Implicit IDA-PBC Design and Implementationfor A Portal Crane System

Pontificia Universidad Católica del Perú
Escuela de Posgrado
Tesis
Implicit IDA-PBC Design and Implementation
for a Portal Crane System
Para obtener el grado académico de:

Magı́ster en Ingenierı́a Mecatrónica
Presentado por: Ing. Jesus Enrique Vidal Sandoval
Tutor Responsable (TU Ilmenau): M. Sc. Oscar B. Cieza Aguirre

Professor Responsable (TU Ilmenau): Prof. Dr.-Ing. Johann Reger
Professor Responsable (PUCP): PhD. Elizabeth Roxana Villota Cerna
Fecha y lugar: 21/10/2019, Ilmenau

Implicit IDA-PBC Design and Implementation
for a Portal Crane System
Master Thesis in Mechatronics / Ingeniería Mecatrónica
submitted by
Enrique Vidal
born on 23. October 1991

in Lima, Peru
in the
Control Engineering Group
Department of Computer Science and Automation

Technische Universität Ilmenau
Advisors: Prof. Dr.-Ing. Johann Reger

M. Sc. Oscar B. Cieza Aguirre
PhD. Elizabeth R. Villota
Submission Date: 21 October 2019

Declaration
I declare that the work is entirely my own and was produced with no assistance from
third parties.
I certify that the work has not been submitted in the same or any similar form for
assessment to any other examining body and all references, direct and indirect, are
indicated as such and have been cited accordingly.
(Enrique Vidal)
Ilmenau, 21 October 2019
To my beloved parents
Carmela and Enrique Vidal.
Abstract
Interconnection and damping assignment passivity-based control (IDA-PBC) is a well-

known technique which regulates the behavior of nonlinear systems, assigning a target
port-Hamiltonian (pH) structure to the closed-loop. In underactuated mechanical sys-
tems (UMSs) its application requires the satisfaction of matching conditions, which in
many cases demands to solve partial differential equations (PDEs). Only recently, the
IDA-PBC has been extended to UMSs modeled implicitly, where the system dynamics
in pH representation are described by a set of differential-algebraic equations (DAEs).
In some system classes this implicit approach allows to circumvent the PDE problem
and also to design an output-feedback law.
The present thesis deals with the design and implementation of the total energy shaping
implicit IDA-PBC on a portal crane system located at the laboratory of the Control
Engineering Group at TU-Ilmenau. The implicit controller is additionally compared
with a simplified (explicit) IDA-PBC [1]. This algorithm shapes the total energy and
avoids the PDE problem. However, this thesis reveales a significant implementation
flaw in the algorithm, which then could be solved.
Master Thesis Enrique Vidal iv

Kurzfassung
Interconnection and damping assignment passivity-based control (IDA-PBC) ist eine

wohlbekannte Methode zur Regelung von nichtlinearen Systemen, die im geschlossenen
Regelkreis eine gewünschte Port-Hamiltonian-Struktur (pH) haben. Die Anwendung
auf unteraktuierte mechanische Systeme (UMS) erfordert die Erfüllung von sogenann-
ten Matching Conditions, die meistens die Lösung partieller Differentialgleichungen
(PDE) benötigt. Erst kürzlich wurde die IDA-PBC auf implizit modellierte UMS erwei-
tert, bei denen die Systemdynamik in pH-Darstellungen durch Differentialalgebraische
Gleichungen (DAE) beschrieben wird. Dieser implizite Ansatz ermöglicht bei einigen
Systemklassen, das PDE-Problem zu umgehen und auch eine Ausgangsrückführung zu
entwerfen.
Die vorliegende Masterarbeit beschäftigt sich mit dem Entwurf und der Implementie-
rung des impliziten IDA-PBC zur Gesamtenergievorgabe auf einem Portalkransystem
im Labor des Fachgebiets Regelungstechnik der TU-Ilmenau. Der implizite Regler wird
mit einem vereinfachten (expliziten) IDA-PBC verglichen [1]. Dieser Algorithmus gibt
ebenso die Gesamtenergie vor und vermeidet das PDE-Problem. In der Masterarbeit
wird in diesem Algorithmus ein wesentlicher Implementierungsfehler offengelegt und
behoben.
Master Thesis Enrique Vidal v

Acknowledgment
I would like to thank all the people who support me during the development of this
thesis. Specially thanks are (i) to my entire family who motivated me to finish this work,
(ii) to my thesis advisor M.Sc. Oscar Cieza at TU Ilmenau, he was always available
to guide me in the right direction, (iii) to the responsible of the Control Engineering
Group Laboratory, Axel Fink, who support me in the well use of the Equipment, and
(iv) to my closest friends.
Master Thesis Enrique Vidal vi

Contents
Abstract iv
Kurzfassung v
Contents vii
1. Introduction 1
1.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1. Portal Crane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.2. Common Algorithms used in Portal Crane control . . . . . . . . 2
1.2.3. Passivity Based Control (PBC) applied to Portal Cranes . . . . . 4
1.2.4. Implicit PBC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3. Contribution of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4. Outline of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2. Theoretical Fundamentals 7
2.1. Analytical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1. Lagrangian Mechanics . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1.1. Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1.2. Newton’s Law . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.1.3. Generalized Coordinates . . . . . . . . . . . . . . . . . 8
2.1.1.4. Virtual Displacement . . . . . . . . . . . . . . . . . . . 9
2.1.1.5. Euler-Lagrange Equations . . . . . . . . . . . . . . . . . 10
2.1.1.6. The Functional and Hamilton’s Principle . . . . . . . . 10
2.1.1.7. Constraints . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1.8. Method of Lagrange Undetermined Multipliers . . . . . 11
2.1.1.9. Canonical Momentum . . . . . . . . . . . . . . . . . . . 11
Master Thesis Enrique Vidal vii

CONTENTS
2.1.2. Hamiltonian Mechanics . . . . . . . . . . . . . . . . . . . . . . . 11

2.1.2.1. Legendre Transformation . . . . . . . . . . . . . . . . . 12
2.1.2.2. The Hamiltonian . . . . . . . . . . . . . . . . . . . . . . 12
2.1.2.3. The Canonical Equations . . . . . . . . . . . . . . . . . 13
2.1.2.4. The Hamiltonian Function with Constraints . . . . . . 14
2.2. Passivity-based control and Energy Shaping . . . . . . . . . . . . . . . . 14
2.2.1. Dissipativity, passivity and stability . . . . . . . . . . . . . . . . 14
2.2.2. Port Hamiltonian Systems . . . . . . . . . . . . . . . . . . . . . . 17
2.2.3. Implicit and Explicit Port Hamiltonian Structure of a Mechanical
System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.4. Underactuated Mechanical Systems . . . . . . . . . . . . . . . . 19
2.2.5. Interconnection and Damping Assignment - PBC . . . . . . . . . 19
3. Implicit IDA-PBC for UMS 24

3.1. Problem formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2. Implicit Interconnection and Damping Assignment - Passivity-Based Con-
trol (IDA-PBC) for holonomic systems . . . . . . . . . . . . . . . . . . . 25
3.2.1. Mechanical systems with constant mass matrix . . . . . . . . . . 29
3.2.2. Position feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4. Explicit and Implicit IDA-PBC applied to a Portal Crane 34

4.1. The Portal Crane System . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1.1. Additional Information of the Crane . . . . . . . . . . . . . . . . 36
4.1.2. 2-D Explicit Model . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.1.3. 3-D Implicit Model . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2. Explicit IDA-PBC Applied to the Portal Crane . . . . . . . . . . . . . . 41
4.2.1. State feedback law from [1] . . . . . . . . . . . . . . . . . . . . . 41
4.2.2. A well-defined state feedback law using IDA-PBC . . . . . . . . 43
4.2.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3. Implicit IDA-PBC Applied to the Portal Crane . . . . . . . . . . . . . . 46
4.3.1. State feedback law . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3.2. An output feedback law using the implicit IDA-PBC approach . 49
4.3.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5. Conclusions and Future Work 57
Appendices 59
A. Euler-Lagrange Equations 60
Master Thesis Enrique Vidal viii

CONTENTS
Abbreviations 64
List of Figures 65
List of Tables 67
Bibliography 68
Master Thesis Enrique Vidal ix

Chapter 1
Introduction
1.1 Motivation
Underactuated Mechanical Systems (UMSs) are widely used in different scenarios in
the industry, such as construction, robotic applications, and transportation. A me-
chanical system is said to be underactuated if it has fewer control inputs than Degrees
of Freedom (DOF). The underactuation can come from different situations, such as
intentional design or due to the failure of some actuators [2]. For this reason, the focus
on developing new control techniques applicable to underactuated nonlinear mechanics
system has grown in the last years.
At the early 2000s, classic Energy-Based Control techniques were presented to reg-
ulate the behaviour of nonlinear systems [3]. Motivated on the passivity properties of
UMSs, different energy-based approaches were useful to overcome the underactuated
control problem. Among them the IDA-PBC approach has been successfully used in
a wide range of UMSs. The keystone of IDA-PBC is to design a closed-loop with
port-Hamiltonian (pH) structure, where the new storage function has a minimum at
the desired equilibrium point. However, the total energy shaping IDA-PBC has a per-
sistent problem: the solution of Partial Differential Equations (PDEs). Recently this
method has been extended theoretically to implicit pH structures [4].
The difference between an explicit model and the implicit one, relies on the consid-
erations of physical constraints. Thus, UMSs can be modeled in explicit or implicit
representation. But, if we use the implicit structure and apply the implicit IDA-PBC
method, it is possible in some system classes to avoid PDEs even when the open- and
closed-loop inertia mass matrices in explicit representation are state dependent. Until
now, the Implicit IDA-PBC technique has remained in the theoretical and simulation
framework with no physical implementation. Therefore, the main task of this master
Master Thesis Enrique Vidal 1

1. Introduction
thesis is to apply the mentioned control algorithm on a portal crane system located at
the laboratory of the Control Engineering Group at TU-Ilmenau.
1.2 Literature Review
1.2.1 Portal Crane

Cranes improve the method of transportation and handling of heavy loads. Their role
in industry enables optimal throughput and logistic [5]. A crane consists of hoisting
and a support mechanism. The hoisting mechanism is essentially a mechanism to lift
something heavy, meanwhile, the support mechanism could be a trolley-girder, a trolley-
jib, or a boom [6]. According to the degrees of freedom of the support mechanism,
cranes can be classified as:
• Rotary cranes: They are also known as tower cranes, and here, the girder1
rotates in a plane parallel to the ground about a fixed vertical axis. In Figure 1.1a
we can see a rotary crane model Potain MDT 98 [7].
• Boom cranes: Their suspension point is fixed at the end of the boom. The main
advantage relies on their structure capable of supporting loads in compression. As
a result, booms are compact and offer the same capacity as the rotary cranes [6].
Figure 1.1b is a boom crane developed by [8].
• Portal cranes: Also known as a gantry or overhead crane. It is composed of a

trolley moving over a girder, and it usually has a setup that allows the movement
of the trolley on a plane parallel to the ground. Figure 1.1c shows a Rubber-Tyred
Gantry Crane Terex by [9].
There exists two main approaches to model a crane, the distributed mass model and
the lumped mass model. The first considers the hoisting line as a distributed mass,
hence the name [10], while the latter considers the hoisting line as a massless cable. In
this thesis, we use the lumped-mass model and consider the cable-hook-payload as a
spherical pendulum.
1.2.2 Common Algorithms used in Portal Crane control

Moving a load from one position to another is an essential task of any crane. However,
an inadequate control in the actuators will increase oscillations on the payload. Con-
sidering that it will move a heavy load, requiring safe movements to protect the workers
1
A girder is a long, thick piece of metal. Normally, the frame of the cranes is composed of a set of
girders.

1. Introduction
(a) Rotary crane [7] (b) Boom crane [8]
(c) Gantry crane [9]
Figure 1.1. – Different kind of cranes
and objects in the workspace. In this sense, it is desired to have minimal residual pen-
dulations en route to the target destination. A primitive idea will be to perform the
motion with minimum velocity. However, this leads in wasting time and thus, lacking
in efficiency and efficacy.
In [6], a detailed survey of the earliest controllers applied to cranes is presented.
The author describes different open and closed loops techniques including linear con-
trol, fuzzy control, optimal control, adaptive control and nonlinear control. Here, we
described some of the most relevant nonlinear control approaches:
Sliding mode control (SMC). In [11] the author models the crane dynamics in 3-D and
proposes two clear objectives: position regulation and anti-swing control. Besides, it is

1. Introduction
implemented an observer contemplating that most of the cranes are not equipped with
velocity sensors. The results of the simulations confirm that using SMC increments
the robustness under uncertain parameters. In [12] the method is extended to Second-
order sliding mode control (SOSMC). The author proposes a SOSMC controller for
a 3-D crane affected by external perturbations. The work seeks a strong Lyapunov
function to enable the use of the twisting algorithm; and, SMC enforcement to deal with
uncertainties and initial swing angle conditions. An extension to Adaptive Dynamic
SMC was presented in [13]. That work proposes to eliminate the effect of chattering
caused by SMC through a dynamic integral sliding surface. An advantage of this
approach is the self-tuning laws which overcome the disturbances and uncertainties.
However, the model is restricted to 2-D.
Partial Feedback Linearization (PFL). In [14] the author uses the PFL combined with
H∞ control theory applied to a 3-D crane model. The approach explores two scenarios:
system without movement and with external perturbations, and trajectory tracking.
The results are implemented in a real system and compared with pole placement and
LQR approach. H∞ and H∞ loop-shaping control show a robustification and good
perfomance.
1.2.3 PBC applied to Portal Cranes

In this subsection, we describe some of the classic energy-based control techniques in
portal crane systems. All of them assume the lumped mass model:
Interconnection and Damping Assignment - Passivity-Based Control (IDA-PBC).
In [15] the authors analyze the crane in 2-D with a holonomic constraint and apply
the potential energy shaping IDA-PBC.2 The main idea is to assign the closed-loop dy-
namics a desired Hamiltonian and solve the matching conditions to obtain a nonlinear
control law. Here the authors take as the holonomic constraint a no-slip restriction in
the pulley/cable model and their controller achieves (local) asymptotic stability.
In [1, 16] it is used a Simplified IDA-PBC approach. The paper follows the proce-
dure of IDA-PBC with a particular parametrization in the closed-loop inertia matrix,
enabling a kinetic energy shaping and a simplification in the PDEs of the matching
conditions. The work considers a 2-D crane model under partial feedback lineariza-
tion. However, the controller has a division by 0 problem in some points or regions of
the state space, e.g. if the system begins at equilibrium. In [17] the author applies a
Combined Flatness and IDA-PBC approach. They develop a point-to-point transfer
of the payload minimizing the cable swing. The crane model is in 2-D and considers
holonomic constraints analyzing the pulley/cable. Besides, they achieve a trajectory
2
The total energy shaping IDA-PBC methodology will be explained in Section 2.2.5.

