1 and s2 are of
asymptotic stability.
Fig. 4.2 System performance by the aggregated SMC method for the overhead crane. a Trolley
position; b Trolley velocity; c Payload angle; d Payload angular velocity
Fig. 4.3 Control performance of the aggregated SMC method. a Control input u; b Subsystem
sliding surface s1 at the rst layer; c Subsystem sliding surface s2 at the rst layer; d Second-layer
sliding surface S
126 4 Overhead Crane Control by Hierarchical Sliding Mode
The control performance is displayed in Fig. 4.3, where the control input, the
subsystem sliding surface s1 at the rst layer, the subsystems sliding surface s2 at
the rst layer, and the aggregated sliding surface S at the second layer are shown in
Fig. 4.3ad, respectively.
From Fig. 4.3a, the control input has a chattering phenomenon. The aggregated
HSMC as a branch of SMC inherits the inner drawback of SMC as well. To
decrease the chattering, saturation function can be employed instead of signum
function in (4.15). In Fig. 4.3bd, all the hierarchical sliding surfaces are stable as
proven in Theorems 4.1 and 4.2. In Fig. 4.3d, the sliding mode of the aggregated
sliding surface at the second layer is reachable at about 2 s. Thereafter, the two
sliding surfaces at the rst layer tend to their subsystems origins along the two
subsystem sliding surfaces.
Although the two sliding surfaces at the rst layer tend to be stable in Fig. 4.3b,
c, the reachability of their sliding modes cannot be theoretically guaranteed.
Theorem 4.2 can just guarantee that the two subsystem sliding surfaces are of
asymptotic stability. However, only the second-layer sliding surface S is reachable
in nite time because the reachability condition of sliding mode is guaranteed in
Theorem 4.1.
The phase curves of the aggregated sliding surfaces are shown in Fig. 4.4, where
the subsystem sliding surface s1 at the rst layer, the subsystem sliding surface s2 at
the rst layer, and the second-layer sliding surface S are displayed in Fig. 4.4ac,
respectively.
Fig. 4.4 Phase portrait of the aggregated sliding surfaces. a Subsystem sliding surface s1 at the
rst layer; b Subsystem sliding surface s2 at the rst layer; c Second-layer sliding surface S
4.2 Aggregated HSMC 127
From Fig. 4.4a, b, the two sliding surface variables s1 and s2 are asymptotically
stable. In other words, the sliding modes of the two subsystem sliding surfaces are
reachable in innite time. In Fig. 4.4c, the phase trajectory of the second-layer
sliding surface tends to the sliding surface. The trajectory remains on the sliding
surface once the control system enters the sliding mode. Thereafter, the two sub-
systems tend to their subsystem origins along the subsystem sliding surfaces s1 and
s2. The three phase curves demonstrate the technical contents in Theorems 4.1 and
4.2.
The theoretical analyses and numerical simulation results show the aggregated
HSMC method is a possible choice to achieve the transport control problem of
single-pendulum-type overhead crane systems. The main contributions of the
aggregated HSMC method are (1) the aggregated hierarchical structure of sliding
surfaces is constructed according to the crane physical property; (2) the sliding
mode of the second layer sliding surface is reachable in nite time; (3) the stability
of the two subsystem sliding surfaces at the rst layer is guaranteed.
Recall the crane structure in Fig. 2.1 and the state Eq. (2.16). There are four state
variables to describe the crane system. As a result, a novel type of sliding surface
structure [5] can be constructed by increasing the number of variables to produce
the hierarchical structure of sliding surfaces. The basic idea behind the type of
incremental hierarchical structure [6, 7] is as follows.
Consider the crane system (2.16). Select arbitrary two state variables to construct
the rst-layer sliding surface. Then, increase one state variable to construct the
second-layer sliding surface with the 1st-layer sliding surface variable. This process
proceeds till all the system state variables are included. The incremental hierarchical
structure is developed by general increment of the system state variables. As the
crane model covers four state variables, the incremental sliding surfaces of the crane
control system have three layers, illustrated in Fig. 4.5.
Consider the aggregated hierarchical sliding surfaces in Sect. 4.2, which is
developed by capturing the cranes physical property. Each sliding surface on the
rst layer is designed for each subsystem. As a result, the aggregated HSMC can
only be employed for mechanical systems composed of several subsystems.
Compared with the aggregated sliding surfaces structure, the development of the
rst-layer sliding surface is dened by arbitrary two state variables in the crane
model, which does not depend on the cranes physical property. The incremental
hierarchical structure is achieved till all the system variables are covered by the
sliding surfaces. Thus, the incremental sliding surfaces structure can be applied to
mechanical systems that are not composed of subsystems.
Concerning the incremental sliding surfaces, two control problems rise . They
are similar to the aggregated HSMC. In the rst-order SMC design, its control
design can guarantee the stability of the traditional single-layer sliding surface. The
128 4 Overhead Crane Control by Hierarchical Sliding Mode
incremental sliding surfaces in Fig. 4.5 contain three layers. How to design a
control law on basis of the incremental sliding surfaces is one crucial issue of such a
HSMC law.
The other control problem is about the stability of the incremental sliding sur-
faces. There are three layers sliding surfaces in the incremental SMC-based crane
control system. It is also eager to know how to theoretically guarantee the stability
of all the sliding surfaces. Besides, the incremental hierarchical structure is uni-
versal because it does not depend on the physical property of cranes. Consider a
system with n state variables. Adopt the incremental sliding surfaces. n 1 layers
sliding surfaces exist. It is also a challenging problem to investigate the stability of
the n 1 layers sliding surfaces of the system with n state variables.
The rst step of the incremental HSMC design is to construct the incremental
structure of sliding surfaces. Recall the crane model (2.16). There are four state
variables in the model. As mentioned, arbitrary two state variables can be employed
to formulate the rst-layer sliding surface. Without loss of generality, the state
variables x1 and x2 in (2.16) are adopted such that the rst-layer sliding surface has
the form
s 1 c1 x1 x2 4:24
where c1 is a positive constant in order to have the stability of the rst-layer surface.
The second layer sliding surface s2 is constructed by the linear combination of
the rst-layer sliding surface variable s1 and one state variable. Without loss of
generality, x3 is picked up such that the second-layer sliding surface can be for-
mulated by
4.3 Incremental HSMC 129
s 2 c2 x3 s 1 4:25
where c2 is constant.
In (4.24), x2 is equal to the derivative of x1. From the viewpoint of stability, c1
should be a positive constant. In (4.25), there is no explicit mathematic function
between x3 and s1. Here c2 is just a constant. Similarly, the third-layer sliding
surface can be written as
s 3 c3 x4 s 2 4:26
In (4.27), ueq is the equivalent control law and usw is the switching control law.
Equation (4.27) has the same expression as the equivalent-control-based SMC law
on basis of the traditional single-layer sliding surface. Compared to the
equivalent-control-based rst-order SMC design, Eq. (4.27) based on the incre-
mental sliding surfaces raises several novel problems, such as: How to design the
equivalent control law among the multiple-layer hierarchical sliding surfaces? How
to guarantee the stability of the multiple-layer hierarchical sliding surfaces?
To attack the issue of the equivalent control design in (4.27), the Lyapunov
function candidate is dened by
1
Vt s23 4:28
2
(4.28) means the control law (4.27) is deduced by making the last-layer sliding
surface asymptotically stable. Differentiating V in (4.28) with respect to time
t yields
dV dV ds3 d(c3 x4 s2
s3 s_ 3 s3 4:29
dt ds3 dt dt
Substituting (2.16), (4.24), (4.25), (4.26), and (4.27) into (4.29) yields
130 4 Overhead Crane Control by Hierarchical Sliding Mode
dV
s3 c3 f2 b2 u c2 x_ 3 s_ 1
dt
s3 c3 f2 b2 u c2 x4 c1 x_ 1 x_ 2 4:30
s3 c3 f2 b2 u c2 x4 c1 x2 f1 b1 u
s3 c3 f2 c2 x4 c1 x2 f1 c3 b2 b1 ueq usw
_
From (4.28), Vt 0. In the sense of Lyapunov, Vt\0 should exist in (4.30)
to have the stability of the last-layer sliding surface. To realize the purpose, dene
c3 f 2 f 1 c2 x4 c1 x2
ueq 4:31
c 3 b2 b1
and
js3 gsgns3
usw 4:32
c 3 b2 b1
Since the derivative of V is just the left side of the reachability condition of the
last-layer sliding surface, (4.33) indicates that the last-layer sliding surface variable
s3 not only has the asymptotic stability, but also its sliding motion is reachable in
nite time.
So far, the rst issue about the equivalent control design has been investigated.
But the stabilities of the other two inner sliding surfaces are kept to be unsolved
because the control law can only guarantee the stability of the last-layer sliding
surface rather than all the incremental sliding surfaces [8].
As far as the incremental HSMC design is concerned, the closed-loop crane
control system can be stabilized if and only if all the hierarchical sliding surfaces
are stabile. To have the stabilities of the other two inner sliding surfaces, more
insights and analyses about the incremental HSMC law (4.27) should be carried
out.
(4.31) and the switching control law (4.32). Then, all the incremental sliding
surfaces are asymptotically stable if (4.34) is satised.
c2 if x3 s1 0
c2 4:34
c2 if x3 s1 \0
Proof
1) Stability of s3: According to the control design, the control law (4.27) can
guarantee the asymptotic stability of s3. Moreover, the last-layer sliding surface
variable not only has the guaranteed stability, but also its sliding mode is
reachable at nite time tr.
2) Stability of s2: Consider the third-layer sliding surface s3 achieves its sliding
mode at tr. Thereafter, there is no discontinuous switching control in the control
system on the time interval (tr, ) and the closed-loop crane control system can
be treated as an autonomous system. According to LaSalles principle, s3
approaches the largest invariant set in
dV
S s 3 2 Sc j 0 4:35
dt
In (4.35), the set Sc is a positively invariant and compact set, dened by
dV
Sc s3 2 R j2
\c; c[0 4:36
dt
S fs3 js3 0 \ s_ 3 0g
fx4 ; s3 jc3 x4 s2 0 \ c3 x4 s_ 2 0g 4:37
fx4 ; s3 js2 c3 x4 const: \ s_ 2 c3 x4 const:g
Assume the two constants in (4.37) are not equal to zero as t . Then, s3
would converge to a point on the phase plane by the axes x4 and s2 except the
origin, which contradicts the fact limt!1 s3 0. Consequently, both the con-
stants in (4.37) must be equal to zero from the proof by contradiction, i.e., s2 is
locally asymptotically stable.
3) Stability of s1: Since s2 is locally asymptotically stable, the limit of s2 in (4.25)
can be drawn from
Provided that the system states are not divergent, the limit of s1 can be deduced
from
132 4 Overhead Crane Control by Hierarchical Sliding Mode
sgnc2 x3 s1 0 4:40
According to (4.39) and (4.40), the constant in (4.39) is only zero rather than
other constants, which indicates
lim x3 0 4:41
t!1
lim s1 0 4:42
t!1
Equation (4.42) means the rst-layer sliding surface variable s1 is locally asymp-
totically stable if and only if the condition (4.34) is satised.
