Nonlinear Dynamics (2006) 45: 149–170

DOI: 10.1007/s11071-006-2425-3 
c Springer 2006

Nonlinear Input-Shaping Controller for Quay-Side

Container Cranes

MOHAMMED F. DAQAQ∗ and ZIYAD N. MASOUD

Department of Engineering Science and Mechanics, MC 0219, Virginia Polytechnic Institute and State University, Blacksburg,
VA 24061, U.S.A.; ∗ Author for correspondence (e-mail: Daqaq@vt.edu)

(Received: 21 September 2004; accepted: 2 August 2005)

Abstract. Input-shaping is one of the most practical open-loop control strategies for gantry cranes, especially those having
predefined paths and operating at constant cable lengths. However, when applied to quay-side container cranes, its performance
is far from satisfactory. A major source of the poor performance can be linked to the significant difference between the gantry
crane model and the quay-side container crane model. Gantry cranes are traditionally modeled as a simple pendulum. However,
a quay-side container crane has a multi-cable hoisting mechanism.
In this paper, a two-dimensional four-bar-mechanism model of a container crane is developed. For the purpose of controller
design, the crane model is reduced to a double pendulum with two fixed-length links and a kinematic constraint. The method of
multiple scales is used to develop a nonlinear approximation of the oscillation frequency of the simplified model. The resulting
frequency approximation is used to determine the switching times for a bang-off-bang input-shaping controller. The performance
of the controller is numerically simulated on the full model of the container crane, and is compared to the performance of
similar controllers based on a nonlinear frequency approximation of a simple pendulum and a linear frequency approximation
of a constraint double pendulum. Results demonstrate a superior performance of the controller based on the nonlinear frequency
approximation of the constraint double pendulum.
The effect of the oscillation frequency on the controller performance is investigated by varying the model’s frequency around
the design value. Simulations revealed that the performance of the controller suffers serious degradation due to small changes
in the model frequency. To alleviate the shortcomings of the input-shaping controller, a delayed-position feedback controller is
successfully applied at the end of each transfer maneuver to eliminate residual oscillations without affecting the commands of
the input-shaping controller.

Key words: container cranes, delayed-position feedback control, input-shaping, nonlinear stability, sway control

1. Introduction

In general, cranes play a very important role in transportation and construction. As a result there is
an increasing demand on faster, bigger, and more efficient cranes to guarantee fast turn-around time,
while meeting safety requirements. This steered recent research into developing more efficient crane
controllers.
Inertial forces on the payload due to crane commanded trajectories or operators commands can cause
the payload to experience large sway oscillations. To avoid exciting these oscillations, crane operators
resolve to slowing down the operations such that the oscillations do not cause safety concerns and
possible damage of the payload. However, slowing down operations increases the cost of loading and
unloading operations.
Traditionally, a gantry crane is modeled as a simple pendulum with a rigid or flexible hoisting cable
and a lumped mass at the end of the cable. However, in the case of quay-side container cranes the
model is significantly different (Figure 1). The actual hoisting mechanism of a container crane con-
sists typically of a set of four hoisting cable arrangement. The cables are hoisted from four different
150 M. F. Daqaq and Z. N. Masoud

Figure 1. Typical quay-side container crane.