1. Introduction
tracking using flatness and considering feedforward control based on flatness. They find
a lack of robustness due to the open-loop nature of flatness based feedforward control
and add the IDA-PBC methodology (solving a PDE) to stabilize the system.
Energy Shaped without solving the PDE. In [18] the authors propose a control law in
two stages. The first stage uses the collocated partial feedback linearization method,
and the second stage is a PI Controller. The technique follows a problem formulation
using a Hamiltonian energy function but does not impose a mechanical pH structure in
the closed-loop. Additionally, they use (Cyclo) passive outputs and consider the model
of the crane in 2-D. Unfortunately, the paper does not shows simulations or a physical
implementation.
Geometric-PBC. In [19] the authors take advantage of the intrinsic geometry of a
3-D underactuated crane with 4-DOF. They use PFL and new passivating outputs to
shape only the potential energy. Finally, they obtain a nonlinear control law without
solving PDEs.
1.2.4 Implicit PBC

The implicit representation has its root in analytical mechanics, where remarkable
contributions of this theory were the Euler-Lagrange equations introduced in 1788, the
Hamilton contributions presented in two essays 1834 and 1835 [20], and also the Jacobi
Contributions of 1837, well-known as the Hamilton-Jacobi method [21]. In [22], the
authors explore the explicit and implicit representations of smooth, finite-dimensional
pH systems. They explain briefly the conditions to model a system in implicit repre-
sentation, i.e., due to physical constraints.
PBC of pH systems [3]. This work is one of the main contributions to PBC with Dirac
structure. Here, the energy balancing PBC, the control by state-modulated source, and
the IDA-PBC methodology, are generalized to the case in which the system is model
from a Dirac-structure perspective. The author focus on electrical circuits.
Implicit IDA-PBC for UMS [4]. The work presents a generalization of IDA-PBC
to UMS described in implicit pH representation. The paper exposes a state-feedback
and output-feedback law using a similar procedure as in (explicit) algebraic IDA-PBC.
Thus, the main contributions of this work are the solution of the IDA-PBC method for
UMS with holonomic constraints and avoiding PDEs. As an example of the proposed
approach, a 3-D model of the crane with holonomic constraints was taken into con-
sideration. The simulations of the output-feedback control law without using partial
feedback linearization shows an (local) asymptotical response.

1. Introduction
1.3 Contribution of this thesis

This master thesis briefly reviews the theory of implicit IDA-PBC for UMS with holo-
nomic constraints. Then, the implicit algorithm is designed, simulated and then imple-
mented in a portal crane system located at the laboratory of the Control Engineering
Group at TU-Ilmenau. The model considered for the implicit IDA-PBC is in 3-D with
the constraint given by a fix pendulum cable length. This work uses a computer with
MATLAB/SIMULINK and an RTI toolbox which enables rapid prototyping in the
implementation thanks to the dSPACE controller board. For comparison, it is also
designed the (explicit) IDA-PBC algorithm to the portal crane modeled in 2-D and,
revealing a significant mistake made by [1]. The (local) asymptotic stability of the
system is investigated through simulations and implementations that were carried out
and finally compared.
1.4 Outline of this thesis

The remainder of the thesis is organized as follows: Chapter 2 gives an overview of
the theoretical fundamentals required to understand the implicit IDA-PBC. It includes
the essential background for explicit and implicit port Hamiltonian (pH) modeling of
mechanical systems, and a concise introduction to passivity based control (PBC) and
IDA-PBC. Chapter 3 briefly provides the theory of IDA-PBC for implicit systems.
In Chapter 4, focused on the portal crane, the design is presented, also simulations and
real implementation for both explicit and implicit IDA-PBC algorithms. This thesis
concludes in Chapter 5 with a summary of the most important remarks and ideas for
future work.

Chapter 2
Theoretical Fundamentals
This chapter presents the theoretical basis, which this thesis is built on. After some
brief mathematical preliminaries based on the authors of [23–25], we introduce sim-
ple mechanical systems in both Lagrangian and Hamiltonian representations in Sec-
tion 2.1.1 and Section 2.1.2, respectively. We then recall some notions of passivity and
passivity-based control theory in Section 2.2.
2.1 Analytical Mechanics

In essence, Analytical Mechanics is the theory on which we base our understanding
of motion [23]. A vast number of mathematicians have contributed to the theory of
Analytical Mechanics, for example, Copernicus, Kepler, Newton, D’Alembert, Euler,
Lagrange, Hamilton and Jacobi. From them we might distinguish Lagrange and Hamil-
ton, which are the fathers of Lagrangian and Hamiltonian Mechanics. In this work,
we repeatedly talk about explicit3 and implicit port Hamiltonian (pH) representations,
but to understand both, we recall Lagrange and Hamiltonian Mechanics.
2.1.1 Lagrangian Mechanics
2.1.1.1 Kinematics
Kinematics is the description of the motion of particles. Commonly, those particles

are called point particles which, in classical mechanics, means that are small enough
compared to the dimensions of the system. The motion of a system consists of a
collection of n point particles with masses mi , i = {1, 2, ..., n}. In order to define the
kinematic description of a particle, i.e., to recognize a position vector and a velocity
vector, we might choose Cartesian coordinates as in Figure 2.1. Here the vector position
3
For simplicity, most authors omit the word explicit when referring to explicit pH systems.

2. Theoretical Fundamentals
is r ∈ Rn and r = r(t) = (x(t), y(t), z(t)). The length from the origin to the vector is
defined by the Euclidean Norm krk2 . Besides, because only the components (x, y, z)
are time dependent, the velocity vector v ∈ Rn is v(t) := dr/dt ≡ ṙ. Similarly, the
acceleration vector a ∈ Rn is a(t) := dṙ/dt ≡ r̈.
2.1.1.2 Newton’s Law
The total force acting on the ith (point) particle is defined by F i = F ext int
i + F i . Where
F ext
i and F int
i are the external and internal forces respectively. Applying Newton’s
Second Law to our Cartesian coordinates (x,y,z), results in a set of three equations,
which we may sum over all particles i to obtain
n
X n
X
mi r̈i = F i. (2.1)
i=1 i=1
2.1.1.3 Generalized Coordinates
Geometric constraints are normally imposed due to physical constraints on the system.
Leading in the possibility of achieving the complete description of the motion in terms
of some dynamical variables. As an example, consider the simple pendulum in Fig-
ure 2.2 where the motion of the mass is constrained to follow the path of a circle of
radius l. The Cartesian coordinates of the system are x = l sin β and z = l cos β. Thus,
the description of the motion is in terms of a single variable β. This variable expresses
the relationship between x and y, and we refer to this variable as the generalized co-
ordinate. In this particular case, the reduction of coordinates results from the single
constraint l2 = x2 + y 2 .
Let us define the generalized coordinates q ∈ Rnq and an element of this set as qk ,
where k = 1 . . . nq . Besides, 1 ≤ nq ≤ N and N is an integer number indicating the
number of degrees of freedom. Thus,
{r1 , r2 , . . . , rn } → {q1 , q2 , . . . , qnq }. (2.2)
To formulate a general case, where the generalized coordinates and the constraints of the
system are together, it will require the method of undetermined multipliers introduced
by Lagrange and later explained in Section 2.1.1.8. From the pendulum example,
it is clear that constraints will always impose geometrical relationships between the
rectangular coordinates r and the generalized coordinates q. Is common to represent

2
r
r (t )
y
Figure 2.1. – Rectangular Cartesian Coordinates
β l
Figure 2.2. – Pendulum with generalized coordinate β
this relationship as
ri = ri (q1 , q2 , . . . , qnq , t) = ri (q, t); i = 1, . . . , n. (2.3)
From above equation we can write

nq
dṙi X ∂ri ∂ri
vi ≡ = q̇k + , (2.4)
dt k=1
∂qk ∂t
which is usually called Pfaffians or Pfaff’s differential forms.
2.1.1.4 Virtual Displacement
The virtual displacement δri is a purely imaginary movement along the path of the
particles. It is called virtual since it does not actually happen. The central assumption
for virtual displacement is that the time is constant and thus dt = 0. From (2.4), with
dt = 0, we have
nq
X ∂ri
δri = δqk . (2.5)
k=1
∂qk

2.1.1.5 Euler-Lagrange Equations
A derivation from d’Alembert’s Principle (see Appendix A) will lead to the well known
Euler-Lagrange equation
∂L d ∂L

− = 0, (2.6)
∂qk dt ∂ q̇k
L(q, q̇, t) = T − V, (2.7)
where L is the Lagrangian, T is the kinetic energy and V represents the potential
energy.
2.1.1.6 The Functional and Hamilton’s Principle
A functional defines an operation on a class of functions {y(x)} that returns a real

number for each function y(x) [23]. The functional is typically denoted as J [ y ] and it
typically has the following form
Zb
d
J [y] = F (y(x), y 0 , x) dx with y0 = y(x). (2.8)
dx
a
Here the functional is a definite integral of a quantity F dependent on the function

y(x), its derivative y 0 (x), and the independent variable x over the interval [a, b].
Theorem 1 (Hamilton’s Principle [23]). Let S [ q ] be a functional
Zt2
S [q] = L(q, q̇, t) dt ,
t1
where L, defined in (2.7), is the Lagrangian function with generalized coordinates q ∈

Rnq , which have continuous first time derivatives q̇ on the interval [t1 , t2 ], and fixed
values at the end points t1 and t2 . Then a necessary condition for S [ q ] to have an
extremum for a given set of generalized coordinates q is that each generalized coordinate
q satisfies the Euler–Lagrange Equation
∂L d ∂L

− =0 k = 1, . . . , nq
∂qk dt ∂ q̇k

2.1.1.7 Constraints
We might reduce the number of generalized coordinates through constraints imposed

on the system. There are different kinds of constraints; however, this thesis focuses on
the holonomic constraints, i.e, constraints that we can write algebraically as
g(q) = 0. (2.9)
2.1.1.8 Method of Lagrange Undetermined Multipliers
The method of Lagrange undetermined multipliers consists in incorporate the con-

straints (2.9) to the system. This method defines λ ∈ Rnλ arbitrary functions of time,
which yield nλ equations
λk gk = 0. (2.10)
Incorporating the integral of the products of Equation (2.10) to the functional in

the Theorem 1 (Hamilton’s Principle) leads to an ‘augmented’ Lagrangian L0 = L +
Pnλ
k=1 λk gk . Theorem 1 then results in the modified Euler-Lagrange equations
λ n
∂L d ∂L ∂gk
X
− + λk = 0. (2.11)
∂qk dt ∂ q̇ k k=1
∂qk
2.1.1.9 Canonical Momentum
The canonical momenta p that conjugate to the generalized coordinates q are defined
as
∂L
pk ≡ . (2.12)
∂ q̇k
For a free particle the momentum that conjugates with x is p = mẋ, and thus canoni-
cally conjugate momentum reduces to the usual definition of momentum [24] which is
well known in classical mechanics as linear momentum [26].
2.1.2 Hamiltonian Mechanics

We can represent the system dynamics described by the Euler-Lagrange equations
in an utterly equivalent form called the Hamiltonian. This section shows that the
Lagrangian will have a transition from a formulation on a single second-order Euler-
Lagrange equation to one based on two first-order equations, one for each generalized
coordinate and one for each conjugate momentum. To achieve this transition we need
to perform a Legendre transformation. Then, we find the fundamental equations of
Analytical Mechanics which are well know as the canonical equations.

2.1.2.1 Legendre Transformation
The Legendre transformation is a very useful mathematical tool since it transforms

functions on a vector space to functions on the dual space. Legendre transformations
are also related to projective duality and tangential coordinates in algebraic geometry.
Formal definitions of this transformation can be found in [25] and in Appendix C of [27].
But, to have a general overview we follow the explanation in [23]. Consider a function
of two variables Ψ(ξ, η) and consider a new variable ζ such that
∂Ψ
ζ= . (2.13)
∂η
Now represent Ψ in Pfaff form, i.e.,
∂Ψ ∂Ψ
dΨ = dξ + dη. (2.14)
∂ξ ∂η
If we define Φ := η ζ − Ψ, and replace Equations (2.13) and (2.14) in its differential, we

obtain
dΦ = dη ζ + η dζ − dΨ
∂Ψ
= η dζ + ζ dη − dξ − ζ dη
∂ξ
∂Ψ
= η dζ − dξ . (2.15)
∂ξ
The function Φ depends on (ψ, ζ) rather than on (ψ, η). In doing so, we have encoun-
tered the legendre transformation.
2.1.2.2 The Hamiltonian
Consider the Lagrangian as Ψ and the variable ζ as the canonical momenta pj . Using
the Legendre Transform we obtain
X
H(p, q, t) = pk q̇k − L(q, q̇, t), (2.16)
k

where H in (2.16) is called the Hamiltonian. The differential of this Hamiltonian is

X
dH = (pk dq̇k + q̇k dpk ) − dL (q, q̇, t)
k
X ∂L ∂L
X
= (pk dq̇k + q̇k dpk ) − pk dq̇k + dqk − dt
k k
∂qk ∂t
X ∂L ∂L

= − dqk + q̇k dpk − dt . (2.17)
k
∂qk ∂t
2.1.2.3 The Canonical Equations
The general form of the Pfaffian of the Hamiltonian H(q, p, t) is
X ∂H
∂H

∂H
dH = dqk + dpk + dt . (2.18)
k
∂qk ∂pk ∂t
We can rewrite the Euler-Lagrange equations (2.6) using the canonical momentum (2.12)
to obtain
∂L d
= pk = ṗk . (2.19)
∂qk dt
Then, (2.17) becomes
X ∂L
dH = [−ṗk qk + q̇k dpk ] − dt . (2.20)
k
∂t
Equating (2.18) and (2.20) we obtain the canonical equations of the Hamilton, that is
>
∂H ∂H

q̇k = → q̇ = , (2.21)
∂pk ∂p
and >
∂H ∂H

ṗk = − → ṗ = − . (2.22)
∂qk ∂q
It is important to notice that the partial derivatives ∂H/∂t and − ∂L/∂t can not
be equated since they were taken considering different variables. However, because
dqk /dt = q̇k and dpk /dt = ṗk is feasible to evaluate − ∂L/∂t from Equation (2.20)
∂L dH X dqk dpk ∂H

− = + ṗk − q̇k = . (2.23)
∂t dt dt dt ∂t
The Hamiltonian is then a constant of the system motion provided the Lagrangian does
not depend explicitly on the time t.

2.1.2.4 The Hamiltonian Function with Constraints
In Section 2.1.1.7 we consider g(q) = 0 where g : Rnq → Rnλ are constraint equations
and λ ∈ Rnλ are arbitrary functions with λk gk = 0. If we incorporate the constraints
in the Hamilton’s Principal Function, we obtain
Zt2 ("
X
#
X
)
S= dt pk q̇k − H + λk gk . (2.24)
t1 k k
Now we can find the δ-variation similar as in the Lagrangian Formulation, which will
lead in the canonical equations with constraints, that is
>
∂H ∂H

q̇k = → q̇ = , (2.25)
∂pk ∂p
and n > >

λ
∂H X ∂gk ∂H ∂g

ṗk = − + λk → ṗ = − + λ. (2.26)
∂qk k
∂qk ∂q ∂q
2.2 Passivity-based control and Energy Shaping

The term PBC was first introduced in the context of motion control of mechanical
systems [28]. The main idea is to define a controller which achieves stabilization by
rendering the system passive with respect to some desired storage function and injecting
damping. Energy shaping is a PBC, which has its roots in the work (Potential energy
shaping) of Takegaki and Arimoto [29] long before it was related to passivity [30]. The
goal of energy shaping is to virtually modify the energy of the system, composed of
kinetic and potential energy, to stabilize the desired equilibrium.
In this thesis, we focus our attention on an energy shaping control method suitable for
mechanical systems: Interconnection and Damping Assignment (IDA).
2.2.1 Dissipativity, passivity and stability

Dissipativity and Passivity properties describe the notion of energy dissipation subject
to a system. The goal is to realize that a system cannot have, at a certain time, more
energy than what was injected into it. To mathematically consolidate the concept we
must define a storage function, i.e., how much energy is stored in the system, and a
supply rate, i.e., a measure of how fast external energy is injected in the System. The
mathematical basis presented here is well explained in [31–35]

Assumption 1. The system
x(t) ∈ X ⊆ Rnx ,
X ẋ = f (x, u)
: , u(t) ∈ U ⊆ Rnu , (2.27)
y = h(x, u)
y(t) ∈ Y ⊆ Rny ,
with f (0, 0) = 0 and h(0, 0) = 0 for all x(0) = x0 and inputs u = u(t) is the solution
x(t) = ϕ(x0 , u(t), t), ∀ t ≥ 0. Where s : U × Y → R is the supply rate of
P
, which
satisfies for all x0 ∈ X and ∀ u(t) ∈ U the condition:
Z t
| s(u(τ ), y(τ ) | dτ < ∞ , ∀ t ≥ 0.
0
P
Definition 1 (Dissipativity [35]). The system is said to be dissipative with respect
to the supply rate s, if there is a non-negative storage function V (x) ≥ 0 , V : X →
R+ , such that ∀ x0 ∈ X and ∀ u(t) ∈ U fulfills the integral dissipation inequality (IDE)
Z t
V (x(t)) − V (x0 ) ≤ s(u(t), y(t)) dt ,∀ t ≥ 0 (2.28)
0
P
where x = ϕ(x0 , u(t), t). In case the IDE becomes a equality, the system is called
loss-free.
Frequently V is continuous and differentiable, then, the derivation of V along the

solution x = x(t) is given by the Differential Dissipativity Inequality (DDI)
∂V
f (x, u) ≤ s(u(t), y(t)) , ∀ t ≥ 0.
∂x
P
Definition 2 (Passivity [35]). A system with nu = ny is called passive if it is
dissipative with respect to the supply rate s(u, y) = y > u and there is a storage function
V satisfying V (0) = 0.
In the literature we can find different types of passive systems. These are shown
in Table 2.1. It is important to notice that a passive system cannot store more energy
than it is supplied from the outside, and a mechanical system satisfies the energy
conservation, that is
Stored energy = Supplied energy − Dissipation
P
Definition 3 (Zero-State observable (ZSO) [35]). The System is zero-state
observable if u(t) = 0, y(t) = 0, ∀ t ≥ 0, implies x(t) = 0, ∀ t ≥ 0.

