ISA Transactions xxx (xxxx) xxx

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Research article

Precision-positioning adaptive controller for swing elimination in

three-dimensional overhead cranes with distributed mass beams
∗
Ling Yang, Huimin Ouyang
College of Electrical Engineering and Control Science, Nanjing Tech University, No.30, Puzhu Road(s), Nanjing, 211816, China

article info a b s t r a c t

Article history: As a type of dispensable transport tool, overhead cranes are extensively applied in large industrial sites
Received 27 June 2021 to transport containers and goods due to their extraordinary convenience and efficiency. However,
Received in revised form 24 August 2021 various transport requirements and working environments may cause some adverse effects for
Accepted 24 August 2021
overhead crane control. In particular, in some transport tasks, distributed-mass beams (DMB) are
Available online xxxx
required to be transported smoothly to a specific position. Hence, most existing controllers designed
Keywords: for crane systems with point-mass payload (PMP) cannot meet the control requirements. For example,
Underactuated system controllers designed for crane systems with PMP do not consider the moment of inertia of DMB, thus,
Overhead crane the residual payload swing caused by the moment of inertia cannot be completely suppressed, which
Distributed mass beam may cause safety risks. Additionally, due to various working environments, inaccurate estimation of
Antiswing control
frictions can result in errors in positioning. To solve the above issues, this paper conducts the dynamic
Adaptive control
analysis of three-dimensional (3-D) overhead crane systems with DMB and designs an improved
adaptive controller which can provide favorable control for state variables. In particular, the proposed
update laws can compensate for the uncertainties of crane systems. Based on a tracking trajectory, the
trolley and guide rail can both reach the preset position, and the swing angles are also well eliminated
throughout the transport process. The stability of the control system is proven theoretically, and the
performance of the proposed controller is verified by comparative experiments.
© 2021 ISA. Published by Elsevier Ltd. All rights reserved.

1. Introduction between the hook and payload is non-negligible; (2) the hook
is heavy and has a great influence on the payload swing. Due
Most of the traditional control methods for cranes rely on the to the above situations, the dynamic characteristic of overhead
operator’s experience, that is, through the human manipulation cranes will present a double-pendulum swing effect. Addition-
of the trolley and guide rail movement to suppress the payload ally, as same as single-pendulum crane systems, the payload of
oscillation as much as possible. Obviously, this method has a double-pendulum crane systems are also regarded as PMP. A
certain lack of flexibility. In recent decades, scholars put forward great number of excellent controllers for overhead cranes have
a series of control methods for overhead cranes with a view to been designed by scholars in a view of this characteristic. Specif-
different control problems. ically, to suppress the swing of payload, Ouyang et al. designed an
Considering the complexity of overhead crane systems, the S-shaped trajectory with the effect of swing-elimination, through
dynamic characteristic of overhead cranes has been generally the optimization algorithm, the time-optimal parameters of the
considered as with single-pendulum or double-pendulum. For trajectory were calculated, and a composite controller based on
crane systems with single-pendulum effect, the mass and volume PD control and disturbance observer was designed to track the
of the hook are completely ignored, and the payload is consid- trajectory, hardware experiments demonstrated that the control
ered as a PMP. A great amount of papers based on controller method can accomplish the control objectives effectively [11].
designed for cranes with single-pendulum effect were developed Based on the neural network theory, Chen et al. proposed an
in past decades, including input-shaping control [1,2], trajectory adaptive control method for overhead crane systems, it does
planning control [3,4], energy-based control [5,6], sliding mode not require precise values of system parameters, and its effec-
control (SMC) [7,8], model predictive control [9] and intelligent tiveness was verified by numerical simulations [12]. Based on
control [10]. However, in some practical applications, the fol- the input–output inverse method, Giacomelli et al. designed a
lowing situations are particularly obvious: (1) the rope length parametric trajectory that can suppress the residual swing of the
payload [13]. Zhang et al. proposed an energy-coupling-based
∗ Corresponding author. control method for double-pendulum crane systems, comparative
E-mail address: [email protected] (H. Ouyang). experiments showed the robustness of the control method in

https://doi.org/10.1016/j.isatra.2021.08.035
0019-0578/© 2021 ISA. Published by Elsevier Ltd. All rights reserved.

Please cite this article as: L. Yang and H. Ouyang, Precision-positioning adaptive controller for swing elimination in three-dimensional overhead cranes with distributed
mass beams. ISA Transactions (2021), https://doi.org/10.1016/j.isatra.2021.08.035.
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

the presence of external disturbances [14]. Tuan et al. applied 1. In this paper, the dynamic model of a 3-D overhead crane
SMC to overhead cranes and designed a first-order sliding mode with DMB is constructed and its characteristics are ana-
surface and a second-order sliding mode surface. The simulation lyzed.
demonstrated that under the two controllers, the state vari- 2. In view of the complex dynamics of such cranes, an im-
ables can effectively reach the expected values [15]. Sun et al. proved adaptive control method is proposed, and it does
designed an energy-based control method for crane systems, not need any linearizing operations on the system model,
which can complete the positioning and swing-elimination tasks and the feedback of swing angles. The new adaptive up-
with a minimum energy consumption [16]. Qian et al. devel- date laws can well compensate for the error in positioning
oped a single-input-rule modules (SIRMs) based fuzzy controller caused by the varying system parameters (e.g., different
for overhead cranes, and the effectiveness of the controller was payload mass and various rope lengths), and it can also
verified by numerical simulations [17]. Jaafar et al. proposed a ensure that the trolley and guide rail can arrive at the
model reference command shaping (MRCS) controller to obtain preset position more accurately.
an adequate suppression of payload oscillations, comparative ex- 3. Comparative experiments show that the controller has a
periments showed that the MRCS controller has great robustness superior performance in swing-elimination, and the sta-
to varying system parameters [18]. Adeli et al. proposed a hybrid bility of the control system is proven by the Lyapunov
controller in view of position regulation and swing-elimination technique and Barbalat’s lemma.
control, and the simulation results verified that the controller can
adequately suppress the positioning errors and angle values [19]. The remainder of this paper is divided as follows: Section 2
Considering parametric uncertainties and external disturbance, establishes the dynamic model of the overhead crane with DMB
Zhang et al. proposed an energy-based control method for crane and describes the control objectives in detail. Subsequently, con-
systems. By analyzing the energy of the system, the tracking error troller design and stability analysis are both provided in Section 3.
of the trolley can be limited within boundary conditions under In Section 4, the structure of the hardware platform is first in-
the controller. The robustness of the controller was demonstrated troduced, followed by experimental results demonstrating the
by numerical simulations [20]. superior performance of the proposed control method. In the end,
Above mentioned controllers are all designed for crane sys- some summaries are presented in Section 5.
tems with PMP. However, due to different transport require-
ments, industrial sites often require distributed mass beams to be 2. Problem statement
transported, which will result in more complex system dynamics.
For example, the coupling relationships between payload swing
2.1. Overhead crane dynamics
angles and system state variables will become more complicated
because of the non-negligible moment of inertia generated by
DMB. In recent years, scholars have designed a number of con- Combining the structure diagram shown in Fig. 1 and the
trollers for crane systems with DMB in response to different Lagrange–Euler modeling method, the following mathematical
control requirements. Wu et al. proposed a trajectory planning dynamic relationship of 3-D overhead cranes with DMB is given:
controller combining with the quasi-periodicity theory, by model M(q)q̈ + C (q, q̇)q̇ + G(q) = U − Fs − D, (1)
decoupling, analytical expressions of the angles are obtained and ]T
θ1 θ2 θ3 θ4 ,
[
combined with the geometry of the system structure, a trajectory q= x y
⎡ ⎤
that can control quasi-periodic motion is achieved. Experiments m11 m12 m13 m14 m15 m16
proved that the trajectory can effectively suppress oscillations in ⎢m21 m22 m23 m24 m25 m26 ⎥
multimodal systems [21]. Huang et al. proposed a novel input ⎢m m32 m33 m34 m35 m36 ⎥
⎢ ⎥
M = ⎢ 31 ⎥,
shaping controller. Experiments showed that the controller can ⎢m41 m42 m43 m44 m45 m46 ⎥
greatly suppress the swing angles of DMB [22]. We can also ⎣m51 m52 m53 m54 m55 m56 ⎦
learn from the controllers designed for tower cranes with DMB. m61 m62 m63 m64 m65 m66
Peng et al. proposed an open-loop controller for tower crane
⎡ ⎤
c11 c12 c13 c14 c15 c16
systems with DMB, after adequate analysis of the dynamic char- ⎢c21 c22 c23 c24 c25 c26 ⎥
acteristics of the crane systems, a new input shaping method ⎢c
⎢
c32 c33 c34 c35 c36 ⎥
⎥
was designed, and the experimental results proved that the con- C = ⎢ 31 ⎥,
⎢c41 c42 c43 c44 c45 c46 ⎥
troller can perform the tasks of positioning and swing-elimination ⎣c51 c52 c53 c54 c55 c56 ⎦
efficiently [23]. c61 c62 c63 c64 c65 c66
Through thorough analysis of the existing literature, it is not ]T
,
[
difficult to summarize the problems of them as the following G = g1 g2 g3 g4 g5 g6
points:
]T
0 ,
[
U = Fx Fy 0 0 0
]T
1. Most of the existing controllers are designed for crane 0 ,
[
Fs = Frx Fry 0 0 0
systems with PMP, which lack specific analysis of DMB, and ]T
d3 θ̇1 d4 θ̇2 d5 θ̇3 d6 θ̇4 ,
[
a large proportion of them need approximate operations on D = d1 ẋ d2 ẏ
the system model. Therefore, they have poor applicability
where M(q) and C (q, q̇) are the inertial matrix and Coriolis matrix,
to crane systems with DMB.
respectively. G(q) is the gravity vector. U, Fs , and D are input
2. Although scholars have also designed some well-performed
vector, friction vector and damping vector, respectively. x(t) and
controllers for crane systems with DMB, most of them re-
y(t) denote the displacement of the trolley in the X direction and
quire accurate system dynamics, which will result in poor
Y direction. θi (t)(i = 1, . . . ,4) denote the swing angles of the hook
robustness of the controllers in the case of varying system
and DMB with the vertical direction. For the sake of simplicity in
parameters.
expression, we use Si and Ci to represent sinθi and cosθi (i = 1,
Specifically, the contributions of this paper can be summarized . . . ,4), respectively. The detailed terms in the above matrices and
as follows: vectors are shown in Appendix A.
2
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

