Journal Européen des Systèmes Automatisés

Vol. 57, No. 2, April, 2024, pp. 425-431

Journal homepage: http://iieta.org/journals/jesa

Position Control and Anti-Sway of Overhead Crane System with Uncertain Nonlinear
Model
Le Van Anh , Vu Thi To Linh*

Faculty of Electrical Engineering, University of Economics – Technology for Industries, Ha Noi 100000, Vietnam

Corresponding Author Email: vttlinh@uneti.edu.vn

Copyright: ©2024 The authors. This article is published by IIETA and is licensed under the CC BY 4.0 license
(http://creativecommons.org/licenses/by/4.0/).

https://doi.org/10.18280/jesa.570212 ABSTRACT

Received: 31 December 2023 Overhead cranes are used to move heavy, bulky objects above the factory floor instead of
Revised: 1 April 2024 along floor walkways. They are commonly used to load and unload goods in factories,
Accepted: 12 April 2024 outdoor warehouses and serve at stations, ports. During operation, chain hoists or cable
Available online: 28 April 2024 hoists are the main equipment, plays the role of hoisting/lowering materials and moving
mechanism along the main beam. Therefore, vibration cannot be avoided during the
process of moving heavy objects, causing danger to people and affecting the product. In
Keywords: addition, the overhead crane is an uncertain nonlinear system, compared to the single
overhead crane, radial basis function neural pendulum type, moving two loads at the same time is much more complicated. That's why
network, adaptive control, Lyapunov theory the author proposes to design a new controller that not only helps balance the cart but also
the two pendulums during operation. First, the system dynamics model is built. Next, an
adaptive controller based on the radial basis function neural network (RBFNN) is designed
and proven to be stable according to Lyapunov theory. Simulation results of the overhead
crane system on MATLAB/Simulink software have shown the effectiveness of the
proposed algorithm even when the working system is affected by model uncertainty.

1. INTRODUCTION Figure 1.
Main beams: These are beams that span the width of the
Overhead crane is a type of equipment used to lift and move overhead crane and support the moving hoist system. In a
goods in factories as well as outdoors. Normally, overhead single girder overhead crane there is one main girder, while in
cranes are often used in factories, but when used outdoors, a double girder overhead crane there are two.
people use gantry cranes. The use of overhead cranes is very Edge beam: This is a beam frame with wheels located at
convenient and highly effective in work, loading and both ends. They allow the overhead crane to move along the
unloading goods and bulky heavy objects such as iron and runway beam.
steel in steel factories. Operated mainly by electric motors, it Rail support beams: These are parallel overhead structural
is widely used in industrial plants, steel plants, hydroelectric beams on which the edge beams move. They will support the
plants, as well as civil works... entire overhead crane system and are often mounted on
The main parts of the overhead crane are shown in detail in support columns or attached to the house frame structure.