Type of passivity Description .

If there is a function ψ(x) > 0 such that the
Strictly passive
supply rate s(u, y) = y > u − ψ(x).
If there is a ψ(x) with u> ψ(u) > 0, ∀ u 6= 0 s.t. the

Strictly input passive
supply rate s(u, y) = y > u − u> ψ(u).
If there is an α > 0 s.t. the

α-input passive
supply rate s(u, y) = y > u − αku(t)k2 .
If there is an S(y) with y > S(y) > 0 ∀ y 6= 0 s.t. the

strictly output passive
supply rate s(u, y) = y > u − y > S(y).
If there is a β > 0 such that the

β-output passive
supply rate s(u, y) = y > u − βky(t)k2 .
Conservative If it is loss-free.
Table 2.1. – Types of passivity
A weaker version of the observability property, used to prove asymptotic stability, is

presented in the next definition.
P
Definition 4 (Zero-State detectable (ZSD) [35]). The System is zero-state
detectable if u(t) = 0, y(t) = 0, ∀ t ≥ 0, implies limt→∞ x(t) = 0.
P
Lemma 1 (Passivity and Lyapunov Stability). Consider a system with storage
function V ∈ C 1. The equilibrium x = 0 of ẋ = f (x, 0) is said to be asymptotically
stable, if one of the two following conditions hold
P
1. The system is strictly passive.
P
2. The system is strictly output passive with V positive definite and ZSO.
In addition, if V is radially unbounded, then x = 0 is globally asymptotically stable.

We can extend this with two additional properties
P P
(i) If is ZSD, then the equilibrium x = 0 of with u = 0 is stable.
(ii) When there is no throughput, i.e., y = h(x), the feedback u = −y achieves asymp-
P
totic stability of x = 0 iff is ZSD.
Proof. see [30, 36]

P
Assumption 2. From now on, it is assumed that has no feedthrough terms, i.e.
y = h(x).
2.2.2 Port Hamiltonian Systems

The class of Hamiltonian systems, i.e., the canonical equations and the Hamiltonian
function, is usually related to network modelling, which extends to include Dirac struc-
tures. In simple words, a Dirac structure4 is a linear space which describes internal
power flows and allows the power exchange between the system and the environment
via a set of input and outputs, called ports, giving rise to the prominent term port-
Hamiltonian (pH) system. These extended models reveal the passivity properties which
we can use to perform the passivity-based control laws. The mathematical concepts
here presented are a brief review of [22, 37, 38]. We consider port-Hamiltonian systems
of the form

= (J(x) − R(x))( ∂H > x(t) ∈ X ⊆ Rnx ,
 ẋ ∂x ) + Gu (x)u,


X
: u(t) ∈ U ⊆ Rnu , (2.29)

G> ∂H >
y(t) ∈ Y ⊆ Rnu ,

 y = h(x) := u (x)( ∂x ) ,
where the continuously differentiable Hamiltonian function H : X → R is bounded

from below, and represents the stored energy H(x) ≥ 0 and follows the IDE. Further,
J : X → Rnx ×nx is the power-conserving internal interconnection structure and it is
skew symmetric, that is J(x) = −J > (x), and R : X → Rnx ×nx is a symmetric matrix
(usually called the resistive structure), which characterizes the energy dissipation, and
is positive semidefinite for physical systems. The input u and the output y are conjugate
P
variables, i.e, their product gives a power quantity. The system with storage function
H(x) is passive, because the passivity inequality
∂H
Ḣ = ẋ
∂x
∂H ∂H >

= (J(x) − R(x))( ) + Gu (x)u
∂x ∂x
∂H ∂H > ∂H ∂H > ∂H
= J(x)( ) − R(x)( ) + Gu (x)u
∂x ∂x ∂x ∂x ∂x
∂H ∂H >
=− R(x)( ) + y>u ≤ y>u (2.30)
∂x ∂x
holds. If u = 0 → Ḣ ≤ 0 and H(x) has an isolated minimum in x? , i.e, x? = arg min H,

then H is a (weak) Lyapunov Function. Besides, if we add damping of the form
4
However, in this work we use modulated Dirac structures which are not linear spaces, see [32, 37].

u = −φ(y), for a function φ : Y → Rny satisfying φ(y) y > 0, ∀ y 6= 0, then asymptotic

stability (in x? ) is guaranteed if (2.29) is zero-state detectable (ZSD).
2.2.3 Implicit and Explicit Port Hamiltonian Structure of a Mechan-

ical System
In Section 2.1.1 we have seen that a mechanical system can be described by generalized
coordinates q ∈ Rnq . Let us represent the potential energy with V = V (q) and the
kinetic energy in the quadratic form T = T (q, q̇) = 1
2 q̇ > M (q) q̇, where M (q) is the
symmetric and positive

definite

mass matrix. The system behavior is defined by the
d ∂L ∂L
Lagrange equations dt ∂ q̇k = ∂qk , where the corresponding Lagrangian is L = T − V .
In Section 2.1.2 we showed how Hamilton simplified the structure of the Lagrange
∂L
equations by introducing the momenta pk ≡ ∂ q˙k and the Hamiltonian as a function of p
and q, i.e. H = H(p, q, q̇) where H = p> q̇ −L(q, q̇). We can pay heed to the momenta
which define for every q a continuously differentiable bijection, i.e. q̇ ↔ p. This map is
called the Legendre Transform, which we achieve since the linear momentum in classical
mechanics is defined as p = M (q) q̇, so the existence of the Legendre transform is
established. By replacing the variable q̇ = M −1 (q) p in the definition of H we obtain
1 >

>
H = p q̇ − q̇ M (q) q̇ − V (q) ,
2
1 > −1
= p M (q) p + V (q). (2.31)
2
If in addition we consider an input matrix of forces G(q) u, i.e., the control forces,
where

u

∈ Rnu , and G : Rnq → Rnq ×nu , then we can rewrite the Lagrange equations as
d ∂L ∂L
dt ∂ q̇k − ∂q k
= G(q) u, and we can present the pH structure by means of the canonical
equations, thus
#  ∂H  "
I ( ∂q )> 
" # " #
q̇ 0 0
= ∂H >
+ u, (2.32)
ṗ −I 0 ( ∂p ) G(q)
1 > −1
H(q, p) = p M (q) q + V (q). (Hamiltonian)
2
Equation (2.32) is well known as the (explicit) pH representation of a mechanical sys-

tem. Similarly, we can consider the constraint equations g(q) = 0, with the arbitrary

functions λ, to find the implicit pH structure of a mechanical system5 , that is

# 
( ∂H )>
" # " " # " #
q̇ 0 Inq  ∂q 
0 0
= >
+ λ+ u, (2.33)
ṗ −Inq 0 ( ∂H
∂P ) b(q) G(q)
∂H >
0 = b> (q) , (2.34)
∂p
1 > −1
H(q, p) = p M (p) q + V (q), (Hamiltonian)
2
where b(q) = ( ∂g(q)/∂q )> and Equation (2.34) is the time derivative of g(q) = 0,
that is
∂g(q) dq

0= = b(q)> q̇
∂q dt
>
∂H

>
= b(q) .
∂p
2.2.4 Underactuated Mechanical Systems

We can identify a fully actuated mechanical system if the input matrix G(q) is invert-
ible, i.e., rank G(q) = nu = nq , and an UMS if rank G(q) = nu < nq , in other words, a
system that has fewer independent inputs than number of degrees of freedom.
h
To check
i
if an implicit system is underactuated we can follow [39] and define S = G(q) b(q) .
Then we say the system is an UMS if rank S < nq .
2.2.5 Interconnection and Damping Assignment - PBC

The IDA-PBC was introduced in [40, 41]. The general idea is to find a passivity-based
control law u on a system with pH structure as in Equation (2.29), such that the
closed-loop has a desired or target pH structure and the equilibrium is asymptotically
stable. To achieve the desired equilibrium (q ? , 0) we need to shape the potential energy.
For the case of UMS, shaping the potential may not be sufficient, thus, requiring to
modify the kinetic energy, which implies solving a PDE [39]. This thesis focuses on
the application of IDA-PBC to UMS. We consider the following desired closed-loop pH
system
?
X
? > ,
: ẋ = (J ? (x) − R? (x))( ∂H
∂x )
(2.35)
5
Note that we are using the same symbols for explicit and implicit structure. However, for the same
physical system, the inertia matrix, the input matrix, the dissipation and the potential energy could
change depending on the representation.

with (R? (x))> = R? (x) ≥ 0, (J ? (x))> = −J ? (x) and equilibrium point in x = x? .

P P?
Equating both systems, i.e., = , gives place to the matching
> >
∂H ? ∂H

? ?
Gu (x)u = (J (x) − R (x)) − (J(x) − R(x)) . (2.36)
∂x ∂x
To ensure (asymptotic) stability in the desired equilibrium x? , we apply Lyapunov’s

direct method. Thus, the following definiteness constraints must be satisfied
1. R? is positive (semi-) definite, and
2. The desire Hamiltonian function H ? has a strict minimum at x = x? .
Lemma 2. Let Gu be a matrix-valued map Gu : X → Rnx ×nu with rank(Gu ) = nu <

nx . Define G⊥
u ∈ R
(nx −nu )×nx as the full rank left annihilator of G , i.e., G⊥ G = 0.
u u u
For any f ∈ Rnx , u ∈ U ⊆ Rnu
(
0 = G⊥
u f (x),
f (x) + Gu (x) u = 0 ⇔ (2.37)
u = −(G> −1 >
u Gu ) Gu f (x).
Proof. The proof is completed using the annihilating property of G⊥

u and noting that
the square matrix G>
u Gu is invertible, see [30].
Remark. As we seek for the control law u we need to to perform the Moore-Penrose
to Equation (2.36), i.e. first multiply both sides by G> > −1
u , then by (Gu Gu ) .
According to Lemma 2 we can choose a control law u, as presented in the following

equation
> > !
∂H ? ∂H

u= (G> −1 >
u Gu ) G u
?
(J (x) − R (x)) ?
− (J(x) − R(x)) , (2.38a)
∂x ∂x
> > !
∂H ? ∂H

0= G⊥
u
?
(J (x) − R (x)) ?
− (J(x) − R(x)) . (2.38b)
∂x ∂x
Therefore the matching problem is solved if and only if the matching condition Equa-
tion (2.38b) is satisfied. For the case of UMS we need to define some restrictions on the
design of J ? , R? and/or on the closed loop Hamiltonian energy function H ? . Solving
the matching condition is unduly a key step of this technique and it can be challenging
to solve.
There are mainly three ways to proceed for the solution of the matching condition (2.38b).
The most common approach is Non-Parameterized IDA, here the idea is to fix the

desired interconnection J ? (x) and dissipation matrices R? (x), hence its name. Addi-
tionally, G⊥ ?
u is fixed and H (x) is obtained from a set of partial differential equations
PDEs given by Equation (2.38b). The energy function H ? (x) is chosen such that it
has a strict minimum at the desired equilibrium x = x? .
The algebraic IDA approach fixes the energy function, then Equation (2.38b) becomes
an algebraic equation in J ? (x), R? (x) and G⊥
u.
The third approach is known as Parameterized IDA. Here, the structure of the desire
energy function is restricted to a certain class. This restriction is motivated for some
physical system, for instance, mechanical systems where the structure of the energy
function is the sum of the kinetic and the potential energy, which are T and V respec-
tively. This choice results in a different (and simpler) set of PDEs, however it also
imposes some constraints in J ? (x) and R? (x).
This thesis centers on the application of the parameterized IDA to UMS. Therefore,
we present the main considerations in order to determine a suitable control law uida .
Consider a pH structure of a mechanical system such as in Equation (2.32), where
>
x = [ q> p> ] , Gu = [ 0 G> ]> and with passive output y = G> >
u ( ∂H/∂x ) .
Definition 5 (Admissible equilibrium [30]). An equilibrium (q ? , 0) (or simply q ? )

of a mechanical system is called admissible if the following equality is satisfied

∂V (q)
G⊥ =0 (2.39)
∂q q?
We assign to the closed-loop a pH structure in accordance with mechanical systems,

that is
#  ∂H ? 
)>
" # "
q̇ 0 J1 (
=  ∂q?  , (2.40)
ṗ −J1> J2 − R2 ( ∂H
∂p )
>
1 > ? −1
H ? (q, p) = p M (q) p + V ? (q), (Desired Energy Function)
2
where M ? = M ? (q) is the desired

h
symmetric
i
and positive definite mass matrix, J2 is
0 0
a skew symmetric matrix, R? = 0 R2 is a symmetric matrix. In this particular case
the matching condition (2.36) becomes
#  ∂H ?  " # 
)> I ( ∂H >
" # "
0 0 J1 ( 0 ∂q ) 
uida =  ∂q?  − . (2.41)
G(q) −J1> J2 − R2 ( ∂H
∂p )
> −I 0 ( ∂H
∂p )
>
Due to the linear momenta q̇ = M −1 p which preserve in the closed loop, we have
the relationship J1 = M −1 M ? . Based on the result of Lemma 2, the control law that

satisfies (2.41) is
> > > !
∂H ∂H ? ∂H ?

> −1 >
uida = (G G) G − J1> + [J2 − R2 ] (2.42)
∂q ∂q ∂p
if and only if the PDE

( > > > )
∂H ∂H ? ∂H ?

⊥
G − J1> + [J2 − R2 ] =0 (2.43)
∂q ∂q ∂p
is satisfied. Assuming R2 = R2 (q), and J2 = J20 (q)+J21 (q, p), with J21 linear in p, the
matching condition Equation (2.43) can be naturally split according to the dependency
on p, i.e. the terms quadratic and independent from p correspond to the kinetic and
potential energies, respectively [42]. Meanwhile the terms linear in p correspond to the
dissipation. Thus, (2.43) can be rewritten as
 !> !> 
∂ p> M −1 p ∂ p> M ? (q)−1 p
G⊥  − J1> + 2 J21 M ? (q)−1 p = 0, (2.44a)
∂q ∂q
> > !
∂V (q) ∂V ? (q)

G⊥ − J1> = 0, (2.44b)
∂q ∂q

G⊥ [J20 − R2 ] M ? (q)−1 p = 0. (2.44c)
Equation (2.44a) is a non-homogeneous, first order quasilinear PDE that has to be

solved to determine the unknown elements of the desired mass matrix M ? (q). For a
given desired mass matrix, Equation (2.44b) becomes a linear PDE for the unknown
function V ? (q), and the third equation, Equation (2.44c) is a simple algebraic equation,
which can be solved by choosing
J20 − R2 = G(Kj − Kv )G> (2.45)
with free parameters Kj and Kv where Kj = −Kj> ∈ Rnu ×nu , and Kv = Kv> > 0. Is
important to note that R2 := G Kv G> , this choice is later explained, but the main
reason is due to the relationship between the time derivative of the desired energy
function
dH ? (q, p) ∂H ? ∂H ?