Fig. 1. Structure diagram of the 3-D overhead crane with DMB.

Frx and Fry are the frictions of the trolley and guide rail in Assumption 1. Taking into account the structure of the actual
their running directions. Inspired by [24], we use the following overhead cranes, the swing angles of the trolley and guide rail
expressions to describe them: always meet the following limitation [4,5,24,25]:
(
ẋ
) π π
Frx = f11 tanh − f12 |ẋ| ẋ, (2) θ1 , θ2 , θ3 , θ4 ∈ (− , ). (6)
εx 2 2
( )
ẏ 2.2. Control objective
Fry = f21 tanh − f22 |ẏ| ẏ, (3)
εy
The control objectives of overhead cranes with DMB can be
where f11 , f12 , f21 , f22 , εx and εy are the friction parameters.
summarized as follows: (1) the trolley and the guide rail can
arrive at the preset position accurately; (2) the oscillation of the
Remark 1. The friction is divided into two parts: static fric- hook and DMB can be fully suppressed during the whole trans-
tion and sliding friction, which are both related to the speed
( ) port process. In general, it can be expressed in a mathematical
ẋ
of the object’s movement. In this paper, we use f11 tanh εx equation as follows:
( )
ẏ
and f21 tanh εy
to describe the Coulombic frictional coefficients. lim q(t) = qd (t), (7)
t →∞
while f12 |ẋ|ẋ and f22 |ẏ|ẏ are used to denote the viscous dissipation. [ ]T
Therefore, the friction model presented in Eqs. (2) and (3) can where qd = xd yd 0 0 0 0 , xd and yd are target posi-
provide a good estimate of the friction in practical applications tions of the trolley and guide rail, respectively.
after reasonable adjustment of the relevant parameters [6,9,11,
14,24]. 3. Main results
By succinct derivations, the following properties of the matri-
ces in Eq. (1) always exist: An adaptive controller is proposed in this section, and the
detailed implementation steps of the controller are illustrated
• M(q) is a positive definite and symmetric matrix, for any in Fig. 2. Further, it is theoretically proven that the proposed
η ∈ R6 , there always exist positive constants λm and λM such controller can stabilize the whole control system.
that [25]
3.1. Controller design
λm ∥η∥2 ≤ ηT M(q)η ≤ λM ∥η∥2 . (4)

• The equation of Ṁ(q) = C (q, q̇) + C T (q, q̇) will always hold The total energy of the overhead cranes with DMB can be
for M(q) and C (q, q̇). constructed using the following expression:
1
In addition, although the actual mass of the hook and DMB are E(t) = q̇T M(q)q̇ + l1 g(1 − C1 C2 )(m1 + m2 ) + m2 glh (1 − C3 C4 ). (8)
not directly available, and the vertical distance between the hook 2
and payload is variable, we can usually achieve some prior infor- According to the properties of overhead crane systems, the deriva-
mation about the lower and upper bounds of them as follows: tive of Eq. (8) with respect to time can be obtained:
1 1 1
Ė(t) = q̈T M q̇ + q̇T Ṁ q̇ +
q̇T M q̈
m1 ≤ m1 ≤ m1 , m2 ≤ m2 ≤ m2 , lh ≤ lh ≤ lh . (5) 2 2 2
Considering the particularity of overhead cranes with DMB, + l1 g(S1 C2 θ̇1 + S2 C1 θ̇2 )(m1 + m2 ) (9)
the following reasonable assumption is adopted: + m2 glh (S3 C4 θ̇3 + S4 C3 θ̇4 )
3
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

+ m2 glh (S3 C4 θ̇3 + S4 C3 θ̇4 )

= q̇Te (U − Fs − D − M q̈r − C q̇r )
= ėx (Fx − γx T ωx ) + ėy (Fy − γy T ωy ) + χ1 + χ2 + χ3 + χ4 ,
where
]T
γx = ẍr tanh( εẋx ) |ẋ| ẋ ẋ ,
[
]T
ωx = m1 + m2 + mx f11 f12 d1 ,
[
[ ]T
ẏ
γy = ÿr tanh( εy ) |ẏ| ẏ ẏ ,
]T
ωy = m1 + m2 + my f21 f22 d2 ,
[

χ1 = −d3 θ̇12 − θ̇1 (ẍr l1 m2 C1 C2 + ẍr l1 m1 C1 C2 ),

χ2 = −d4 θ̇22 + θ̇2 (ÿr l1 m1 C2 + ÿr l1 m2 C2
+ ẍr l1 m1 S1 S2 + ẍr l1 m2 S1 S2 ),
χ3 = −d5 θ̇32 − θ̇3 ẍr lh m2 C3 C4 ,
χ4 = −d6 θ̇42 + θ̇4 (ÿr lh m2 C4 + ẍr lh m2 S3 S4 ).
Inspired by the passive nature of the system, and the charac-
teristics of overhead cranes with DMB, we design the following
controller:

Fx = γx T ω̂x − k1p (x − xd ) − k1d (ẋ − ẋd ), (12)