(a) Structure of the overhead crane (b) Operating principle of each part

Figure 1. Structure diagram and operating principle of the overhead crane

425
 Hoist: A hoist is a device that lifts, lowers and holds loads. 2. MATHEMATICAL MODEL OF THE OVERHEAD
It can be electric, hand-pulled or pneumatic. Includes motor, CRANE SYSTEM
gearbox, cable reel, cable or chain and hook.
Cart: The cart is the part that carries the hoist and allows the We can represent a simple model of a crane system as
hoist to move along the main beam. It includes frame, wheels shown in Figure 2. Suppose the cart, pendulum and payload
and transmission mechanism. work on a two-dimensional plane, the cart moves on a
Control: Overhead cranes can be controlled by wire control, horizontal line.
remote control or cabin control. These systems allow operators
to control overhead crane movements including hoisting,
lowering and positioning loads.
Electrical system: This system includes power supply
components such as wires, contact brushes, power rails that
supply power to the motor and overhead crane control system.
These parts are shown in detail in Figure 1.
During overhead crane operation, the goal is to design the
controller so that the goods are transported to the desired
location without vibration or vibration. Usually, parameters
such as cable length and load volume change, these are
uncertain parameters of the overhead crane. In addition, the
system is also affected by external disturbances such as
friction, wind... which also make it difficult for the vehicle to
reach the desired position and causes the pendulum to vibrate.
Therefore, designing a controller for the overhead crane
system to ensure stable working system in all situations is
always of interest to research by scientists.
In the studies [1, 2], the authors proposed a closed-loop Figure 2. Simple overhead crane system model
feedback system control structure using a proportional integral
derivative (PID) controller to control the position of the cart The overhead crane system model can be built according to
and reduce the vibration angle. The results show good control the Newton-Euler or Euler-Lagrange method. In this study, the
performance, fast response, but traditional PID controllers authors use the Euler-Lagrange method based on kinetic and
easily lose control when noise occurs, and adjustment depends potential energy to establish the equations of the mechanical
on the operating engineer. Linear quadratic regulation (LQR) system. The Euler-Lagrange equation is represented as follows:
optimal controller proposed in studies [3, 4]. When the system
has uncertain model parameters, adaptive control methods 𝑑 𝜕𝐿 𝜕𝐿 𝜕𝑃
( )− + =𝑢 (1)
have been proposed by authors in the studies [5-12]. A robust 𝑑𝑡 𝜕𝑞̇ 𝜕𝑞 𝜕𝑞
control method based on sliding mode control (SMC) is also
often applied to nonlinear systems, which is very useful for where, 𝐿 and 𝑃 are kinetic energy and potential energy,
overhead crane systems [13-16]. However, SMC has respectively; 𝑢 denotes the driving force of the cart; 𝑞 =
disadvantages that cause chattering, affecting output response [𝑥 𝜑1 𝜑2 ]𝑇 , 𝑥 is the position of the cart, 𝜑1 and 𝜑2 are the
performance as well as reducing the lifespan of the devices. positions of pendulum 1 and pendulum 2, respectively, relative
Model predictive control (MPC) is studied for its advantages to the vertical axis.
in handling constraints as shown in the studies [17, 18]. In The total potential energy of the mechanical system is
addition to the above methods, many intelligent controllers written as:
such as fuzzy control [19, 20], neural networks [21, 22] have
also been proposed to implement for overhead crane systems. 𝑃 = 𝑚1 𝑙1 𝑔 (1 − 𝑐𝑜𝑠(𝜑1 (𝑡)))
From the analysis of the works published above, we see that (2)
most of these methods only design control for structural +𝑚2 𝑙2 𝑔 (1 − 𝑐𝑜𝑠(𝜑2 (𝑡)))
systems with one load. For a system consisting of two loads,
research mainly proposes traditional control methods and The total kinetic energy of a mechanical system is written
represents the model in linear form. Therefore, the authors as:
proposed an adaptive control method based on the RBF neural
network for the uncertain nonlinear overhead crane system. 1 1 1
In summary, the main contributions of this paper are 𝐿= 𝑚1 𝑣12 + 𝑚2 𝑣22 + 𝑀𝑥̇ 2
2 2 2
presented in the following ideas. 1
- Designing a new controller not only helps balance the cart = 𝑚1 ((𝑙1 𝜑̇ 1 𝑠𝑖𝑛𝜑1 )2 + (𝑥̇ − 𝑙1 𝜑̇ 1 𝑐𝑜𝑠𝜑1 )2 )
2 (3)
but also helps balance the two pendulums during operation. 1
- The effects of model uncertainty are regulated by an + 𝑚2 ((𝑙2 𝜑̇ 2 𝑠𝑖𝑛𝜑2 )2 + (𝑥̇ − 𝑙2 𝜑̇ 2 𝑐𝑜𝑠𝜑2 )2 )
2
adaptive law based on the RBF neural network. This is 1
+ 𝑀𝑥̇ 2
completely suitable for real-life applications when overhead 2
cranes operate.
- Data about overhead crane systems can be measured where, 𝑥̇ is the linear velocity of the cart; 𝜑̇ 1 is the angular
through output feedback, reducing costs when designing velocity of the payload 1; 𝜑̇ 2 is the angular velocity of the
hardware. payload 2; 𝑣1 , 𝑣2 are the velocity components of the pendulum
in the x and y directions calculated as follows:

426
 𝑣12 = (𝑥̇ − 𝑙1 𝜑̇ 1 𝑐𝑜𝑠𝜑1 )2 + (𝑙1 𝜑̇ 1 𝑠𝑖𝑛𝜑1 )2 ; 𝑥̇1 = 𝑥̇ = 𝑥2
𝑣22 = (𝑥̇ − 𝑙2 𝜑̇ 2 𝑐𝑜𝑠𝜑2 )2 + (𝑙2 𝜑̇ 2 𝑠𝑖𝑛𝜑2 )2 . 𝑥̇ 2 = 𝑥̈ = 𝑓1 (𝑋) + 𝑔1 (𝑋)𝑢
𝑥̇ 3 = 𝑥4 = 𝜑̇ 1
The parameters of the mechanical system model are (8)
𝑥̇ 4 = 𝜑̈ 1 = 𝑓2 (𝑋) + 𝑔2 (𝑋)𝑢
described in Table 1. 𝑥̇ 5 = 𝑥6 = 𝜑̇ 2
{𝑥̇ 6 = 𝜑̈ 2 = 𝑓3 (𝑋) + 𝑔3 (𝑋)𝑢
Table 1. Parameters of the mechanical system model
Eq. (8) is briefly written as follows:
Symbol Meaning
𝑀 Denotes the mass of the cart
𝑚1 Mass of payload 1 𝑋̇ = 𝐹(𝑋) + 𝐺(𝑋)𝑢 (9)
Cable length between the center of the cart and the
𝑙1
payload 1
𝜑1 Payload 1 swing angle 3. DESIGN A CONTROLLER FOR AN OVERHEAD
𝑚2 Mass of payload 2 CRANE SYSTEM
Cable length between the center of the cart and the
𝑙2
payload 2 3.1 Adaptive RBFNN controller
𝜑2 Payload 2 swing angle
𝑔 Gravitational constant
Recently, the RBF neural network has attracted the research
attention of scientists. With a simple three-layer structure
Partial derivative of Eq. (1) with respect to the variables 𝑥, including input layer, hidden layer, output layer and good
𝜑1 , 𝜑2 respectively. generalization ability, lengthy and unnecessary calculations
Partial derivative with respect to variable 𝑥 we get: are avoided compared to multi-layer feed forward networks.
This section introduces an adaptive control method based
𝑑 𝜕𝐿 𝜕𝐿 𝜕𝑃
( )− + =𝑢 on neural approximation with unknown parameters [21-23].
𝑑𝑡 𝜕𝑥̇ 𝜕𝑥 𝜕𝑥 Defining error vector
⇒ 𝑚1 𝑙1 (𝜑̇ 12 𝑠𝑖𝑛𝜑1 − 𝜑̈ 1 𝑐𝑜𝑠𝜑1 )
+𝑚2 𝑙2 (𝜑̇ 22 𝑠𝑖𝑛𝜑2 − 𝜑̈ 2 𝑐𝑜𝑠𝜑2 ) 𝑥1 − 𝑥𝑟𝑒𝑓 𝑥 − 𝑥𝑟𝑒𝑓
+(𝑚1 + 𝑚2 + 𝑀)𝑥̈ = 𝑢 (4) 𝑒(𝑡) = [𝑥3 − 𝜑1𝑟𝑒𝑓 ] = [𝜑1 − 𝜑1𝑟𝑒𝑓 ] = 𝑞 − 𝑞𝑟𝑒𝑓 (10)
𝑚1 𝑙1 (𝜑̈ 1 𝑐𝑜𝑠𝜑1 − 𝜑̇ 12 𝑠𝑖𝑛𝜑1 ) 𝑥5 − 𝜑2𝑟𝑒𝑓 𝜑2 − 𝜑2𝑟𝑒𝑓
⇒ 𝑥̈ =
𝑚1 + 𝑚2 + 𝑀
𝑚2 𝑙2 (𝜑̈ 2 𝑐𝑜𝑠𝜑2 − 𝜑̇ 22 𝑠𝑖𝑛𝜑2 ) 𝑢 where, 𝑞𝑟𝑒𝑓 = [𝑥𝑟𝑒𝑓 𝜑1𝑟𝑒𝑓 𝜑2𝑟𝑒𝑓 ]𝑇 , 𝑥𝑟𝑒𝑓 is the desired
+ + position and 𝜑1𝑟𝑒𝑓 , 𝜑2𝑟𝑒𝑓 are the desired rotation angles.
𝑚1 + 𝑚2 + 𝑀 𝑚1 + 𝑚2 + 𝑀
The control law is designed as follows:
Partial derivative with respect to variable 𝜑1 we get:
1
𝑢= [𝑞̈ − 𝐹(𝑋) + 𝐾 𝑇 𝔢] (11)
𝑑 𝜕𝐿 𝜕𝐿 𝜕𝑃 𝐺(𝑋) 𝑟𝑒𝑓
( )− + =0
𝑑𝑡 𝜕𝜑̇ 1 𝜕𝜑1 𝜕𝜑1
⇒ 𝑚1 𝑙1 (𝑔𝑠𝑖𝑛𝜑1 − 𝑐𝑜𝑠𝜑1 𝑥̈ + 𝑙1 𝜑̈ 1 ) = 0 where, 𝑒 = [𝑒 𝑒̇ ]𝑇 , 𝐾 = [𝐾𝑝 𝐾𝑑 ]𝑇 .
(5) Substituting Eq. (11) into (9), we get the error equation of
(𝑔𝑠𝑖𝑛𝜑1 − 𝑐𝑜𝑠𝜑1 𝑥̈ + 𝑙1 𝜑̈ 1 ) = 0
𝑐𝑜𝑠𝜑1 𝑥̈ − 𝑔𝑠𝑖𝑛𝜑1 the system as follows:
⇒ 𝜑̈ 1 =
𝑙1
𝑒̈ + 𝐾𝑑 𝑒̇ + 𝐾𝑝 𝑒 = 0 (12)
Similarly, partial derivative with respect to variable 𝜑2 we
have: We have the characteristic equation of the system as:

𝑑 𝜕𝐿 𝜕𝐿 𝜕𝑃 𝑠 2 + 𝐾𝑑 𝑠 + 𝐾𝑝 = 0 (13)
( )− + =0
𝑑𝑡 𝜕𝜑̇ 2 𝜕𝜑2 𝜕𝜑2
⇒ 𝑚2 𝑙2 (𝑔𝑠𝑖𝑛𝜑2 − 𝑐𝑜𝑠𝜑2 𝑥̈ + 𝑙2 𝜑̈ 2 ) = 0 where, 𝐾𝑝 and 𝐾𝑑 are designed so that the characteristic Eq.
(6) (13) has a solution on the left side of the complex plane,
(𝑔𝑠𝑖𝑛𝜑2 − 𝑐𝑜𝑠𝜑2 𝑥̈ + 𝑙2 𝜑̈ 2 ) = 0
𝑐𝑜𝑠𝜑2 𝑥̈ − 𝑔𝑠𝑖𝑛𝜑2 meaning when 𝑡 → ∞ then 𝑒(𝑡) → 0 and 𝑒̇ (𝑡) → 0.
⇒ 𝜑̈ 2 = To find the control law in Eq. (10), we need to know the
𝑙2
function 𝐹(𝑋) . However, 𝐹(𝑋) is an uncertain nonlinear
Substituting Eqs. (6) and (5) into Eq. (4) we have: function. Therefore, the authors use the RBFNN to
approximate 𝐹(𝑋).
−𝑚1 𝑔𝑠𝑖𝑛𝜑1 𝑐𝑜𝑠𝜑1 − 𝑚2 𝑔𝑠𝑖𝑛𝜑2 𝑐𝑜𝑠𝜑2 The RBFNN is described as follows:
𝑥̈ =
𝑚1 𝑠𝑖𝑛2 (𝜑1 ) + 𝑚2 𝑠𝑖𝑛2 (𝜑2 ) + 𝑀 2
−𝑚1 𝑙1 𝜑̇ 12 𝑠𝑖𝑛𝜑1 − 𝑚2 𝑙2 𝜑̇ 22 𝑠𝑖𝑛𝜑2 ‖𝔢 − 𝑐𝑖𝑗 ‖
+ (7) ℎ𝑗 = 𝑒𝑥𝑝 ( ) (14)
𝑚1 𝑠𝑖𝑛2 (𝜑1 ) + 𝑚2 𝑠𝑖𝑛2 (𝜑2 ) + 𝑀 𝑏𝑗2
𝑢
+
𝑚1 𝑠𝑖𝑛 (𝜑1 ) + 𝑚2 𝑠𝑖𝑛2 (𝜑2 ) + 𝑀
2
𝐹(𝑋) = 𝑊 𝑇 ℎ(𝔢) + 𝜀 (15)
Let 𝑋 = [𝑥1 𝑥2 𝑥3 𝑥4 𝑥5 𝑥6 ]𝑇 . Then the state where, 𝑒 is the input vector of the neural network;
space equation of the nonlinear system is:

427
 𝑖 represents the number of input layer neurons; 𝑊 ∗ = 𝑎𝑟𝑔𝑚𝑖𝑛(𝑠𝑢𝑝|𝐹̂ (𝑋) − 𝐹(𝑋)|) (22)
𝑗 represents the number of hidden layer neurons;
ℎ = [ℎ1 , ℎ2 , ℎ3 , . . . , ℎ𝑛 ]𝑇 represents the output of the hidden Set the model approximation error calculated by the optimal
layer; RBFNN weights to be:
𝑐𝑖𝑗 represents the coordinate vector of the center point of the
Gaussian basis function; 𝐹̃ (𝑋) = 𝐹̂ (𝑋/𝑊 ∗ ) − 𝐹(𝑋) (23)
𝑏𝑗 represents the width of the Gaussian function;
𝑊 is the weight value of the neural network; ⇒ 𝐹̃ (𝑋) − 𝐹̂ (𝑋/𝑊 ∗ ) = −𝐹(𝑋) (24)
𝜀 is the approximation error of the neural network.
Then the function 𝐹(𝑋) is approximated as follows: Substituting Eq. (24) into (21) we have:

𝐹̂ (𝑋) = 𝑊
̂ 𝑇 ℎ(𝑒) (16) 𝑒̇ = 𝐴𝔢 + 𝐵(𝐹̂ (𝑋) + 𝐹̃ (𝑋) − 𝐹̂ (𝑋/𝑊 ∗ )) (25)

where, 𝐹̂ (𝑋) is the output of the neural network; 𝑊̂ is an ⇒ 𝑒̇ = 𝐴𝔢 + 𝐵(𝐹̂ (𝑋) + 𝐹̃ (𝑋) − 𝐹̂ (𝑋/𝑊 ∗ ))
approximation of the neural network weight 𝑊; input vector = 𝐴𝔢 + 𝐵[(𝐹̂ (𝑋) − 𝐹̂ (𝑋/𝑊 ∗ ))] + 𝐹̃ (𝑋) (26)
of the neural network 𝑒 = [𝑒 𝑒̇ ]𝑇 . ̂ − 𝑊 ∗ )ℎ(𝔢) + 𝐹̃ (𝑋))
At this point, the control law (11) will become: = 𝐴𝔢 + 𝐵 ((𝑊