= q̇ + ṗ (2.46)
dt ∂q ∂p
and the so-called passive output of the closed-loop, which is defined by
∂H ? >
y ? = G> ( ) . (2.47)
∂p

Corollary 1 (Stability of the closed-loop system [30]). Consider the system (2.32)
with desired cloosed loop system (2.40), R2 := G Kv G> , and control law (2.42). If
q ? = arg min V ? , M ? (q) = M ? (q)> > 0 and R2 ≥ 0 (2.48)
in a neighborgood of q ? , then the equilibrium q ? is (locally) stable with Lyapunov func-

tion H ? . Asymptotic stability follows if the closed loop system is zero-state detectable
from the output y ? .
Proof. Replacing q̇ and ṗ from the closed loop system (2.40) in (2.46) yields
dH ? (q, p) ∂H ? ∂H ? >
=( ) (J2 − R2 ) ( ) . (2.49)
dt ∂p ∂p
with J2 = −J2> and R2 := G Kv G> . Then, replacing the passive output in Equa-
tion (2.49) results in
dH ? (q, p)
= −(y ? )> Kv y ? ≤ −λmin {Kv } ky ? k22 ≤ 0 (2.50)
dt
Asymptotic stability, under the zero-state detectability condition, is established invok-

ing Barbashin-Krasovskii, see [36].

Chapter 3
Implicit IDA-PBC for UMS
This chapter addresses the systematic design of passivity-based controllers for implicit
port Hamiltonian structures of underactuated mechanical systems (UMS)s. Therefore,
the energy shaping method will be introduced to implicit pH structures. The mathe-
matical concepts here presented are borrowed from [4, 37].
3.1 Problem formulation

In Section 2.2.5 the IDA-PBC methodology was presented for UMSs, where we found
a state feedback (2.42) through the solution of the matching problem (2.41). We also
observed that it is possible to represent a mechanical system in implicit pH structure.
In view of this, three questions naturally arise:
1. Is it possible to find a suitable control law for UMS in implicit pH representation?
2. If we shape the energy of a mechanical system with constraints, is it possible to

find a general structure for the desire mass matrix?
3. Can we find an output-feedback law?
This three question are answered by Cieza and Reger in [4]. The purpose of this chapter
is to present the general idea of the Implicit IDA-PBC approach and later in Chapter 4,
implement it on a physical system, namely, the portal crane.

3. Implicit IDA-PBC for UMS
3.2 Implicit IDA-PBC for holonomic systems

As presented in Section 2.2.3, let us briefly recapitulate the implicit6 representation of
a mechanical system given by
# 
( ∂H )>
" # " " # " #
ṙ 0 Inq  ∂r 
0 0
= >
+ λ+ u, (3.1a)
ρ̇ −Inq 0 ( ∂H
∂ρ ) b(r) G(r)
∂H >
b> (r) = 0, (3.1b)
∂ρ
1 >
H(r, ρ) = ρ M(r)−1 ρ + V(r), (Hamiltonian)
2
where r ∈ Rnr and ρ ∈ Rnr are implicit generalized coordinates (position and its
canonical momenta, respectively), u ∈ Rnu is the input and G : Rnr → Rnr ×nu is
the implicit full rank input matrix. The nλ holonomic constraints g(r) = 0 and the
arbitrary functions (or implicit variables) λ ∈ Rnλ are related by the constraints forces
b(r) λ where ( ∂g(r)/∂r )> = b(r) : Rnr → Rnr ×nλ . The velocity in implicit coordinates
is ṙ ≡ ( ∂H/∂ρ )> = M(r)−1 ρ, where M(r) : Rnr → Rnr ×nr is the symmetric positive
definite mass matrix and the Hamiltonian H : Rnr × Rnr → R represents the total
energy function. We say that (3.1) is an UMS if rank S < nr , where S = [ G(r) b(r) ].
Proposition 1 (Well-posedness [4]). Consider the holonomic system (3.1). Then for
all r ∈ X = {r ∈ Rnr | rank ∆ = nλ }, ∆ := b> (r) M(r)−1 b(r), the constrained
state-space set
∂H >
Xc = {(r, ρ) ∈ X × Rnr | b> (r) ( ) = 0, gk (r) = 0}
∂ρ
is a regular manifold embedded in Rnr × Rnr , and (3.1) described by Differential-

Algebraic Equations (DAEs) has differential index 1 with unique solution for λ. Here,
Z r
gk (r) = b>
k (s) ds + ci ≡ 0, (3.2)
0
is the general expression for the integrated constrains, i.e., b(r) = ( ∂g(r)/∂r )> , bk is
the kth column vector of b(r), ci is a constant and ∂H/∂ρ b(r) differentiable.
6
We write now r instead of q, ρ instead of p, H instead of H, V instead of V , M instead of M and G
instead of G with the purpose to be aware that we are dealing with the implicit pH representation.

Similar to the explicit IDA-PBC, we define the desired closed-loop pH system

#  ∂H  "
d >
" # " #
ṙ 0 J (r) ( ∂r ) 0
= 
∂H > +
 λd , (3.3a)
ρ̇ −J > (r) −W(r, ρ) ( ∂ρ
d
) bd (r)
∂Hd >
0 = b>
d (r) , (3.3b)
∂ρ
1 >
Hd (r, ρ) = ρ Md (r)−1 ρ + Vd (r), (Desired Hamiltonian)
2
where the desired dynamics are well defined for all
r ∈ Xd = {r ∈ Rnr | rank ∆d = nλ }, ∆d := b>

d (r) Md (r)
−1
bd (r).
The desired Hamiltonian Hd = Hd (r, ρ) is the new shaped energy function, λd ∈ Rnλ
are the new implicit variables in closed-loop, Md = Md (r) is the nonsingular symmetric
desired mass matrix, J : Xd → Rnr ×nr is nonsingular, and W = W(r, ρ) : Xd × Rnr →
Rnr ×nr .
As before we are dealing with mechanical systems, thus, the momenta are preserved
in the closed-loop, i.e. ( ∂Hd /∂ρ )> = M−1
d ρ ⇒ J = M
−1 M . Besides, the physical
d
property (3.1b) is equivalently represented as (3.3b), then equating both results in
bd = J > b(r). Next, we equate (3.1a) and (3.3a), resulting in
" #
i u > > >
∂H ∂Hd ∂Hd
h
>
G(r) b(r) = −J −W + J > bd (r) λd . (3.4)
| {z } λ ∂r ∂r ∂ρ
=: S
Then, similar to Lemma 2, we have

" # > > > !
u ∂H ∂Hd ∂Hd

> −1 > > >
= (S S) S −J −W +J bd (r) λd (3.5a)
λ ∂r ∂r ∂ρ
> > > !
∂H ∂Hd ∂Hd

⊥ > >
0=S −J −W +J bd (r) λd . (3.5b)
∂r ∂r ∂ρ
Where S ⊥ is the full rank left annihilator of S. The implicit matching problem is solved
through the following propositions
Proposition 2 (Implicit Matching [43]). The implicit feedback u = uI ,

> > > !
∂H ∂Hd ∂Hd

† > >
uI = S −J −W +J b(r) λd (3.6)
∂r ∂r ∂ρ

transforms the system (3.1) into (3.3) for any trajectory of r that remains in X ∩ Xd ,
whenever the implicit matching conditions
 !> !> 
∂M(r)−1 ρ ∂Md (r)−1 ρ
S⊥  −J> − W1 Md (r)−1  ρ = 0, (3.7a)
∂r ∂r
> > !
∂V(r) ∂Vd (r)

⊥ >
S −J = 0, (3.7b)
∂r ∂r

S ⊥ J > b(r) = 0, (3.7c)
−1
are satisfied with J = M−1 Md , bd (r) = J > b(r), S † = [ Inu 0 ](S > S) S>,
1
W(r, ρ) = W1 (r, ρ) + S Ku (r) S > , (3.8)
2
Ku = Ku (r) ∈ R(nu +nλ )×(nu +nλ ) and W1 = W1 (r, ρ) ∈ Rnr ×nr .
Remark. We can compute λ or λd through, the hidden constraints, i.e., the time
derivative of (3.1b) or (3.3b).

d b> M−1 ρ ∂ b> M−1 ρ ∂ b> M−1 ρ
= ṙ + ρ̇ = 0
dt ∂r ∂ρ
  
∂ b> M−1 ρ
∂H
> !!
⇒ λ = ∆−1 − ṙ  − b> M −1 G(r) uI −  (3.9)
∂r ∂r
Proposition 3 (Implicit stability [43]). Assume that the conditions of Proposition 2

are satisfied for an holonomic system, and define the new domain
XI = ({r ∈ Xd | b⊥ M M−1 ⊥ > nr

d M (b ) 0} × R ) ∩ Xc .
The closed-loop system (3.3) has a stable equilibrium in

( > )
∂V(r)

? ? ⊥
x = (r , 0) ∈ Xa = x ∈ XI | S =0
∂r
for any Ku (r) + Ku> (r) 0 if
r∗ = arg min Vd (3.10)

r ∈ XI

is an isolated minimum and

>
M−1 −1

0= ρ d W1 Md ρ . (3.11)
XI
Furthermore, if
1
yI = (Ku (r) + Ku> (r)) 2 M−1
d ρ (3.12)
is a detectable output of (3.3), x? is asymptotically stable.
Proof. The definition of Xd assures well-defined dynamics of (3.3a) under (3.3b). The
transformation of the implicit pH structure (3.1) to a closed-loop implicit structure (3.3)
is possible by noting that the control law (3.6) and the implicit matching equations (3.7)
are sufficient conditions for (3.5). Besides, such mechanical structure transformation
implies J = M−1 Md , bd = J > b. Since all constraints are integrable, define the
Lagrangian function
Ld (r, ρ, ν, µ) := Hd + ν > b> M−1 ρ + µ> g (3.13)
with Lagrange multipliers µ and ν and constraints g. Then, seeking for a minimum
(or maximum) of Hd |Xc , the following expression should be satisfied
>
∂Ld (r ? , ρ? , ν ? , µ? )

= 0, (3.14)
∂x
with
   
> > M−1
∂Ld (r, ρ, ν, µ) ( ∂L d >
∂r ) ( ∂Vd )> + ( ∂ρ bν >
) + b µ

=  ∂Ld >  =  ∂r ∂r . (3.15)
∂x ( ∂ρ ) M−1 b ν + M−1
d ρ
h i>
Multiplying on the left by b (b⊥ )> M−1 Md , and taking advantage of the full rank
condition of ∆d , Equation (3.14) is reduced to
>
∂Vd (r ? )

ρ? = 0, ν ? = 0, and, + b(r ? )µ? = 0. (3.16)
∂r
Replacing x? = (r ? , ρ? ), (3.14) and (3.7c) in (3.7b), yields

>
∂V(r ? )

⊥ ?
S (r ) = 0, (3.17)
∂r

which represents the attainable set Xa . We might now employ r ? = arg min Vd |XI and
Finsler’s Lemma on ρ> M−1
d (r)ρ subject to (3.3b), obtaining
Hd > 0 iff x ∈ XI and Vd (r ? ) = 0. (3.18)
Stability in x? can be demonstrated if we use Hd as Lyapunov Function for x ∈ XI

with minimum at x? and time derivative

> −1 −1

Ḣd (r, ρ) = −ρ Md (r) W Md (r) ρ ≤ 0. (3.19)
XI
Finally, applying Barbashin-Krasovskii-LaSalle’s Theorem implies convergence of yI

to 0. Therefore, asymptotically stability in x? is reached if yI is a detectable output
of (3.3).
Remark. Md is a full rank matrix but not necessarily Md 0. Also, it is important

to notice that the matching condition (3.7c) introduces conservativeness and allows to
reduce complexity solving (3.7a) and (3.7b) independently of λd , even though the latter
is required for the controller (3.6).
3.2.1 Mechanical systems with constant mass matrix

One may claim that the implicit matching conditions are quite complicated to solve.
However, the main advantage lies on systems that have a constant mass matrix. That
means that the mass matrix modeled in the Euclidean space has no dependence on the
generalized coordinates. However, in the explicit representation, it may indeed possess
that dependence. The following proposition avoids solving the kinetic PDE (3.7a).
Proposition 4 (Algebraic Implicit IDA-PBC [4]). Consider an implicit pH system

(3.1) with only holonomic constraints. Assume, additionally, that (3.1) is well-posed
and posesses a constant mass matrix M, a linear potential energy V(r), and polynomial
G(r) and g(r). The feedback (3.6) stabilizes the closed-loop (3.3) at the equilibrium
x? = (r ? , 0), with Ku + Ku> 0 if, there
(i) exist a constant vector µ? ∈ Rnλ ,
(ii) exist a matrix A ∈ R(nr −nλ )×(nr −nλ ) with A = A> 0,
(iii) exist a matrix C ∈ R(nr −nλ )×nλ ,
(iv) exist a non-singular matrix D ∈ Rnλ ×nλ with D = D> , and
(v) exist a matrix S̄(r) ∈ R(nu +nλ )×nψ , where nψ ≤ nu + nλ and S i is the ith column
of S

such that
> !
∂V

⊥ ?> ?
S + M Zc D b µ = 0, (3.20a)
∂r

S ⊥ M (b?⊥ )> A b?⊥ + Zc D Zc> b = 0, (3.20b)
!>
∂(S S i ) −1 ∂(S S i )
M Md − Md M−1 = 0, (3.20c)
∂r ∂r
Z ⊥ Za (Z ⊥ )> 0, (3.20d)
where
" #
? A + CDC > CD
Md := M B B ?> M, (3.21)
DC > D
1 >
Vd := ψ Kψ ψ + r̃ > b? µ? , (3.22)
2
b?⊥
h i Rr > −1
B ?> = , ψ(r) = r? S̄i (s) Md M ds, r̃ := r ? − r, b? = b(r ? ) , Zc = (b∗⊥ )> C +
b?>
b? , Za = A b?⊥ ∂b µ? /∂r |r=r? (b?⊥ )> A, Z = [ Inr −nλ −C ]B ?−1 M−1 S(r) S(r ? ) has
full rank,
h i>
(S ⊥ )> M−1
d W1 = 0, (3.23a)

Kψ −Z † Za − Za (Z ⊥ )> (Z ⊥ Za (Z ⊥ )> )−1 Z †> , (3.23b)
Kψ ∈ Rnψ ×nψ is symmetric, and Z † = (Z > Z)−1 Z > . Moreover if yI is a detectable

output, then (3.3) is asymptotically stable.
Proof. It begins by observing that Vd exists due to the existence of ψ by the integrability
condition (3.20c). Similar to the proof of Proposition 3 we need a Lagrange function
Ld as presented in (3.13), thus µ∗ ∈ Rnλ . Then, direct substitution of Hd in (3.7) with
(3.23a), results in (3.20a)–(3.20b) and fulfills (3.7a) and (3.11). The next step is to
make x? a strict (local) minimum of Hd |Xc . The necessary and sufficient conditions
(see [44]) for this are (3.14) and
yρ> M−1
d yρ > 0, ∀yρ with yρ> M −1 b? = 0, and,
!
∂ 2 Vd ∂(bµ? )
yr> + yr > 0, ∀yr with yr> b? = 0
∂r 2

∂r
r=r ?