points on a trolley and are attached on the payload side to four points on a spreader bar used to lift
containers. The two most commonly used modeling approaches for gantry cranes are the lumped-mass
and distributed-mass models. In the distributed-mass model, the hoisting cable is modeled as a dis-
tributed mass, and the hook and payload are lumped as a point mass and are applied as a boundary
condition to this system. d’Andrea-Novel et al. [1, 2] and d’Andrea-Novel and Boustany [3] used
the distributed-mass model. They ignored the inertia of the payload and modeled the cable as per-
fectly flexible, inextensible body using the wave equation. Others [4–6] extended the model to include
the inertia of the payload by changing the boundary conditions at the payload end. However, the
most widely used approach for crane modeling, is the lumped-mass approach. The hoisting line is
treated as a massless rigid link, and the payload is lumped with the hook and modeled as a point
mass.
For gantry cranes, the level of control of payload oscillations varies according to the application at
hand. In some applications, oscillations are acceptable while the payload is on its way to target, while
the settling time and residual oscillations are kept very small to allow for accurate payload positioning.
In other applications, such as nuclear reactors, or where the space around the crane is populated, the
safety requirements are very strict. Thus, large oscillations are not acceptable during and at the end of
a transfer maneuver.
Input-shaping is one of the practical open-loop control strategies for gantry cranes. Controllers using
various forms of input-shaping are incorporated into gantry cranes and currently used in ports [7]. This
technique moves the crane a known distance along a set path, and depends significantly on system
parameters such as cable length and system delays.
Alsop et al. [8] were the first to propose a controller based on input-shaping. The controller ac-
celerates the trolley in steps of constant acceleration then kills the acceleration when the payload
reaches zero oscillation angle (after multiples of a full period). The trolley then coasts at constant
speed along the path for a period of time. A replicate of the acceleration procedure is used in the
deceleration stage. Alsop et al. used constant-amplitude acceleration/deceleration steps, and a linear
frequency approximation of a simple pendulum model. Their results showed very little residual os-
cillations, while transient oscillation angles were of order of 10◦ during the acceleration/deceleration
stages.
 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 151
Jones and Petterson [9] extended the work of Alsop et al. [8] using a nonlinear approximation of the
payload period to generate an analytical expression for the duration of the coasting stage as a function
of the amplitude and duration of the constant acceleration steps. This analytical expression is then used
to generate a two-step acceleration profile. Numerical simulations using various acceleration profiles
show that this technique was not able to dampen out initial disturbances of the payload and could even
amplify them.
Starr [10] used a symmetric two-step acceleration/deceleration profile to transport a suspended object
with minimal oscillations. The duration of the constant acceleration steps is assumed to be negligible
compared to the period of the payload. A linear approximation of the period of the payload is used to
generate analytically the acceleration profile. Strip [11] extended this work by employing a nonlinear
approximation of the payload frequency to generate one-step and two-step symmetric acceleration
profiles.
Kress et al. [12] showed analytically that input-shaping is equivalent to a notch filter applied to a
general input signal and centered around the natural frequency of the payload. Based on that, they
proposed a Robust Notch Filter, a second-order notch filter, applied to the acceleration input. Numerical
simulation and experimental verification of this strategy on an actual bidirectional crane, moving at an
arbitrary step acceleration and changing cable length at a slow constant speed, showed that the strategy
was able to suppress residual payload oscillation.
Parker et al. [13] developed a command shaping notch filter to reduce payload oscillation on rotary
cranes excited by the operator commands. They reported that in general, there was no guarantee that
applying such filter to the operator’s speed commands would result in excitation terms having the desired
frequency content, and that it only works for small speed and acceleration commands. Parker et al. [14]
experimentally verified their simulation results.
Input-shaping techniques are limited by the fact that they are sensitive to variations in the parameter
values about the nominal values and changes in the initial conditions and external disturbances and that
they require “highly accurate values of the system parameters” to achieve satisfactory system response
[15–17]. While a good design can minimize the controller’s sensitivity to changes in the payload mass,
it is much harder to alleviate the controller’s sensitivity to changes in the cable length. Singhose et al.
[18] developed four input-shaping controllers. Simulations of their best controller produced a reduction
of 73% in transient oscillations over the time-optimal rigid-body commands. However, they reported
that “transient deflection with shaping increases with hoist distance, but not as severely as the residual
oscillations”. Their simulations showed that “the percentage in reduction with shaping is dependent on
system parameters”. As a result, their controllers suffered significant degradation in crane maneuvers
that involved hoisting.
Alzinger et al. [19] showed that a two-step acceleration/deceleration profile results in significant
reductions in travel time over a one-step acceleration profile. Testing on an actual crane has shown
that the two-step acceleration profile can deliver both faster travel and minimal payload oscillations
at the target point. However, any deviation from the prescribed parameters, causes significant payload
oscillations.
Dadone and Vanlandingham [20] used the method of multiple scales to generate a nonlinear approx-
imation of the oscillation period of a simple pendulum model. They produced numerical simulations
and compared the response of the nonlinear controller to the linear controller. They found that the
acceleration profile based on the nonlinear frequency approximation could damp the residual oscilla-
tions 2 orders of magnitude more than the profile based on the linear approximation. The enhanced
performance of the nonlinear strategy was most pronounced for longer coasting distances and higher
accelerations.
152 M. F. Daqaq and Z. N. Masoud
In addition to feedforward controllers, researchers used linear and nonlinear feedback techniques
for anti-sway control of suspended objects [21–23]. In an application to quay-side container cranes,
Yong-Seok et al. [24] and Yong-Seok, Keum-Shik, and Seung-Ki [25] developed a state feedback
controller to control oscillations on a rail-mounted quay-side crane. They proposed a novel technique
to measure the sway angle. This technique is based on mounting an inclinometer on the top surface
of the spreader bar, then using the geometry of the crane and the angle relations to recover the actual
sway angle. Although, the sway angle measurement was based on the actual geometry of the crane, the
authors used a simple pendulum model to express the dynamics of the payload, and verified the results
experimentally using a single-cable rubber-tired gantry crane.
Researchers [26, 27] have extensively studied the possibility of using time delay to control mechanical
systems. It has been noticed that systems with time delays exhibit interesting complex responses.
Time delay has the capability of stabilizing or destabilizing dynamic systems. For this reason, they
have been used as simple switches to control the behavior of systems, either by damping out the
oscillations or creating chaotic responses that are sometimes desirable to secure communication signals
[28].
Cheng and Chen [29] were among the first to use time delay to control gantry cranes. They proposed
a control strategy which employs time delay control and feedback linearization to move a crane along
a predefined path and to eliminate residual oscillations. Their results showed that the strategy is ca-
pable of delivering the payload to its goal with minimal transient oscillations and almost no residual
oscillations.
Masoud and Nayfeh [30] introduced a two-dimensional four-bar-mechanism model of a container
crane. They approximated the model with a constrained double pendulum to find a linear frequency
approximation of the payload oscillation. The resulting frequency was then used to find the delay and
gain of a nonlinear delayed-position feedback controller. The resulting controller was applied to the
full crane model. Masoud et al. [31] verified the modeling approach and the controller performance
experimentally on a 1/10 scale model of a container crane.
Despite their practicality, the performance of open-loop input-shaping controllers designed for gantry
cranes is far from satisfactory when applied to quay-side container cranes. One of the main reasons
behind this poor performance, is the lack of a good model that resembles the dynamics and approximates
the frequency of a real container crane. In this paper, we analyze the nonlinear dynamics of the traversing
motion of the crane payload, then develop a one-step input-shaping controller based on a four-bar-
mechanism model of a quay-side container crane. We use the method of multiple scales to develop
a nonlinear frequency approximation of the payload oscillation. To eliminate residual oscillations,
the open-loop input-shaping controller is augmented with a closed-loop delayed-position feedback
controller applied at the end of each transfer maneuver.

2. Mathematical Models

In this section, a four-bar-mechanism is developed to model the actual hoisting mechanism of the quay-
side container crane. This model is further simplified to a double pendulum model with a kinematic
constraint between the angles of both links of the pendulum. The simplified model is used for the
purpose of controller design, however, numerical simulations are performed on the full model of the
crane.
To study the performance of input-shaping controllers developed using a simple pendulum model, a
nonlinear version of the traditional simple pendulum model is used.
 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 153

Figure 2. A schematic model of a simple pendulum model of a container crane.

2.1. SIMPLE PENDULUM

We start by deriving the nonlinear equation of motion of a constant length simple pendulum model
(Figure 2). The position vector to the center of mass of the payload is

r = ( f + L sin θ )i − L cos θj (1)

where L is the length of the cable, and f is the position of the trolley. The equation of motion is derived
using the Euler–Lagrangian equation
 
d ∂L ∂L
− =0 (2)
dt ∂ q̇ ∂q

where q = θ and L is the Lagrangian and is defined as

L=T −V (3)

T and V are the kinetic and potential energy of the payload, which can be expressed as

1 1
T= m ṙṙT = m( ḟ 2 + L 2 θ̇ 2 + 2 ḟ θ̇ L cos θ ) (4)
2 2
V = mgr y = −mgL cos θ (5)

where g is the gravitational acceleration. Substituting Equations (3)–(5) into Equation (2) yields the
following nonlinear equation of motion:

θ̈ + η cos θ + 2◦ sin θ = 0 (6)

where η = f¨/L is the normalized trolley acceleration, and 2◦ = g/L is the linear frequency of the
pendulum.
154 M. F. Daqaq and Z. N. Masoud

Figure 3. A schematic model of a container crane.