Fig. 4.6 System performance by the incremental SMC method for the overhead crane. a Trolley
position; b Trolley velocity; c Payload angle; d Payload angular velocity
Fig. 4.7 Control performance of the aggregated SMC method. a Control input u; b First-layer
sliding surface s1; c Second-layer sliding surface s2; d Third-layer sliding surface s3
134 4 Overhead Crane Control by Hierarchical Sliding Mode
sliding surface s3 are shown in Fig. 4.7ad, respectively. As proven in Theorem 4.3,
the three layers sliding surfaces have the guaranteed stabilities. s3 enters its sliding
mode at about 1 s. Thereafter, s2 tends to be stable and the switch of c2 is excited to
make s1 asymptotically stable. The switch of c2 is shown in Fig. 4.8.
The phase curves of the incremental sliding surfaces are shown in Fig. 4.9,
where the rst-layer sliding surface s1, the second-layer sliding surface s2, and the
Fig. 4.9 Phase portrait of the incremental sliding surfaces. a First-layer sliding surface s1;
b Second-layer sliding surface s2; c Third-layer sliding surface s3
4.3 Incremental HSMC 135
Concerning the crane control problems, it is hard to depict the single-layer sliding
surface of the rst-order SMC method by phase portrait because of the constraint of
the phase-plane method. The aggregated and incremental HSMC methods over-
come the drawback such that the phase portraits of the two HSMC methods can be
intuitively illustrated in phase plane.
Besides the aggregated and incremental HSMC methods, the combining hier-
archical structure of the sliding surfaces [9, 10] can be constructed for the crane
control problem. On basis of the combining sliding surfaces, the combining HSMC
design can be investigated. As suggested by its name, the challenge of the com-
bining HSMC is how to combine the hierarchical structure of the sliding surfaces.
The basic idea behind the combining HSMC method [9, 10] is as follows. Recall
the mathematic model of the single-pendulum-type overhead crane (2.16). There
are four state variables in (2.16). They can be divided into two groups. One group is
composed of x1 and x3 and the other group covers x2 and x4, where x2 and x4 are the
derivatives of x1 and x3, respectively.
The derivative relations among the four state variables inspire us to dene an
intermediate variable z by combining x1 and x3 such that a sliding surface variable
can be constructed by the intermediate variable and the derivative of the intermediate
variable. The schematic of the combining sliding surfaces is illustrated in Fig. 4.10.
The advantage of such a combination is that the sliding mode of the sliding
surface variable can be illustrated by the phase-plane method. However, the com-
bination of the state variables raises the stability problem that the four state vari-
ables have no guaranteed stability, because the sliding mode motion can only make
the intermediate variable stable.
Recall the crane model (2.16). Consider the group of state variables x1 and x3.
Dene the intermediate variable z as
z x1 cx3 4:43
In (4.43), c is a positive constant. Since the intermediate variable has the form of
linear combination of x1 and x3, it is the source that the novel hierarchical structure
of sliding surfaces is entitled the combining SMC. From Fig. 4.10, the top lay
sliding surface s is constructed by the intermediate variables and its derivative,
which is determined by
s az z_ 4:44
here should be positive to have the stability the top lay sliding surface.
In Fig. 4.10, there is only the top lay sliding surface. However, such a structure is
also treated as a hierarchical one because the intermediate variable and its derivative
can be treated as the rst-layer sliding surfaces. Adopt the equivalent-control-based
SMC design and dene the combining SMC law as
In (4.45), ueq is the equivalent control law and usw is the switching control law,
where the switching control law is employed to drive the system states moving
toward the specic sliding surface (4.44) and the equivalent control law guarantees
the system states keep sliding on the sliding surface and converge to the origin
along the sliding surface.
To obtain the equivalent control law in (4.45), differentiate s in (4.46) with
respect to time t and substitute the crane model (2.16) into the derivative of s. Then,
the equivalent control law (4.46) can be deduced from s_ 0.
In order to have the stability of the top-layer sliding surface s, select the
Lyapunov function candidate as
Vt s2 =2 4:47
V_ s_s sa_z z
sax2 acx4 x_ 2 c_x4 4:48
sfax2 acx4 f1 x cf2 x b1 x cb2 xug
Further, substituting the control law (4.45) and the expression of the equivalent
control law (4.46) into (4.48) yield
Let
Theorem 4.4 Consider the overhead crane system (2.16), construct the interme-
diate variable (4.43), design the combining sliding surfaces (4.44), adopt the
control law (4.45), then all the state variables are bounded.
138 4 Overhead Crane Control by Hierarchical Sliding Mode
E fsjs 0 \ s_ 0g
4:53
fzjaz z_ 0 \ a_z z 0g fzjz z_ z 0g
The set E is attracting. This case indicates that the largest invariant set in
E contains no sets other than the coordinate origin constructed by the intermediate
variable and its derivative. As a result, the sliding surface s and the intermediate
variable z are asymptotically stable as t by means of Lasalles principle.
From (4.43) and (4.53), the limits of z and z_ are formulated by
For arbitrary time t0 on the interval (tf, ), (4.55) can be deduced from (4.54).
Compared with the asymptotically stable s and z, all the state variables x1, x2, x3,
and x4 are just bounded on the time interval [0, ) on account of the denition of
the combining structure of the sliding surfaces.
Theorem 4.5 Consider the overhead crane system (2.16), construct the interme-
diate variable (4.43), design the combining sliding surfaces (4.44), and adopt the
control law (4.45). Then, all the state variables are asymptotically stable if (4.58) is
satised on the time interval (tf, ), where tf is the time when the sliding mode of
the sliding surface variable s takes place.
c if x1 x3 0
c 4:58
c if x1 x3 \0
Proof The time tf divides the time interval [0, ) into two parts, i.e., [0, tf] and (tf,
). According to Theorem 4.4, all the state variables are bounded on [0, tf].
Subsequently, all the state variables are bounded on (tf, ), i.e.,
here i = 1, 2, 3, 4.
Equation (4.59) can be rewritten by
here j = 1, 3.
Equation (4.61) can be drawn from (4.60).
xj 2 L 1 and x_ j 2 L1 4:61
Z1 Z1 Z1
2
z dt
2
x1 cx3 dt x21 cx23 2cx1 x3 dt\1 4:62
tf tf tf
Further,
Z1 Z1 Z1
2
4cx1 x3 dt\ x1 cx3 dt z2 dt 4:64
tf tf tf
R1
Equation (4.63) comes into existence as long as tf 4cx1 x3 dt [ 0, which can be
guaranteed by choosing the sign of c by means of (4.58). As a result, (4.65) can be
deduced from (4.63).
Z1 Z1
x21 dt\1 and x23 dt\1 4:65
tf tf
Fig. 4.11 System performance by the combining SMC method without parameter switch.
a Trolley position; b Trolley velocity; c Payload angle; d Payload angular velocity
Fig. 4.12 Control performance of the combining SMC method without parameter switch.
a Control input u; b Intermediate variable z; c Derivative of the intermediate variable; d Top-layer
sliding surface s
142 4 Overhead Crane Control by Hierarchical Sliding Mode
Fig. 4.13 System performance by the combining SMC method with parameter switch. a Trolley
position; b Trolley velocity; c Payload angle; d Payload angular velocity
Fig. 4.14 Control performance of the combining SMC method with parameter switch. a Control
input u; b Intermediate variable z; c Derivative of z; d Top-layer sliding surface s
4.4 Combining HSMC 143
Fig. 4.15 Switch process of the parameter c during the system dynamics
The crane performance under the combining SMC design without the switch of
c is illustrated in Fig. 4.11, where the trolley position, the trolley velocity, the
payload angle, and the payload angular velocity are displayed in Fig. 4.11ad,
respectively. As proven in Theorem 4.4, all the state variables of the crane system
are bounded. Such a linear combination of the state variables results in the bounded
control system. Although the crane control system is stable in the sense of
Lyapunov, the theoretical design cannot be put into practice because any real
applications of crane systems cannot endure the harmonic motion of the trolley in
Fig. 4.11a. The payload curve in Fig. 4.11c also has the harmonic motion. Such
swag is very small and it is acceptable in most real applications, which can be
treated as residual oscillations of the crane system.
The control performance under the combining SMC design without the switch of
c is illustrated in Fig. 4.12, where the control input, the intermediate variable, the
derivative of the intermediate variable, and the top-layer sliding surface are dis-
played in Fig. 4.12ad, respectively.
The harmonic motions of the trolley and payload make the control input curve
harmonic in Fig. 4.12a. According to the combining HSMC design, the control law
can guarantee the asymptotic stability of the intermediate variable z. and the
top-layer sliding surface s. The theoretical analysis coincides with the curves in
Fig. 4.12bd.
From Fig. 4.12d, the sliding mode takes place at about 2.4 s. This time point is
very important for the switch of c in Theorem 4.5. As displayed in Figs. 4.11 and
4.12, the combining SMC design arouses the novel problem of the crane control by
SMC. The asymptotic stability of the intermediate variable does not indicate the
asymptotic stability of the state variables.
144 4 Overhead Crane Control by Hierarchical Sliding Mode
The crane performance under the combining SMC design with the switch of c is
illustrated in Fig. 4.13, where the trolley position, the trolley velocity, the payload
angle, and the payload angular velocity are displayed in Fig. 4.13ad, respectively.
As proven in Theorem 4.5, all the state variables of the crane system can be
asymptotically stable when the switch mechanism of the parameter c is carried out
after the sliding mode of s is reached. Once the switch of c is carried out, the state
variables of the crane control system become asymptotically stable in Fig. 4.13 but
rather get bounded in Fig. 4.11.
The control performance under the combining SMC design with the switch of
c is illustrated in Fig. 4.14, where the control input, the intermediate variable, the
derivative of the intermediate variable, and the top-layer sliding surface are dis-
played in Fig. 4.14ad, respectively.
From Fig. 4.14d, the switch of c enters the crane control system at 3 s.
Thereafter, the switch mechanism is triggered once the sign of x1 x 3 changes. The
process continues till the crane control system arrives at the desired position.
The switch process of the parameter c is shown in Fig. 4.15. Compared with the
numerical results in Sects. 4.2 and 4.3, the crane control system by the combining
HSMC law can arrive the desired position at about 10 s, which is slower than the
aggregated and incremental HSMC methods. The fact means that the crane per-
formance by the combining HSMC law is not as effective as the other two HSMC
methods.
So far, the three kinds of hierarchical structure of the sliding surfaces have been
constructed for single-pendulum-type overhead cranes by employing the crane state
variables. The stabilities of the three kinds of crane control systems are also pre-
sented. Some numerical results are illustrated. Compared with the classic SMC
designs in Chap. 3, the HSMC designs capture the physical nature of the cranes and
present some novel control structures, which enrich the contents of SMC. Based the
basic hierarchical structures, some extensions can be explored, i.e., adaptive law
design, complex crane systems, and so on.
The aggregated, incremental and combining HSMC designs are basic types of
hierarchical structures. In the designs, all the sliding surface parameters are con-
sidered as time-invariant ones such that all the hierarchical sliding surfaces are
linear. Under any hierarchical structures, the sliding surfaces are xed and they wait
for the state trajectories entering the sliding mode. In this sense, these types of
sliding surfaces are passive.
To improve the crane control performance, one possible choice is to make xed
sliding surfaces active. Design of the adaptive sliding surfaces is just such a choice.