Now, we can use Finsler’s Lemma and replace ψ, J and Md in the above inequalities7
which results in

⊥
M M−1 ⊥ >
= A−1 ,

b d M (b ) (3.24a)
r=r ?
>
Z Kψ Z + Za 0. (3.24b)
Inequality (3.24a) is equivalent to A 0. We can obtain (3.20d) and (3.23b) if we mul-

tiply (3.24b) on both sides by adequate full rank matrices and use Schur’s complement.
Straightforward calculations in r? show that ∆d (r ? ) = (b? )> b? D (b? )> b? is full rank,
which implies that x? is a strict (local) minimum of Hd |XI ⊆Xc . Thus, the desired pH
structure in closed-loop is (asymptotically) stable in a neighborhood of x? .
Figure 3.1 shows how to perform the algorithm8 . The algorithm begins by selecting r∗
and adequate full rank left annihilators S ⊥ and b⊥ . Afterwards, we find a nummerical
solution of C (if possible) and constraints inequalities in A, D and µ? aided by the
implicit matching conditions (3.20a) and (3.20b). In the next step, assisted by (3.20c),
we select S̄, calculate Z and then select its full rank left annihilator Z ⊥ . Finally,
use (3.20d), (3.23) and Ku + Ku> 0 to choose A 0, W1 , non-singular D, µ? , Kψ
and Ku .
Remark. It is possible to consider Kψ as a matrix function of r, i.e., Kψ = Kψ (r).

However, Kψ (r ? ) has to fulfill Equation (3.23b). Although, taking this into practice
may influence the region of attraction.
Start
Select r , b⊥ and S ⊥
Find C and possible Use (3.20a)

constraints of A, D and µ and (3.20b)
Select S̄ , Z and Z ⊥ Use (3.20c)
Choose W1 , A 0 , Use (3.20d)

D (nonsingular ) , µ, Kψ , Ku and (3.23)
Figure 3.1. – Algorithm for the algebraic implicit IDA-PBC
7
The inversion of Md can be reached if the Schur complement is performed.
8
A requirement is to have the system in implicit pH structure. Thus, we can easily identify G, b and
S.

3.2.2 Position feedback

As we mentioned in Section 3.1 it is possible to produce a simple dynamic feedback
that obviates the velocity measurement. In order to do this, we need some additional
requirements.
Proposition 5 (Output-feedback [4]). Let the conditions of Proposition 4 be satisfied

with S S = G and W1 = 0 if, additionally, b⊥ J > b = 0, the new control law
> > !
∂H ∂Vd

† >
uN = S −J − G K̄u (ξ + ψ̄(r)) , (3.25a)
∂r ∂r
ξ˙ = −Λξ (r) K̄u (ξ + ψ̄(r)) (3.25b)
−1
with K̄u = K̄u> , Λξ + Λ> > nu
ξ 0, ∂ψ/∂r = G Md M, ξ ∈ R , Λξ ,K̄u ∈ R
nu ×nu ,
and Hd as in Prop. 4, stabilizes the system at x? . Moreover, the closed-loop is asymp-

totically stable if the system (3.3a) is zero state detectable with respect to the ouput
yN = G > M−1
d ρ.
Proof. Assume the conditions of Propositon (4) are met. Then, closing the loop of the
implicit pH system (2.2.3) with uN , b⊥ J > b = 0 or its equivalent J > b = bk (for some
square matrix k), S S = G, and ψ̇ = G > M−1
d ρ, results in a ‘new’ structure for the
implicit closed loop system, that is
∂ H̄d >
 
     
ṙ 0 J 0  ∂r  0
ρ̇ = −J >  ∂ H̄d >  > 
       
   0 −G 
  ∂ρ  J b λd ,
+ (3.26a)
ξ̃˙ G> −Λξ ∂ H̄d >
 
0 0
∂ξ
!>
∂ H̄d
0 = b> J , (3.26b)
∂ρ
1 1 > 1 >
H̄d (r, ρ) = ρ> M−1
d ρ+ ψ Kψ ψ + r̃ > b? µ? + ξ̃ K̄u ξ̃, (Hamiltonian)
2 2
| {z } 2
Vd (r)
| {z }
Hd
where ξ̃ = ξ + ψ̄. Afterwards, the time derivative of the desire energy function
1 > >
H̄˙ d (r, ρ) = − ξ̃ K̄u> Λξ + Λ>

ξ K̄u ξ̃ ≤ 0
2
reveals that H̄ is a (weak) Lyapunov function. Therefore, invoking Barbashin-Krasovskii-

Lasalle, ξ̃ goes to zero as time goes to infinity. Eventually, if ξ̃ = 0 we return
to the original desire implicit structure, i.e., (3.26) reduces to (3.3) with W = 0,

G > M−1
d ρ = 0 ≡ yn . Asymptotical stability can be shown if (3.3) has a zero state
detectable output yn .
Remark. The condition J > b = bk can also be stated as

0 = b⊥ M (b∗⊥ )> A b∗⊥ + Zc D Zc> b, (3.27)
replacing (3.20b).

Chapter 4
Explicit and Implicit IDA-PBC applied

to a Portal Crane
In this chapter, it is presented the implementation of the explicit and implicit IDA-PBC
approaches in a portal crane9 system, whose main task consists of moving a payload
in a smooth path and deposit the payload at the desired position. It is not a big
surprise to find within the scope a vast literature with different linear and non-linear
control approaches for this system [11, 15, 17, 19, 45–47]. However, the motivation for
this system is in its relative complexity which becomes a valuable resource to show
the advantages and disadvantages of the implicit non-linear controller. This chapter
starts by describing the system in Section 4.1. Then in Section 4.2 and Section 4.3, we
analyze, simulate and implement the IDA-PBC control laws in the real setup located
at the Laboratory of Control Engineering Group at Computer Science and Automation
Department, Technische Universität Ilmenau.
4.1 The Portal Crane System

Cranes are widely used in transportation and construction. Commonly they consist of
a hoisting mechanism (traditionally a hoisting line and hook) and a support mechanism
(trolley-girder) [6]. There are three classifications of the cranes: the portal crane, the
rotary crane and the boom crane. The main difference between them lies on the degrees
of freedom the support mechanism offers to the suspension point.
This work focuses on the portal crane system shown in Figure 4.1. Here, the support
mechanism (frame) is attached to the floor, i.e., not all the portal crane moves. In the
top, there are a set of girders and belts that move a bridge (Figure 4.1.b) through the
x-axis motor. Meanwhile, the trolley is mounted under the bridge, and its movement is
9
also known as gantry crane or overhead crane.

4. Explicit and Implicit IDA-PBC applied to a Portal Crane
x-Axis y-Axis
Motor Motor
Bridge
z-Axis
Motor
Trolley mc
Payload mp
(a) Frontal view of the whole system (b) Bridge, trolley and payload view
Figure 4.1. – Subfigures showing the portal crane system at the laboratory.
restricted to the y-direction. The x-axis, y-axis and z-axis motors are fixed in the frame,
bridge and trolley, respectively. A sketch of this configuration is presented in Figure 4.2,
where also, the frame dimensions are drawn. The maximum displacements of the trolley
in each axis are drawn in brackets, e.g. for the x-axis is [0 − 2000], where 0 represents
the origin. It is clear that the displacements should be less than the frame dimensions.
A closer look of the trolley in Figure 4.3 shows other elements such as encoders, a
pulley, and a winch or rope drum. A basket hangs from this winch, and thus it can also
move in the z-direction. The load can be attached to the basket in various ways and
then transported with it. We will refer to the basket and the additional weight (load)
as the payload. Gravity produces a constant force on the payload. For this reason a
driver must be equipped with a holding brake, preventing the payload movement (in
z [−1300 − 0]
[0 − 2000] x
00
y [0 − 1300]
15
mc
l
z x-Axis Motor
x mp
y-Axis Motor
Controller
1400
y
units z-Axis Motor

in mm
Figure 4.2. – Sketch of the portal crane.

the z-axis) without the servomotor being controlled. In case of failure, it also prevents
the free fall of the payload to the ground.
y-
z- Axis
Ax
is
Winchdrive
Bridge z-Axis
servomotor
x-Axis Gear drive
Trolley mc
Encoder α
Encoder β
l
Pulley
Payload mp
Figure 4.3. – Trolley and its elements
4.1.1 Additional Information of the Crane

Certainly, an advantage of this underactuated mechanical system is that it can be
modelled in 2-D or 3-D space as we will see in Sections 4.2 and 4.3. Regardless of the
model the goal is clear: To move the hanging load from one stationary position to a
new, target. However, it is crucial to be aware of how the components achieve this goal.
The physical setup is composed of a PC, a DS1103 controller board by the dSPACE
company [48], actuators, sensors, and the trolley and pendulum. The Laboratory uses a
fast prototyping approach. In essence, a full model of the system is built and simulated
in the PC before a hardware implementation is generated. Thus, the computer is
equipped with Matlab/Simulink and an RTI-toolbox, which compiles and export the
algorithm to the controller. The controller is connected to the servo drivers who enables
the control of the motor. There are two options to control them (i) by means of an
internal velocity loop or (ii) through a current control that could be approximate to
force control. However, the equipment do not have a sensor to measure the force applied
to the x- or y-axis. Thus, implementing a force control would require many unknown
parameters, e.g. the weights of the trolley, bridge, motors and other measurements in
the encoders. To avoid this, we use option (i) which is almost equivalent to the PFL
approach, where is usual to take as new input the acceleration of the system, see [49].
Thus, we can integrate the acceleration in order to use the velocity loop.

Let us remark, that the system is equipped with encoders (see Fig. 4.4a) to measure
the angles α(t) and β(t). Moreover, the system is able to calculate the measurement
of the trolley positions xc (t) and yc (t). Besides, the positions of the payload are xp (t),
yp (t) and zp (t), where the geometric restriction shown in Fig. 4.4b shows that
xp = l sin β, yp = l cos β sin α, zp = l cos β cos α
The velocities of the angles α̇, β̇ and positions x˙c , y˙c can also be measured through the
encoders. Thus, the velocities of the generalized coordinates can be easily computed
if we realized that each one can be defined as a function of the angles, and then the
df (β) ∂f (β)
time derivative of each can be computed, e.g., xp = f (β) → dt = ∂β β̇. Figure 4.5
illustrates the interconnection among the devices. The DS1103 specification are shown
in Table 4.1.
( xc , yc )
l cos β cos α
β α
α l
β
x
l cos
if α = 0 α
then β sin
o sβ
measures the ( xp , y p , zp ) lc
angle in the l sin β
x -direction β y
(a) Angles α and β (b) Generalized coordinates
Figure 4.4. – Subfigures showing the relationship between the angles and the
generalized coordinates
Actuators
Sensors
DS1103 Portal
PC
Controller Crane
Figure 4.5. – PC, dSPACE and Portal Crane
4.1.2 2-D Explicit Model

Before implementing the classical (explicit) IDA-PBC in an UMS, it is required first,
to model the system dynamics, using, e.g., Newton-Euler or Euler-Lagrange equations,

DS1103
Processor 933 MHz
I/O Channels 50 Bit - I/O Channels
A/D Channels 36
D/A Channels 8
Table 4.1. – Controller Board Specifications
and then, represent the system in port-Hamiltonian structure. The mathematical 2-D
model for the crane is given by [50] and represented in Figure 4.6. Where mc is the mass
of the trolley, mp is the mass of the payload, xc denotes the trolley displacement10 , l is
the length of the rope, β is the payload swing angle w.r.t the vertical, F is the resultant
force imposed on the trolley, fr is the forced caused by the girder friction and gr is the
gravity force.
x
g mc
F
fr
β l
mp
Figure 4.6. – 2-D crane
Assumption 3. The payload is always under the trolley in the sense that
π π
− <β< ∀t ≥ 0. (4.1)
2 2
Neglecting the friction, the dynamics of the system are described by
(mc + mp )ẍc + mp l β̈ cos(β) − mp l β̇ 2 sin(β) = F, (4.2a)

mp l2 β̈ + mp l cos(β)ẍc + mp l gr sin(β) = 0, (4.2b)
10
Note that it is possible to measure the displacement of the trolley in the y-axis, that is yc . If so the
angle to be considered w.r.t. the vertical is α.

mc +mp F
Let us divide (4.2) by mp l2 and define ḡ := gr /l, a := 1/l, m̄ := mp l 2
and F̄ := mp l2
.
Thus, we can represent (4.2) in compact form
" #" # " # " # " #
1 a cos β β̈ 0 ḡ sin β 0
+ + = F̄ (4.3)
a cos β m̄ u −aβ̇ 2 sin β 0 1
Is possible to extend (4.3) to the pH structure [1] by using the partial feedback lin-
earization method (see [49]) and taking
F̄ = m̄ u − a2 cos2 β u − a β̇ 2 sin β − aḡ sin β cos β,
where the new input is the acceleration u = ẍc . Thus, we are able to represent (4.3) as
" # " # " #
β̈ −ḡ sin β −a cos β
I2 = + u, (4.4)
ẍc 0 1
h i> h i>
where q = q1> q2> = β > x> c are the generalized coordinates. To represent (4.4)
in pH structure it is essential to recognize the mass matrix and the potential energy.
Due to the partial feedback linearization, the mass matrix is now M = I2 and the linear
momenta become equal to the time derivative of the generalized coordinates, that is
q̇ = M −1 p = p. The potential energy is calculated from − ∂V /∂q1 = ḡ sin q1 ⇒
V (q) = ḡ cos q1 . Therefore the pH system representation is
# 
>
I ( ∂H
" # " " #
q̇ ∂q ) 
0 0
= ∂H >
+ u (4.5)
ṗ −I 0 ( ∂p ) G(q)
| {z }
Gu
1
H(q, p) = p> M −1 p + V (q) (Hamiltonian)
2
h i> h i
with G> >
u = 0 G (q) , where G> (q) = −a cos β 1 .
4.1.3 3-D Implicit Model

Similar to the explicit case, to perform the implicit IDA-PBC approach, we obtain the
3-D implicit model of the portal crane (as presented in (3.1)) which is presented in
Figure 4.7. Where mc is the mass of the trolley, mp is the mass of the payload, xc and
yc denotes the trolley displacements, meanwhile xp , yp and zp denotes the pendulum
displacements relative to the trolley, bλ is a constraint force, l is the length of the

z z
x xc x
y y
u1
mcc u2 mc yc
l
bλ l z zp z zp

x x xp
mp y mp yp
y
(a) 3-D crane (b) Generalized coordinates in the crane
Figure 4.7. – Subfigures showing the portal crane 3-D model
rope11 , and u1 , u2 are the trolley hexternal forces imposed

i
on the x- and y-axis. The
>
generalized coordinates are r = xp yp zp xc yc , and the constraint, resulting
from a fixed rope length, can be selected as g(r) := 21 (x2p + yp2 + zp2 − l2 ).
The kinetic and the potential energy composes the Lagrangian, thus
1 1
T = mc (ẋ2c + ẏc2 ) + mp ((ẋc + ẋp )2 + (ẏc + ẏp )2 + żp2 ) (4.6)
2 2
and
V = mp gr zp . (4.7)
The resulting Lagrange equations with external forces G u and constraints g(r) = 0 are
> >
d ∂L ∂L

− = b(r) λ + G u, L = T − V, (4.8)
dt ∂ ṙ ∂r
h i>
where b(r) = ( ∂g/∂r )> , G = 02×3 I2 . According to (4.8), we have
   
∂V > ∂g >
" #" # " #
M1 M2 r̈ p ( ∂r ) ( ∂r ) 03×2
= − p + p  λ+ u, (4.9)
M2> M3 r̈ c ∂V >
( ∂r c
) ∂g >
( ∂r c
) I2
h i h i h i h i
I2
where r > > >
p = xp yp zp , r c = xc yc , u = u1 u2 , M1 = mp I3 , M2 = mp 01×2
and M3 = (mc + mp ) I2 . As explained in Section 4.1.1 we need to use partial feedback
11
It will be considered as a fixed value; otherwise, the constraint would not exist.