Fy = γy ω̂y − k2p (y − yd ) − k2d (ẏ − ẏd ),
T
(13)

where k1d , k2d , k1p , k2p are positive control gains to be determined.
ω̂x , ω̂y are online estimated values for ωx and ωy , the detailed
expressions of them are as follows:
]T
ω̂x = m1 +ˆ ,
[
m2 + m3 fˆ11 fˆ12 d̂1 (14)
]T
ω̂y = m1 +ˆ .
[
m2 + m3 fˆ21 fˆ22 d̂2
Considering the complex dynamic characteristics of overhead
cranes with DMB, and to improve its accuracy in positioning, we
Fig. 2. Controller implementation steps. design the following adaptive update laws:
ex
ω̂˙ x = −Πx −1 γx (αėx + δ ( )), (15)
1 2
= q̇T M q̈ + q̇T (C + C T )q̇ + l1 g(S1 C2 θ̇1 + S2 C1 θ̇2 )(m1 + m2 ) ω̂˙ y = −Πy −1 γy (αėy + δ (
ey
)),
2
2
+ m2 glh (S3 C4 θ̇3 + S4 C3 θ̇4 )
where Πx and Πy are positive definite diagonal matrices, the
= q̇T (U − Fs − D − G) + l1 g(S1 C2 θ̇1 + S2 C1 θ̇2 )(m1 + m2 ) detailed terms of them will be designed. α is a positive param-
+ m2 glh (S3 C4 θ̇3 + S4 C3 θ̇4 ) eter introduced to compensate for the tracking errors. δ (∗) is a
saturation function designed as follows:
= ẋ(Fx − Frx − d1 ẋ) + ẏ(Fy − Fry − d2 ẏ) − d3 θ̇12
π
∗>
⎧
1
− d4 θ̇22 − d5 θ̇32 − d6 θ̇42 . ⎪
⎪
⎪ 2
⎨
π
It can be seen that the first derivative of Eq. (8) is the total δ (∗) = sin(∗) |∗| ≤ 2
. (16)
power of the acting forces. It is not difficult to summarize that ⎪
⎪
π
⎪
the overhead crane systems with DMB is passive. −1 ∗<
⎩
2
By analyzing Eq. (8), and combining the control objectives in
Eq. (7), the following energy function is selected as a Lyapunov Through brief mathematical calculations, it is not difficult to
function candidate: draw the property of Eq. (16) that |δ (∗)| ≤ |∗|, which will be
1 applied in the following section.
Ve = q̇Te M q̇e + l1 g(1 − C1 C2 )(m1 + m2 ) + m2 glh (1 − C3 C4 ), (10) The detailed parameters in Eqs. (12), (13) and (15) satisfy the
2
following limitations:
where qe = [ex ey θ1 θ2 θ3 θ4 ]T , ex = x − xd , ey = y − yd . By taking ⎧( )
the derivative of Eq. (10) with respect to time, we can obtain a 8 + 8l1 2 l̄2 m̄2 +
⎨ 4l 2 m̄ +h 4l2 2 m̄ + λ
⎪
⎪
similar equation as Eq. (9), that is, 1 1 1 2 M
a ≥ max , (17)
1 1 1 ⎪ 2d3
V̇e = q̈Te M q̇e + q̇Te Ṁ q̇e + q̇Te M q̈e
⎪
⎩
2 2 2
+ l1 g(S1 C2 θ̇1 + S2 C1 θ̇2 )(m1 + m2 ) (11) 12 + 9l1 2 l̄2h m̄22 + 6l1 2 m̄1 + 6l1 2 m̄2 + λM
,
+ m2 glh (S3 C4 θ̇3 + S4 C3 θ̇4 ) 2d4
1 36 + 4l̄2h m̄2 + λM 42 + 6l̄2h m̄2 + λM
= q̇Te M q̈e + q̇Te (C + C T )q̇e + l1 g(S1 C2 θ̇1 + S2 C1 θ̇2 )(m1 + m2 ) , ,
2 2d5 2d6
4
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx
√
2l1 2 m̄21 + 2l1 2 m̄22 + 2l̄2h m̄22 + λM ,
√ ⎫
l1 2 m̄21 + l1 2 m̄22 + l̄2h m̄22
⎬
+ λM ,
2 ⎭
1 1
k1d = k2d = α, k1p ≥ , k2p ≥ .
4 4

Remark 2. From the above results and Fig. 2, we give the detailed
implementation steps of the proposed control method as follows:
step 1: Combining the control objectives, the specific system
energy function shown in Eq. (10) is first constructed.
step 2: To enhance the positioning performance of the overhead
crane with DMB, the new adaptive update laws in Eq. (15) can be
obtained.
step 3: Considering the control task of precise positioning and
Fig. 3. Experiment setup.
system parameter identification, the Lyapunov function candidate
in Eq. (30) can be constructed and analyzed. The specific con-
Table 1
straints of the parameters α , kip , kid (i = 1, 2) shown in Eq. (17)
Overhead crane parameters.
are obtained, and also the specific expressions of the controllers
mx [kg] 0.5 my [kg] 2.0 m1 [kg] 0.5
Fx , Fy can be achieved as well. l1 [m] 0.34 g [m/s2 ] 9.8 d1 0.1
d2 0.12 d3 0.07 d4 0.08
d5 0.08 d6 0.08 − −
3.2. Stability analysis

Theorem 1. Under the controller in Eqs. (12), (13) and the new
4.2. Experimental parameters
adaptive update laws in Eq. (15), the trolley and guide rail will both
arrive at the preset position, meanwhile, swing angles of the DMB In this section, we replace xd and yd in the proposed con-
and hook will also be eliminated, that is, troller with an S-shaped trajectory which has the effect of swing-
]T ]T elimination, the mathematical expression of which can be de-
θ1 θ2 θ3 θ4 0 ,
[ [
lim ex ey = 0 0 0 0 0 scribed as follows:
t →∞
]T [ ]T
θ̇1 θ̇2 θ̇3 θ̇4 = 0 0 .
[
sin( t 2π t )
( )
lim ėx 0 0 0 0
⎧
ėy q(i)d
t →∞
⎪ t
⎨(q(i)f − q(i)0 ) tq(i) −
⎪
2π
⎪
d
(18) q(i)d = , (19)
⎪
⎪
⎪ +q(i)0 , t ∈ [0, tq(i)d )
Proof. See Appendix B.
q(i)f , t ∈ [tq(i)d , tf ]
⎩

where i = 1, 2, q(i)f and q(i)0 are the target position and initial
4. Experiments and discussion position of the trajectory, respectively. tq(i)d denotes the reaching
time of the trajectory. In addition, q(1)f and q(2)f are set as 0.4
This section will be divided into three parts. Firstly, for com- [m], and txd = tyd = 4 [s], respectively.
pleteness, the composition of the experimental platform and The parameters of the overhead crane are stored in Table 1.
the way of data interaction will be introduced. Then, system In this study, the mass of the hook is known as 0.5 [kg]. The
and control parameters are listed. The detailed expressions of range of DMB mass changes from 0.5 [kg] to 1.5 [kg], and the
compared controllers will also be shown. Finally, the existing range of vertical distance between the hook and DMB changes
linear–quadratic-regulator (LQR) and SMC were utilized to con- from 0.15 [m] to 0.25 [m]. therefore we determine the upper and
duct the comparison experiments. Additionally, the experiments lower bounds of them as follows:
of anti-disturbance verification will also be introduced. m1 = 0.4 [kg], m1 = 0.6 [kg], (20)
m2 = 0.4 [kg], m2 = 1.6 [kg],
4.1. Experimental equipment lh = 0.14 [m], lh = 0.26 [m].
Combining with the parameters in Table 1 and the selec-
The structure diagram of the experiment platform is illus-
tion method of λM in Eq. (35), it is not difficult to achieve the
trated in Fig. 3. Specifically, the hardware part is composed of an-
controller gains through Eq. (17) as follows:
gle encoders, motor (encoder embedded), motion control board,
and PC. In the horizontal direction, two DC motors which are α = k1d = k2d = 273, k1p = 6.8, k2p = 11.5, (21)
embedded with an absolute encoder (550 PPR) are used to drive and the detailed terms in Eq. (15) are selected as Πx = diag
the trolley and guide rail. The angle data of the hook and DMB (25, 35, 15, 10), Πy = diag(30, 45, 20, 5).
is collected by four encoders (16 384 PPR). The data collected
by the above six encoders is transferred to the PC by a motion 4.3. Effectiveness comparison
control board. After software data calculations, the control signal
of the controller is also transmitted to the motors. Additionally, For effectiveness verification of the proposed controller, and
the Matlab/Simulink is used to calculate the output signal in fully consider the rationality of comparison, we choose LQR and
real-time, and the sampling period is set as 5 ms. SMC [24] as the comparison controllers.
5
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