The adaptive sliding surfaces can move in the phase plane according to the
4.5 Adaptive Control Design Based on Hierarchical Sliding Surfaces 145
designed adaptive law so that they become active to search the state trajectories.
Such an adaptive design can make the state trajectories may enter the sliding mode
as soon as possible.
To realize the purpose, adaptive control design based on hierarchical sliding sur-
faces [11] is taken into consideration. Without loss of generality, the aggregated
hierarchical sliding surfaces are employed for the adaptive law design. The design
method can be generalized to the incremental and combining sliding surfaces as well.
Recall the crane model (2.16). Design the sliding surfaces of the trolley and payload
subsystems as
s 1 c1 x1 x2 4:66
s 2 c2 x3 x4 4:67
Here c1 and c2 are positive constants. According to the aggregated structure, both
the subsystem sliding surfaces (4.66) and (4.67) are located at the rst layer. In
(4.66) and (4.67), x2 and x4 are the derivatives of x1 and x3 with respect to time t,
respectively. Consequently, the subsystem sliding surfaces s1 and s2 must be in the
second and fourth quadrants to stabilize the sliding surface variables.
To construct the higher layer sliding surface, aggregate the two sliding surfaces
and design the second-layer sliding surface as
S as1 s2 4:68
where the equivalent control laws of the two subsystems are formulated by
1
ueq1 c1 x2 f1 x 4:70
b1 x
and
1
ueq2 c2 x4 f2 x 4:71
b2 x
So far, some parts of the adaptive control based on the aggregated sliding
surfaces have been done. The adaptive law of and the switching control law usw
will be drawn from the system stability analysis.
jS g sgnS
usw 4:72
ab1 x b2 x
b1 xueq2 s1 b1 xueq1 s1
a_ a 4:73
jjs1 jj2 d jjs1 jj2 d
1
Vt S2 4:74
2
_
Vt _ 1 s_ 2
Sa_s1 as 4:75
_
Vt _ 1 ac1 x_ 1 x_ 2 c2 x_ 3 x_ 4
Sas 4:76
_
Vt _ 1 ac1 x2 f1 x b1 xu c2 x4 f2 x b2 xug
Sfas 4:77
_
Vt _ 1 ac1 x2 f1 x b1 xueq1 ueq2 usw
Sfas
c2 x4 f2 x b2 xueq1 ueq2 usw g
4:78
_ 1 ab1 xueq2 usw b2 xueq1 usw g
Sfas
_ 1 ab1 x b2 xusw ab1 xueq2 b2 xueq1 g
Sfas
_
From (4.74), Vt 0. In the sense of Lyapunov, Vt\0 should exist in (4.78)
to make the crane control system asymptotically stable. To achieve this purpose,
dene the switching control law as (4.72) and select the adaptive law of the
_
parameter as (4.73). Then Vt\0 can be deduced from (4.78).
According to the control design of the adaptive control system, the 2nd-layer
sliding surface S is of asymptotic stability in the sense of Lyapunov.
Theorem 4.7 Consider the single-pendulum-type overhead crane system (2.16),
design the hierarchical sliding surfaces (4.66), (4.67), and (4.68), adopt the
aggregated HSMC law (4.69), dene the equivalent control laws of the two sub-
systems as (4.70) and (4.71), dene the switching control law as (4.72) and select
the adaptive law of the parameter as (4.73). Then, the rst-layer sliding surfaces
s1 and s2 in (4.70) and (4.71) are asymptotically stable.
Proof The proof of Theorem 4.7 is similar to the proof of Theorem 4.2.
To verify the feasibility of the adaptive control based on the hierarchical sliding
surfaces, some numerical simulation results will be demonstrated. The physical
parameters of the overhead crane system and the initial and desired states are kept
unchanged from Table 4.1.
148 4 Overhead Crane Control by Hierarchical Sliding Mode
Fig. 4.16 System performance by adaptive control based on the aggregated HSMC. a Trolley
position; b Trolley velocity; c Payload angle; d Payload angular velocity
The parameters of the two subsystem sliding surfaces at the rst layer are
selected as c1 = 0.7 and c2 = 8.2, which are kept unchanged from Sect. 4.2. The
coefcients of the switching control law are determined by = 3 and = 0.1. The
initial value of is determined by 2 and the value of in (4.73) is set by 0.5. Some
numerical results are displayed in Figs. 4.16, 4.17, 4.18, and 4.19 and are obtained
by a Simulink model of MATLAB in Appendix D.
The crane performance under the adaptive control law based on the aggregated
structure of sliding surfaces is illustrated in Fig. 4.16, where the trolley position, the
trolley velocity, the payload angle, and the payload angular velocity are displayed
in Fig. 4.16ad, respectively.
As illustrated in Fig. 4.16a, the trolley can arrive the desired position at about
6 s. Meanwhile, the maximum payload deviation during the crane transport is about
0.12 rad. Such a small deviation is good enough in practice to increase crane
effectiveness and throughput.
Compared with the numerical results by the aggregated HSMC system in
Fig. 4.2, the designed adaptive law does not have the crane transport efciency
because the trolley also arrives the desired position at about 6 s in Fig. 4.16a. But
the adaptive law can apparently decrease the payload deviation. In Fig. 4.16c, the
maximum angular deviation is just 0.12 rad, which is 0.02 rad smaller than the
maximum deviation in Fig. 4.2c.
4.5 Adaptive Control Design Based on Hierarchical Sliding Surfaces 149
Fig. 4.17 Control performance of adaptive control based on the aggregated HSMC. a Control
input u; b Subsystem sliding surface s1; c Subsystem sliding surface s2; d Sliding surface S
The control performance is displayed in Fig. 4.17, where the control input, the
subsystem sliding surface s1 at the rst layer, the subsystems sliding surface s2 at
the rst layer, and the aggregated sliding surface S at the second layer are shown in
Fig. 4.17ad, respectively. In Fig. 4.17a, the maximum control input is about
100 N. The aggregated sliding surface at the top layer in Fig. 4.17d is of asymptotic
stability, which coincides with the results proven in Theorem 4.6. On the other
150 4 Overhead Crane Control by Hierarchical Sliding Mode
Fig. 4.19 Phase portrait of the sliding surface S at the top layer
hand, the subsystem sliding surfaces in Fig. 4.17b, c are asymptotically stable as
proven in Theorem 4.7.
The adaptive process of the parameter at the top-layer sliding surface is dis-
played in Fig. 4.18. From Fig. 4.18, the value of is adaptively governed by the
adaptive law designed in (4.73). The value of tends to be stabilized at about 3.6
as t .
In Fig. 4.19, the phase trajectory of the top-layer sliding surface is shown, where
the solid line is the phase trajectory and the dash line is the adaptive top-layer
sliding surface. From Fig. 4.19, the adaptive sliding surface can change its slope in
the phase plane, indicating that the adaptive law can accelerate the phase trajectory
to reach the sliding mode.
[12], design the sliding surface vector s to construct the aggregated structure of
sliding surfaces and the vector s is described by
2 3 2 3
s1 k1 x1 xd x2
s 4 s2 5 4 k2 x 3 x 4 5 4:79
s3 k3 x 5 x 6
From (4.80), the equivalent control laws of the three subsystems are written as
1
ueq1 k1 x2 f1 x
b1 x
1
ueq2 k2 x4 f2 x 4:81
b2 x
1
ueq3 k3 x6 f3 x
b3 x
In (4.80), the equivalent control law ueq1 attracts a pair of state variables (x1, x2)
to component s1. Similarly, the duties of ueq2 and ueq3 are to force the pairs of state
variables (x3, x4) and (x5, x6) to approach components s2 and s3, respectively.
Therefore, to drive all state trajectories to reach the rst layer of the aggregated
structure of sliding surfaces, a total equivalent control law is dened by
In (4.82), all the parts are continuous. To guarantee the reachability of sliding
mode, the switching control law usw must be introduced such that the HSMC law
based on the aggregated sliding surfaces becomes
In (4.83), usw has a switching action with sufciently high frequency to retain all
state trajectories moving on manifold S = 0, where S is referred as the aggregated
sliding surface located at the top layer and the sliding surface is dened by
152 4 Overhead Crane Control by Hierarchical Sliding Mode
S lT s l1 s1 l2 s2 l3 s3 4:84
where = [1 2 3]T and 1, 2, and 3 are constants. From (4.84), the aggregated
structure of sliding surfaces of the crane control systems can be demonstrated in
Fig. 4.20.
In (4.84), S denotes the switching line that all system states are attracted to and
retained on. To have the reachability of sliding mode at the aggregated sliding
surface S, a Lyapunov function candidate can be dened by
1
V S2 4:85
2
V_ SS_ 4:86
V_ Sl1 s_ 1 l2 s_ 2 l3 s_ 3 4:87
_
From (4.85), Vt 0. In the sense of Lyapunov, Vt\0 should exist in (4.87)
to make the crane control system asymptotically stable. For this purpose, the
switching control usw can be dened by
4.6 HSMC Design for Double-Pendulum-Type Overhead Cranes 153
From the control design, the aggregated sliding surface located at the top layer is
asymptotically stable. The stabilities of the three subsystem sliding surfaces have to
be considered because they have no guaranteed stabilities from the control design.
Fig. 4.21 System performance by the HSMC law for double-pendulum-type overhead cranes.
a Trolley position; b Trolley velocity; c Hook angle; d Hook angular velocity; e Payload angle;
f Payload angular velocity
results are displayed in Figs. 4.21, and 4.22 and are obtained by a Simulink model
of MATLAB in Appendix E.
The crane performance under the HSMC law is illustrated in Fig. 4.21, where the
trolley position, the hook angle, the hook angular velocity, the trolley velocity, the
payload angle, and the payload angular velocity are displayed in Fig. 4.21af,
respectively.
From Fig. 4.21, the trolley arrives the desired position at about 12 s. Meanwhile,
both the maximum angular deviations of the hook and payload are not larger than
0.1 rad. Such deviations are small enough in real crane applications. On the other
hand, the performance of the two angular curves by the HSMC method has no
residual oscillations under the condition that the crane model has no modeling
errors. However, such a perfect model is too ideal to obtain. As a result, residual
oscillations of the double-pendulum dynamics are hard to be completely suppressed
by the HSMC method in practice.
The control performance is displayed in Fig. 4.22, where the control input, the
subsystem sliding surface s1, the subsystems sliding surface s2, the subsystems
4.6 HSMC Design for Double-Pendulum-Type Overhead Cranes 155
Fig. 4.22 Control performance by the HSMC law for double-pendulum-type overhead cranes.
a Control input u; b Subsystem sliding surface s1; c Subsystem sliding surface s2; d Subsystem
sliding surface s3; e Top-layer sliding surface S
sliding surface s3, and the aggregated sliding surface S are shown in Fig. 4.22ae,
respectively.
From Fig. 4.22a, the control input has chattering, which is an inner drawback of
the SMC method. The whole sliding surfaces in Fig. 4.22be are asymptotically
stable as proven in Theorem 4.8. From Fig. 4.22e, the aggregated sliding surface at
the top layer enters its sliding mode at about 3 s. Thereafter, the trajectories of the
three subsystems in phase plane enter their private subsystem sliding surfaces under
the actions of their private subsystem equivalent controls.