1 ̂ − 𝑊 ∗ )𝑇 ℎ(𝔢) + 𝐹̃ (𝑋)) , then Eq. (26) is

Set 𝑁 = 𝐵 ((𝑊
𝑢= [𝑞̈ − 𝐹̂ (𝑋) + 𝐾 𝑇 𝑒] (17)
𝐺(𝑋) 𝑟𝑒𝑓 rewritten as:
Substituting Eq. (16) into (17), the system control law is 𝑒̇ = 𝐴𝔢 + 𝑁 (27)
expressed as follows:
Choose the Lyapunov function:
1
𝑢= [𝑞̈ ̂ 𝑇 ℎ(𝔢) + 𝐾 𝑇 𝑒]
−𝑊 (18)
𝐺(𝑋) 𝑟𝑒𝑓 1 1
𝐿 = 𝑒 𝑇 𝑃𝑒 + ̂ − 𝑊 ∗ )𝑇 (𝑊
(𝑊 ̂ − 𝑊 ∗) (28)
2 2𝛾
The neural network weight adaptation law is chosen as:
The matrix P must satisfy the Lyapunov equation:
̂̇ = −𝛾𝔢𝑇 𝑃𝑏ℎ(𝑒)
𝑊 (19)
𝑃𝐴 + 𝐴𝑇 𝑃 = −𝑄 (29)
where, 𝑃 is a positive definite symmetric matrix; 𝛾 > 0.
The block diagram of the closed-loop control system is 0 𝐼 0
shown in Figure 3. where, 𝐴 = [−𝐾 −𝐾𝑑 ]; 𝐵 = [ 𝐼 ]; 𝑄 ≥ 0.
𝑝

1 1 1
𝐿̇ = 𝑒̇ 𝑇 𝑃𝑒 + 𝔢𝑇 𝑃𝑒̇ + (𝑊 ̂ − 𝑊 ∗ )𝑇 𝑊
̂̇
2 2 𝛾
1 1
= (𝑒 𝑇 𝐴𝑇 + 𝑁 𝑇 )𝑃𝑒 + 𝑒 𝑇 𝑃(𝐴𝐸 + 𝑁)
2 2
1
+ (𝑊 ̂ − 𝑊 )𝑇 𝑊
∗ ̂̇
𝛾
1 1 1
= 𝑒 𝑇 (𝐴𝑇 𝑃 + 𝑃𝐴)𝑒 + 𝑁 𝑇 𝑃𝑒 + 𝑒 𝑇 𝑃𝑁 (30)
2 2 2
1 ∗ 𝑇 ̂̇
̂
+ (𝑊 − 𝑊 ) 𝑊
𝛾
1 𝑇 1
= − 𝑒 𝑄𝑒 + (𝑁 𝑇 𝑃𝑒 + 𝑒 𝑇 𝑃𝑁)
2 2
1 𝑇 ̇
̂
+ (𝑊 − 𝑊 ∗ ) 𝑊 ̂
Figure 3. Block diagram of closed loop control system 𝛾

3.2 Stability analysis 1 1

̂ − 𝑊 ∗ )𝑇 𝑊
⇒ 𝐿̇ = − 𝑒 𝑇 𝑄𝑒 + 𝑒 𝑇 𝑃𝑁 + (𝑊 ̂̇ (31)
2 𝛾
Substituting Eq. (11) into (9) we have:
Substituting the 𝑁 value set above into Eq. (31), we get:
𝑒̈ = (𝐹̂ (𝑋) − 𝐹(𝑋)) − 𝐾 𝑇 𝑒 (20)
1
𝐿̇ = − 𝑒 𝑇 𝑄𝔢 + 𝑒 𝑇 𝑃𝐵 ((𝑊 ̂ − 𝑊 ∗ )𝑇 ℎ(𝑒) + 𝐹̃ (𝑋))
Transforming Eq. (20) we have: 2
1
+ (𝑊 ̂ − 𝑊 ∗ )𝑇 𝑊̂̇
𝑒̇ = 𝐴𝔢 + 𝐵(𝐹̂ (𝑋) − 𝐹(𝑋)) (21) 𝛾
1 (32)
= − 𝑒 𝑇 𝑄𝑒 + 𝑒 𝑇 𝑃𝐵(𝑊 ̂ − 𝑊 ∗ )𝑇 ℎ(𝑒) + 𝑒 𝑇 𝑃𝐵𝐹̃ (𝑋)
Assuming W is the optimal weight of RBFNN, then W is 2
calculated according to the following equation. 1
+ (𝑊 ̂ − 𝑊 ∗ )𝑇 𝑊
̂̇
𝛾