linearization. Setting as new input ac = r̈ c (trolley acceleration), we obtain

> 
∂g >
" #    " #
r̈ p −1 ∂V −1 G1
I5 = mp ∂r p + mp ∂r p  λ+ ac , (4.10)
r̈ c 0 0 I2
| {z }
Ḡ
where G1 = −m−1
p M2 and M̆ = I5 is the new mass matrix. Now, it is possible to
represent the system in an implicit pH representation, that is
# 
H̆ >
" # "
I5 ( ∂∂r
" # " #
ṙ 0 ) 0 0
= ∂ H̆ >
+ λ̆ + ac , (4.11a)
ρ̇ −I5 0 ( ∂ρ ) b(r) Ḡ
>
> ∂ H̆
b (r) = 0, (4.11b)
∂ρ
1 >
H̆(r, ρ) = ρ M̆(r)−1 ρ + V̆(r), (Hamiltonian)
2
h i >
where Ḡ > = G>
1 I2 , V̆ = gr zp is the new potential energy (taken from 1
mp
∂V
∂r p ≡
>
∂ V̆ 1
∂r p ) and λ̆ = mp λ is the new implicit variable.
4.2 Explicit IDA-PBC Applied to the Portal Crane

We are now ready to present the steps to find a well-defined state feedback law from (4.5)
of the 2-D crane model. As a reference, we follow the procedure of [1]. However, in
that paper, the author has made a flaw while implementing the IDA-PBC method.
4.2.1 State feedback law from [1]

The desired closed-loop pH structure has the same form as in (2.40), that is
#  ∂H ? 
)>
" # "
q̇ 0 J1 (
=  ∂q?  , (4.12)
ṗ −J1> J2 − R2 ( ∂H
∂p )
>
1 > ? −1
H ? (q, p) = p M (q) p + V ? (q), (Desired Energy Function)
2
h i
where we pick a full rank left annihilator of G, that is G⊥ = 1 −a cos(q1 ) it is clear
that the desired mass matrix has the following form
" #
? m1 (q) m2 (q)
M (q) = . (4.13)
m2 (q) m3 (q)

Then, the potential energy matching condition (2.44b) results in
∂V ? (q) ∂V ? (q)
(−a m2 cos(q1 ) − m1 ) + (−a m3 cos(q1 ) − m2 ) + ḡ sin(q1 ) = 0.
∂q1 ∂q2
(4.14)
To reduce the complexity of (4.14), it is possible to choose m2 = −m3 a cos(q1 );

thus, (4.14) is a simple integral in q1 that can be solved, e.g in Maple, and results
in
−ḡ arctanh( a m
√3 cos(q1 ) )
m1 m3
?
V (q) = √ + Υ(q2 ), (4.15)
a m1 m3
where Υ(q2 ) is an arbitrary differentiable function s.t q ∗ = arg min V ? (q) is an isolated
minimum. By taking Υ(q2 ) = 1/2 Kp (q2 − q2? )2 , the necessary and sufficient conditions
for V ? to posess a local strict minimum in q ? = (0, q2? ) are
− a2 m3ḡ−m1
" #
∂V ? ∂ 2 V ? 0

= 0, = 0. (4.16)
∂q 2 q∗

∂q q∗ 0 Kp
The matrix in the second condition is clearly positive definite picking m1 = m3 a2 + ε,

ε > 0 and Kp > 0. Now the shaped energy function is
−ḡ arctanh( √a m3 cos(q1 )

)
m3 (a2 m3 +ε) 1
V ? (q) = p + Kp (q2 − q2? )2 , (4.17)
a m3 (a2 m3 + ε) 2
and " #
? a2 m3 + ε −a m3 cos q1
M (q) = . (4.18)
−a m3 cos q1 m3
Next step is to solve

h
thei matching condition related to the kinetic energy, i.e., (2.44a).
0 j2
By taking J2 = −j2 0 it results in
a m3 sin(q1 ) (p1 1 + (p2 a cos(q1 )) (a m3 p1 cos(q1 )) + p2 (a2 m3 + ε)

j2 = (4.19)
p2 (a2 m3 cos2 (q1) − a2 m3 − ε)
However, taking (4.19) and trying to implement the control law uida , defined by (2.42),
will lead to the mistake of [1]. First, the matching condition (2.44a) is not satisfied
whenever the momenta p2 = 0, and second, uida has p2 in its denominator; thus, the
closed-loop is not well-defined for all (r, ρ) ∈ {(r, ρ) | p2 = 0}.

4.2.2 A well-defined state feedback law using IDA-PBC

To avoid the aforementioned problems, we find different M ? and V ? . The starting
point is to take a look at the skew symmetric property of J2
>
∂H ? ∂H ?

J2 = 0. (4.20)
∂p ∂p
Now define
" #
∂H ? h i 0 −s h i
h := = q̇ > J1−> = h1 h2 , J2 := h> 1 b̄ ,
∂p s 0
with s an arbitrary function of q. Replacing them in (2.44a) results in

 !> 
i ∂H ? >
" #
⊥ ∂ p> M ? (q)−1 p 0 −s h
G −J1> +2 h >
1 b̄  = 0. (4.21)
∂q s 0 ∂p
Using the identity

!> 2
∂ p> M ? (q)−1 p ∂(M ? ) ?−1
ek p> M ?−1
X
=− M p,
∂q k=1
∂qk
in (4.21), where ek represents the unit column vector of size 2, i.e., the k-th column of
the identity matrix I2 , gives
2
" # " #" #
h i ∂(M ? ) h 0 −s h1
X 1
G⊥ J1 ek h1 h2 + 2 G⊥ (h1 + b̄ h2 ) = 0. (4.22)
k=1
∂qk h2 s 0 h2
Now, if we pick b̄ = 0,
" #
? ? m1 (q1 ) m2 (q1 )
M (q) = M (q1 ) = ,
m2 (q1 ) m3
and define
" #
m1 (q1 ) + a m2 (q1 ) cos(q1) m2 (q1 ) + a m3 cos(q1)
K := G⊥ J1 = | {z } | {z } . (4.23)
K1 K2

Equation (4.22) becomes

" # " #" #
h i ∂(M ? ) h 0 −s h1
X 1
0=K e k h1 h2 + 2 G⊥ h1,
∂qk h2 s 0 h2
" #

∂m1 (q1 ) ∂m2 (q1 )
h i −s h
2
= K1 h21 + 2 h1 h2 + 2h1 1 a cos(q1) ,
∂q1 ∂q1 s h1
∂m1 (q1 ) ∂m2 (q1 )
= h21 K1 + 2 h1 h2 K1 + h21 2 a s cos(q1) − 2 h1 h2 s,
∂q1 ∂q1
∂m1 (q1 ) ∂m2 (q1 )

= h21 K1 + 2 a s cos(q1) + 2 h1 h2 K1 −s . (4.24)
∂q1 ∂q1
Then, the only possible solution of (4.24) for arbitrary h1 and h2 is that in the last
equality both terms in parenthesis are equal to zero. We select, as before, m2 (q1 ) =
−a m3 cos(q1 ) to simplify the potential energy PDE and proceed to calculate s from
the term that multiplies 2 h1 h2 , obtaining
s = a m3 sin(q1) K1 .
Then, from the term multiplying h21 , it follows that
∂m1 (q1 ) 1 2
K1 = −2 a s cos(q1) ⇒ m1 (q1 ) = a m3 cos(2 q1 ) + c1 ,
∂q1 2
where c1 is an arbitrary constant. Thus, the new shaped mass matrix is

" # " #
1
? m1 (q1 ) m2 (q1 ) 2 a2 m3 cos(2 q1 ) + c1 −a m3 cos(q1 )
M (q) = = . (4.25)
m2 (q1 ) m3 −a m3 cos(q1 ) m3
As mentioned previously, replacing M ? (q) (with m2 (q1 ) = −a m3 cos(q1 )) in (2.44b),

results in a simplification of the matching condition, that is
1 2 ∂V ? (q)
(a2 m3 cos2 (q1 ) − a m3 cos(2 q1 ) − c1 ) + +ḡ sin(q1 ) = 0,
2 ∂q1
which is again an integral and can be computed with Maple to obtain
2 ḡ cos(q1 )
V ? (q) = + Υ(q2 ) (4.26)
2 c1 − a2 m3

Eventually, the necessary and sufficient conditions for q ? = (0, q2? ) to be a strict local
?
minimum of V ? (q) with Υ(q2 ) := 1/2 Kp (q2 − q2? )2 , are ∂V
∂q ∗ = 0 and q
− a2 m2ḡ
" #
∂ 2 V ? 3 −2 c1
0
= 0. (4.27)
∂q 2 q∗

0 Kp
Inequality (4.27) is verified if Kp > 0 and c1 > a2 m3 /2. Thus, it is possible to select
c1 = a2 m3 . Finally, taking R2 = G Kv G> where Kv is a positive scalar, and J20 = 0,
i.e., J2 ≡ J21 , we see that the (explicit) IDA-PBC controller (2.42) is reduced to
2 m3 sin(q1 ) (ḡ + a p1 p2 ) cos(q1 ) + p21 − a m23 Kp (q2 − q2? ) + Kv p2

uida = . (4.28)
a m3
It is clear that the control law (4.28) is well defined for any p2 . Moreover, it is inde-
pendent of the weights of the trolley and the pendulum mc and mp .
4.2.3 Results
The parameters for the following simulations are shown in Table 4.2 and Table 4.3.
Figure 4.8 shows the block diagram of the system, where the input is the target po-
sition. As the real system has a velocity loop, the uida is integrated to obtain the
desired velocity. The next block is the crane with partial feedback linearization, whose
components are shown in Figure 4.9. Besides the state feedback, it needs the constants
l, gr , the tuning parameter m3 , and the tuning values Kp and Kv . The measurement
of p1 and p2 can be obtained thanks to the encoder and the controller (since p = q̇),
see Section 4.1.1.
symbol description value unit

mp payload mass 4.975 kg
l rope length 1.00 m
gr gravitational constant 9.81 m/s2
Table 4.2. – Crane constant parameters
The implementation is made in the y-axis, where the system was at rest, i.e. initial
velocities are zero. The measured angled in this direction was α, see Figure 4.4a.
Figure 4.10 show the portal crane response to a set point yc? that changes between
0, 0.2, −0.2 and 0.4. Figure 4.11 shows the swing angle behaviour and Figure 4.12
illustrates the control law. It is easy to note that simulation and result measurement
are quite similar. The response shows an asymptotically stable behaviour. However,
due to uncertain parameters like the vibration of the frame, friction, and encoder lack

symbol description value unit

m3 constant in M? 0.42 kg
c1 constant in V? a2 m3 kg/m2
Kv constant in uida 1.80
Kp constant in uida 15.00
Table 4.3. – (Explicit) IDA-PBC tuning paramaters
q uida
IDA-PBC
q CRANE + PFL
controller
q̇
q̇
Figure 4.8. – (Explicit) IDA-PBC block diagram
uida v q
PID velocity
−
.q˙
Plant
ẋc , ẏc
Figure 4.9. – Block crane plus partial feedback linearization (CRANE + PFL)
of resolution, we see some noise in the control law.
4.3 Implicit IDA-PBC Applied to the Portal Crane

We follow the algorithm steps in Figure 3.1 to obtain the implicit control law. Besides,
considering the additional requirements it is possible to get a position feedback law.
To close this section, we present the simulation and implementation results.

0.6
Trolley Pos. yc
0.4
0.2
yc [m]
0
yc? [m]
−0.2 yc measured
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t [s]
Figure 4.10. – Simulated and measured trolley displacement in y-axis

Pendulum Swing Angle
0.1
−0.1 α [rad]
α measured
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t [s]
Figure 4.11. – Simulated and measured swing angle α in the y-axis.

Note that is assumed that β ≡ 0
4
uIDA measured
2 uIDA [m/s2 ]
uIDA -y
−2
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t [s]
Figure 4.12. – Simulated (Explicit) IDA-PBC control law uIDA applied in the
y-direction and the measured control law
4.3.1 State feedback law

We proceed to employ the algebraic IDA-PBC algorithm. We have already selected the
coordinates and function g. Thus, the first step is to select r ? = [ 0 0 −l x?c yc? ] ∈ Xa ,
> h i
calculate b(r) = ∂g
∂r = [ xp yp zp 0 0 ]> , b? >
= [ 0 0 −l 0 0 ] , S = Ḡ b , select the full

rank left annihilators

" # " #
−zp 0 xp −zp 0 I2 02×1 02
S⊥ = , (b∗ )⊥ =
0 −zp yp zp 02 02×1 I2
and define Dx := D l2 . Matrix Zc can be reduced to b? selecting C = 04×1 . From (3.20a)

and (3.20b), we have
" # " #
xp (gr − Dx l µ? ) 0
= , (4.29)
yp (gr − Dx l µ? ) 0
" #
((a11 + a31 ) l µ? − gr ) xp + (a12 + a32 ) l µ? yp
= 0. (4.30)
(a21 + a41 ) l µ? xp + ((a22 + a42 ) l µ? − gr ) yp
where aij is an element of A located in the ith row and jth column. Solving for arbitrary
gr
xp and yp with symmetric A, leads to Dx = l µ? , a14 = a23 , a14 = −a12 , a31 = Dx − a1
and a42 = Dx − a3 . After renaming a2 = a12 , a11 = a1 , a22 = a3 , a33 = a4 , a34 = a43 =
a5 and a44 = a6 , matrix A has the following form
 
a1 a2 Dx − a1 −a2
 
 a
2 a3 −a2 Dx − a3 
A= . (4.31)
 
Dx − a1 −a2 a4 a5 
 
−a2 Dx − a3 a5 a6
We might also consider the stronger condition12 b⊥ J > b = 0, which results in a2 = 0

and a1 = a3 = Dx . Replacing them in (4.31), yields
 
Dx 0 0 0
 
 0 Dx 0 0
A= .
 
 0
 0 a4 a5 

0 0 a5 a6
With this A, (3.21) becomes

 
Dx 0 0 0 0
 
 0
 Dx 0 0 0

Md =  0 0 Dx 0 0 .
 
 
 0
 0 0 a4 a5 

0 0 0 a5 a6
12
We take this condition because it will be later used in Section 4.3.2. However, this condition reduces
the tunning capabilities.