Further, concerning these sliding surfaces of the aggregated hierarchical struc-
ture, only the reachability condition of the sliding surface S at the top layer is
guaranteed during the control design. The three subsystem sliding surfaces s1, s2,
and s3 are just asymptotically stable rather than being reachable in nite time.
156 4 Overhead Crane Control by Hierarchical Sliding Mode
Appendices
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 0;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 4;
sizes.NumInputs = 4;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlOutputs(t,x,u)
% parameter of system
g=9.81;M=37.32;m=5;l=1.05;
A=(M+m);B=(m*l*cos(u(3)));C=(cos(u(3)));D=l;E=(m*l*sin(u(3)));
% parameter of controller
e(1)=(u(1));e(2)=u(2);e(3)=u(3);e(4)=u(4);
a =-2.3;afa =[0.7 8.2];bata=[1 1];
k=3;ita=0.1;
s1=afa(1)*e(1)+bata(1)*e(2);S1=s1;
s2=afa(2)*e(3)+bata(2)*e(4);S2=a*S1+s2;s=S2;
b1 =D*inv(A*D-B*C);
f1 =(m*l*u(4)^2*sin(u(3))*D+g*B*sin(u(3)))*inv(A*D-B*C);
b2 =C*inv(B*C-A*D);
f2 =(m*l*u(4)^2*sin(u(3))*C+g*A*sin(u(3)))*inv(B*C-A*D);
den=a*bata(1)*b1+bata(2)*b2;
num=a*(afa(1)*u(2)+bata(1)*f1)+(afa(2)*u(4)+bata(2)*f2);
ds=-k*s-ita*sign(s);
uslaw=-num*inv(den)+ds*inv(den);
sys(1)=uslaw;
sys(2)=s1;
sys(3)=s2;
sys(4)=S2;
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 0;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 5;
sizes.NumInputs = 4;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlOutputs(t,x,u)
% system prameters
g=9.81;M=37.32;m=5;l=1.05;
f1=(m*l*u(4)^2*sin(u(3))+m*g*sin(u(3))*cos(u(3)))/(M+m-m*(cos(u(3)))^2);
b1=1/(M+m-m*(cos(u(3)))^2);
f2=(m*l*u(4)^2*sin(u(3))*cos(u(3))+(M+m)*g*sin(u(3)))/...
(m*l*(cos(u(3)))^2-(M+m)*l);
b2=(cos(u(3)))/(m*l*(cos(u(3)))^2-(M+m)*l);
s1=c(1)*u(1)+c(2)*u(2);
s2=s1+c(3)*u(3);
s=s2+c(4)*u(4);
Appendices 159
if u(3)*s1<0
c(3)=-c(3);
else
c(3)=c(3);
end
ds=-k*s-ita*sign(s);
den=c(2)*b1+c(4)*b2;num=c(1)*u(2)+c(2)*f1+c(3)*u(4)+c(4)*f2;
uslaw=-num*inv(den)+ds*inv(den);
sys(1) = uslaw;
sys(2) =s1;
sys(3)=s2;
sys(4)=s;
sys(5)=c(3);
C Simulink Model to Plot Figs. 4.11, 4.12, 4.13, 4.14 and 4.15
160 4 Overhead Crane Control by Hierarchical Sliding Mode
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 0;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 5;
sizes.NumInputs = 4;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlOutputs(t,x,u)
% system prameters
M=1;m=0.8;L=0.305;g=9.81;
c = [0.242 0.487];
c1=c(1);c2=c(2);k0=8;eta=0.04;
f1=(m*L*u(4)^2*sin(u(3))+m*g*sin(u(3))
*cos(u(3)))*inv(M+m*sin(u(3))*sin(u(3)));
b1=1*inv(M+m*sin(u(3))*sin(u(3)));
f2=-((M+m)*g*sin(u(3))+m*L*u(4)^2*sin(u(3))*cos(u(3)))
*inv((M+m*sin(u(3))*sin(u(3)))*L);
b2=-cos(u(3))*inv((M+m*sin(u(3))*sin(u(3)))*L);
Appendices 161
z=(u(1))+c1*u(3);
dz=u(2)+c1*u(4);
s=c2*z+dz;
ds=-k0*s-eta*sign(s);
den=c1*b2+b1;
num=(c1*f2+f1+c2*u(2)+c2*c1*u(4));
uslaw=-num*inv(den)+ds;
sys(1) = uslaw;
sys(2) =z;
sys(3) =dz;
sys(4) =s;
sys(5) =c1;
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 1;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 5;
sizes.NumInputs = 4;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlDerivatives(t,x,u)
% parameter
g=9.81;M=37.32;m=5;l=1.05;
A=(M+m);B=(m*l*cos(u(3)));C=(cos(u(3)));D=l;E=(m*l*sin(u(3)));
delta=0.6;
b1 =D*inv(A*D-B*C);
f1 =(m*l*u(4)^2*sin(u(3))*D+g*B*sin(u(3)))*inv(A*D-B*C);
b2 =C*inv(B*C-A*D);
f2 =(m*l*u(4)^2*sin(u(3))*C+g*A*sin(u(3)))*inv(B*C-A*D);
afa =[0.7 8.2];
ueq2=-(afa(2)*u(4)+f2)*inv(b2);
ueq1=-(afa(1)*u(2)+f1)*inv(b1);
s1=afa(1)*u(1)+u(2);
sys(1)=-x(1)*b1*ueq2*s1*inv(s1^2+delta)-(b2)*ueq1*s1*inv(s1^2+delta);
Appendices 163
function sys=mdlOutputs(t,x,u)
% parameter of system
g=9.81;M=37.32;m=5;l=1.05;
A=(M+m);B=(m*l*cos(u(3)));C=(cos(u(3)));D=l;E=(m*l*sin(u(3)));
e(1)=(u(1));e(2)=u(2);e(3)=u(3);e(4)=u(4);
k=3; ita=0.1;
s1=afa(1)*e(1)+bata(1)*e(2);
S1=s1;
s2=afa(2)*e(3)+bata(2)*e(4);
S2=x(1)*S1+s2;
s=S2;
b1 =D*inv(A*D-B*C);
f1 =(m*l*u(4)^2*sin(u(3))*D+g*B*sin(u(3)))*inv(A*D-B*C);
b2 =C*inv(B*C-A*D);
f2 =(m*l*u(4)^2*sin(u(3))*C+g*A*sin(u(3)))*inv(B*C-A*D);
den=x(1)*bata(1)*b1+bata(2)*b2;
num=x(1)*(afa(1)*u(2)+bata(1)*f1)+(afa(2)*u(4)+bata(2)*f2);
ds=-k*s-ita*sign(s);
uslaw=-num*inv(den)+ds*inv(den);
sys(1)=uslaw;
sys(2)=s1;
sys(3)=s2;
sys(4)=S2;
sys(5)=x(1);
164 4 Overhead Crane Control by Hierarchical Sliding Mode
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 0;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 5;
sizes.NumInputs = 8;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlOutputs(t,x,u)
%system parameter
mt=50;mc=2;mh=10;g=9.81;b=0;l1=3;l2=0.3;
A=(mt+mh+mc);B=(mh+mc);
%desired parameter
dis=4;ddis=0;phi=0;dphi=0;theta=0;dtheta=0;
%designed parameter
K=1;ante=0.02;a1=0.35;a2=22;a3=50;b1=1.18;b2=1.2;b3=-0.35;
%control law
fq=b1/A-b2*cos(u(3))/(l1*(B*(cos(u(3)))^2-A))-
b3*cos(u(6))/(l2*(mc*(cos(u(6)))^2-A));
gq=b1*(B*l1*cos(u(3))*u(5)+mc*l2*cos(u(6))*u(8)-b*u(2)-
.B*l1*sin(u(3))*(u(4))^2-mc*l2*sin(u(6))*(u(7))^2)/A...
+b2*(mc*l2*(mt*cos(u(3)-
u(6))+B*sin(u(3))*sin(u(6)))*u(8)+B*A*g*sin(u(3))...
+mc*l2*(mt*sin(u(3)-u(6))+B*sin(u(3))*cos(u(6)))*(u(7))^2-...
l1*B^2*cos(u(3))*sin(u(3))*(u(4))^2
+B*cos(u(3))*b*u(2))/(l1*B*(B*(cos(u(3)))^2-A))...
+b3*(l1*(mt*cos(u(3)-u(6))+B*sin(u(3))*sin(u(6)))*u(5)
+cos(u(6))*b*u(2)+A*g*sin(u(6))...
+l1*(B*sin(u(6))*cos(u(3))-mt*sin(u(3)-
u(6)))*(u(4))^2+mc*l2*cos(u(6))*sin(u(6))*(u(7))^2)/...
(l2*(mc*(cos(u(6)))^2-A))+b1*a1*(u(2))+b2*a2*(u(4))+b3*a3*(u(7));
control_law=-inv(fq)*(gq+K*S+ante*sign(S));
sys(1)=control_law;
sys(2)=s1;
sys(3)=s2;
sys(4)=s3;
sys(5)=S;
References
1. Saridis GN (1983) Intelligent robotic control. IEEE Trans Autom Control 28(5):547557
2. Wang W, Yi J, Zhao D, Liu D (2004) Design of a stable sliding-mode controller for a class of
second-order underactuated systems. IEE Proc Control Theory Appl 151(6):683690
3. Wang W, Liu XD, Yi JQ (2007) Structure design of two types of sliding-mode controllers for
a class of under-actuated mechanical systems. IET Control Theory Appl 1(1):163172
4. Qian DW, Yi JQ (2010) Fuzzy aggregated hierarchical sliding mode control for underactuated
systems, In: Proceedings of 2010 international conference on mechatronics and automation,
Xian, China, pp 196201
5. Hirschorn RM (2002) Incremental sliding mode control of the ball and beam. IEEE Trans
Autom Control 47(10):16961700
6. Hao YX, Yi JQ, Zhao DB, Wang W (2006) Proposal of incremental sliding mode control. In:
Proceedings of rst international conference on innovative computing, information and
control, Beijing, China, pp 340343
7. Hao YX, Yi JQ, Zhao DB, Qian DW (2007) Incremental sliding mode controller for
large-scale underactuated system. In: Proceedings of IEEE international conference on
networking, sensing and control, London, UK, pp 8792
8. Qian DW, Yi JQ, Ma YF (2010) Fuzzy incremental hierarchical sliding mode control for
underactuated systems. In: Proceedings of international conference on articial intelligence
and computational intelligence, Sanya, China, pp 276280
9. Qian DW, Yi JQ, Zhao DB (2011) Control of overhead crane systems by combining sliding
mode with fuzzy regulator. In: Proceedings of the 18th IFAC World Congress, Milano, Italy,
pp 93209325
10. Qian DW, Yi JQ (2013) Design of combining sliding mode controller for overhead crane
systems. Int J Control Autom 6(1):131140
11. Qian DW, Liu XJ, Yi JQ (2012) Adaptive control based on hierarchical sliding mode for
under-actuated systems. In: Proceedings of international conference on mechatronics and
automation, Chengdu, China, pp 10501055
166 4 Overhead Crane Control by Hierarchical Sliding Mode
12. Tuan LA, Lee SG (2013) Sliding mode controls of double-pendulum crane systems. J Mech
Sci Technol 27(6):18631873
13. Utkin V, Guldner J, Shi J (2009) Sliding mode control in electromechanical systems, 2nd edn.
Taylor & Francis, USA
Chapter 5
Compensator Design Based on Sliding
Mode for Uncertain Overhead Cranes
Abstract Crane systems have uncertainties. The uncertainties cover both matched
and unmatched ones. The methodology of sliding mode control (SMC) can sup-
press matched uncertainties in light of the invariance of sliding mode. Concerning
the crane control problem, unmatched uncertainties become challenging. This
chapter focuses on compensator design based on sliding mode for uncertain
overhead cranes. Three design methods are taken into account, i.e., compensator
design based on hierarchical sliding surfaces, compensator design based on fuzzy
logic, and compensator design based on neural networks (NNs). For the purpose of
illustration, an overhead crane is adopted as benchmark, and each control method is
carried out by the transport control problem of the crane. In addition, the chapter
proceeds with source codes of all the simulations.