2.2. FOUR-BAR-MECHANISM MODEL

The two-dimensional model of a quay-side container crane is shown in Figure 3. The container is
grabbed using a spreader bar. The spreader bar is then hoisted from the trolley by means of four cables,
two of which are shown in the figure. The cables are spaced a distance d at the trolley and a distance w
at the spreader bar. The hoisting cables in the model are considered as rigid massless links with constant
lengths.
The position vector to the combined center of mass of the payload and the spreader bar is

r = xi − yj (7)

The unconstrained equations of motion of the payload mechanism can be derived using the Euler–
Lagrangian Equation (2), where q = [x, y, θ ]T . The kinetic energy and potential energy of the payload
and spreader bar are given by

1 1 1 1
T= m ṙṙT + J θ̇ 2 = m(ẋ 2 + ẏ 2 ) + J θ˙2 (8)
2 2 2 2
V = mgr y = −mgy (9)

where m is the combined mass of the payload and the spreader bar, and J is the combined moment of
inertia of the payload and the spreader bar about point Q.
Since this is a single-degree-of-freedom system, two constraints are used governing the distances AB
and DC. The constraints are

  2 2
1 1 1
x + R sin θ − w cos θ − f + d+ y − R cos θ − w sin θ − L 2 = 0 (10)
2 2 2
 2  2
1 1 1
x + R sin θ + w cos θ − f − d + y − R cos θ + w sin θ − L 2 = 0 (11)
2 2 2
 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 155
We multiply these two constraint by the Lagrange multipliers λ1 and λ2 then append them to the
Lagrangian L to form the augmented Lagrangian La

1 1
La = m(ẋ 2 + ẏ 2 ) + J θ˙2 + mgy
2 2
   2 
1 1 2 1
+ λ1 x + R sin θ − w cos θ − f + d + y − R cos θ − w sin θ − L 2
(12)
2 2 2
 2  2 
1 1 1
+ λ2 x + R sin θ + w cos θ − f − d + y − R cos θ + w sin θ − L 2
2 2 2

Substituting Equation (12) into (2), we get the following three nonlinear differential equations of motion:
  
1 1
m ẍ − 2λ1 x + R sin θ − w cos θ − f + d
2 2
 
1 1
+ 2λ2 x + R sin θ + w cos θ − f − d =0 (13)
2 2
    
1 1
m ÿ − 2λ1 y − R cos θ − w sin θ + 2λ2 y − R cos θ + w sin θ − mg = 0 (14)
2 2
  
1 1 1
J θ̈ − 2λ1 R cos θ + w sin θ x − w cos θ − f + R sin θ + d
2 2 2
  
1 1
+ R sin θ − w cos θ y − w sin θ − R cos θ
2 2
  
1 1 1
− 2λ2 R cos θ − w sin θ x + w cos θ − f + R sin θ − d
2 2 2
  
1 1
+ R sin θ + w cos θ y − R cos θ + w sin θ =0 (15)
2 2

In order to solve the equations of motion with the two constraint equations, we differentiate the con-
straints Equations (10) and (11) twice with respect to time to obtain

α11 ẍ + α12 ÿ + α13 θ̈ + 2ẋ 2 + 2 ẏ 2 + α14 θ̇ 2 + α15 θ̇ ẋ + α16 θ̇ ẏ + α17 θ̇ ḟ + α18 f¨ − 4 f˙ ẋ + 2 f˙2 = 0
(16)
α21 ẍ + α22 ÿ + α23 θ̈ + 2ẋ + 2 ẏ + α24 θ̇ + α25 θ̇ ẋ + α26 θ̇ ẏ + α27 θ̇ ḟ + α28 f¨ − 4 ḟ ẋ + 2 f˙2 = 0
2 2 2

(17)

where

αk1 = −(−1)k d + (−1)k w cos θ − 2 f + 2R sin θ + 2x

αk2 = −2R cos θ + (−1)k w sin θ + 2y
1
αk3 = −(−1)k d R cos θ − 2R f cos θ + dw sin θ
2
+ (−1)k w f sin θ + 2Rx cos θ − (−1)k wx sin θ
+ (−1)k wy cos θ + 2Ry sin θ
156 M. F. Daqaq and Z. N. Masoud
1
αk4 = (−1)k d R sin θ + 2R f sin θ + dw cos θ (18)
2
+ (−1)k w f cos θ − 2Rx sin θ − (−1)k wx cos θ
− (−1)k wy sin θ + 2Ry cos θ
αk5 = 4R cos θ − 2(−1)k w sin θ
αk6 = 4R sin θ + 2(−1)k w cos θ
αk7 = −4R cos θ + 2(−1)k w sin θ
αk8 = −2R sin θ − 2x + 2 f + (−1)k d − (−1)k w cos θ

Equations (13)–(17) are the full nonlinear equations of motion of the four-bar-mechanism of the con-
tainer crane.

2.3. CONSTRAINED DOUBLE PENDULUM

For the purpose of controller design, the four-bar-mechanism model is simplified to a double pendulum
system with two fixed-length links, and a kinematic constraint relating the angles φ and θ as shown in
Figure 4.
In Figure 3, point O is the midpoint between points A and D, and point P is the midpoint between
points B and C. The closing constraints of the loop ABPO are
1 1
l sin φ − w cos θ + d − L sin φ1 = 0 (19)
2 2
1
l cos φ − w sin θ − L cos φ1 = 0 (20)
2
Similarly, the closing constraints of the loop ODCP are written as

1 1
l sin φ + w cos θ − d − L sin φ2 = 0 (21)
2 2
1
l cos φ + w sin θ − L cos φ2 = 0 (22)
2

Figure 4. A schematic model of a constrained double pendulum model of a container crane.