σ = ex θ1 θ2 θ3 θ4 θ̇1 θ̇2 θ̇3 θ̇4 ,

[ ]
The dynamic relationship in Eq. (1) can be linearized into the ey ẋ ẏ
matrix form as follows:
then we [choose ] Q = diag(50, 50, 50, 50, 50, 50, 2, 2, 2, 2, 2, 2)
q̈ = β + ηU , (22) and R = 1 1 .
[
q̈ = ẍ ÿ θ̈1 θ̈2 θ̈3 θ̈4
]T
, Through Eq. (26) and off-line calculations in Matlab/Simulink,
]T the specific terms in Eq. (25) are determined as follows:
β = β1 β2 β3 β4 β5 β6 ,
[
]T k11 = 50, k12 = 18, k13 = −26, k14 = 9.6, (27)
η η2 η3 η4 η5 η6
[
η= 1 , k15 = 1.5, k16 = −0.7, k21 = 50, k22 = 22,
η7 η8 η9 η10 η11 η12
]T k23 = 28, k24 = −12, k25 = 1.4, k26 = −0.7.
U = Fx Fy ,
[
Then, we conducted a group of comparative experiments with
the detailed terms in β and η are as follows: m2 = 0.5 [kg] and lh = 0.15 [m] to analyze the control effect
gm1 θ1 + gm2 θ1 −gm1 θ2 − gm2 θ2 of the proposed controller. The related results are all shown in
β1 = , β2 = , Fig. 4, such as the swing angles of the hook and payload θi (t)(i
mx my
⎛ ⎞ = 1, . . . ,4), the displacement of the trolley and guide rail, x and
−θ1 (Igm1 2 + Igm2 2 + glh 2 m1 2 y, and the control output forces, τx and τy . It is obvious that
⎜+2Igm1 m2 + Igm1 mx + Igm2 mx ⎟ the proposed control method has better control effect than other
⎝+gm1 m2 lh 2 + gm1 mx lh 2 − gm2 mx lh 2 )⎠
⎜ ⎟
control methods (e.g., LQR and SMC). Specifically, compared with
+gm2 mx lh 2 θ3 other control methods, the proposed controller can drive the
β3 = , trolley and guide rail to the preset position at a relatively short
l1 mx (m1 lh 2 + Im1 + Im2 ) time, and it does not exist residual oscillations of swing angles.
( )
−θ2 (gm1 2 + gm1 m2 + gm1 my + gm2 my ) In addition, from the values of τx and τy shown in Fig. 4, we can
+gm2 my θ4 see that the proposed controller consumes less energy than other
β4 = , methods.
l1 + my + m1
The detailed values of some significant parameters are stored
glh (m1 + m2 )(θ1 − θ3 ) g(m1 + m2 )(θ2 − θ4 )
β5 = , β6 = , in Table 2, such as the maximum swing angles of the hook
m1 lh 2 + Im1 + Im2 m1 l h and payload θ1max , θ2max , θ3max , θ4max , and the reaching time
1 1 of the trolley and guide rail, txr and tyr . As we can summarize
η1 = , η2 = 0, η3 = 0, η4 = ,
mx my from Table 2, the reaching time of the trolley and guide rail are
both 4.00 [s] under the proposed control method, and the above
−Im1 − Im2 − lh 2 m1
η5 = , η6 = 0, η7 = 0, parameters under SMC method and LQR method are 7.86 [s], 7.55
l1 mx (m1 lh 2 + Im1 + Im2 ) [s], 6.10 [s], 7.29 [s], respectively. Compared with SMC method,
m1 the reaching time of the trolley and guide rail under the proposed
η8 = , η9 = 0, η10 = 0, η11 = 0, η12 = 0.
l1 + my + m1 control method decreased about 49% and 47%, respectively. As
Then, we set to LQR, it decreased about 34% and 45%, respectively. In the
swing-elimination part, the maximum values of θi (i = 1 · · · 4)
S = e + ϕė, (23) under the proposed controller are 1.45[deg], 1.66[deg], 1.05[deg],
Ṡ = −µS − ρ tanh(S), 0.55[deg], respectively. Compared with LQR method, the values
of the above parameters decreased about 42%, 43%, 59%, 68%,
where respectively. Compared with SMC method, the proposed method
e = q − qd , also has some advantages in swing-elimination. In conclusion, the
]T proposed controller can accomplish the tasks of positioning and
θ1 θ2
θ4 , θ3
[
q= x y payload swing-elimination effectively.
]T
qd = xd yd 0 0 0 0 ,
[
4.4. Robustness comparison
ϕ = diag(ϕ1 , ϕ2 , ϕ3 , ϕ4 , ϕ5 , ϕ6 ),
µ and ρ are positive gains of the convergence law, and the Two groups of payload with m2 = 1.0 [kg] to 1.5 [kg], and lh
specific terms of ϕ will be designed. from 0.2 [m] to 0.25 [m] were utilized to conduct the robustness
Hence, we have the specific form of SMC as follows: comparative experiments, and the experiment results are illus-
trated in Figs. 5 and 6. As we can see, under different parameters,
Usmc = ηT ϕ T (−µS − ρ tanh(S) − q̇) − β ,
[ ]
(24) the proposed control method still has an excellent performance
where ϕ1 = 0.7, ϕ2 = 0.5, ϕ3 = 0.05, ϕ4 = 0.02, ϕ5 = 0.05, in positioning and swing-elimination. The detailed indicators are
ϕ6 = 0.02, µ = 35, ρ = 8. stored in Tables 3 and 4. Obviously, in the case of different system
Then, from the linearized equation in Eq. (22) and the system parameters, there is little difference in the control effect of the
parameters stored in Table 1, the following LQR controller can be proposed controller. LQR and SMC have high requirements for
achieved: the precision of the system parameters, therefore, the control
performance of LQR and SMC changes greatly in the case of
Fqx = − k11 ex − k12 ẋ − k13 θ1 − k14 θ3 − k15 θ̇1 − k16 θ̇3 + Frx , (25) different system parameters.
Fqy = − k21 ey − k22 ẏ − k23 θ2 − k24 θ4 − k25 θ̇2 − k26 θ̇4 + Fry . Then, another group of comparative experiments in terms
of non-zero initial swing angles of the hook and payload were
After analysis of a number of experiments, we choose f11 = conducted, and the results are illustrated in Fig. 7. Specifically,
3.5, f12 = 0.7, f21 = 4.2, f22 = 1.2 and ε = 0.01 to formulate the the initial angles of θi (i = 1 · · · 4) under the proposed con-
frictions in Eq. (25). The gain parameters of LQR are calculated by troller were set as 0.3[deg], −0.9[deg], −0.4[deg] and 0.8[deg],
the following cost function: respectively, and the corresponding values of SMC and LQR were
∫ ∞ ]T set as 1.8[deg], 0.2[deg], −0.6[deg], 0.1[deg], 0.3[deg], 0.4[deg],
(σ T Q σ + R F x ),
[
J = Fy (26) −0.3[deg] and 0.2[deg], respectively. From Fig. 7, it is obvious
0
6
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

Fig. 4. Experiment results with m2 = 0.5 [kg], lh = 0.15 [m].