428
 1
𝐿̇ = − 𝑒 𝑇 𝑄𝑒 + 𝑒 𝑇 𝑃𝐵𝐹̃ (𝑋)
2
1
+ (𝑊 ̂ − 𝑊 ∗ )𝑇 (𝑊
̂̇ + 𝛾𝑒 𝑇 𝑃𝐵ℎ(𝑒))
𝛾

With the adaptation law Eq. (19) the Lyapunov function Figure 7. Scenario 1 control input when 𝑥 = 5
becomes:
The initial and desired positions of the system are:
1
𝐿̇ = − 𝑒 𝑇 𝑄𝑒 + 𝑒 𝑇 𝑃𝐵𝐹̃ (𝑋) ≤ 0 (33)
2 𝑋0 = [5 0 𝜋⁄4 0 − 𝜋⁄4 0]𝑇
{
𝑋𝑟𝑒𝑓 = [10 0 0 0 0 0]𝑇
Complete proof.

4. SIMULATION VERIFICATION

To verify the effectiveness of the proposed algorithm. We

perform system simulation on Matlab/Simulink software with
different scenarios. The selected RBF neural network has 6 Figure 8. Scenario 1 cart position when 𝑥 = 10
input layer neurons, 13 hidden layer neurons and 1 output layer
neuron. The hidden layer parameters 𝑐𝑖𝑗 and 𝑏𝑗 of the neural
network are: [-3, -2.5, -2, -1.5, -1, -0.5, 0, 0.5, 1, 1.5, 2, 2.5, 3]
and 2.
The parameters of the selected controller are: 𝐾𝑝 = 30 ,
𝐾𝑑 = 50, 𝛾 = 500. The initial neural network weight values
are 0. Figure 9. Scenario 1 swing angle of load 1 when 𝜑1 = 𝜋⁄4
Scenario 1: The parameters of the overhead crane system
are: 𝑀 = 1000 (𝑘𝑔) , 𝑚1 = 50 (𝑘𝑔) , 𝑚2 = 100 (𝑘𝑔) , 𝑙1 =
10 (𝑚), 𝑙2 = 20 (𝑚).
The initial and desired positions of the system are:

𝑋0 = [3 0 𝜋⁄3 0 − 𝜋⁄3 0]𝑇

{ Figure 10. Scenario 1 swing angle of load 2 when 𝜑2 =
𝑋𝑟𝑒𝑓 = [5 0 0 0 0 0]𝑇
− 𝜋 ⁄4
The simulation results from Figures 4-19 show that the
proposed controller stabilizes all responses of the overhead
crane system with good control performance. The responses of
the cart's position and the cable's swing angle always track the
set value with a small response time.
Figure 11. Scenario 1 control input when 𝑥 = 10

Scenario 2: The parameters of the overhead crane system

are: 𝑀 = 100 (𝑘𝑔) , 𝑚1 = 5 (𝑘𝑔) , 𝑚2 = 10 (𝑘𝑔) , 𝑙1 =
1 (𝑚), 𝑙2 = 2 (𝑚).
The initial and desired positions of the system are:
Figure 4. Scenario 1 cart position when 𝑥 = 5
𝑋0 = [3 0 𝜋⁄3 0 − 𝜋⁄3 0]𝑇
𝑋𝑟𝑒𝑓 = [5 0 𝜋⁄3 0 − 𝜋⁄3 0]𝑇

Figure 5. Scenario 1 swing angle of load 1 when 𝜑1 = 𝜋⁄3

Figure 12. Scenario 2 cart position when 𝑥 = 5

Figure 6. Scenario 1 swing angle of load 2 when 𝜑2 =

− 𝜋 ⁄3 Figure 13. Scenario 2 swing angle of load 1 when 𝜑1 = 𝜋⁄3

429
 network (overshoot 0.217%) show that the system quality is
improved in terms of overshoot compared to the fuzzy logic
controller [24] (overshoot 0.227%). In addition, fuzzy logic
controllers have the disadvantage of depending on the expert's
understanding of the object. Specifically, the operating range
Figure 14. Scenario 2 swing angle of load 2 when 𝜑2 = of input and output variables needs to be surveyed first
− 𝜋 ⁄3 because they greatly affect the quality of the system when
using a fuzzy logic controller.