 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 157
Squaring and adding Equations (19) and (20), and squaring and adding Equations (21) and (22), we can
eliminate L from the resulting equations and obtain the following relation between φ and θ:

d sin φ = w sin(θ + φ) (23)

Equation (23) represents the kinematic constraint between the angles φ and θ, which reduces the
two-degrees-of-freedom double pendulum model to a single-degree-of-freedom model [32]. Using
Equation (23), we simplify the closing constraints of the loop ABPO to obtain

1
l 2 = L 2 − (d 2 + w 2 − 2dw cos θ ) (24)
4

The position vector to the center of mass of the payload of the constrained double pendulum is

r = ( f + l sin φ − R sin θ )i − (l cos φ + R cos θ)j (25)

Substituting the position vector into Equations (8) and (9), the kinetic and potential energies of the
constrained double pendulum can be written as

1 2 2 1 1
T= ml φ̇ + (J + m R 2 )θ̇ 2 + m ḟ 2 − m Rl φ̇ θ̇ cos(φ + θ) + ml φ̇ ḟ cos φ − m R θ̇ ḟ cos θ (26)
2 2 2
V = −mg(l cos φ + R cos θ ) (27)

The equation of motion expressed in terms of the angle φ is obtained by substituting the constraint
Equations (23) and (24) into Equations (26) and (27), then applying the Euler–Lagrange equation to the
Lagrangian (3). Due to the lengthy expressions in this equation, we only show a Taylor series expansion
of the resulting equations up to cubic terms.

φ̈ + ω◦2 φ + c1 φ 2 φ̈ + c1 φ φ̇ 2 − c2 ξ φ 2 − c3 φ 3 + c4 ξ = 0 (28)

where
  
mg k32 + 2k42 + k12 k3 R
ω◦ =  
k3 J k12 + m(k3 − k1 R)2
 
m 4k42 + (k1 (1 + k1 )2 − 6k2 )k33 R
c1 =  
k32 J k12 + m(k3 − k1 R)2
 
2mk1 (3 + 2k1 )k3 k5 R − 2k32 (k4 − 3k1 k2 R 2 )
+  
k3 2 J k12 + m(k3 − k1 R)2
6J k1 k2 k3 2
+   (29)
k3 2
+ m(k3 − k1 R)2
J k12
  
m 6k4 + k3 k3 − k13 R + 6k2 R
c2 =  
2 J k12 + m(k3 − k1 R)2
158 M. F. Daqaq and Z. N. Masoud
  
gm 12k4 + k4 (k3 − 24k3 k5 + k1 k13 − 24k2 R
c3 = −  
6k3 J k12 + m(k3 − k1 R)2
mk3 (k3 − k1 R)
c4 =
J k12 + m(k3 − k1 R)2
f¨
ξ=
k3

and
d −w
k1 =
w
d(d 2 − w 2 )
k2 = 3
6w
1
k3 = L 2 − (d − w)2 (30)
4
dw
k4 =
8
dw(d 2 − 4L 2 + dw + w 2 )
k5 =
24((d − w)2 − 4L 2 )2

3. Nonlinear Analysis

A nonlinear frequency approximation of the simple pendulum and constrained double pendulum models
will be derived. These frequency approximations along with the linear frequency ω◦ of the constrained
double pendulum model will be used to generate acceleration profiles for the same input-shaping
controller.

3.1. CONSTRAINED DOUBLE PENDULUM

To find a nonlinear frequency approximation we use the method of multiple scales [33]. First we scale
Equation (28) by introducing a bookkeeping parameter  which is set to one at the end of this analysis.
ξ and θ are scaled at order , that is

φ = φ ξ = ξ (31)

Substituting the scaled parameters into Equation (28) yields

φ̈ + ω◦2 φ = −c4 ξ −  2 (c1 φ 2 φ̈ + c1 φ φ̇ 2 − c2 ξ φ 2 + c3 φ 3 ) (32)

The time dependence is expanded into multiple time scales as following

t = T0 + T1 +  2 T2 + O( 3 )
d
= D0 +  D1 +  2 D2 + O( 3 ) (33)
dt
d2
= D02 + 2 D0 D1 +  2 D12 + 2 2 D0 D2 + O( 3 )
dt2
 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 159
∂
where Dn = ∂ Tn
. We then seek a solution in the form

φ(T0 , T1 , T2 ) = φ0 (T0 , T1 , T2 ) + φ1 (T0 , T1 , T2 ) +  2 φ2 (T0 , T1 , T2 ) + O( 3 ) (34)

Substituting Equations (33) and (34) into Equation (32) and equating coefficients of like powers of 
we obtain

O(1) : D02 φ0 + ω◦2 φ0 = −c4 ξ

O() : D02 φ1 + ω◦2 φ1 = −2D0 D1 φ0
(35)
O( 2 ) : D02 φ2 + ω◦2 φ2 = −2D0 D1 φ1 − 2D0 D2 φ0
− D12 φ0 − c2 ξ φ02 − c1 φ0 (D0 φ0 )2 − c1 φ02 D02 φ0 − c3 φ03

The solution of the O(1) equation can be expressed as

c4 ξ
φ0 (T0 , T1 , T2 ) = A(T1 , T2 )eiω◦ T0 − + cc (36)
ω02

where A is complex. Substituting Equation (36) into O() of Equation (35) and eliminating the terms
that lead to secular terms, we obtain

∂A
= 0 ⇒ A = A(T2 ) (37)
∂ T1

Now, substituting Equations (36) and (37) into O( 2 ) of Equation (35) and eliminating the coefficients
of eiω◦ T0 which leads to secular terms, we obtain the solvability condition
 
∂A ξ 2 c4 c 3 c4  
−2iω◦ + 2 c1 c4 − 3 2 − 2c2 A + 2ω◦2 c1 − 3c3 A2 Ā = 0 (38)
∂ T2 ω◦ ω◦

Introducing the polar transformation

1
A= a(T2 )eiβ(T2 ) (39)
2

into Equation (38) and separating real and imaginary parts, we obtain the following two modulation
equations:

∂a
=0 (40)
∂ T2
   
∂β ξ2 3c3 c42 a2 3c3
= 2c 2 c 4 + − c1 c4
2
− − + c 1 ω◦ (41)
∂ T2 2ω◦3 ω◦2 4 2ω◦

Solving Equations (41) and (41) we get the following approximate solution of Equation (32):

c4 ξ
φ = a cos(ωt + β0 ) − (42)
ω◦2
160 M. F. Daqaq and Z. N. Masoud
5
Numerical Solution
Nonlinear Frequency
Linear Frequency

-5
Angle [deg]

-10

-15

-20
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 5. Payload sway angle for the constrained double pendulum model of a container crane. The results are obtained for
L = 17.5 m, R = 2.5 m, ξ = 0.1 s−2 , θ0 = 0, and θ˙0 = 0.

where
    
3c3 c1 ξ 2 −3c3 c42
ω = ω◦ 1 − a 2 + + + 2c c
2 4 − c c
1 4
2
(43)
8ω◦2 4 2ω◦4 2ω◦2

is the nonlinear frequency approximation of the constrained double pendulum.