Table 2
Experimental results with m2 = 0.5 [kg], lh = 0.15 [m].
Control method Reaching time txr [s] Reaching time tyr [s] Maximum value θ1 max [◦ ]
Proposed 4.00 4.00 1.45
SMC 7.86 7.55 1.11
LQR 6.10 7.29 2.54
Maximum value θ2 max [◦ ] Maximum value θ3 max [◦ ] Maximum value θ4 max [◦ ]
Proposed 1.66 1.05 0.55
SMC 1.50 1.75 0.96
LQR 2.98 2.57 1.23

Fig. 5. Experiment results with m2 = 1.0 [kg], lh = 0.20 [m].

that the proposed controller can still perform the positioning 5. Conclusion
task accurately and can rapidly eliminate payload oscillations.
Hence, the proposed control method has a great performance in To achieve positioning and swing-elimination for 3-D over-
robustness. head cranes with DMB, an adaptive control method based on
7
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

Fig. 6. Experiment results with m2 = 1.5 [kg], lh = 0.25 [m].

Table 3
Experimental results with m2 = 1.0 [kg], lh = 0.20 [m].
Control method Reaching time txr [s] Reaching time tyr [s] Maximum value θ1 max [◦ ]
Proposed 4.00 4.00 1.05
SMC 4.87 4.90 1.43
LQR 5.12 6.54 2.38
Maximum value θ2 max [◦ ] Maximum value θ3 max [◦ ] Maximum value θ4 max [◦ ]
Proposed 0.7 1.73 1.26
SMC 0.89 1.70 1.31
LQR 1.87 2.79 2.50

Table 4
Experimental results with m2 = 1.5 [kg], lh = 0.25 [m].
Control method Reaching time txr [s] Reaching time tyr [s] Maximum value θ1 max [◦ ]
Proposed 4.00 4.00 0.94
SMC 7.39 7.46 1.49
LQR 5.26 7.91 3.05
Maximum value θ2 max [◦ ] Maximum value θ3 max [◦ ] Maximum value θ4 max [◦ ]
Proposed 1.52 1.93 0.77
SMC 1.38 1.73 1.05
LQR 2.99 2.91 1.21

improved adaptive update laws was proposed. By using the Lya- 3. In order to adapt to multiple faults in industrial fields,
punov technique and Barbalat’s lemma, the stability of the con- we will study the controller based on fault-tolerance to
troller was theoretically proven. Finally, sufficient verification ex- improve the practicability of crane controllers.
periments were conducted. However, under some certain hook- 4. We will study the nonlinear control method for a cate-
to-payload connection structures, there is a phenomenon of DMB gory of crane systems, and improve the robustness and
rotation during transportation, which is an inadequate point con- anti-turbulence of the corresponding controller.
sidered in this paper. Hence, combined with the experimental
results, our future work will concentrate on the following points: Declaration of competing interest
1. In order to make the proposed algorithm more widely
applicable, our future research will further analyze the The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared
crane model considering the DMB rotation and analyze its
to influence the work reported in this paper.
dynamic properties.
2. From the experimental results, the values of angles are
generally small. Therefore, we will study the controller Acknowledgment
considering strong disturbances in the next stage and in-
vestigate the nature of angle convergence in the case of This work is supported by the National Natural Science Foun-
large angle values. dation of China under Grant 61703202.
8
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

Fig. 7. Experiment results with non-zero initial hook and payload swing angles with m2 = 1.0 [kg], lh = 0.20 [m].

Appendix A c21 = 0, c22 = 0, c23 = 0, c24 = θ̇2 l1 m1 S2 + θ̇2 l1 m2 S2 , c25 = 0,

c26 = θ̇4 lh m2 S4 ,
The terms of the matrices and vectors in Eq. (1) are as follows:
c31 = 0, c32 = 0, c33 = −θ̇2 l1 2 m1 C2 2 C2 S2 − θ̇2 l1 2 m2 C2 2 C2 S2
m11 = m1 + m2 + mx , m12 = 0, m13 = l1 m1 C2 C2 + l1 m2 C2 C2 ,
− θ̇2 l1 2 m1 C2 S1 2 S2 ,
m14 = −l1 m1 S1 S2 − l1 m2 S1 S2 ,
c34 = −θ̇1 l1 2 m1 C2 2 C2 S2 − θ̇1 l1 2 m2 C2 2 C2 S2 − θ̇1 l1 2 m1 C2 S1 2 S2
m15 = lh m2 C3 C4 , m16 = −lh m2 S3 S4 , m21 = 0,
− θ̇1 l1 2 m2 C2 S1 2 S2 ,
m22 = m1 + m2 + my , m23 = 0,
c35 = −θ̇3 l1 lh m2 C2 C2 C4 S3 + θ̇3 l1 lh m2 C2 C3 C4 S1 − θ̇4 l1 lh m2 C2 C2 C3 S4
m24 = −l1 m1 C2 − l1 m2 C2 , m25 = 0, m26 = −lh m2 C4 ,
m31 = l1 m1 C2 C2 + l1 m2 C2 C2 , − θ̇4 l1 lh m2 C2 S1 S3 S4 ,
m32 = 0, m33 = l1 2 m1 C2 2 C2 2 + l1 2 m2 C2 2 C2 2 + l1 2 m1 C2 2 S1 2 c36 = −θ̇4 l1 lh m2 C2 C2 C4 S3 + θ̇4 l1 lh m2 C2 C3 C4 S1 − θ̇3 l1 lh m2 C2 C2 C3 S4

+ l1 2 m2 C2 2 S1 2 , m34 = 0, − θ̇3 l1 lh m2 C2 S1 S3 S4 ,
m35 = l1 lh m2 C2 C2 C3 C4 + l1 lh m2 C2 C4 S1 S3 , c41 = 0, c42 = 0, c43 = θ̇1 l1 2 m1 C2 2 C2 S2 + θ̇1 l1 2 m2 C2 2 C2 S2
m36 = −l1 lh m2 C2 C2 S3 S4 + l1 lh m2 C2 C3 S1 S4 , + θ̇1 l1 2 m1 C2 S1 2 S2 + θ̇1 l1 2 m2 C2 S1 2 S2 ,
m41 = −l1 m1 S1 S2 − l1 m2 S1 S2 , m42 = −l1 m2 C2 − l1 m1 C2 , m43 = 0, c44 = −θ̇2 l1 2 m2 C2 S2 − θ̇2 l1 2 m1 C2 S2 + θ̇2 l1 2 m1 C2 2 C2 S2
m44 = l1 2 m1 C2 2 S2 2 + l1 2 m2 C2 2 S2 2 + l1 2 m1 S1 2 S2 2 + l1 2 m2 S1 2 S2 2 + θ̇2 l1 2 m2 C2 2 C2 S2
+ l1 2 m1 C2 2 + l1 2 m2 C2 2 , + θ̇2 l1 2 m1 C2 S1 2 S2 + θ̇2 l1 2 m2 C2 S1 2 S2 ,
m45 = l1 lh m2 C2 C4 S2 S3 − l1 lh m2 C3 C4 S1 S2 , m46 = l1 lh m2 C2 C3 S2 S4 c45 = θ̇3 l1 lh m2 C4 S1 S2 S3 + θ̇3 l1 lh m2 C2 C3 C4 S2
+ l1 lh m2 S1 S2 S3 S4 − θ̇4 l1 lh m2 C2 S2 S3 S4 + θ̇4 l1 lh m2 C3 S1 S2 S4 ,
+ l1 lh m2 C2 C4 , m51 = lh m2 C3 C4 , m52 = 0, m53 = l1 lh m2 C2 C2 C3 C4 c46 = −θ̇4 l1 lh m2 C2 S4 + θ̇4 l1 lh m2 C2 C3 C4 S2
+ l1 lh m2 C2 C4 S1 S3 , + θ̇4 l1 lh m2 C4 S1 S2 S3 − θ̇3 l1 lh m2 C2 S2 S3 S4 + θ̇3 l1 lh m2 C3 S1 S2 S4 ,
m54 = l1 lh m2 C2 C4 S2 S3 − l1 lh m2 C3 C4 S1 S2 , m55 = lh 2 m2 C3 2 C4 2 c51 = 0, c52 = 0, c53 = θ̇1 l1 lh m2 C2 C2 C4 S3 − θ̇1 l1 lh m2 C2 C3 C4 S1
+ lh 2 m2 C4 2 S3 2 + Im2 , − θ̇2 l1 lh m2 C2 C3 C4 S2
m56 = 0, m61 = −lh m2 S3 S4 , m62 = −lh m2 C4 , − θ̇2 l1 lh m2 C4 S1 S2 S3 , c54 = θ̇2 l1 lh m2 C2 C2 C4 S3 − θ̇2 l1 lh m2 C2 C3 C4 S1
m63 = −l1 lh m2 C2 C2 S3 S4 + l1 lh m2 C2 C3 S1 S4 , − θ̇1 l1 lh m2 C2 C3 C4 S2 − θ̇1 l1 lh m2 C4 S1 S2 S3 , c55 = −θ̇4 lh 2 m2 C3 2 C4 S4
m64 = l1 lh m2 C2 C3 S2 S4 + l1 lh m2 S1 S2 S3 S4 + l1 lh m2 C2 C4 , m65 = 0,
− θ̇4 lh 2 m2 C4 S3 2 S4 ,
m66 = lh 2 m2 C3 2 S4 2 + lh 2 m2 S3 2 S4 2 + lh 2 m2 C4 2 ,
c56 = −θ̇3 lh 2 m2 C3 2 C4 S4 − θ̇3 lh 2 m2 C4 S3 2 S4 , c61 = 0, c62 = 0,
c11 = 0, c12 = 0, c13 = −θ̇1 l1 m1 C2 S1 − θ̇1 l1 m2 C2 S1 − θ̇2 l1 m1 C2 S2
c63 = θ̇1 l1 lh m2 C2 S1 S3 S4 + θ̇1 l1 lh m2 C2 C2 C3 S4 + θ̇2 l1 lh m2 C2 S2 S3 S4
− θ̇2 l1 m2 C2 S2 ,
− θ̇2 l1 lh m2 C3 S1 S2 S4 ,
c14 = −θ̇2 l1 m1 C2 S1 − θ̇2 l1 m2 C2 S1 − θ̇1 l1 m1 C2 S2 − θ̇1 l1 m2 C2 S2 ,
c64 = −θ̇2 l1 lh m2 C4 S2 + θ̇2 l1 lh m2 C2 C2 C3 S4 + θ̇2 l1 lh m2 C2 S1 S3 S4
c15 = −θ̇3 lh m2 C4 S3 − θ̇4 lh m2 C3 S4 , c16 = −θ̇4 lh m2 C4 S3
+ θ̇1 l1 lh m2 C2 S2 S3 S4
− θ̇3 lh m2 C3 S4
9
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