5. CONCLUSIONS

In this study, the authors have transformed the dynamic

Figure 15. Scenario 2 control input when 𝑥 = 5 equation of the overhead crane system in accordance with the
proposed control method. The proposed controller is stable for
The initial and desired positions of the system are: all outputs of the system: the position of the cart and the
rotation angle of the cable accurately track the desired values,
𝑋0 = [5 0 𝜋⁄4 0 − 𝜋⁄4 0]𝑇 the vibration of the goods is completely suppressed. The
{ adaptability of the system is guaranteed when changing the
𝑋𝑟𝑒𝑓 = [10 0 0 0 0 0]𝑇
system parameters to suit the reality when the crane operates.
Finally, in the near future the author will conduct experiments
to check the simulation results.
A potential future research direction is to install the
algorithm on real application systems. In addition, we need to
integrate additional obstacle avoidance features into the
controller to ensure safety for people and goods during
Figure 16. Scenario 2 cart position when 𝑥 = 10 operation.

ACKNOWLEDGMENT

This study was supported by University of Economics -

Technology for Industries (UNETI), Ha Noi – Viet Nam;
Website: http://www.uneti.edu.vn/.
Figure 17. Scenario 2 swing angle of load 1 when 𝜑1 = 𝜋⁄4

REFERENCES

[1] Fantuzzi, N., Rustico, A., Formenti, M., Ferreira, A.J.

(2022). 3D active dynamic actuation model for offshore
cranes. Computer-Aided Civil and Infrastructure
Figure 18. Scenario 2 swing angle of load 2 when 𝜑2 = Engineering, 37(7): 864-877.
− 𝜋 ⁄4 https://doi.org/10.1111/mice.12690
[2] Vaughan, J., Karajgikar, A., Singhose, W. (2011). A
study of crane operator performance comparing PD-
control and input shaping. In Proceedings of the 2011
American Control Conference, San Francisco, CA, USA,
pp. 545-550.
https://doi.org/10.1109/ACC.2011.5991506
[3] Yang, B., Xiong, B. (2010). Application of LQR
Figure 19. Scenario 2 control input when 𝑥 = 10 techniques to the anti-sway controller of overhead crane.
Advanced Materials Research, 139: 1933-1936.
The simulation results from Figures 4-19 show that the https://doi.org/10.4028/www.scientific.net/amr.139-
proposed controller stabilizes all responses of the overhead 141.1933
crane system with good control performance. The responses of [4] Jafari, J., Ghazal, M., Nazemizadeh, M. (2014). A LQR
the cart's position and the cable's swing angle always track the optimal method to control the position of an overhead
set value with a small response time. crane. IAES International Journal of Robotics and
The adaptability of the controller is studied in relation to Automation, 3(4): 252.
changes in the parameters of the overhead crane system. [5] Sun, N., Fang, Y., Chen, H. (2015). Adaptive antiswing
Usually the volume of the load is very diverse and depends on control for cranes in the presence of rail length
each individual operating condition. By varying the mass of constraints and uncertainties. Nonlinear Dynamics, 81(1-
the load and the cable length, we obtain the system response 2), 41-51. https://doi.org/10.1007/s11071-015-1971-y
as shown in Figures 12-19. Note that the controller parameters [6] Ouyang, H., Tian, Z., Yu, L., Zhang, G. (2020). Adaptive
remain the same in the two scenarios. tracking controller design for double-pendulum tower
The results obtained with the controller using RBF neural