We tested the resulting approximation against a numerical solution of the full nonlinear equations of
motion Equations (13)–(17) for different values of ξ . The results showed that this approximation holds
for values of ξ approximately less than 0.1 s−2 . This value sets a limit above which the nonlinear solution
starts to deteriorate. However, container cranes normally operate at lower accelerations (0.015–0.03).
Figure 5 shows that the numerical integration and the multiple scale approximate solution closely match,
while the linear frequency approximate solution quickly drifts away from the numerical solution as a
result of the inaccurate frequency approximation.

3.2. SIMPLE PENDULUM

To compare controllers based on the constrained double pendulum model with those based on the simple
pendulum model, we develop a nonlinear frequency approximation of the payload oscillations for the
traditional simple pendulum model of the container crane. Using Equation (6) the procedure described
in Section 3.1 is followed to obtain the following nonlinear frequency approximation:
 
a2 η2
 = ◦ 1 − + (44)
16 44◦

where η = f¨ /L is the normalized acceleration, and ◦ is the linear frequency approximation of the
payload oscillations of a simple pendulum. We tested the resulting approximation against a numerical
 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 161

Numerical Simulation
10 Nonlinear Frequency
Linear Frequency

-10
Angle [deg]

-20

-30

-40

-50
0 2 4 6 8 10 12 14 16 18 20

Time [sec]
Figure 6. Payload sway angle for a simple pendulum model of a container crane. The results are obtained for L = 20 m,
η = 0.2 s−2 , θ0 = 0, and θ˙0 = 0.

solution of Equation (6) for different values of η. The results showed that this approximation holds
for values of η approximately less than 0.2 s−2 . Figure 6 shows that the solution based on the multiple
scales method has an excellent agreement with the numerical solution.

4. Controllers Design

4.1. INPUT-SHAPING CONTROLLER

A bang-off-bang input-shaping controller is shown in Figure 7. The controller works by generating an

acceleration profile designed to cancel only its own oscillations. The controller will be used to perform
a transfer maneuver that produces zero residual oscillations while taking into consideration any delays
in the system.
Assuming that both the cable length and the system delays are known, the controller determines the
magnitude of the constant acceleration and the switching times to reach the target point with zero residual
oscillations. We will follow the procedure developed by Dadone et al. [12]. First the trolley accelerates
for the period of a half swing cycle ta = Ta /2, where Ta is the period of the sway oscillation in the
acceleration mode. The acceleration is then switched off for a period of time (necessary to accomplish
the load transfer). This period is called the coast mode. To bring the load to complete stop a negative
acceleration is applied taking into account the known system delays.
For known system delays τs , the length of τs will determine the minimum time that the trolley must
spend in the coast mode. By design the coast time must be an odd multiple of half the period of the
sway oscillation Tc in the coast mode. The coast time tc depends also on the maximum magnitude of
acceleration achievable by the trolley drives. Thus, we can define the coast time as

2n + 1
tc = Tc , n = 0, 1, 2, 3, . . . (45)
2
162 M. F. Daqaq and Z. N. Masoud

Figure 7. A schematic drawing showing the bang-off-bang acceleration profile.

where n is known as the number of coasting cycles. To calculate the least number of cycles the trolley
needs to spend in the coast mode, the constraint tc ≥ τs must be satisfied. Thus, we can write n as
 
1 2τs
n≥ −1 (46)
2 Tc

here n is rounded up to the nearest integer including zero.

The total distance travelled by the trolley in a full maneuver S derived from Figure 7 is

S = f¨( ta )2 + f¨( ta )( tc ) (47)

Using the normalized acceleration ξ = f¨/k3 and ta = Ta /2, and substituting Equation (45) into (47),
the normalized travel distance δ = S/k3 is

1 2 2n + 1
δ= ξT + ξ Ta Tc (48)
4 a 4

To adapt the input-shaping controller to the constrained double pendulum, the nonlinear frequency
approximation, Equation (43) is used. Assuming zero initial conditions we solve Equation (42) for a
and then substitute back into Equation (43) to get
  
2π ξ2 15c3 c42 3c1 c42 −1
Ta = 1+ 2c c
2 4 + − (49)
ω◦ 2ω◦4 4ω◦2 2

The period in the coast mode can be obtained by setting ξ to zero and substituting a = −2c4 ξ/ω◦2 in
Equation (43). a here is the amplitude of oscillation in the coast mode and is found by substituting
t = π/ω in Equation (42)
  −1
2π 4ξ 2 c42 c1 3c3
Tc = 1− − (50)
ω◦ ω◦4 4 8ω◦2
 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 163
Now we can summarize the controller design process as following:
• Using the known values for τs and Tc , determine the minimum number of coast cycles n from
Equation (46).
• Solve Equation (48) for ξ and compare it with ξmax . ξmax is the minimum of two values; the first being
the maximum acceleration that the trolley drives can provide and the second being the maximum
acceleration above which the nonlinear solution starts to deteriorate, which in this case is equal to
0.1 s−2 . If ξ is greater than ξmax increment n up by one and recompute a new ξ .
• Compute the switching times t1 , t2 , and t3 according to the following equations:

Ta 2n + 1
t1 = , t 2 = t1 + Tc , t 3 = t2 + t 1 (51)
2 2
The design of the input-shaping controller based on the simple pendulum model follows the same proce-
dure. However, in Equation (48) δ = LS is used and η and ηmax are substituted for ξ and ξmax , respectively.
Assuming zero initial conditions, Equation (44) is used to calculate the nonlinear approximation of the
periods in the bang mode as follows

 
2π η2 −1
Ta = 1+ (52)
◦ 44◦

In the coast mode, the normalized acceleration η is set to zero and a to η/ 2◦ , which is the linear
approximation of the amplitude of oscillation at the end of the bang mode obtained from Equation (6)