Recall uncertain crane models (2.20) and (2.47). Whatever types of overhead crane
systems are, but overhead crane models cover uncertain terms. Without loss of
generality, the model of single-pendulum-type overhead cranes is employed to
illustrate the design methods of compensator. These design methods can be
extended to double-pendulum-type overhead cranes as well.
Consider the crane mode (2.20). 1 and 2 in (2.20) cover both matched and
unmatched uncertainties. Due to the invariance of sliding mode control (SMC),
matched parts of uncertainties in the crane model can be overcome. As far as crane
control is concerned, a crucial issue is how to deal with unmatched uncertainties.
This chapter attacks the issue. Some design methods will be developed to achieve
the robustness of crane control based on SMC.
Crane systems are inherently uncertain and nonlinear. Model uncertainties are
inevitable in practice. The crane dynamics are subject to uncertainties. The
uncertainties, including parameter fluctuations, model uncertainties and random
external disturbances, cover both matched and unmatched uncertainties, where the
unmatched uncertainties cannot be suppressed by the invariance of SMC.
The uncertain crane mode (2.20) can be divided into two parts, i.e., nominal
crane model and uncertainties. Consider the nominal crane model (2.16). Many
design methods based on SMC have been introduced in Chaps. 3 and 4. Section 5.2
investigates the problem of compensator design based on the aggregated hierar-
chical SMC.
Generally speaking, there are two methods to design a compensator on basis of the
hierarchical sliding surfaces [1]. One is to design a distributed compensator and
compensate the unmatched uncertainties at every layer of the hierarchical sliding
mode surfaces [2]. Two disadvantages of this idea are that this makes the controller
structure complex and that if the compensator at a lower layer does not eliminate
the uncertainties, it will affect the stabilities of higher layers. The other method is to
design a lumped compensator and compensate the unmatched uncertainties at the
top layer. Its advantage is that this method simplies the control design. Thus, a
lumped sliding mode compensator at the top layer is designed in Sect. 5.2.
Consider the crane mode (2.20). Design the aggregated hierarchical sliding
surfaces illustrated in Fig. 4.1 and formulate the sliding surfaces as (4.1), (4.2) and
(4.3). To overcome the uncertainties, the control input u of the uncertain crane
system is dened by
u un ucn 5:1
where ucn is the law of compensation based on the aggregated hierarchical sliding
surfaces; un is the aggregated HSMC law and its design and analysis have been
introduced in detail in Sect. 4.2.
In (5.1), some expressions about the control law un for the nominal crane system
will be directly employed without more explanations. However, ucn is kept
unknown, and its expression will be deduced from the stability analysis of the crane
control system.
Because the lumped compensator is located at the top layer of the hierarchical
sliding surfaces, the stability of the top-layer sliding surface S should be rst
considered. To have the stability of the sliding surface S, a Lyapunov function
candidate is dened by
5.2 Compensator Design Based on HSMC 169
S2
Vt 5:2
2
Differentiate V with respect to time t in (5.2) and substitute (4.3) into the
derivative of V. Then, (5.3) can be obtained.
_
Vt SS_ Sa_s1 s_ 2 5:3
Consider the crane mode (2.20). Substituting (4.1) and (4.2) into (5.3) yields
_
Vt Sac1 x_ 1 x_ 2 c2 x_ 3 x_ 4
Sac1 x2 af1 x ab1 xu 5:4
c2 x4 f2 x b2 xu an1 n2
Substitute the control input (5.1) into (5.4). Then, (5.4) becomes
_
Vt Sfac1 x2 af1 x ab1 xun c2 x4 f2 x b2 xun
5:5
ab1 x b2 xucn an1 n2 g
Substituting the aggregated HSMC law of the nominal crane system (4.15) into
(5.5) yields
_
Vt SfjS g sgnS ab1 x b2 xucn an1 n2 g
5:6
jS2 gjSj Sfab1 x b2 xucn an1 n2 g
d0 sgnS
ucn 5:7
ab1 x b2 x
_
Vt jS2 gjSj Sd0 sgnS an1 n2
5:8
jS2 gjSj Sd0 sgnS jjan1 n2 jj
_
Vt\0 can be derived from (5.8) because of d0 sup jan1 j sup jn2 j. As a result,
the aggregated sliding surface S located at the top layer is of asymptotic stability in
the crane control system with unmatched uncertainties.
By means of the control scheme, the HSMC law of the nominal crane system in
(4.15) and the compensation law of the unmatched uncertainties in (5.7) work
together to achieve the strong robustness of the crane control system against
170 5 Compensator Design Based on Sliding Mode
unmatched uncertainties. At last, the nal control aw of the uncertain crane system
has the form of
u un ucn
ueq1 ueq2 usw ucn
5:9
ab1 xueq1 b2 xueq2 jS gsgnS d0 sgnS
ab1 x b2 x ab1 x b2 x
unchanged from Sect. 4.2 and the coefcients of the switching control law are also
kept unchanged from Sect. 4.2.
In the uncertain crane model (2.20), both the uncertain terms 1 and 2 are set by
0.1 [2 rand()1], where rand() is a Matlab command to generate a random
number drawn from the standard uniform distribution on the open interval (0, 1).
Concerning the compensator design, only a parameter d0 has to be set according to
the two supremums of unmatched uncertainties. Here d0 is selected as 0.2.
Some results are displayed in Figs. 5.1 and 5.2, obtained by a Simulink model of
MATLAB in Appendix A. The crane performance is illustrated in Fig. 5.1, where
the trolley position, the trolley velocity, the payload angle, and the payload angular
velocity are displayed in Fig. 5.1ad, respectively. From Fig. 5.1a, the trolley can
arrive at the origin from the initial position at about 6 s. The maximum payload
deviation is about 0.14 rad in Fig. 5.1c. During the crane transportation, such a
small deviation is good enough in practice to increase crane effectiveness and
throughout.
In Fig. 5.1c, d, there are some slightly residual oscillations because of the
existence of uncertainties. Although the compensator plus controller structure
cannot realize the zeroth residual oscillations compared with the nominal crane
control system in Sect. 4.2, the control scheme can still resist unmatched uncer-
tainties. In Fig. 5.1, the aggregated HSMC law of the nominal crane system works
together with the compensation law of uncertainties to realize the robust control of
the uncertain crane system.
The control performance is displayed in Fig. 5.2, where the control input u, the
compensation law ucn, the subsystem sliding surface s1 at the rst layer, the
Fig. 5.1 System performance by the compensator design based on the aggregated HSMC method.
a Trolley position; b Trolley velocity; c Payload angle; d Payload angular velocity
172 5 Compensator Design Based on Sliding Mode
Fig. 5.2 Control performance by the compensator design on basis of the aggregated HSMC
method. a Control input u; b Compensation law; c Subsystem sliding surface s1 at the rst layer;
d Subsystem sliding surface s2 at the rst layer; e Second-layer sliding surface S
subsystems sliding surface s2 at the rst layer, and the aggregated sliding surface
S at the second layer are shown in Fig. 5.2ae, respectively. In Fig. 5.2a, the
maximum of the absolute control input is about 100 N. The maximum of the
absolute compensation law is about 2.3 N. The fact indicates that the part of the
compensation law is very small in the total control law. In Fig. 5.2e, the sliding
mode of the aggregated sliding surface at the top layer is reachable at about 2 s.
Thereafter, the two sliding surfaces at the rst layer tend to their subsystems origins
along the two subsystem sliding surfaces.
A critical assumption in Sect. 5.2 is that the system uncertainties must have a
known boundary in (5.7). However, the boundary value is hard to exactly know in
practice. The lack of knowing the important information may cause several prob-
lems such as deciency of the system stability, decrease of the system robustness,
and deterioration of the system performance. To turn SMC-based compensator
methods into practical account on the transport control problem of uncertain crane
systems, it is necessary to approximate and compensate the uncertainties.
5.3 Sliding Mode-Based Fuzzy Compensator Design 173
The methodology of fuzzy logic has been proven to be effective in dealing with
complex nonlinear systems containing uncertainties that are otherwise difcult to
model. The fuzzy logic method has been applied to many practical applications,
especially in MEMS triaxial gyroscope system [3], mechanical manipulator [4],
vector voltmeter [5], etc. The fuzzy logic method is also a possible choice to attack
the uncertainties of the SMC-based crane control problem.
A fuzzy inference system (FIS) [6] consists of four parts: knowledge base, fuzzier,
fuzzy inference engine, and defuzzier, where the knowledge base is composed of
some fuzzy if-then rules and the fuzzy inference engine employs the rules.
Consider a multi-input multi-output FIS. The structure of this FIS is displayed in
Fig. 5.3, where the input vector is a = [a1, a2, , am]T 2 <m , the output vector
b = [b1, b2, , bn]T 2 <n , and every element in the vectors a and b is scalar. The
multi-input multi-output if-then rules can be expressed by
R [M
l1 Rl 5:10
Alp and Blq in (5.11) are the linguistic variables of the fuzzy sets, where p = 1, 2,
, m and q = 1, 2, , n. Both the linguistic variables are determined by their
membership functions lAlp ap and lBlq bq , here l = 1, 2, , M.
Adopting the singleton fuzzier mapping, the product inference engine, and the
center-average defuzzier mapping, the qth output of the FIS has a form of
X
M
bq bl wl a HT wa 5:12
q q
l1
Hq b1q ; . . .;
bM T l
q 2 < is named the parameter vector, where bq is an adaptively
M
PB, denoting negative big, negative small, zero, positive small, and positive big,
respectively. The membership function of Alp is Gaussian, dened by
!
ap cp 2
lAlp ap exp 5:14
2o2p
where cp and op are the center and width of the Gaussian function and p = 1, 2, 3, 4.