430
 cranes. Mechanism and Machine Theory, 153: 103980. [19] Sun, Z., Bi, Y., Zhao, X., Sun, Z., Ying, C., Tan, S.
https://doi.org/10.1016/j.mechmachtheory.2020.103980 (2018). Type-2 fuzzy sliding mode anti-swing controller
[7] Abdullahi, A.M., Mohamed, Z., Selamat, H., Pota, H.R., design and optimization for overhead crane. IEEE
Abidin, M.Z., Ismail, F.S., Haruna, A. (2018). Adaptive Access, 6: 51931-51938.
output-based command shaping for sway control of a 3D https://doi.org/10.1109/access.2018.2869217
overhead crane with payload hoisting and wind [20] Leite, D., Aguiar, C., Pereira, D., Souza, G., Škrjanc, I.
disturbance. Mechanical Systems and Signal Processing, (2019). Nonlinear fuzzy state-space modeling and lmi
98: 157-172. fuzzy control of overhead cranes. In 2019 IEEE
https://doi.org/10.1016/j.ymssp.2017.04.034 International Conference on Fuzzy Systems (FUZZ-
[8] Sun, N., Yang, T., Chen, H., Fang, Y., Qian, Y. (2017). IEEE), New Orleans, LA, USA, pp. 1-6.
Adaptive anti-swing and positioning control for 4-DOF https://doi.org/10.1109/FUZZ-IEEE.2019.8858968
rotary cranes subject to uncertain/unknown parameters [21] Chen, Q., Cheng, W., Gao, L., Fottner, J. (2021). A pure
with hardware experiments. IEEE Transactions on neural network controller for double-pendulum crane
Systems, Man, and Cybernetics: Systems, 49(7): 1309- anti-sway control: Based on Lyapunov stability theory.
1321. https://doi.org/10.1109/tsmc.2017.2765183 Asian Journal of Control, 23(1): 387-398.
[9] Ouyang, H., Wang, J., Zhang, G., Mei, L., Deng, X. https://doi.org/10.1002/asjc.2226
(2019). Novel adaptive hierarchical sliding mode control [22] Yang, T., Sun, N., Chen, H., Fang, Y. (2019). Neural
for trajectory tracking and load sway rejection in double- network-based adaptive antiswing control of an
pendulum overhead cranes. IEEE Access, 7: 10353- underactuated ship-mounted crane with roll motions and
10361. https://doi.org/10.1109/access.2019.2891793 input dead zones. IEEE Transactions on Neural
[10] Zhang, M., Ma, X., Rong, X., Tian, X., Li, Y. (2016). Networks and Learning Systems, 31(3): 901-914.
Adaptive tracking control for double-pendulum overhead https://doi.org/10.1109/tnnls.2019.2910580
cranes subject to tracking error limitation, parametric [23] Qiu, Y., Cai, B., Cui, G. (2020). Research on suspension
uncertainties and external disturbances. Mechanical control strategy based on adaptive RBF neural network
Systems and Signal Processing, 76: 15-32. for MVAWT. In 2020 Chinese Automation Congress
https://doi.org/10.1016/j.ymssp.2016.02.013 (CAC), Shanghai, China, pp. 1029-1034.
[11] He, W., Zhang, S., Ge, S.S. (2013). Adaptive control of https://doi.org/10.1109/CAC51589.2020.9326947
a flexible crane system with the boundary output [24] Esleman, E.A., Önal, G., Kalyoncu, M. (2021). Optimal
constraint. IEEE Transactions on Industrial Electronics, PID and fuzzy logic based position controller design of
61(8): 4126-4133. an overhead crane using the Bees Algorithm. SN Applied
https://doi.org/10.1109/tie.2013.2288200 Sciences, 3(10): 811. https://doi.org/10.1007/s42452-
[12] Wang, Y., Chen, J., Yan, F., Zhu, K., Chen, B. (2019). 021-04793-0
Adaptive super-twisting fractional-order nonsingular
terminal sliding mode control of cable-driven
manipulators. ISA transactions, 86: 163-180. NOMENCLATURE
https://doi.org/10.1016/j.isatra.2018.11.009
[13] Li, X., Geng, Z., Peng, X. (2018). A double-sliding 𝑀 Denotes the mass of the cart, kg
surface based control method for underactuated cranes. 𝑚1 Mass of payload 1, kg
In 2018 37th Chinese Control Conference (CCC), 𝑙1 Cable length between the center of the cart and the
Wuhan, China, pp. 2648-2652. payload 1, m
https://doi.org/10.23919/ChiCC.2018.8483857 𝑚2 Mass of payload 2, kg
[14] Yang, C., Du, C., Liao, L. (2022). Design and 𝑙2 Cable length between the center of the cart and the
implementation of finite time sliding mode controller for payload 2, m
fuzzy overhead crane system. ISA Transactions, 124: 𝑔 Gravitational constant, m/s2
374-385. https://doi.org/10.1016/j.isatra.2019.11.037 𝔢 The input vector of the neural network
[15] Wang, T., Tan, N., Zhang, X., et al. (2021). A time- 𝑖 Represents the number of input layer neurons
varying sliding mode control method for distributed- 𝑗 Represents the number of hidden layer neurons
mass double pendulum bridge crane with variable ℎ Represents the output of the hidden layer
parameters. IEEE Access, 9: 75981-75992. 𝑐𝑖𝑗 Represents the coordinate vector of the center point
https://doi.org/10.1109/access.2021.3079303 of the Gaussian basis function
[16] Wu, X., Xu, K., Lei, M., He, X. (2020). Disturbance- 𝑏𝑗 Represents the width of the Gaussian function
compensation-based continuous sliding mode control for 𝑊 The weight value of the neural network
overhead cranes with disturbances. IEEE Transactions 𝑋 State space variables
on Automation Science and Engineering, 17(4): 2182-
2189. https://doi.org/10.1109/tase.2020.3015870 Greek symbols
[17] Chen, H., Fang, Y., Sun, N. (2016). A swing constraint
guaranteed MPC algorithm for underactuated overhead 𝜑1 Payload 1 swing angle, rad
cranes. IEEE/ASME Transactions on Mechatronics, 𝜑2 Payload 2 swing angle, rad
21(5): 2543-2555. 𝜀 The approximation error of the neural network
https://doi.org/10.1109/tmech.2016.2558202
[18] Wu, Z., Xia, X., Zhu, B. (2015). Model predictive control Subscripts
for improving operational efficiency of overhead cranes.
Nonlinear Dynamics, 79: 2639-2657. ref reference
https://doi.org/10.1007/s11071-014-1837-8

431