 
2π η2 −1
Tc = 1− (53)
◦ 164◦

4.2. DELAYED-POSITION FEEDBACK CONTROLLER

The delayed-position feedback controller creates damping in the system by adding a delayed feedback
component to the operator command to the trolley drives. This component is proportional to the swing
angle of the hoist line. The controller takes the following general form

f = f ◦ + k̂ sin φτd (54)

where k̂ is the feedback gain, τd is the delay time, f ◦ is the operator input, and φτd = φ(t − τd ). To
study the linear stability of the controller and to make a proper choice of the gain-delay combination,
we substitute Equation (54) into (28) and add a linear damping term to account for the damping in the
system. Assuming that the operator input is slowly varying and keeping only linear terms yield to

k̂  
φ̈ + ω◦2 φ + 2μφ̇ + φ̈ τd + 2μφ̇τd = 0 (55)
k3

where μ is a linear damping coefficient. Now we seek an exponentially damped solution of the equation
in the form

φ = ceσ t cos(ωt + ψ) (56)

164 M. F. Daqaq and Z. N. Masoud
where c, σ , ω, and ψ are real constants. Substituting Equation (56) into (55), and setting the coefficients
of cos(ωt + ψ) and sin(ωt + ψ) equal to zero independently, we get the following two equations

K (σ 2 + 2μσ − ω2 ) sin(ωτd ) − 2K ω(μ + σ ) cos(ωτd ) − 2ω(σ + μ)eσ τd = 0 (57)

K (σ + 2μσ − ω ) cos(ωτd ) + 2K ω(μ + σ ) sin(ωτd ) + (σ + 2μσ − ω +
2 2 2 2
ω◦2 )eσ τd =0 (58)

where k̂/k3 = K . The stability of the system depends on the value of the parameter σ . The system is
asymptotically stable for σ < 0 and unstable for σ > 0. To determine the boundaries of linear stability
and instability we set σ equals to zero in Equations (57) and (58) to obtain

K ω2 sin(ωτd ) + 2K μ cos(ωτd ) + 2μω = 0 (59)

2K ωμ sin(ωτd ) − ω (1 + K cos(ωτd )) +
2
ω◦2 =0 (60)

Equations (59) and (60) can be normalized by dividing them by ω◦2 , and set the time delay τd proportional
to the linear period of oscillation T = 2π/ω◦ . This yields to

K λ2 sin(2πλγ ) + 2K νλ cos(2πλγ ) + 2νλ = 0 (61)

2K νλ sin(2πλγ ) − λ (1 + K cos(2πλγ )) + 1 = 0
2
(62)

where λ = ω/ω◦ , γ = τd /T , and ν = μ/ω◦ . Manipulating Equations (61) and (62), K and γ as
functions of λ are

4ν 2 + λ2 (λ2 + 4ν 2 − 1)2
K (λ) = − (63)
λ(λ2 + 4ν 2 )
 
1 2ν
γ (λ) = arctan + jπ j = 0, 1, 2, . . . (64)
2πλ λ(λ2 + 4ν 2 − 1)

The stability boundaries are determined by varying λ in Equations (63) and (64) and solving for K and
γ . Figure 8 shows the stable and the unstable regions as predicted by the linear theory. Any gain-delay
combination that lies inside the shaded areas of Figure 8 leads to an asymptotically stable response. It is
worth mentioning that there are infinite number of stability pockets, which decrease in size as the time
delay of the controller τd increases.
To determine the magnitude σ of the damping factor resulting from each gain-delay combination, τd
and K are varied in Equation (57) and (58) and calculate σ . Figure 9 shows a contour plot of the value
of σ in the first pocket of stability.

5. Simulations

To verify the accuracy of our nonlinear approximation of the simple pendulum frequency, an input-
shaping profile is generated using the linear and the nonlinear frequency approximations of the simple
pendulum model. The following values are used for the model parameters in the simulation: R = 2.5 m,
d = 2.825 m, w = 1.4125 m, and m = 50000 kg, and for the IS controller n = 1 and S = 50 are used.
Figure 10 shows the acceleration profiles and the payload sway simulated on the full nonlinear model of
the simple pendulum. The figure shows that the residual oscillations associated with the linear simple
 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 165

Figure 8. A stability plot of the delayed-position feedback controller for a relative damping ν = 0.0033. The shaded areas
represent the pockets of stability.

Figure 9. A contour plot of the damping as a function of the gain K and the delay τd , where τd is given in terms of the natural
period of the uncontrolled system. The darker the areas are the higher the damping is.

pendulum (LSP) profile are about 8 orders of magnitude larger than those associated with the nonlinear
simple pendulum model (NSP). Although the residual oscillations associated with both models are small
and can be neglected for the given system parameters, thorough investigation shows that the residual
oscillations associated with a LSP increase for shorter cables and longer tracks.
Similarly, to verify the accuracy of the nonlinear frequency approximation of the constrained double
pendulum, acceleration profiles based on the constrained double pendulum model are generated using
both a linear and a nonlinear frequency approximations and then applied to the same constrained double
pendulum model. Figure 11 shows the shaped acceleration profiles for the linear and the nonlinear cases.
It can be easily seen that the residual oscillations associated with the linear double pendulum model
(LDP) profile are orders of magnitude larger than those associated with the nonlinear double pendulum
(NDP) profile.
166 M. F. Daqaq and Z. N. Masoud
1.5 4
LSP LSP
NSP NSP
3
1

0.5
Acceleration[m/s ]

1
2

Sway[m]
0 0

-1
-0.5

-2

-1
-3

-1.5 -4
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40

Time [sec] Time [sec]

LSP
0.06 NSP

0.04

0.02
Sway [m]

-0.02

-0.04

-0.06

24 26 28 30 32 34 36 38
Time [sec]

Figure 10. Sway response of a simple pendulum model of a container crane to shaped operator commands. Results are obtained
for L = 20 m.