− θ̇1 l1 lh m2 C3 S1 S2 S4 , c65 = θ̇3 lh 2 m2 C4 S3 2 S4 + θ̇3 lh 2 m2 C3 2 C4 S4 , 1 1 1

+ α k1p ex 2 + α k2p ey 2 + ω̃xT Πx ω̃x
2 2 2
c66 = −θ̇4 lh 2 m2 C4 S4 + θ̇4 lh 2 m2 C3 2 C4 S4 + θ̇4 lh 2 m2 C4 S3 2 S4 ,
1 e
g11 = 0, g21 = 0, g31 = gl1 m1 C2 S1 + gl1 m2 C2 S1 , + ω̃yT Πy ω̃y + δv ( )M ė,
2 2
g41 = gl1 m1 C2 S2 + gl1 m2 C2 S2 , where ω̃x = ωx − ω̂x , ω̃y = ωy − ω̂y .
g51 = glh m2 C4 S3 , g61 = glh m2 C3 S4 , Then, by taking the time derivative of Eq. (30), we can obtain
where mx represents the mass of the trolley, and my is the total ex ey
V̇ = − α k1d ė2x − α k2d ė2y + ω̃xT γx δ ( ) + ω̃yT γy δ ( ) + αχ1 + αχ2
mass of the trolley and guide rail. m1 and m2 denote the mass 2 2
of the hook and DMB, respectively. I represents the moment of (31)
inertia of DMB. l1 denotes the rope length between the trolley e T e T e T
and hook, and l2 denotes the rope length between the hook and
+ αχ3 + αχ4 + δ̇v ( ) M ė + δv ( ) Ṁ ė + δv ( ) M ë,
2 2 2
the fixed point of DMB. lh is the distance between the hook and
by substituting Eq. (28) into Eq. (31), we have
the centroid of DMB, and lp is the length of DMB. Fx and Fy are
driving forces in the X and Y directions, respectively. di (i = 1, V̇ = − α k1d ė2x − α k2d ė2y + αχ1 + αχ2 + αχ3 (32)
. . . ,6) represents the resistance in the six state dimensions. ex ey θ1
+ αχ4 − δ ( )k1p ex − δ ( )k2p ey − δ ( )d3 θ̇1
2 2 2
Appendix B
θ2 θ3 θ4
− δ( )d4 θ̇2 − δ ( )d5 θ̇3 − δ ( )d6 θ̇4
2 2 2
e e e
Proof. To simplify the proof steps, we first make some mathe- − δv ( )T C q̇ − δv ( )T G + δ̇v ( )T M ė
matical arrangements for several significant terms: 2 2 2
e T
e + δv ( ) Ṁ ė + Ω1 + Ω2 ,
δv ( )T M ë (28) 2
2 θ1
e Ω1 = δ ( )(−ẍr (l1 m1 C1 C2 + l1 m2 C1 C2 ))
= δv ( )T (M q̈ − M q̈r ) 2
2 θ2
e + δ( )(ẍr (l1 m1 S1 S2 + l1 m2 S1 S2 )
= δv ( )T (U − Fs − D − C q̇ − G − M q̈r ) 2
2
θ3
−γx T ω̃x − k1p ex − k1d ėx + ÿr (l1 m1 C2 + l1 m2 C2 )) + δ (
⎡ ⎤
)(−ẍr lh m2 C3 C4 )
−γy ω̃y − k2p ey − k2d ėy
T 2
θ4
⎢ ⎥
− d θ̇ − ẍ (l m C C + l m C C ) + δ( )(ẍr lh m2 S3 S4 + ÿr lh m2 C4 ),
⎢ ⎥
3 1 r 1 1 1 2 1 2 1 2
e T ⎢(
⎢ )⎥
= δv ( ) ⎢ −d4 θ̇2 + ẍr (l1 m1 S1 S2 + l1 m2 S1 S2 ) ⎥
⎥ 2
2 ⎢ +ÿ (l m C + l m C ) ex ey
r 1 1 2 1 2 2
⎥ Ω2 = − k1d ėx δ ( ) − k2d ėy δ (
).
2 2
⎢ ⎥
⎣ −d5 θ̇3 − ẍr lh m2 C3 C4 ⎦
Combined with Eq. (29), the following inequality about Eq. (32)
−d6 θ̇4 + ẍr lh m2 S3 S4 + ÿr lh m2 C4
can be obtained:
e T e
− δv ( ) C q̇ − δv ( )T G V̇ ≤ − α k1d ė2x − α k2d ė2y + αχ1 + αχ2 + αχ3 + αχ4 (33)
2 2
ex ex ey θ1 θ2
= δ ( )(−γx T ω̃x − k1p ex − k1d ėx ) − 2k1p δ 2 ( ) − 2k2p δ 2 ( ) − δ( )d3 θ̇1 − δ ( )d4 θ̇2
2 2 2 2 2
ey
+ δ ( )(−γy T ω̃y − k2p ey − k2d ėy ) θ3 θ4 e e
2 − δ( )d5 θ̇3 − δ (
)d6 θ̇4 + ėT C δv ( ) − δv ( )T C q̇r
2 2 2 2
θ1 e e
+ δ( )(−d3 θ̇1 − ẍr (l1 m1 C1 C2 + l1 m2 C1 C2 )) − δv ( )T G + δ̇v ( )T M ė + Ω1 + Ω2 ,
2 2 2
θ2
+ δ( )(−d4 θ̇2 + ẍr (l1 m1 S1 S2 + l1 m2 S1 S2 ) by utilizing the conclusion in Eq. (4), one obtains
2
e
θ3 δ̇v ( )T M ė (34)
+ ÿr (l1 m1 C2 + l1 m2 C2 )) + δ ( )(−d5 θ̇3 − ẍr lh m2 C3 C4 ) 2
2 e e
θ4 = δ̇v ( )T M q̇ − δ̇v ( )T M q̇r
+ δ( )(−d6 θ̇4 + ẍr lh m2 S3 S4 + ÿr lh m2 C4 ) 2 2
2 1 e
e e
− δv ( )T C q̇ − δv ( )T G, ≤ λM (ẋ2 + ẏ2 + θ̇12 + θ̇22 + θ̇32 + θ̇42 ) − δ̇v ( )T M q̇r ,
2 2 2 2
]T for simplicity, we directly give the selection method of λM after
where δv ( 2e ) = δ ( e2x ) δ ( 2y ) δ ( θ21 ) δ ( θ22 ) δ ( θ23 ) δ ( θ24 ) de-
e
[
analyzing the dynamic relationship in Eq. (1), that is,
notes the saturated function vector with δv (∗) defined in Eq. (16). {( )
m̄1 + m̄2 + mx + 4l1 m̄1
Further, combining with the property of the saturation func- λM = max , (35)
+41 l1 m̄2 + 4l̄h m̄2
tion in Eq. (16), we can elicit an inequality as follows:
m1 + m̄2 + my + 2l1 m1 + 2l1 m̄2 + 2l̄h m̄2 ,
e e
δ ( )k1p e ≥ 2k1p 2 δ 2 ( ). (29) m̄1 + m̄2 + my + 2l1 m̄1 + 2l1 m̄2 + 2l̄h m̄2 ,
2 2
By combining Eq. (10) and the new update laws designed l1 m̄1 l1 m̄2
2l1 2 m̄1 + 2l1 2 m̄2 + + + l1 l̄h m̄2 ,
in Eq. (15), the following Lyapunov function candidate can be 2 2
obtained: 3l1 l̄h m̄2
3l1 2 m̄1 + 3l1 2 m̄2 + l1 m̄1 + l1 m̄2 + ,
1 2
V = α q̇Te M q̇e + α l1 g(1 − C1 C2 )(m1 + m2 ) + α m2 glh (1 − C3 C4 )
2 l̄h m̄2
(30) 2l̄2h m̄2 + I m̄2 + + 4l1 l̄h m̄2 ,
2
10
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