>
The next step is select S̄. Since Ḡ is constant, selecting S S̄ = Ḡ → S̄ = [ I2 02×1 ]
meets Equation (3.20c). Now it is possible to calculate the matrix Z and choose Z ⊥ =
[ 10 01 10 01 ]. Then, W1 = 05×5 is a valid selection for (3.23a). At this point, we are able to
calculate ψ, that is
 x a6 (xc −x∗c ) a5 (yc −yc∗ ) 
− Dpx + a6 a4 −a25
− a6 a4 −a25
ψ= .
 
y a5 (xc −x∗c ) a4 (yc −yc∗ )
− Dpx + a6 a4 −a25
+ a6 a4 −a25
Using (3.20d) and (3.23b), the problem is reduced to select A 0, µ? > 0, Kψ 0 and
Ku + Ku> 0. Finally, all elements of the control law (3.6) are on hand.
4.3.2 An output feedback law using the implicit IDA-PBC approach

As b⊥ J > b = 0 and W1 = 0 hold, Proposition 5 can be used selecting A 0, Kψ 0,
Λξ + Λ>
ξ 0 and symmetric K̄u 0.
4.3.3 Results
The parameters for the implicit IDA-PBC state feedback simulations are shown in Ta-
ble 4.2 and Table 4.4. Figure 4.13 shows the block diagram of the system, where the
input is the target position r ? . The block CRANE plus PFL is similar as the one shown
in Figure 4.9 and thus avoided. Figure 4.14 shows the simulated system when x?c and
symbol description value

a4 constant in A 82.5
a5 constant in A 0
a6 constant in A 62.5
µ? constant in Dx 9.81/36
Ku constant in uI 80 diag(1.1, 1, 0)
Kψ constant in uI 29 diag(0.9, 1)
Table 4.4. – Parameters for implicit state feedback IDA-PBC
yc? are 0.8m. In the response, we can see a settling time of 12 seconds in the x-axis and
11 seconds in the y-axis; in both axis the maximum overshoot are minimal.However, it
doesn’t mean that is possible to find more suitable values to enhance the perfomance.
Figure 4.15 and Figure 4.16 show the 3D crane response and controller behaviour,
respectively, comparing it with the measured data. Responses achieve asymptotic sta-
bility. Besides, regarding some noise in the real control action, it is clear how it follows
a similar path to the simulated one. The noise can be caused due to the vibration of

r (implicit) uI
r IDA-PBC CRANE + PFL
ṙ controller
ṙ
r
Figure 4.13. – Implicit algebraic IDA-PBC block diagram
the frame, and the lack of accuracy in the encoders. The positions of the pendulum
are derived from the angles and the length of the rope. In this case, the angles veloci-
ties α̇ and β̇ are measured thanks to the encoders and the generalized coordinates are
computed as explained in Section 4.1.1.
The parameters for the implicit IDA-PBC output-feedback simulations are shown
in Table 4.2 and Table 4.5. Figure 4.17 displays the block diagram of the system, where
the input is the target position r ? and the feedback is only r. Figure 4.18 shows the
inner elements of the implicit output-feedback controller where it requires to calculate
ξ˙ using (3.25b).
Figure 4.19 show the response of the crane using the output-feedback controller (3.25)
symbol description value

a4 constant in A 29
a5 constant in A 0
a6 constant in A 29
µ? constant in Dx 9.81/12
Λξ constant in uN 0.10I2
K̄u constant in uN 175I2
Kψ constant in uN 0.0045I2
Table 4.5. – Parameters for implicit output-feedback IDA-PBC
and compares it with the measured data. It is easy to note that this approach guar-
antees asymptotic stability without overshooting and settling time of 9 and 10 seconds
in x- and y-axis. It is also possible to see a constant error in the measurement of
the trolley position xc , due to lack of accuracy in the encoder. The behaviour of the
output-feedback control law is shown in Figure 4.21. We can see a reduction of the
noise since we don’t measure the velocities.
We see great flexibility of using the implicit IDA-PBC approach. Compared to the
explicit one, the advantage is to avoid the PDE, and moreover without difficulties to

Pend. Pos. xp [m]

0.01000 xp
0.00000
−0.01000
−0.02000
0.02000
Pend. Pos. yp [m]
yp
0.00000
−0.02000
−0.99940
Pend. Pos. zp [m]
zp
−0.99960
−0.99980
−1.00000
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t [s]
Trolley Pos. yc [m] Trolley Pos. xc [m]
0.8 xc [m]
0.6 x?c [m]
0.4
0.2
0.8 yc [m]
0.6 yc? [m]
0.4
0.2
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t [s]
Figure 4.14. – Simulation with state-feedback law uI for desire xc and yc
achieve the energy shaping. The (explicit) IDA-PBC was implemented successfully to
the Portal Crane in 2-D and simulations show the results of the (explicit) controller

Pend. Pos. xp [m] 0.01000

xp meas.
0.00500 xp sim.
0.00000
−0.00500
−0.01000
Pend. Pos. yp [m]
yp meas.
0.01000 yp sim.
0.00000
−0.01000
Pend. Pos. zp [m]
zp meas.
−0.99990
zp sim.
−0.99995
−1.00000
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t [s]
Trolley Pos. xc [m]
0.2 x?c
0.1 xc sim.
xc meas.
0
−0.1
−0.2
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t
Trolley Pos. yc [m]
0.2 yc?
0.1 yc sim.
yc meas.
0
−0.1
−0.2
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t
Figure 4.15. – Implementation implicit IDA-PBC and response of r

1 uI -x meas.
uI -x sim.
uI -x [m/s2 ]
0
−1
1 uI -y meas.
uI -y sim.
uI -y [m/s2 ]
−1
0 5 10 15 20 25 30
Figure 4.16. – Implicit control law uI
r (implicit) uN r
IDA-PBC CRANE + PFL
r controller
Figure 4.17. – Implicit Output-feedback IDA-PBC applied to the Portal Crane

Block Diagram
r r
.
. Output-
feedback
uN
Calculation ξ̇ ξ Controller
of ξ̇
Figure 4.18. – Implicit Output-feedback IDA-PBC controller diagram
when the crane moves in the x-axis. However, we witnessed that the solution for a
well-defined controller is not an easy task and in this case, lies in a mathematical move
which simplifies the PDEs. Let us remark, that it is possible to apply the same (explicit)
control law (4.28) in each axis x and y, to obtain complete control of the workspace.

Pend. Pos. xp [m] 0.0400

xp meas.
0.0200 xp sim.
0.0000
−0.0200
−0.0400
0.0400
Pend. Pos. yp [m]
yp meas.
yp sim.
0.0200
0.0000
−0.0200
Pend. Pos. zp [m]
−0.9992 zp meas.
zp sim.
−0.9994
−0.9996
−0.9998
−1.0000
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t [s]
Trolley Pos. yc [m] Trolley Pos. xc [m]
0.2 x?c
0.1 xc sim.
xc meas.
0
−0.1
−0.2
0.2 yc?
0.1 yc sim.
yc meas.
0
−0.1
−0.2
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t [s]
Figure 4.19. – Output-feedback measurement
However, its performance can be affected since it is not modelled in 3-D, i.e. wrong
measurement of the angle β see Figure 4.4b. On the other hand, the implicit IDA-PBC

uN -x meas.
uN x-axis [m/s2 ] 1
uN -x sim.
0
−1
−2
uN -y meas.
uN y-axis [m/s2 ]
1
uN -y sim.
0
−1
−2
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time t [s]
Figure 4.20. – Implicit control law output-feedback uN
approach has the advantage to have more tuning parameters, and the responses show
less overshoot than the explicit method, a trade-off between doing it smoothly and
doing it faster, with the selected values13 .
Furthermore, it was possible to obtain an output-feedback law. Using this latter, we
could appreciate a controller with less noise in the measurements, as show in a brief
comparison between the control laws in Fig. (4.21). We can also appreciate that the
measured pendulum positions xp , yp and zp have peaks of ≈ 0.01m, ≈ 0.02m and
≈ 0.9999m respectively for a state-feedback law, meanwhile that for an output-feedback
law we can appreciate peaks of ≈ 0.04m, ≈ 0.04m and ≈ −0.9992m. Thus, we can
appreciate more oscillations in x− and y−axis and an improvement in the z-axis by
using the selected parameter in the output-feedback law. A brief comparison between
the control laws is found in Fig. (4.21).Moreover, The computational algorithm has an
improvement since we are not computing the velocities. To illustrate this latter, we
show the performance measurement in Table 4.6 where the identification is developed
in a computer with i5-4210H processor, 16 GB of RAM and MATLAB/Simulink in
Windows 10.
13
Let us remark that these values were the best the author of this thesis have found, however we do
not claim that those are the best ones

uI -x meas. uI -x sim. uI -y meas. uI -y sim.

1
uI [m/s2 ]
−1
uN -x meas. uN -x sim. uN -y meas. uN -y sim.
1
uN [m/s2 ]
−1
−2
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time t [s] Time t [s]
Figure 4.21. – Comparison between uN and uI
Implicit IDA-PBC Sampling Simulation Computation

Solver
Technique time [s] time [s] time [s]
State-feedback Runge-kutta ode4 0.001 36 14.2243
State-feedback Runge-kutta ode4 0.01 36 3.1152
State-feedback Domand-Prince ode8 0.001 36 35.4607
State-feedback Domand-Prince ode8 0.01 36 4.4686
Output-feedback Runge-kutta ode4 0.001 36 13.0429
Output-feedback Runge-kutta ode4 0.01 36 2.9170
Output-feedback Domand-Prince ode8 0.001 36 32.4135
Output-feedback Domand-Prince ode8 0.01 36 5.6275
Table 4.6. – Performance Measurement

Chapter 5
Conclusions and Future Work
In this work, we put into practice a novel implicit IDA-PBC method developed by Cieza
and Reger, where the total energy shaping is extended to underactuated mechanical
systems modeled in implicit pH representation with the primary objective to avoid
the persistent demand of solving PDEs in the classical (explicit) IDA-PBC. The first
part of the thesis was devoted to showing that a mechanical system can be modeled
in an explicit- or implicit-pH structure. Nevertheless, we witnessed that applying the
IDA-PBC method might result in the presence of complex PDEs. Later, in Chapter
03, the propositions show an algebraic solution to achieve asymptotic stability (with
IDA-PBC) avoiding PDEs in a class of systems modeled implicitly.
This thesis implements the total energy shaping explicit and implicit IDA-PBC tech-
niques on a real portal crane system. However, in the first case, the model is restricted
to 2-D (only one axis), and the solution of the matching conditions depends on a
shrewd mathematical move. Simulations and experimentation show that the system
in closed-loop is asymptotically stable, i.e., the standard IDA-PBC technique works.
Despite the control law being designed only in the y−axis, we found that parallel ap-
plication in the x−axis still achieves asymptotically stable results on the whole 3-D
portal crane. In the second case, using the implicit IDA-PBC reduces the effort to
design a suitable controller in the specific case of the portal crane; because, it fulfills
the proposition conditions under simple algebra. Moreover, there is high flexibility in
the tuning parameters, and the application of an output-feedback law was possible.
In the simulations, the implicit IDA-PBC yields an asymptotically stable response for
both cases, state- and output-feedback laws.
Synthesizing both approaches (implicit and explicit IDA-PBC) under partial feed-
back linearization results in controllers that are independent of the trolley and payload

5. Conclusions and Future Work
mass, which are typically unknown or variable (in the payload case). Measured data
in the implementation reflects a good behaviour of the system even though it was per-
turbed for parameters like vibration of the frame, friction, and lack of accuracy in the
sensors. Lastly, this method validates the implicit IDA-PBC approach experimentally.
Unfortunately, it was not possible to make an objective comparison between 2-D and
3-D IDA-PBC or even other controllers because a method that can contrast both with
an appropriate reference parameters has not yet been defined and it goes beyond the
scope of this thesis. Regardless of that, it shows a clear advantage in comparison with
linear controllers from the perspective that is not limited to a specific operating point.
However, the complexity of modeling the dynamics of a mechanical system in the im-
plicit structure can be considered a trade-off.
Finally, it would be interesting if: (1) May be feaseable to develop a technique

which can serve as a tool for contrasting the implicit IDA-PBC with other controllers,
perhaps through local linearization. (2) The implicit LMI approach presented in [43]
can be implemented and compared with the results presented in this work. (3) Some
uncertainties can be addressed, maybe through the extension of this work to sliding
mode control or adaptive backstepping. (4) This method could be experimentally
validated on non-holonomic systems, e.g. to a homogeneous ball on a rotating table.

Appendices

Appendix A
Euler-Lagrange Equations
This Appendix is the demonstration of the Euler-Lagrange Equations. The demonstra-

tion is extracted from [23–25]
The well known Euler-Lagrange equations are a derivation from d’Alembert’s Princi-
ple. Equation (2.1) can be written such as the path followed by the system,
d2
mi ri − Fi = 0. (A.1)
dt2
The virtual displacement δri is a possible next infinitesimal part of the path, consistent
with the constraints on the system,
" #
X d2
mi 2 ri − Fi δri = 0. (A.2)
i
dt
Equation (A.2) is well know as d’Alembert’s Principle, which is normally expressed as

a scalar equation involving what is termed virtual work Fi δri , i.e. the work that would
be done in the mass mi in the virtual displacement δri .
Then, using the virtual displacement Equation (2.5), Equation (A.2), becomes
X ∂ri ∂ri

mi r̈ i − Fi δqk = 0. (A.3)
i,k
∂qk ∂qk
Equation (A.3) is of the form

X
αk δqk = 0. (A.4)
k
Since the generalized coordinates q are independent of one another, the δqk are arbi-
trary. Therefore Equation (A.4) is valid if and only if each αk is independently zero for

A. Euler-Lagrange Equations
each component k. Then,
n nq nq
∂ri ∂ri
X X X
mi r̈i = Fi . (A.5)
i=1 k=1
∂qk k=1
∂qk
Because the Cartesian coordinates are functions of the generalized coordinates and the
time, the time derivative of the coordinate ri is
nq
dri X ∂ri ∂ri
ṙi = = q̇k + (A.6)
dt k=1
∂qk ∂t
If we take the partial derivative of Equation (A.6) with respect to q̇k we will find what
is often called cancellation of the dots because it appears as though we have simply
cancelled the dots (time derivatives) in ∂ ṙi /∂ q̇k to obtain ∂ri /∂qk . Now, if we analyze
the left part of Equation (A.5) we can notice that
∂ri ∂ ṙi d ∂ ṙi d ∂ri

r̈i = r̈i = ṙi − ṙi . (A.7)
∂qk ∂ q̇k dt ∂ q̇k dt ∂qk
Since the partial derivative of ri with respect to generalized coordinates qk depends on

(q, t), as does ri , the time derivative of ∂ri /∂qk has the same form as Equation (A.6),
that is nq
d ∂ri ∂ 2 ri ∂ 2 ri
X
= q̇j + (A.8)
dt ∂qk j=1
∂qj ∂qk ∂t ∂qk
where we can notice that Equation (A.6) appears in Equation (A.8)

 
nq
d ∂ri ∂ X ∂ri ∂ri 

= q̇j + .
dt ∂qk ∂qk j=1 ∂qj ∂t 
| {z }
=ṙi
Therefore,
d ∂ri ∂ ṙi

= (A.9)
dt ∂qk ∂qk
Using Equation (A.9) in the right hand side of Equation (A.7)
∂ri d ∂ ṙi ∂ ṙi

r̈i = ṙi − ṙi (A.10)
∂qk dt ∂ q̇k ∂qk
with
∂ ṙi 1 ∂ 2 ∂ ṙi 1 ∂ 2
ṙi = ṙ and ṙi = ṙ (A.11)
∂ q̇k 2 ∂ q̇k i ∂qk 2 ∂qk i

we can arrange Equation (A.10) in
∂ri d ∂ ∂ 1 2

r̈i = − ṙ . (A.12)
∂qk dt ∂ q̇k ∂qk 2 i
Then using Equation (A.12), the left hand side of Equation (A.5) becomes
n nq
∂ri
X X
mi r̈i
i=1 k=1
∂qk
nq
( n )
X d ∂ ∂ X 1 2

= − mi ṙi (A.13)
k=1
dt ∂ q̇k ∂qk i=1 2
and from Equation (A.13) we recognize the kinetic energy, which we shall designate as
X1
T = mi ṙ2i
i
2
X1
= mi v 2i . (A.14)
i
2
Equation (A.5) becomes

nq nq
d ∂ ∂ ∂ri
X X
− T = Fi . (A.15)
k=1
dt ∂ q̇k ∂qk k=1
∂qk
We now recall that the remaining forces are conservative and are those arising from
external fields. These forces are assumed to be equal to the negative gradient of a scalar
called the potential Energy V , which is a function only of spatial coordinates. That is
F i = − ∂V /∂ri . Therefore, using the chain rule, becomes
∂V ∂ri ∂V
= (A.16)
∂ri ∂qk ∂qk
since the potential energy V depends only on the coordinates and not on the velocities.
We also can notice that now
V {r1 , r2 , . . . , rn } → V {q1 , q2 , . . . , qnq }. (A.17)
With Equation (A.16), Equation (A.15) becomes
∂ d ∂

− . (T − V ) = 0 (A.18)
∂qk dt ∂ q̇k

Equation (A.18) are the Euler-Lagrange Equations. Meanwhile the combination of the
Potential energy V and the Kinetic energy T is called the Lagrangian
L = T − V. (A.19)
It is important to notice that the Lagrangian is a scalar function of the generalized

coordinates q, the time derivatives of the generalized coordinates q̇, and possibly the
time t. To obtain the Lagrangian we only need the kinetic energies of the interacting
bodies and the potential energies of the external fields (if they have a potential). Using
the Lagrangian we obtain the final form of the Euler-Lagrange Equations
∂L d ∂L

− = 0. (A.20)
∂qk dt ∂ q̇k

List of Acronyms
PBC Passivity Based Control
UMS Underactuated Mechanical System
DOF Degrees of Freedom
IDA-PBC Interconnection and Damping Assignment - Passivity-Based