The shaped acceleration profiles based on the NSP, the LDP, and the NDP frequency approxima-
tions are then applied to the full four-bar-mechanism model as shown in Figure 12. Results show
that the NSP acceleration profile, will not reduce the residual oscillations. On the contrary, this
shaped profile will amplify residual oscillations to magnitudes that are even larger than the transient
oscillations.
The LDP and NDP models did reduce the residual oscillations significantly. However, the figure
shows that the NDP acceleration profile has a much better performance than the LDP profile.
Figure 13 demonstrates the sensitivity of the input-shaping controller to changes in system parameters.
The NDP acceleration profile shown in Figure 12(a) is applied to the full model while varying the length
of the hoisting cables. Simulations show that a change of 1.0 m in the cables length causes significant
degradation in the controller performance.
To overcome the problem of the open-loop input-shaping controller sensitivity, and to eliminate
the residual oscillations, we apply a delayed-position feedback controller at the end of the transfer
 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 167

1.5 4
LDP LDP
NDP NDP
3
1
2

0.5
1
Acceleration[m/s2]

Sway[m]
0 0

-1
-0.5
-2

-1
-3

-4
-1.5 0 5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40
Time[sec] Time [sec]

Figure 11. Sway response of a constrained double pendulum model of a container crane to shaped operator commands. Results
are obtained for L = 17.5 m.

1.5 4
NSP NSP
LDP LDP
NDP
3 NDP
1

2
0.5
Acceleration[m/s2]

1
Sway[m]

0 0

-1
-0.5

-2
-1
-3

-1.5
0 5 10 15 20 25 30 35 40 -4
0 5 10 15 20 25 30 35 40
Time [sec] Time [sec]

Figure 12. Sway response of the full model of the container crane to shaped operator commands. Results are obtained for
L = 17.5 m.

maneuver. The choice of delayed-position feedback controller is based on its ability to handle systems
with inherent time delays. The controller can incorporate these delays in its parametric delay.
A number of factors were considered behind the choice of applying a feedback controller only at
the end of the transfer maneuver. A significant factor is reducing the power required to perform a
transfer maneuver with minimal residual oscillations. This stems from the fact that for large sway
angles feedback control systems may require input accelerations that are beyond the normal oper-
ating accelerations, which may overload the trolley motors during the acceleration and deceleration
stages. When applied at the end of the transfer maneuver performed using an input-shaping con-
trol system, feedback controllers require lower accelerations. This is due to the fact that a well-
designed input-shaping controller is expected to produce significant reduction in residual payload
oscillations.
168 M. F. Daqaq and Z. N. Masoud
4
16.5 m
17.5 m
3 18.5 m

1
Sway[m]

-1

-2

-3

-4
0 5 10 15 20 25 30 35 40
Time [sec]
Figure 13. Sway response of the full model of the container crane to the shaped operator commands shown in Figure 12(a).

4
1.5 IS
IS
IS + FB
IS + FB
3
1
2

0.5
1
Acceleration[m/s ]
2

Sway[m]

0
0

-1
-0.5
-2

-1
-3

-4
-1.5 0 5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40
Time [sec] Time [sec]

Figure 14. Sway response of the full model of the container crane with input-shaping (IS) and combination (IS+FB (feedback))
controllers. Results are obtained for L = 16.5 m.

To demonstrate the performance of the combination controller, an NDP acceleration profile based
on a 17.5 m cable is applied to a 16.5 m cable model. Figure 14 shows that an input-shaping controller
based on the NDP frequency approximation results in a residual sway of approximately 1.25 m, while
a combination of input-shaping and delayed-position feedback controller suppresses the residual sway
to a magnitude less than 0.05 m within 4.5 s of the end of the transfer maneuver. The delayed-position
feedback controller parameters used are K = 0.4 and τd = 0.28 T.
 Nonlinear Input-Shaping Controller for Quay-Side Container Cranes 169
6. Concluding Remarks

A bang-off-bang input-shaping controller based on a simple pendulum model falls short of satisfying
the goal of eliminating residual oscillations on quay-side container cranes. The residual oscillations are
many orders of magnitude larger than those when the controller is applied to a simple pendulum model.
To achieve satisfactory performance, input-shaping controllers should be based on accurate math-
ematical models. Our results show that a constrained double pendulum model based on a four-bar-
mechanism model of a container crane is a considerably accurate model for quay-side container cranes.
For predefined system parameters, the input-shaping acceleration profiles based on a nonlinear fre-
quency approximation of this model are capable of reducing the residual oscillations to magnitudes less
than 0.01 m.
A delayed-position feedback controller is used to minimize the sensitivity of the input-shaping
controller to variations in system parameters and to improve its robustness. The addition of such feedback
controller will account for system delays, external disturbances, and will compensate for uncertainties
in the system. The delayed-position feedback controller is shown to be capable of reducing the residual
oscillations of the payload to less than 0.05 m within 5 s of the end of the transfer maneuver.
Feedback controllers produce continuously changing acceleration profiles, which affects the crane
operator performance and comfort. Therefore, applying an efficient feedback controller, such as the
delayed-position feedback, at the end of the transfer maneuver for a very short period of time eliminates
excessive trolley motion, thus maintaining comfortable working conditions for the crane operator.

Acknowledgement

The authors are indebted to Dr. Ali H. Nayfeh for his guidance and continuous support.