θ3 (t) θ3 (0)
( )
3l̄2h m̄2 + l̄h m̄2 + 6l1 l̄h m̄2 ,
}
+ 2d5 cos − cos
2 2
combined with the AM–GM inequality, an inequality about Ω2
θ4 (t) θ4 (0)
( ) ∫ t
can be obtained:
+ 2d6 cos − cos + Υ3 d τ
k1d 2 1 ex k2d 2 1 ey ex 2 2 0
Ω2 ≤ ẋ2 + δ2 ( )+ ẏ2 + δ2 ( ) + k1d ẋr δ ( ) ∫ t
2 2 2 2 2 2 2
ey ≤ V (0) + 4d3 + 4d4 + 4d5 + 4d6 + Υ3 dτ ,
+ k2d ẏr δ (
), (36) 0
2
from the detailed expression of Υ3 in Eq. (38) and the restricted
by using the same method mentioned above, we can also have
( ) conditions in Eq. (17), we have the following conclusion:
e 4 + 4l1 2 lh 2 m2 2
ėT C δv ( ) ≤ θ̇12 (37) V (t) ∈ L∞ . (41)
2 +2l1 2 m1 + 2l1 2 m2
+ θ̇22 (6 + 4.5l1 2 lh 2 m2 2 + 3l1 2 m1 + 3l1 2 m2 ) From Eqs. (30) and (41), we have
+ θ̇ 2
3 (18 +2lh 2 m2 ) 2
4 (21 + θ̇
3lh 2 m2 ) + ėx , ėy , θ̇1 , θ̇2 , θ̇3 , θ̇4 , ex , ey , ω̃x , ω̃y ∈ L∞ , (42)
2 2 2 2 2 2 2
+ ẋ (l1 m1 l1 m2
+ lh m2 ) + combined with Eqs. (12), (13), (42) and the property of the
l1 2 m1 2 + l1 2 m2 2 + lh 2 m2 2 e trajectory, one can be easily obtained:
+ ẏ2 ( ) − q̇Tr C δv ( ).
4 2
ẋ, ẏ ∈ L∞ , (43)
Combined with Eqs. (34)–(37), (33) can be rewritten into the
following structure: Fx = γx ω̂x − k1p ex − k1d ėx ∈ L∞ ,
T

Fy = γy T ω̂y − k2p ey − k2d ėy ∈ L∞ ,

V̇ ≤ Υ1 + Υ2 + Υ3 , (38)
Υ1 = − α k1d ẋ2r − α k2d ẏ2r + 2α k1d ẋr ẋ + 2α k2d ẏr ẏ besides, combining Eqs. (1) and (43), yields,

+ αχ1 + αχ2 + αχ3 + αχ4 + α θ̇ d3 12 ẍ, ÿ, θ̈1 , θ̈2 , θ̈3 , θ̈4 ∈ L∞ . (44)
e
+ α θ̇ + α θ̇ + α θ̇ −
d4 22 d5 32 d6 42 δ
q̇Tr C v ( ) ∫ t From Eqs. (40) and (41), we know V (0)+4d3 +4d4 +4d5 +4d6 +
2
e e ex 0
Υ3 dτ ∈ L∞ , further, combined with the detailed expression of
− δv ( )T C q̇r − δ̇v ( )T M q̇r + Ω1 + k1d ẋr δ ( ) Υ3 , we have the following two significant conclusions:
2 2 2
ey ∫ t
ex ey
+ k2d ẏr δ ( ), Υ3 dτ ∈ L∞ , ẋ, ẏ, θ̇1 , θ̇2 , θ̇3 , θ̇4 , δ ( ), δ ( ) ∈ L2 , (45)
2 2 2
0
θ1 θ2 θ3 θ4
Υ2 = − δ ( )d3 θ̇1 − δ ( )d4 θ̇2 − δ ( )d5 θ̇3 − δ ( )d6 θ̇4 , from Eqs. (44), (45) and the knowledge of Barbalat’s lemma, we
2 2 2 2
1 have
Υ3 = θ̇12 (4 + 4l1 2 lh 2 m2 2 + 2l1 2 m1 + 2l1 2 m2 + λM − α d3 ) ex ey
2 lim ẋ, ẏ, θ̇1 , θ̇2 , θ̇3 , θ̇4 , δ ( ), δ ( ) = 0, lim γx , γy = 0. (46)
1 t →∞ 2 2 t →∞
+ θ̇22 (6 + 4.5l1 2 lh 2 m2 2 + 3l1 2 m1 + 3l1 2 m2 + λM
2 Then, we rewrite Eq. (1) into the following structure:
1
− α d4 ) + θ̇ 2
3 (18 + 2lh m2 + λM − α d5 )
2
q̈ = M −1 (U − Fs − D − C q̇ − G), (47)
2
1 further, by some matrix operations on Eq. (47), an implicit func-
+ θ̇42 (21 + 3lh 2 m2 + λM − α d6 )
2 tion of ẍ can be obtained:
k1d 2 1
+ ẋ2 (l1 2 m1 2 + l1 2 m2 2 + lh 2 m2 2 + + λM ẍ = faux1 {ẋ, ẏ, θ̇1 , θ̇2 , θ̇3 , θ̇4 , ẋr , ẏr , γx T ω̂x }
2 2 +faux1 {C1 , S1 , C2 , S2 , C3 , S3 , C4 , S4 , x, y, θ1 , θ2 , θ3 , θ4 , xr , yr }
l1 2 m1 2 + l1 2 m2 2 + lh 2 m2 2 k2d 2
− α k1d ) + ẏ2 ( + (48)
4 2
1 ex ey 1 1 from the conclusions in Eqs. (42) and (46), and combined with
+ λM − α k2d ) + δ 2 ( )( − 2k1p ) + δ 2 ( )( − 2k2p ),
the Barbalat’s lemma [26], the following results can be achieved:
2 2 2 2 2
by utilizing the property of the S-shaped trajectory defined in
Eq. (19), the following result can be obtained when t > txd and
lim faux1 = 0, lim f˙aux2 ∈ L∞ , lim ẍ = 0, (49)
t > tyd : t →∞ t →∞ t →∞