From (5.12), the approximate values b ^n1 ; ^n2 T can be calculated by
" # " #
^n1 _
HT1 wx; x_ ; h; h
T _
H wx; x_ ; h; h 5:15
^n2 H wx; x_ ; h; h
T _
2
_ is 625 1.
where the approximate matrix H is 625 2 and wx; x_ ; h; h
5.3 Sliding Mode-Based Fuzzy Compensator Design 175
Provided that an optimal parameter matrix H exists, the matrix has a form of
" " # #
^n n1
1
H arg min sup 5:16
H2v0 ^n n 2
2
Consider the uncertain crane system (2.20), adopt the aggregated structure of
sliding surfaces, design the HSMC law for the nominal crane system, and com-
pensate the uncertainties by the lumped compensator (5.7) at the top layer of the
hierarchical sliding surfaces. However, the boundary values of the uncertainties are
kept unknown so that the control law of the uncertain crane (5.1) cannot be directly
employed here. To approximate the boundary values, the designed FIS is adopted,
where the approximation values of the boundary values are the network outputs.
Finally, the control law can be rewritten by
In (5.18), and are positive scalars; d^0 determined by the network outputs has
the form of d^0 ja^n1 j j^n2 j.
H _ and
_ 1 a1 C1 S wx; x_ ; h; h _
_ 2 C1 S wx; x_ ; h; h
H 5:19
1 2
1 1X 2
~ T Cq H
~q
Vt S2 H 5:20
2 2 q1 q
~ q H Hq .
where H q
Differentiating (5.20) with respect to time t yields
X
2
_
Vt SS_ H ~_ q
~ T Cq H 5:21
q
q1
S_ a_s1 s_ 2
ac1 x_ 1 x_ 2 c2 x_ 3 x_ 4 5:22
ac1 x2 f1 b1 u n1 c2 x4 f2 b2 u n2
Consider the control law for the uncertain crane system (5.18) and the HSMC
law of the nominal crane system. Equation (5.23) can be deduced from (5.22).
Consider the FIS output (5.12). Substituting (5.13) into (5.21) yields
X
2
_
Vt SjS gsgnS d^0 sgnS an1 n2 H ~_ q
~ T Cq H
q
q1
X
2
SjS gsgnS aS^n1 n1 S^n2 n2 ~_ q
~ T Cq H
H 5:24
q
q1
X
2
jS2 gjSj H ~_ q SH
~ T Cq H _
~ T wx; x_ ; h; h
q q
q1
_
Vt jS2 gjSj aq1 q2 S
5:25
jS2 gjSj aq1 q2 jSj
_
Vt\0 in (5.25) exist because of |1| + |2|. In the sense of Lyapunov, the
crane control system is asymptotically stable. As a result, the adaptive law (5.19) of
the designed FIS is convergent and the aggregated sliding surface at the top layer is
of asymptotic stability. h
Theorem 5.3 Consider the single-pendulum-type overhead crane system with
uncertainties (2.20), design the hierarchical sliding surfaces (4.1), (4.2) and (4.3),
construct the FIS (5.12), and dene the control law (5.18). Then, the subsystem
sliding surfaces s1 and s2 are of asymptotic stability.
Proof The proof of Theorem 5.3 is very similar to the proof of Theorem 4.2. h
To verify the feasibility and validity of the fuzzy compensator design based on
HSMC for uncertain crane systems, some numerical results are illustrated in
Figs. 5.4, 5.5 and 5.6, obtained by a Simulink model of MATLAB in Appendix B.
Some physical parameters of the crane are determined in Table 4.1. The initial
position of the trolley is located at the origin and the desired position is assigned to
x = 1 m. The control law based the hierarchical sliding surfaces for the nominal
crane system is adopted, where the parameters of the hierarchical sliding surfaces
are kept unchanged from Sect. 4.2 and the coefcients of the switching control law
are also kept unchanged from Sect. 4.2.
The centers of the Gaussian function (5.14) are set by /2, /4, 0, /4, and /2.
p
The width of the Gaussian function is set by 2 /8. Both 1 and 2 in (5.19) are
178 5 Compensator Design Based on Sliding Mode
set by 0.5. In the uncertain crane model (2.20), both the uncertain terms 1 and 2
are set by 0.1 [2 rand() 1], where rand() is a MATLAB command to generate
a random number drawn from the standard uniform distribution on the open
interval (0, 1).
The crane performance is illustrated in Fig. 5.4, where the trolley position, the
trolley velocity, the payload angle, and the payload angular velocity are displayed
in Fig. 5.4ad, respectively. From Fig. 5.4, the trolley can arrive at the desired
position from the origin at about 6 s. During the crane movement, the maximum
payload deviation is about 0.05 rad. Such a small deviation is good enough in real
crane applications to increase effectiveness and throughout. However, the payload
angle has some slight residual oscillations from the angle and angular velocity
curves in Fig. 5.4c, d. From Fig. 5.4, the aggregated HSMC law of the nominal
crane system works together with the fuzzy compensation law of uncertainties
against the uncertainties of the crane system. Such combinations can realize the
strong robustness against uncertainties of crane control systems.
The curves of the aggregated sliding surfaces are displayed in Fig. 5.5, where the
sliding surface at the top layer S, the subsystem sliding surface s1 at the rst layer
and the subsystem sliding surface s2 at the rst layer are shown in Fig. 5.5ac,
respectively. From Fig. 5.5, the sliding surface at the top layer rst arrives at its
sliding mode at about 1 s. Then, the state trajectories of the two subsystems slide
Fig. 5.4 System performance by the fuzzy compensator design based on the aggregated HSMC
method. a Trolley position; b Trolley velocity; c Payload angle; d Payload angular velocity
5.3 Sliding Mode-Based Fuzzy Compensator Design 179
Fig. 5.5 Hierarchical sliding surfaces. a Sliding surface S at the top layer; b Subsystem sliding
surface s1 at the rst layer; c Subsystem sliding surface s2 at the rst layer
along their private sliding surface and converge to their subsystem origins. As
proven in Theorems 5.2 and 5.3, only the reachability of the sliding surface at the
top layer is guaranteed. The sliding modes of the subsystem sliding surfaces cannot
be reached at nite time and their sliding modes are asymptotic.
Figure 5.6 shows the control performance by the fuzzy compensation law based
on the aggregated HSMC method, where the control input u, the compensator
output ^
n1 , and the compensator output ^n2 are displayed in Fig. 5.6ac, respectively.
From Fig. 5.6a, the maximum control input is about 40 N. However, u has to switch
in high frequency to resist the system uncertainties. From Fig. 5.6b, c, the two
compensators can adaptively track the system uncertainties according to the
adaptive law (5.14) till the compensator outputs tend to be stable.
Neural networks [7] use algorithms in their programming, and they use weights to
change the parameters of the throughput and the varying connections to the neu-
rons. The neural network types vary a lot, where the type of radial basis function
(RBF) neural networks (NNs) is often employed as a universal approximator [8].
In [9], a gradient-type method on basis of RBF NNs is proposed to deal with the
180 5 Compensator Design Based on Sliding Mode
Fig. 5.6 Control performance by the fuzzy compensator design based on the aggregated HSMC
method. a Control input u; b Compensator output ^
n1 ; c Compensator output ^
n2
dead band nonlinearity. Provided that the crane uncertainties have an unknown
boundary, the RBF NNs in Sect. 5.4 are designed to compensate the crane
uncertainties. Weight update formulas of the networks are deduced from Lyapunov
direct method, so the weight convergence and system stability are simultaneously
guaranteed in the sense of Lyapunov.
RBF networks are a kind of three-layer feed-forward networks, where the mapping
from the input layer to the output layer is inherently nonlinear but the mapping from
the hidden layer to the output layer is linear. Since RBF NNs have the ability to
approximate complex nonlinear mapping directly from inputoutput data with a
simple topological structure, such kind of NNs is adopted here to realize the
compensator design [10].
5.4 Sliding Mode-Based Neural Compensator Design 181
The prototype of the designed RBF networks is illustrated in Fig. 5.7. In Fig. 5.7,
there are m neurons at the input layer and each element xq is employed as an input
element, where q = 1, , m. There are l neurons in the hidden layer. At the output
layer, only 1 neuron is assigned and the network output is written by y. In other
words, the designed RBF networks have m inputs and 1 output.
Associated with the application of the RBF networks, the purpose is to
approximate the uncertainties in the crane dynamics (2.20) so that the approximate
values are denitely selected as the network output. To approximate the two
uncertain terms in (2.20), two RBF networks have to be employed and the ith
network output is described by yi ^ni (i = 1, 2). Moreover, the system uncertainties
have a direct effect on the crane performance so that the state variables x
_ T in the crane dynamics (2.20) are picked up as the network inputs.
x; x_ ; h; h
From Fig. 5.7, the network output of the ith network can be calculated by
^ni x; xi xT hx 5:26
i
where xi 2 <l is the weight vector of the ith RBF network; hx 2 <l is the
Gaussian function vector and the pth element hp(x) of the vector hp(x) is dened by
!
jjx cp jj2
hp x exp 5:27
2o2p
Consider the uncertain crane system (2.20), adopt the aggregated structure of
sliding surfaces, design the HSMC law for the nominal crane system, and com-
pensate the uncertainties by the lumped compensator (5.7) at the top layer of the
hierarchical sliding surfaces. However, the boundary values of the uncertainties are
kept unknown so that the control law of the uncertain crane (5.1) cannot be directly
employed here.
To approximate the boundary values, both of the designed RBF networks are
adopted and the control law can be rewritten by
u un ucn
ueq1 ueq2 usw ucn
5:29
ab1 xueq1 b2 xueq2 jS gsgnS d^0 sgnS
ab1 x b2 x ab1 x b2 x
Assumption 5.1 For either of the designed RBF networks, there exists an optimal
weight vector xi such that the network output satises
T
x hx sup jni j ex\ei 5:30
i 0
Assumption 5.2 The boundary value sup jni j minus the norm of the system
uncertainties satises the following formula
x_ 1 aC1
1 jSj hx
5:32
x_ 2 C1
2 jSj hx
Proof Take the positive denite function (5.33) into account as a Lyapunov can-
didate function
1 X2
C1 T
Vt S2 ~ x
x ~i 5:33
2 i1
2 i
~ i is determined by
In (5.33), x
~ i xi xi
x 5:34
X
2
_
Vt SS_ ~ Ti x_ i
Ci x 5:35
i1
S_ a_s1 s_ 2 ac1 x_ 1 x_ 2 c2 x_ 3 x_ 4
5:36
ac1 x2 f1 b1 u n1 c2 x4 f2 b2 u n2
Consider the control law for the uncertain crane system (5.29) and the HSMC
law of the nominal crane system. Equation (5.37) can be deduced from (5.36).
S_ ac1 x2 f1 c2 x4 f2 ab1 b2 un
ab1 b2 ucn an1 n2
5:37
jS gsgnS ab1 b2 ucn an1 n2
jS gsgnS d^0 sgnS an1 n2
Consider the outputs (5.28) of the RBF neural networks. Substituting (5.37) into
(5.35) yields
X
2
_ SjS gsgnS a^n1 n1 ^n2 n2
Vt ~ Ti x_ i
Ci x 5:38
i1
_
Concerning Assumptions 5.1 and 5.2, Vt\0 in (5.39) exists. In the sense of
Lyapunov, the crane control system is asymptotically stable. Consequently, the
adaptive law (5.32) of the designed RBF networks is convergent and the aggregated
sliding surface at the top layer is of asymptotic stability. h
Theorem 5.5 Consider the single-pendulum-type overhead crane system with
uncertainties (2.20), design the hierarchical sliding surfaces (4.1), (4.2) and (4.3),
construct the RBF networks (5.26), and dene the control law (5.29). Then, the
subsystem sliding surfaces s1 and s2 are of asymptotic stability.