References

1. d’Andrea-Novel, B., Boustany, F., and Conrad, F., ‘Control of an overhead crane: Stabilization of flexibilities’, in Boundary
Control and Boundary Variation: Proceedings of the IFIP WG 7.2 Conference, Sofia Antipolis, France, 1990, pp. 1–26.
2. d’Andrea-Novel, B., Boustany, F., Conrad, F., and Rao, B. P., ‘Feedback stabilization of a hybrid PDE-ODE system: Appli-
cation to an overhead crane’, Mathematics of Controls, Signals, and Systems 7, 1994, 1–22.
3. d’Andrea-Novel, B. and Boustany, F., ‘Control of an overhead crane: Feedback stabilization of a hybrid PDE-ODE system’,
in Proceedings of the 1st European Control Conference: ECC 91, Grenoble, France, 1991, pp. 2244–2249.
4. Joshi, S. and Rahn, C. D., ‘Position control of a flexible cable of a gantry crane: Theory and experiment’, in Proceedings of
the American Control Conference, Seattle, WA, 1995, pp. 301–305.
5. Martindale, S. C., Dawson, D. M., Zhu, J., and Rahn, C. D., ‘Approximate nonlinear control for a two degree of freedom
overhead crane: Theory and experimentation’, in Proceedings of the American Control Conference, Seattle, WA, 1995, pp.
301–305.
6. Rahn, C. D., Zhang, F., Joshi, S., and Dawson, D. M., ‘Asymptotically stabilizing angle feedback for a flexible cable gantry
crane’, Journal of Dynamic Systems, Measurement, and Control 15, 1999, 563–566.
7. Hubbel, J. T., Koch, B., and McCormic, D., ‘Modern crane control enhancements’, in Ports 92, Seattle, WA, 1992, pp.
757–767.
8. Alsop, C. F., Forster, G. A., and Holmes, F. R., ‘Ore unloader automation – a feasibility study’, in Proceedings of IFAC
Workshop on Systems Engineering for Control Systems, Tokyo, Japan, 1965, pp. 295–305.
9. Jones, J. F. and Petterson, B. J., ‘Oscillation damped movement of suspended objects’, in Proceedings of the IEEE International
Conference on Robotics and Automation, Philadelphia, PA, Vol. 2, 1988, pp. 956–962.
10. Starr, G. P., ‘Swing-free transport of suspended objects with a path-controlled robot manipulator’, Journal of Dynamic
Systems, Measurement, and Control 107, 1985, 97–100.
170 M. F. Daqaq and Z. N. Masoud
11. Strip, D. R., ‘Swing-free transport of suspended objects: A general treatment’, IEEE Transactions on Robotics and Automation
5(2), 1989, 234–236.
12. Kress, R. L., Jansen, J. F., and Noakes, M. W., ‘Experimental implementation of a robust damped-oscillation control algorithm
on a full-sized, two-degree-of-freedom, AC induction motor-driven crane’, in Proceedings of the 5th International Symposium
on Robotics and Manufacturing: Research, Education, and Applications, ISRAM’94, Maui, HI, 1994, pp. 585–592.
13. Parker, G. G., Groom, K., Hurtado, J., Robinett, R. D., and Leban, F., ‘Command shaping boom crane control system with
nonlinear inputs’, in Proceedings of the IEEE International Conference on Control Applications, Kohala Coast, HI, Vol. 2,
1999, pp. 1774–1778.
14. Parker, G. G., Groom, K., Hurtado, J. E., Feddema, J., Robinett, R. D., and Leban, F., ‘Experimental verification of a command
shaping boom crane control system’, in Proceedings of the American Control Conference, San Diego, CA, Vol. 1, 1999,
pp. 86–90.
15. Zinober, A. S. I., ‘The self-adaptive control of an overhead crane operations’, in Proceedings of the 5th IFAC Symposium on
Identification and System Parameter Estimation, Darmstadt, East Germany, 1979, pp. 1161–1167.
16. Virkkunen, J. and Marttinen, A., ‘Computer control of a loading bridge’, in Proceedings of the IEE International Conference:
Control ’88, Oxford, UK, 1988, pp. 484–488.
17. Yoon, J. S., Park, B. S., Lee, J. S., and Park, H. S., ‘Various control schemes for implementation of the anti-swing crane’, in
Proceedings of the ANS 6th Topical Meeting on Robotics and Remote Systems, Monterey, CA, 1995, pp. 472–479.
18. Singhose, W. E., Porter, L. J., and Seering, W. P., ‘Input shaped control of a planar crane with hoisting’, in Proceedings of
the American Control Conference, Albuquerque, NM, 1997, pp. 97–100.
19. Alzinger, E. and Brozovic, V., ‘Automation and control system for grab cranes’, Brown Boveri Review 7, 1983, 351–356.
20. Dadone, P. and Vanlandingham, H. F., ‘Load transfer control for a gantry crane with arbitrary delay constraints’, Journal of
Vibration and Control 7, 2001, 135–158.
21. Hazlerig, A. D. G., ‘Automatic control of crane operations’, in Proceedings of the IFAC 5th World Congress, Paris, France,
Vol. 1, 1972, Paper no. 11.3.
22. d’Andrea-Novel, B. and Lenvine, J., ‘Modeling and nonlinear control of an overhead crane’, in Proceedings of the Interna-
tional Symposium MTNS-89, Vol. 2, pp. 523–529.
23. Ebeid, A. M., Moustafa, K. A. F., and Emara-Shabaik, H. E., ‘Electromechanical modeling of overhead cranes’, International
Journal of Systems Science 23(12), 1992, 2155–2169.
24. Yong-Seok, A., Hyungbo, S., Hidemasa, Y., Naoki, F., Hideshi, K., and Seung-Ki, S., ‘A new vision-sensoreless anti-sway
control system for container cranes’, in IEEE Industry Applications Conference, Salt Lake City, UT, Vol. 1, 2003, pp. 262–269.
25. Yong-Seok, A., Keum-Shik, H., and Seung-Ki, S., ‘Anti-sway control of container cranes: Inclinometer, observer, and state
feedback’, International Journal of Control, Automation, and Systems 2(4), 2004, 435–449.
26. Pyragas, K., ‘Continuous control of chaos by self-controlling feedback’, Physics Letters A 170, 1992, 421–428.
27. Gregory, D. V. and Rajarshi, R., ‘Chotic communication using time-delayed optical systems’, International Journal of
Bifuractions and Chaos 9, 1999, 2129–2156.
28. Pacora, L. M. and Carroll, T. L., ‘Synchronization in chaotic systems’, Physics Review Letter 64, 1990, 821–824.
29. Cheng, C. C. and Chen, C. Y., ‘Controller design of an overhead crane system with uncertainties’, Control Engineering
Practise 4, 1996, 645–653.
30. Masoud, Z. and Nayfeh, A. H., ‘Sway reduction on container cranes using delayed feedback controller’, in Proceedings of the
43rd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, Denver, CO, April 22–25,
2002.
31. Masoud, Z. N., Nayfeh, N. A., and Nayfeh, A. H., ‘Sway reduction on container cranes using delayed feedback controller:
Simulations and experiments’, in Proceedings of the 19th Biennial ASME Conference on Mechanical Vibrations and Noise,
Chicago, IL, September 2–6, 2003.
32. Nayfeh, N. A., ‘Adaption of the Delayed Position Feedback to the Reduction of Sway of Container Cranes’, Master’s Thesis,
Virginia Polytechnic Institute and State University, Blacksburg, VA, 2002.
33. Nayfeh, A. H., Introduction to Perturbation Techniques, Wiley, New York, 1981.