Control
IDA Degrees of Freedom
IDA Interconnection and Damping Assignment
PFL Partial Feedback Linearization
pH port-Hamiltonian
PDE Partial Differential Equation
IDE Integral Dissipation Inequality
DDI Differential Dissipativity Inequality
DAE Differential-Algebraic Equation

List of Figures
1.1. Different kind of cranes . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1. Rectangular Cartesian Coordinates . . . . . . . . . . . . . . . . . . . . . 9

2.2. Pendulum with generalized coordinate β . . . . . . . . . . . . . . . . . . 9
3.1. Algorithm for the algebraic implicit IDA-PBC . . . . . . . . . . . . . . . 31
4.1. Subfigures showing the portal crane system at the laboratory. . . . . . . 35

4.2. Sketch of the portal crane. . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3. Trolley and its elements . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4. Subfigures showing the relationship between the angles and the general-
ized coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.5. PC, dSPACE and Portal Crane . . . . . . . . . . . . . . . . . . . . . . . 37
4.6. 2-D crane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.7. Subfigures showing the portal crane 3-D model . . . . . . . . . . . . . . 40
4.8. (Explicit) IDA-PBC block diagram . . . . . . . . . . . . . . . . . . . . . 46
4.9. Block crane plus partial feedback linearization (CRANE + PFL) . . . . 46
4.10. Simulated and measured trolley displacement in y-axis . . . . . . . . . . 47
4.11. Simulated and measured swing angle α in the y-axis. Note that is as-
sumed that β ≡ 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.12. Simulated (Explicit) IDA-PBC control law uIDA applied in the y-direction
and the measured control law . . . . . . . . . . . . . . . . . . . . . . . . 47
4.13. Implicit algebraic IDA-PBC block diagram . . . . . . . . . . . . . . . . 50
4.14. Simulation with state-feedback law uI for desire xc and yc . . . . . . . . 51
4.15. Implementation implicit IDA-PBC and response of r . . . . . . . . . . . 52
4.16. Implicit control law uI . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.17. Implicit Output-feedback IDA-PBC applied to the Portal Crane Block
Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.18. Implicit Output-feedback IDA-PBC controller diagram . . . . . . . . . . 53

LIST OF FIGURES
4.19. Output-feedback measurement . . . . . . . . . . . . . . . . . . . . . . . 54

4.20. Implicit control law output-feedback uN . . . . . . . . . . . . . . . . . . 55
4.21. Comparison between uN and uI . . . . . . . . . . . . . . . . . . . . . . . 56

List of Tables
2.1. Types of passivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.1. Controller Board Specifications . . . . . . . . . . . . . . . . . . . . . . . 38

4.2. Crane constant parameters . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3. (Explicit) IDA-PBC tuning paramaters . . . . . . . . . . . . . . . . . . 46
4.4. Parameters for implicit state feedback IDA-PBC . . . . . . . . . . . . . 49
4.5. Parameters for implicit output-feedback IDA-PBC . . . . . . . . . . . . 50
4.6. Performance Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . 56

Bibliography
[1] L. Xue and G. Zhiyong, “Control of underactuated bridge cranes: A simplified

ida-pbc approach,” in 11th Asian Control Conference (ASCC), 2017, pp. 717–722.
[2] M. W. Spong, “Underactuated mechanical systems,” in Control problems in

robotics and automation. Springer, 1998, pp. 135–150.
[3] A. Macchelli, “Passivity-based control of implicit port-Hamiltonian systems,”

SIAM Journal on Control and Optimization, vol. 52, no. 4, pp. 2422–2448, 2014.
[4] O. Cieza Aguirre and J. Reger, “IDA-PBC for underactuated mechanical systems
in implicit port-hamiltonian representation,” European Control Conference (ECC)
2019, pp. 614–619.
[5] Siemens. Heavy lifting in the industry. [Online]. Available: https://new.siemens.

com/global/en/markets/cranes/industry-cranes.html
[6] E. M. Abdel-Rahman, A. H. Nayfeh, and Z. N. Masoud, “Dynamics and control

of cranes: A review,” Modal Analysis, vol. 9, no. 7, pp. 863–908, 2003.
[7] M. cranes, “Portain mdt 98 data sheet.” [Online]. Available: www.manitowoc.com
[8] Ale. The latest crane lifting technology for any project challenge. [Online].
Available: https://www.ale-heavylift.com/services/heavy-crane-lifting/
[9] L. Aimix Group Co. Rubber-Tyred gantry crane Terex. [Online]. Available:
https://gantrycranesmanufacturer.com/rubber-tyred-gantry-crane/
[10] B. d’Andréa Novel, F. Boustany, F. Conrad, and B. Rao, “Feedback stabilization

of a hybrid PDE-ODE system: Application to an overhead crane,” Mathematics
of Control, Signals and Systems, vol. 7, no. 1, pp. 1–22, 1994.
[11] N. B. Almutairi and M. Zribi, “Sliding mode control of a three-dimensional over-

head crane,” Journal of vibration and control, vol. 15, no. 11, pp. 1679–1730, 2009.

BIBLIOGRAPHY
[12] C. Vázquez, L. Fridman, J. Collado, and I. Castillo, “Second-order sliding mode

control of a perturbed-crane,” Journal of Dynamic Systems, Measurement, and
Control, vol. 137, no. 8, 2015.
[13] X. Weimin, Z. Xiang, L. Yuqiang, Z. Mengjie, and L. Yuyang, “Adaptive dynamic

sliding mode control for overhead cranes,” in 2015 34th Chinese Control Conference
(CCC). IEEE, 2015, pp. 3287–3292.
[14] J. L. Zarate Moya, “Tracking controller design for a nonlinear model of a gantry
crane based on dynamic extension and robustification,” mathesis, Technische Uni-
versität Ilmenau, 2015.
[15] R. Banavar, F. Kazi, R. Ortega, and N. Manjarekar, “The IDA-PBC methodology

applied to a gantry crane,” in Proceedings of the Mathematical Theory of Networks
and Systems. Kyoto, Japan, 2006, pp. 143–147.
[16] M. Ryalat and D. S. Laila, “A simplified IDA-PBC design for underactuated me-
chanical systems with applications,” European Journal of Control, vol. 27, pp.
1–16, 2016.
[17] F. Kazi, R. Banavar, R. Ortega, and N. Manjarekar, “Point-to-point control of

a gantry crane: A combined flatness and ida-pbc strategy,” in European Control
Conference (ECC), 2007, pp. 5815–5820.
[18] A. Donaire, R. Mehra, R. Ortega, S. Satpute, J. G. Romero, F. Kazi, and N. M.

Singh, “Shaping the energy of mechanical systems without solving partial differ-
ential equations,” in 2015 American Control Conference (ACC). IEEE, 2015, pp.
1351–1356.
[19] R. Mehra, S. Satpute, F. Kazi, and N. Singh, “Geometric-PBC based control

of 4-dof underactuated overhead crane system,” in International Symposium on
Mathematical Theory of Networks and Systems, 2014, pp. 1232–1237.
[20] W. R. Hamilton, The Mathematical Papers of Sir William Rowan Hamilton. CUP
Archive, 1931, vol. 2.
[21] C. Lanczos, The variational principles of mechanics. Courier Corporation, 2012.
[22] F. Castaños, D. Gromov, V. Hayward, and H. Michalska, “Implicit and explicit

representations of continuous-time port-Hamiltonian systems,” Systems & Control
Letters, vol. 62, no. 4, pp. 324–330, Apr. 2013.
[23] C. S. Helrich, Analytical Mechanics. Springer International Publishing, 2017.

BIBLIOGRAPHY
[24] L. N. Hand and J. D. Finch, Analytical mechanics. Cambridge University Press,

1998.
[25] V. Arnold, Mathematical methods of classical mechanics, second edition ed.

Springer, 1989.
[26] J. L. Meriam, L. G. Kraige, and B. J. N, Engineering mechanics: dynamics. John

Wiley & Sons, 2015, vol. 2.
[27] K. Andreas, Mathematical Physics: Classical Mechanics. Springer, 2018.
[28] R. Ortega and M. W. Spong, “Adaptive motion control of rigid robots: A tutorial,”
Automatica, vol. 25, no. 6, pp. 877–888, 1989.
[29] M. Takegaki and S. Arimoto, “A new feedback method for dynamic control of
manipulators,” Journal of Dynamic Systems, Measurement, and Control, vol. 103,
no. 2, pp. 119–125, 1981.
[30] S. Delgado, “Total energy shaping for Underactuated Mechanical Systems: Dis-
sipation and nonholonomic constraints,” PhD Thesis, Technische Universität
München, 2016.
[31] R. Ortega, J. A. L. Perez, P. J. Nicklasson, and H. J. Sira-Ramirez, Passivity-based

control of Euler-Lagrange systems: mechanical, electrical and electromechanical
applications. Springer, 2013.
[32] A. J. van der Schaft and A. Van Der Schaft, L2-gain and passivity techniques in
nonlinear control. Springer, 2000.
[33] R. Sepulchre, M. Jankovic, and P. V. Kokotovic, Constructive nonlinear control.

Springer Science, 2012.
[34] A. Astolfi, R. Ortega, and R. Sepulchre, “Stabilization and disturbance attenuation

of nonlinear systems using dissipativity theory,” European journal of control, vol. 8,
no. 5, pp. 408–431, 2002.
[35] C. I. Byrnes, A. Isidori, and J. C. Willems, “Passivity, feedback equivalence, and

the global stabilization of minimum phase nonlinear systems,” IEEE Transactions
on automatic control, vol. 36, no. 11, pp. 1228–1240, 1991.
[36] J. A. Popayán, O. B. Cieza, and J. Reger, “Adaptive ida-pbc for a class of umss:
The iwip analysis,” IFAC Symposium on Nonlinear Control Systems (NOLCOS),
2019.

BIBLIOGRAPHY
[37] A. Ilchmann and T. Reis, “Surveys in Differential-Algebraic Equations i,”

A. van der Schaft, Ed. Springer, 2013, vol. I, ch. Port-Hamiltonian Differential-
Algebraic Systems, pp. 173–224.
[38] V. Duindam, A. Macchelli, S. Stramigioli, and H. Bruyninckx, Modeling and Con-

trol of Complex Physical Systems: The Port Hamiltonian Approach. Springer,
2009.
[39] F. Castaños and D. Gromov, “Passivity-based control of implicit port-Hamiltonian

systems with holonomic constraints,” Systems & Control Letters, vol. 94, pp. 11–
18, 2016.
[40] R. Ortega, A. Van Der Schaft, B. Maschke, and G. Escobar, “Interconnection

and damping assignment passivity-based control of port-controlled Hamiltonian
systems,” Automatica, vol. 38, no. 4, pp. 585–596, 2002.
[41] R. Ortega, A. Van der Schaft, B. Maschke, and G. Escobar, “Energy-shaping of

port-controlled Hamiltonian systems by interconnection,” in Conference on Deci-
sion and Control (CDC), 1999, pp. 1646–1651.
[42] R. Ortega, M. W. Spong, F. Gómez-Estern, and G. Blankenstein, “Stabilization

of a class of underactuated mechanical systems via interconnection and damping
assignment,” IEEE transactions on automatic control, vol. 47, no. 8, pp. 1218–
1233, 2002.
[43] O. Cieza Aguirre and J. Reger, “Implicit IDA-PBC for underactuated mechanical
systems: An lmi-based approach,” 2019, accepted for the Conference on Decision
and Control - (CDC 2019).
[44] D. P. Bertsekas, Constrained optimization and Lagrange multiplier methods. Aca-

demic press, 1996.
[45] Z. Zhang, Y. Wu, and J. Huang, “Differential-flatness-based finite-time anti-swing

control of underactuated crane systems,” Nonlinear Dynamics, vol. 87, no. 3, pp.
1749–1761, 2017.
[46] R. Singhal, R. Patayane, and R. N. Banavar, “Tracking a trajectory for a gantry

crane: Comparison between ida-pbc and direct lyapunov approach,” in Interna-
tional Conference on Industrial Technology, 2006, pp. 1788–1793.
[47] X. Wu and X. He, “Partial feedback linearization control for 3-d underactuated
overhead crane systems,” ISA transactions, vol. 65, pp. 361–370, 2016.

BIBLIOGRAPHY
[48] “dSPACE company website.” [Online]. Available: https://www.dspace.com/de/

gmb/home.cfm
[49] M. W. Spong, “Partial feedback linearization of underactuated mechanical sys-

tems,” in International Conference on Intelligent Robots and Systems (IROS),
1994, pp. 314–321.
[50] N. Sun and Y. Fang, “An efficient online trajectory generating method for under-
actuated crane systems,” International Journal of Robust and Nonlinear Control,
vol. 24, no. 11, pp. 1653–1663, 2014.

Implicit IDA-PBC Design and Implementationfor A Portal Crane System

Uploaded by

Copyright:

Available Formats

Implicit IDA-PBC Design and Implementationfor A Portal Crane System

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Implicit IDA-PBC Design and Implementationfor A Portal Crane System

Uploaded by

Copyright:

Available Formats

Pontificia Universidad Católica del Perú

Para obtener el grado académico de:

Presentado por: Ing. Jesus Enrique Vidal Sandoval

Tutor Responsable (TU Ilmenau): M. Sc. Oscar B. Cieza Aguirre

Fecha y lugar: 21/10/2019, Ilmenau

Master Thesis in Mechatronics / Ingeniería Mecatrónica

born on 23. October 1991

Control Engineering Group

Department of Computer Science and Automation

Advisors: Prof. Dr.-Ing. Johann Reger

Submission Date: 21 October 2019

Interconnection and damping assignment passivity-based control (IDA-PBC) is a well-

Master Thesis Enrique Vidal iv

Interconnection and damping assignment passivity-based control (IDA-PBC) ist eine

Master Thesis Enrique Vidal v

Master Thesis Enrique Vidal vi

Master Thesis Enrique Vidal vii

2.1.2. Hamiltonian Mechanics . . . . . . . . . . . . . . . . . . . . . . . 11

3. Implicit IDA-PBC for UMS 24

4. Explicit and Implicit IDA-PBC applied to a Portal Crane 34

5. Conclusions and Future Work 57

Master Thesis Enrique Vidal viii

Master Thesis Enrique Vidal ix

Master Thesis Enrique Vidal 1

1.2 Literature Review

1.2.1 Portal Crane

• Portal cranes: Also known as a gantry or overhead crane. It is composed of a

1.2.2 Common Algorithms used in Portal Crane control

Master Thesis Enrique Vidal 2

(a) Rotary crane [7] (b) Boom crane [8]

(c) Gantry crane [9]

Figure 1.1. – Different kind of cranes

Master Thesis Enrique Vidal 3

1.2.3 PBC applied to Portal Cranes

Master Thesis Enrique Vidal 4

1.2.4 Implicit PBC

Master Thesis Enrique Vidal 5

1.3 Contribution of this thesis

1.4 Outline of this thesis

Master Thesis Enrique Vidal 6

2.1 Analytical Mechanics

2.1.1 Lagrangian Mechanics

Kinematics is the description of the motion of particles. Commonly, those particles

Master Thesis Enrique Vidal 7

2.1.1.2 Newton’s Law

2.1.1.3 Generalized Coordinates

{r1 , r2 , . . . , rn } → {q1 , q2 , . . . , qnq }. (2.2)

Master Thesis Enrique Vidal 8

Figure 2.1. – Rectangular Cartesian Coordinates

Figure 2.2. – Pendulum with generalized coordinate β

ri = ri (q1 , q2 , . . . , qnq , t) = ri (q, t); i = 1, . . . , n. (2.3)

From above equation we can write

which is usually called Pfaffians or Pfaff’s differential forms.

2.1.1.4 Virtual Displacement

Master Thesis Enrique Vidal 9

2.1.1.5 Euler-Lagrange Equations

2.1.1.6 The Functional and Hamilton’s Principle

A functional defines an operation on a class of functions {y(x)} that returns a real

Here the functional is a definite integral of a quantity F dependent on the function

and n > >