Proof The proof of Theorem 5.5 is very similar to the proof of Theorem 4.2. h
To verify the feasibility and validity of the neural compensator design based on
HSMC for uncertain crane systems, some numerical results are illustrated in
Figs. 5.8, 5.9 and 5.10, obtained by a Simulink model of MATLAB in Appendix C.
Some physical parameters of the crane are determined in Table 4.1. The initial
position of the trolley is located at x = 2 m and the desired position is assigned to
the origin. The control law based the hierarchical sliding surfaces for the nominal
crane system is adopted, where the parameters of the hierarchical sliding surfaces
are kept unchanged from Sect. 4.2 and the coefcients of the switching control law
are also kept unchanged from Sect. 4.2.
Both the centers and width of the Gaussian function (5.27) [11] are set by 1. The
initial weights of the two neural networks are designed as random numbers on the
open interval (0, 1). Both 1 and 2 in (5.19) are set by 1079. In the uncertain crane
model (2.20), both the uncertain terms 1 and 2 are set by 0.1 [2 rand() 1],
where rand() is a MATLAB command to generate a random number drawn from
the standard uniform distribution on the open interval (0, 1).
The crane performance is illustrated in Fig. 5.8, where the trolley position, the
trolley velocity, the payload angle, and the payload angular velocity are displayed
in Fig. 5.8ad, respectively. From Fig. 5.8a, the trolley can arrive at the origin from
the initial position at about 6 s. The maximum angular deviation is about 0.2 rad.
From Fig. 5.8b, d, the payload angle has some residual oscillations because of the
adverse effects of uncertainties. From Fig. 5.8, the aggregated HSMC law of the
nominal crane system works together with the neural network-based compensation
law of uncertainties against the uncertainties of the crane system. Such combina-
tions can realize the strong robustness against uncertainties of crane control
systems.
The curves of the aggregated sliding surfaces are displayed in Fig. 5.9, where the
sliding surface at the top layer S, the subsystem sliding surface s1 at the rst layer
and the subsystem sliding surface s2 at the rst layer are shown in Fig. 5.9ac,
5.4 Sliding Mode-Based Neural Compensator Design 185
Fig. 5.8 System performance by the neural compensator design based on the aggregated HSMC
method. a Trolley position; b Trolley velocity; c Payload angle; d Payload angular velocity
Fig. 5.9 Hierarchical sliding surfaces. a Sliding surface S at the top layer; b Subsystem sliding
surface s1 at the rst layer; c Subsystem sliding surface s2 at the rst layer
186 5 Compensator Design Based on Sliding Mode
Fig. 5.10 Control performance by the neural compensator design based on the aggregated HSMC
method. a Control input u; b Compensator output ^
n1 ; c Compensator output ^n2
respectively. From Fig. 5.9, the sliding surface at the top layer rst arrives at its
sliding mode at about 1.5 s. Then, the state trajectories of the two subsystems slide
along their private sliding surface and converge to their subsystem origins. As
proven in Theorems 5.4 and 5.5, only the reachability of the sliding surface at the
top layer is guaranteed. The sliding modes of the subsystem sliding surfaces cannot
be reached at nite time and their sliding modes are asymptotic.
Figure 5.10 shows the control performance by the neural compensation law
based on the aggregated HSMC method, where the control input u, the compensator
output ^ n1 , and the compensator output ^n2 are displayed in Fig. 5.10ac, respec-
tively. From Fig. 5.10a, the maximum control input is about 50 N. However, u has
to switch in high frequency to resist the system uncertainties. From Fig. 5.10b, c,
the two neural-network-based compensators can adaptively track the system
uncertainties according to the adaptive law (5.32) till the compensator outputs tend
to be stable. Compared with the results in Fig. 5.6b, c, the results in Fig. 5.10b, c
illustrate that the neural-network-based compensators have different outputs. The
fuzzy logic-based compensator design has constant outputs as approximation values
of uncertainties. However, the neural-network-based compensators have the zeroth
outputs as t tends to innite.
Appendices 187
Appendices
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 4;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 4;
sizes.NumInputs = 1;
sizes.DirFeedthrough = 0;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
%parameter
g=9.81;M=37.32;m=5;l=1.05;
A=(M+m);B=(m*l*cos(x(3)));C=(cos(x(3)));D=l;
sys(1)=x(2);
sys(2)=((u(1)+m*l*x(4)^2*sin(x(3)))*D+g*B*sin(x(3)))*inv(A*D-
B*C)+0.1*(2*rand()-1);
b1 =D*inv(A*D-B*C);
f1 =(m*l*x(4)^2*sin(x(3))*D+g*B*sin(x(3)))*inv(A*D-B*C);
sys(3)=x(4);
sys(4)=((u(1)+m*l*x(4)^2*sin(x(3)))*C+g*A*sin(x(3)))*inv(B*C-
A*D)+0.1*(2*rand()-1);
b2 =C*inv(B*C-A*D);
f2 =(m*l*x(4)^2*sin(x(3))*C+g*A*sin(x(3)))*inv(B*C-A*D);
function sys=mdlOutputs(t,x,u)
sys = x;
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 0;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 5;
sizes.NumInputs = 4;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlOutputs(t,x,u)
% parameter of system
g=9.81;M=37.32;m=5;l=1.05;
A=(M+m);B=(m*l*cos(u(3)));C=(cos(u(3)));D=l;E=(m*l*sin(u(3)));
% parameter of controller
Appendices 189
%e=x-xd;
e(1)=(u(1));e(2)=u(2);e(3)=u(3);e(4)=u(4);
s1=afa(1)*e(1)+bata(1)*e(2);
S1=s1;
s2=afa(2)*e(3)+bata(2)*e(4);
S2=a*S1+s2;
s=S2;
b1 =D*inv(A*D-B*C);
f1 =(m*l*u(4)^2*sin(u(3))*D+g*B*sin(u(3)))*inv(A*D-B*C);
b2 =C*inv(B*C-A*D);
f2 =(m*l*u(4)^2*sin(u(3))*C+g*A*sin(u(3)))*inv(B*C-A*D);
den=a*bata(1)*b1+bata(2)*b2;
num=a*(afa(1)*u(2)+bata(1)*f1)+(afa(2)*u(4)+bata(2)*f2);
ds=-k*s-ita*sign(s);
dd0=0.2;
uslaw=-num*inv(den)+ds*inv(den)-dd0*inv(den);
sys(1)=uslaw;
sys(2)=s1;
sys(3)=s2;
sys(4)=S2;
sys(5)=-dd0*inv(den);
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 1250;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 6;
sizes.NumInputs = 6;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 1; % at least one sample time is needed
sys = simsizes(sizes);
x0 = [0.1*ones(1250,1)];
str = [];
ts = [0 0];
function sys=mdlDerivatives(t,x,u)
q1=u(1);dq1=u(2);q2=u(3);dq2=u(4);
fsd1=0;
for l1=1:1:5
gs1=-[(q1+pi/2-(l1-1)*pi/4)/(pi/4)]^2;
u1(l1)=exp(gs1);
end
for l11=1:1:5
gs11=-[(dq1+pi/2-(l11-1)*pi/4)/(pi/4)]^2;
U1(l11)=exp(gs11);
end
fsd2=0;
for l2=1:1:5
gs2=-[(q2+pi/2-(l2-1)*pi/4)/(pi/4)]^2;
u2(l2)=exp(gs2);
end
Appendices 191
for l22=1:1:5
gs22=-[(dq2+pi/2-(l22-1)*pi/4)/(pi/4)]^2;
U2(l22)=exp(gs22);
end
for l1=1:1:5
for l11=1:1:5
fsu1=u1'*U1;
end
end
AA=fsu1';BB=AA(:);
for l2=1:1:5
for l22=1:1:5
fsu2=u2'*U2;
end
end
CC=fsu2';DD=CC(:);
EE=BB*DD';FF=EE';GG=FF(:);
for L1=1:1:625
fsd1=fsd1+EE(L1);
end
fs1=GG'/(fsd1+0.001);
for L2=1:1:625
fsd2=fsd2+EE(L2);
end
fs2=GG'/(fsd2+0.001);
%state error;
e1=(u(1))-1;e2=u(2);
e3=u(3);e4=u(4);
lamad1=1;lamad2=10;alpha1=-3;alpha2=1;
s1=e2+lamad1*e1;
s2=e4+lamad2*e3;
s=alpha1*s1+alpha2*s2;
Gama1=0.5;Gama2=0.5;
S1=1/Gama1*s*fs1;
S2=1/Gama2*s*fs2;
for i=1:1:625
sys(i)=S1(i);
end
for j=626:1:1250
sys(j)=S2(j-625);
end
function sys=mdlOutputs(t,x,u)
ita=0.1;k=1;lamad1=0.7;lamad2=8.2;alpha1=-2.3;alpha2=1;
% parameter of system
g=9.81;M=37.32;m=5;l=1.05;
192 5 Compensator Design Based on Sliding Mode
A=(M+m);B=(m*l*cos(u(3)));
C=(cos(u(3)));D=(m*l*sin(u(3)));
%state error;
e1=(u(1))-1;e2=u(2);
e3=u(3);e4=u(4);
s1=e2+lamad1*e1;
s2=e4+lamad2*e3;
s=alpha1*s1+alpha2*s2;
ds=ita*sign(s)+k*s;
f1=(D*(u(4))^2+C*D*g/l)/(A-B*C/l);
b1=1/(A-B*C/l);
f2=(D*C*(u(4))^2+A*g*sin(u(3)))/(B*C-A*l);
b2=C/(B*C-A*l);
for i=1:1:625
thta1(i,1)=x(i);
end
for i=1:1:625
thta2(i,1)=x(i+625);
end
q1=u(1);dq1=u(2);q2=u(3);dq2=u(4);
fsd1=0;
for l1=1:1:5
gs1=-[(q1+pi/2-(l1-1)*pi/4)/(pi/4)]^2;
u1(l1)=exp(gs1);
end
for l11=1:1:5
gs11=-[(dq1+pi/2-(l11-1)*pi/4)/(pi/4)]^2;
U1(l11)=exp(gs11);
end
fsd2=0;
for l2=1:1:5
gs2=-[(q2+pi/2-(l2-1)*pi/4)/(pi/4)]^2;
u2(l2)=exp(gs2);
end
for l22=1:1:5
gs22=-[(dq2+pi/2-(l22-1)*pi/4)/(pi/4)]^2;
U2(l22)=exp(gs22);
end
for l1=1:1:5
for l11=1:1:5
fsu1=u1'*U1;
end
end
AA=fsu1';BB=AA(:);
for l2=1:1:5
for l22=1:1:5
fsu2=u2'*U2;
end
end
CC=fsu2';DD=CC(:);
Appendices 193
EE=BB*DD';
FF=EE';GG=FF(:);
for L1=1:1:625
fsd1=fsd1+EE(L1);
end
fs1=GG'/(fsd1+0.001);
for L2=1:1:625
fsd2=fsd2+EE(L2);
end
fs2=GG'/(fsd2+0.001);
Fp1=thta1'*alpha1*fs1';
Fp2=thta2'*alpha1*fs2';
con_law=inv(alpha1*b1+alpha2*b2)*(-alpha1*f1-alpha2*f2-
abs(alpha1*Fp1)*sign(s)-abs(alpha2*Fp2)*sign(s)...