V̇ ≤ Υ2 + Υ3 , (39) by similar procedures as Eqs. (48) and (49), limt →∞ ÿ = 0 can

also be obtained.
then, by integrating both sides of Eq. (39) with respect to time, Then, from Eq. (47), we have
one obtains
θ̈1 = faux3 ẍ, ÿ, ẋ, ẏ, θ̇1 , θ̇2 , θ̇3 , θ̇4
⎛ ⎞ { }
θ1
sin( 2 )d3 θ̇1
⎜+ sin( θ2 )d4 θ̇2 ⎟
∫ t⎜ ⎟ ∫ t + faux4 {g3 , C1 , S1 , C2 , S2 , C3 , S3 , C4 , S4 } , (50)
⎟dτ + Υ3 dτ
V (t) ≤ V (0) − ⎜ 2 ⎟ (40)
θ
0 ⎝+ sin( 3 )d5 θ̇3 ⎠ where g3 is a term of the gravity matrix G in Eq. (1). From Eqs. (1),
⎜
0
2
θ4 (46) and (49), the following conclusions can be obtained:
+ sin( 2 )d6 θ̇4
θ1 (t) θ1 (0) lim faux3 = 0, lim f˙aux4 ∈ L∞ , (51)
( )
t →∞ t →∞
= V (0) + 2d3 cos − cos
2 2
by utilizing the Barbalat’s lemma, we have
θ2 (t) θ2 (0)
( )
+ 2d4 cos − cos lim θ̈1 = 0, lim g3 = 0. (52)
2 2 t →∞ t →∞

11
L. Yang and H. Ouyang ISA Transactions xxx (xxxx) xxx

From the assumption in Eq. (6) and the conclusions in Eq. (52), [11] Ouyang H, Hu J, Zhang G, Mei L, Deng X. Decoupled linear model
we have and s-shaped curve motion trajectory for load sway suppression con-
trol in overhead cranes with double-pendulum effect. J Mech Eng Sci
lim θ1 = 0, (53) 2019;233(10):3678–89.
t →∞ [12] Chen Q, Cheng W, Gao L, Fottner J. A pure neural network controller for
similarly, limt →∞ θ2 , θ3 , θ4 = 0 can also be obtained. double-pendulum crane anti-sway control: based on Lyapunov stability
theory. Asian J Control 2021;23(1):387–98.
Finally, combined with Eqs. (1), (25), (46) and (53), it is clear [13] Giacomelli M, Padula F, Simoni L, Visioli A. Simplified input–output inver-
that sion control of a double pendulum overhead crane for residual oscillations
reduction. Mechatronics 2018;56:37–47.
lim ex , ey = 0. (54) [14] Zhang M, Ma X, Rong X, Song R, Tian, Li Y. A novel energy-coupling-based
t →∞
control method for double-pendulum overhead cranes with initial control
Hereto, through the discussions above, we have demonstrated force constraint. Adv Mech Eng 2018;10(1):1–13.
the stability of the control system under the proposed controller. [15] Tuan L, Lee S. Sliding mode controls of double-pendulum crane systems.
J Mech Sci Technol 2013;27(6):1863–73.
In particular, with the proposed controller, the positioning error
[16] Sun N, Wu Y, Chen H, Fang Y. An energy-optimal solution for trans-
of the trolley and guide rail will converge to zero, and the oscil- portation control of cranes with double pendulum dynamics: Design and
lation of the hook and payload will be completely suppressed. experiments. Mech Syst Signal Process 2018;102:87–101.
[17] Qian D, Tong S, Lee S. Fuzzy-logic-based control of payloads sub-
References jected to double-pendulum motion in overhead cranes. Autom Constr
2016;65:133–43.
[18] Jaafar H, Mohamed Z, Shamsudin M, Mohd N, Ramli L, Abdullahi A. Model
[1] Ramli L, Mohamed Z, Jaafar H. A neural network-based input shaping for
reference command shaping for vibration control of multimode flexible
swing suppression of an overhead crane under payload hoisting and mass
systems with application to a double-pendulum overhead crane. Mech Syst
variations. Mech Syst Signal Process 2018;107:484–501.
Signal Process 2019;115:677–95.
[2] Xie X, Huang J, Liang Z. Vibration reduction for flexible systems by
[19] Adeli M, Zarabadipour H, Zarabad S, Shoorehdeli M. Anti-swing control of
command smoothing. Mech Syst Signal Process 2013;39:461–70.
a double-pendulum-type overhead crane using parallel distributed fuzzy
[3] Sun N, Fang Y. An efficient online trajectory generating method
LQR controller. Int J Phys Sci 2011;6(35):306–11.
for underactuated crane systems. Internat J Robust Nonlinear Control
[20] Zhang M, Ma X, Rong X, Tian X, Li Y. Adaptive tracking control for
2014;24(11):1653–63.
double-pendulum overhead cranes subject to tracking error limitation,
[4] Chen H, Fang Y, Sun N. Optimal trajectory planning and tracking control
parametric uncertainties and external disturbances. Mech Syst Signal
method for overhead cranes. IET Control Theory Appl 2016;10(6):692–9.
Process 2016;76–77:15–32.
[5] Wu X, He X. Nonlinear energy-based regulation control of three-
[21] Wu Q, Wang X, Hua L, Xia M. Improved time optimal anti-swing control
dimensional overhead cranes. IEEE Trans Autom Sci Eng 2017;14(2):1297–
system based on low-pass filter for double pendulum crane system with
308.
distributed mass beam. Mech Syst Signal Process 2021;151:107444.
[6] Sun N, Wu X, Fang Y. An enhanced coupling nonlinear control method for
[22] Huang J, Xie X, Zhang L. Control of bridge cranes with distributed-mass
bridge cranes. IET Control Theory Appl 2014;8(13):1215–23.
payload dynamics. IEEE/ASME Trans Mechatronics 2015;20(1):481–6.
[7] Almutairi N, Zribi M. Sliding mode control of a three-dimensional overhead
[23] Peng J, Huang J, Singhose W. Payload twisting dynamics and oscillation
crane. IEEE Trans Automat Control 2009;15(11):1679–730.
suppression of tower cranes during slewing motions. Nonlinear Dynam
[8] Tian Z, Yu L, Ouyang H, Zhang G. Sway and disturbance rejection control
2019;98(2):1041–8.
for varying rope tower cranes suffering from friction and unknown
[24] Ouyang H, Tian Z, Yu L, Zhang G. Adaptive tracking controller design for
payload mass. Nonlinear Dynam 2021. http://dx.doi.org/10.1007/s11071-
double-pendulum tower cranes. Mech Mach Theory 2020;153:103980.
021-06793-6, (in press).
[25] Sun N, Fang Y, Chen H, He B. Adaptive nonlinear crane control with
[9] Chen H, Fang Y, Sun N. A swing constraint guaranteed MPC algo-
load hoisting/lowering and unknown parameters: design and experiments.
rithm for underactuated overhead cranes. IEEE/ASME Trans Mechatronics
IEEE/ASME Trans Mechatronics 2015;20(5):2107–19.
2016;21(5):2543–55.
[26] Khalil H. Nonlinear systems. 3rd ed.. Englewood Cliffs, NJ, USA:
[10] Zhang C, Wang C, Wang J, Li C. Neuro-adaptive trajectory tracking control
Prentice-Hall; 2002.
of underactuated autonomous surface vehicles with high-gain observer.
Appl Ocean Res 2020;97:102051.

12