-alpha1*lamad1*e2-alpha2*lamad2*e4-ds);
sys(1)=con_law;
sys(2)=s;
sys(3)=Fp1; %out of compensator
sys(4)=Fp2; %out of compensator
sys(5)=s1;
sys(6)=s2;
end
function [sys,x0,str,ts]=mdlInitializeSizes
sizes = simsizes;
sizes.NumContStates = 12;
sizes.NumDiscStates = 0;
sizes.NumOutputs = 6;
sizes.NumInputs = 4;
sizes.DirFeedthrough = 1;
sizes.NumSampleTimes = 0; % at least one sample time is needed
sys = simsizes(sizes);
function sys=mdlDerivatives(t,x,u)
we1=1*ones(4,6);we2=1*ones(4,6);bb1=1*ones(6,1);bb2=1*ones(6,1);
xstate=[u(1);u(2);u(3);u(4)];
ita=0.2;k=3;lamad1=0.7;lamad2=8.2;alpha1=-2.3;alpha2=1;
s1=u(2)+lamad1*u(1);
s2=u(4)+lamad2*u(3);
s=alpha1*s1+alpha2*s2;
ds=ita*sign(s)+k*s;
for j=1:6
h1(j)=exp(-norm(xstate-we1(:,j))^2/(2*bb1(j)^2));
end
for j=1:6
h2(j)=exp(-norm(xstate-we2(:,j))^2/(2*bb2(j)^2));
end
Appendices 195
eq0=0.002;
eq1=0.001;
xite1=alpha1*10^81;
xite2=10^81;
for i=1:6
sys(i)=xite1*abs(s)*h1(i);
end
for i=1:6
sys(i+6)=xite2*abs(s)*h2(i);
end
function sys=mdlOutputs(t,x,u)
we1=10*ones(4,6);we2=10*ones(4,6);bb1=1*ones(6,1);bb2=1*ones(6,1);
% parameter of system
g=9.81;M=37.32;m=5;l=1.05;
A=(M+m);B=(m*l*cos(u(3)));C=(cos(u(3)));D=l;E=(m*l*sin(u(3)));
% parameter of controller
%e=x-xd;
e(1)=(u(1));e(2)=u(2);e(3)=u(3);e(4)=u(4);
ita=0.2;k=3;lamad1=0.7;lamad2=8.2;alpha1=-2.3;alpha2=1;
s1=u(2)+lamad1*u(1);
s2=u(4)+lamad2*u(3);
s=alpha1*s1+alpha2*s2;
ds=ita*sign(s)+k*s;
% c=[1.6 2.3 -8.7 1];
% s=c*u;
% k=10;
% eta=0.01;
b1 =D*inv(A*D-B*C);
f1 =(m*l*u(4)^2*sin(u(3))*D+g*B*sin(u(3)))*inv(A*D-B*C);
b2 =C*inv(B*C-A*D);
f2 =(m*l*u(4)^2*sin(u(3))*C+g*A*sin(u(3)))*inv(B*C-A*D);
den=alpha1*b1+b2;
num=alpha1*lamad1*u(2)+lamad2*u(4)+alpha1*f1+f2;
% we1=1*ones(4,6);
% we2=1*ones(4,6);
% b1=0.20*ones(6,1);
% b2=0.20*ones(6,1);
w1=[x(1);x(2);x(3);x(4);x(5);x(6)];
w2=[x(7);x(8);x(9);x(10);x(11);x(12)];
xstate=[u(1);u(2);u(3);u(4)];
for j=1:6
196 5 Compensator Design Based on Sliding Mode
h1(j)=exp(-norm(xstate-we1(:,j))^2/(2*bb1(j)^2));
end
for j=1:6
h2(j)=exp(-norm(xstate-we2(:,j))^2/(2*bb2(j)^2));
end
Up1=w1'*h1';
Up2=w2'*h2';
ds=-k*s-(ita)*sign(s);
uslaw=-num*inv(den)+ds*inv(den)-(alpha1*Up1+Up2)*sign(s)*inv(den);
sys(1)=uslaw;
sys(2)=Up1;
sys(3)=Up2;
sys(4)=s;
sys(5)=s1;
sys(6)=s2;
References
1. Qian DW, Liu XJ, Yi JQ (2009) Robust sliding mode control for a class of underactuated
systems with mismatched uncertainties. Proc Inst Mech Eng Part I J Syst Control Eng 223
(6):785795
2. Lin CM, Mon YJ (2005) Decoupling control by hierarchical fuzzy sliding-mode controller.
IEEE Trans Control Syst Technol 13(4):593598
3. Fei JT, Zhou J (2012) Robust adaptive control of MEMS triaxial gyroscope using fuzzy
compensator. IEEE Trans Syst Man Cybern, B Cybern 42(6):15991607
4. Singh HP (2014) Simulation of neural network based adaptive compensator control scheme
for multiple mobile manipulators with uncertainties. Int J Nonlinear Sci Numer Simul 15(3
4):181188
5. Chatterjee A, Sarkar G, Rakshit A (2011) A reinforcement-learning-based fuzzy compensator
for a microcontroller-based frequency synthesizer/vector voltmeter. IEEE Trans Instrum Meas
60(9):31203127
6. McNeill FM, Thro E (2014) Fuzzy logic: a practical approach. Academic Press, USA
7. Baughman DR, Liu YA (2014) Neural networks in bioprocessing and chemical engineering.
Academic Press, USA
8. Park J, Sandberg IW (1991) Universal approximation using radial-basis-function networks.
Neural Comput 3(2):246257
9. Wu YL, Sun FC, Zheng JC, Song Q (2010) A robust training algorithm of discrete-time
MIMO RNN and application in fault tolerant control of robotic system. Neural Comput Appl
19(7):10131027
10. Qian DW, Zhao DB, Yi JQ, Liu XJ (2013) Neural sliding-mode load frequency controller
design of power systems. Neural Comput Appl 22(2):279286
11. Qian DW, Yi JQ, Liu XJ (2011) A robust sliding mode controller based on RBF neural
networks for overhead crane systems with uncertain dynamics. ICIC Express Lett 5(6):1995
2000
Chapter 6
Conclusions and Open Problems
Keywords Hierarchical sliding mode control Overhead crane Transport control
6.1 Conclusions
Overhead crane systems are under-actuated because they have a lower number of
control inputs than the number of degrees of freedom to be controlled. Such an
inherent property challenges their control design. Sliding mode control (SMC) is
recognized as one of the efcient tools to design robust controllers for complex
nonlinear systems. Concerning the SMC-based design for overhead crane systems,
it is one of the most active elds of research in control community. For such a wide
research area, this book focuses on the structure design of sliding surfaces.
The physical characteristic is that an overhead crane is composed of several
subsystems. By capturing the characteristic, the methodology of hierarchical sliding
mode control (HSMC) is presented. The HSMC methodology employs the hier-
archical structure design of sliding surfaces for overhead cranes. This methodology
covers three basic types of hierarchical structures, entitled aggregated HSMC,
incremental HSMC and combing HSMC.
Compared with the traditional single-layer sliding surface of several SMC
methods, i.e., integral SMC, 1st-order SMC, 2nd-order SMC, terminal SMC, etc.,
the HSMC methodology has signicantly contributed to reveal the sliding motion
by the phase-plane method because the hierarchical sliding surfaces consists of a
series of 2nd-order sliding surfaces. Such hierarchical surfaces give an insight into
the sliding motion. This book investigates the three kinds of HSMC methods for
single-pendulum-type overhead crane systems. Numerical simulations show the
crane state variables, the hierarchical sliding surfaces, the control performance and
the sliding modes based on HSMC. The applications of HSMC can also be
extended to double-pendulum-type overhead cranes. Some numerical results are
also illustrated.
Any real overhead crane systems are subject to uncertainties. Uncertainties can
be categorized by matched uncertainties and unmatched ones. Although SMC is of
invariance again matched uncertainties, uncertainties involved in crane dynamics
cover both matched and unmatched uncertainties. To overcome the adverse effects
of unmatched uncertainties, the compensator design on basis of hierarchical sliding
surfaces is touched in this book. The main goal of implementing the HSMC
methods again uncertainties is to guarantee the crane control systems is of
asymptotic stability in the presence of uncertain overhead crane systems.
This book touches this topic and adopts the structure of controller plus com-
pensator to deal with the transport control problem of uncertain overhead cranes.
The hierarchical structure of sliding surfaces offers two possible compensator
structures. To simplify the problem, the lumped compensator structure is adopted.
To guarantee the crane control systems is of asymptotic stability in the presence of
uncertain, a common assumption is that uncertainties should have a known
boundary. However, the assumption is too strict to satisfy under some conditions.
This book attempts the technologies of fuzzy logic and neural networks to make the
assumption mild. The theoretical analysis and numerical simulations demonstrate
the feasibility and robustness of the designed methods against uncertainties.
Control of overhead crane systems by HSMC is open and it is difcult to include all
the works in this book. However, we believe that this book can provide readers with
the most up-to-date knowledge in the eld. Finally, several important issues and
possible future research directions are drawn:
Hierarchy. Pointed out by Lima and Saridis [2], control systems are also of
hierarchy. Current contributions about HSMC are hierarchical structures of sliding
surfaces. Such hierarchy is only located at the controller level rather than at the
control-system level. This indicates there are still some spaces for current HSMC
methods to improve their intelligence.
Payload angle. This book assumes that the payload angle is measurable such
that payload angle and its angular velocity are directly adopted for control design
without more explanations. But payload angle may be hard to obtain in practice [3].
It is meaningful and signicant to investigate transport control problems of over-
head cranes under the limited measurable state variables.
Modeling of uncertainties. This book analyzes that overhead crane systems are
subject to uncertainties. These uncertainties challenge the design of HSMC meth-
ods. However, these uncertainties are not accurately modeled. To suppress the
uncertainties and compensate them, it is necessary to investigate this eld.
6.2 Extensions and Open Problems 199
Parameter tuning. Controller parameters have direct effects on the system sta-
bility and robustness. But parameter tuning is a time-consuming business. In this
book, controller parameters are tuned after trial and error. Since there are several
controller parameters for each control method, it is possible to improve HSMC
methods by employing evolutional algorithms on this aspect.
Extensions of other types of cranes. There are three types of cranes. To dem-
onstrate HSMC methods, this book only takes overhead cranes as an example to
illustrate the control design and stability analysis. The feasibility and validity of
HSMC methods have been demonstrated by overhead cranes. It is possible and
expected to extend HSMC methods to the other two types of cranes. Since the other
two types of cranes are described by non-Cartesian coordinates, some technical
details about HSMC design remain problematic and challenging.
References
1. Utkin V, Guldner J, Shi J (2009) Sliding mode control in electromechanical systems, 2nd edn.
Taylor & Francis, USA
2. Lima PU, Saridis GN (1996) Design of intelligent control systems based on hierarchical
stochastic automata. World Scientic, USA
3. Singhose W (2009) Command shaping for flexible systems: a review of the rst 50 years. Int J
Precis Eng Manuf 10(4):153168