Available online at www.sciencedirect.

com

ScienceDirect
Advances in Space Research 72 (2023) 3645–3665
www.elsevier.com/locate/asr

Dynamic modeling and tracking control strategy for flexible

telescopic space manipulator based on neural network compensation
Dongyang Shang a, Xiaopeng Li a,⇑, Meng Yin b, Jiaqi Liu a, Sainan Zhou a
a
School of Mechanical Engineering and Automation, Northeastern University, Shenyang 110819, PR China
b
Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China

Received 11 March 2023; received in revised form 18 June 2023; accepted 18 July 2023
Available online 24 July 2023

Abstract

Flexible telescopic space manipulator (FTSM) is playing an increasingly important role in assisting astronauts to accomplish space
missions. The telescopic structure leads to time-varying dynamical parameters of the FTSM, thereby reducing the tracking control accu-
racy. At the same time, the lightweight and large length-diameter ratios are very likely to cause the vibration phenomenon of FTSM
during the motion. To solve the above problems, in the dynamics modeling, this paper uses the assumed modal method (AMM) and
Lagrange principle to establish the dynamics model considering two-dimensional deformation and disturbance torque. Furthermore,
the dynamics models are simplified and derived by ignoring the nonlinear terms (INTs). Besides, the influence of the simplified dynamics
models on the accuracy of FTSM’s deformation is analyzed by simulation. It is found that the INTs simplified dynamics model has
higher modeling accuracy. Based on the INTs simplification model, the control law is designed. More importantly, the RBF neural net-
work is used to identify and compensate for the time-varying terms and disturbance torque in the FTSM. Then the sliding mode control
strategy is proposed by the hyperbolic tangent function as the approximation rate. Finally, the effectiveness of the RBF neural network
compensated sliding mode control strategy is illustrated by simulation and ground control experiments of the FTSM.
Ó 2023 COSPAR. Published by Elsevier B.V. All rights reserved.

Keywords: Flexible telescopic space manipulator; Neural network compensation; Dynamic modeling; Tracking control strategy

1. Introduction vice life of space equipment and save costs, many space
agencies have carried out research into on-orbit services
With the rapid development of the space industry, the (Misra and Bai, 2017; Jia and Shan, 2020). The Japan
exploration of space has become more and more in- Aerospace Exploration Agency (JAXA) has developed
depth. More and more spacecraft, space stations, satellites, the Engineering Test Satellite and completed a successful
and other space equipment are placed in orbit (Wang et al. mission of assembling a manipulator on a spacecraft
2020). Some large spacecraft can only be launched sepa- (Kaori et al. 2023). The on-orbit services missions include
rately and assembled in orbit due to the limitations of the the assembly and maintenance of space stations (Xiao
means of transport. Some spacecraft and satellites have et al. 2020; Angelini and Costa, 2002), the refueling and
to break down before they reach their life expectancy due upgrading of satellites (Ge et al. 2020), the assisted docking
to the deterioration of some parts or the depletion of fuel of spacecraft, and the clearing of space debris (Lu et al.
(Hastings et al. 2016; Ding et al. 2020). To increase the ser- 2019; Sharma and Rout, 2022). Because of the harsh envi-
ronment of strong radiation and a large range of tempera-
tures in space, astronauts cannot work in space for a long
⇑ Corresponding author.
time. But space manipulators have become an effective tool
E-mail address: [email protected] (X. Li).

https://doi.org/10.1016/j.asr.2023.07.038
0273-1177/Ó 2023 COSPAR. Published by Elsevier B.V. All rights reserved.
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Nomenclature

h FTSM rotation angle (rad) l Telescopic flexible link length (m)

M Tip payload mass (kg) q Linear density (kg/m)
Tf Disturbance torque (Nm) cx(x, t) Lateral deformation
x Abscissa in the XlOlYl coordinate system XmOmYm Coordinate system attached to the motor
cy(x, t) Longitudinal deformation shaft
XOY Inertial reference frame n1(t) First-order modal coordinate
XlOlYl Coordinate system attached to the flexible link a1 First-order modal function characteristic root
d1(x) First-order modal function c 0, c 1, c 2
EI Bending stiffness (Nm2) Parameters in LuGre friction model
z Internal friction state Fc, Fs Coulomb and static friction torque
hs Stribeck speed q Displacement vector
M(q) Inertia matrix C(q) Damping matrix
K(q) Stiffness matrix D Disturbance
u Input vector x RBF neural network input
^
e Error vector q Desired displacement vector
^
h Desired flexible manipulator rotation angle ^n1 Desired first-order modal coordinate
ci Gaussian function center vector ti Normalizing constant
h^ j Hidden nodes’ output w Neural networks weight coefficients
f ðxÞ Neural network output Z Positive matrix in sliding mode function
v Sliding mode function dN Parameter in control law
f Uncertain term vector w* Optimal weight coefficient
_
d Infinitesimal error vector w Weight coefficients error
j Coefficient in the hyperbolic tangent function Kv Positive matrix in control law
Tm Input torque (Nm)
m Flexible link mass (kg)

for on-orbit services missions (Lavı́n-Delgado et al. 2022) grasping tasks. The grasping mass change will cause the
due to their high flexibility, high stability, and numerous rotational speed fluctuation which affects the FTSM con-
functions. Many scientific research institutions have pro- trol accuracy. The FTSM control accuracy is affected by
posed space manipulators and applied them to space envi- the accuracy of the dynamic model. Therefore, the FTSM
ronments. The six degrees-freedom space manipulator research needs to propose an accurate dynamic model
Rocket Technology Experiment (ROTEX) developed by and design a control law to compensate for dynamic
the German Aerospace Centre is used to complete on- uncertainties.
orbit assembly tasks (Hirzinger et al. 1998). Besides, the The FTSM is a complex nonlinear system with dynamic
seven degrees-of-freedom redundant manipulator called parameters and time-varying characteristics. The premise
the Space Station Remote Manipulator System (SSRMS) and basis for high-accuracy tracking control is an accurate
is currently used on the International Space Station FTSM dynamic model. The dynamic modeling method of
(Zhao et al. 2023). The traditional space manipulator is the FTSM is similar to that of flexible manipulators. To
composed of rigid structures. To solve the disadvantages establish the dynamic model of flexible manipulators,
of traditional rigid space manipulators with heavy weight, scholars have conducted a lot of research (Shafei and
high energy consumption, and low speed, scholars have Shafei, 2020). Most scholars equivalence the flexible
proposed flexible manipulators to solve this problem. manipulators to the beam model. Thus, the Euler-
Shang et al. (2022a) proposed a fixed-length flexible space Bernoulli beam theory and the Timoshenko beam theory
manipulator to capture aircraft. Similarly, this paper pro- are proposed to establish the beam dynamic models
poses a flexible telescopic space manipulator (FTSM) with (Korayem et al. 2014; Yang et al. 2022). Korayem et al.
a larger range of motion. Due to of characteristics of light- (2016) established an n-viscoelastic-link robotic manipula-
ness and large span, flexible space manipulators will gener- tors dynamics model based on the Timoshenko beam the-
ate vibration during motion. With the motion of the ory. With scholars’ in-depth study of flexible beam
telescopic arm, the parameters such as the length and den- modeling, it has been found that the dynamic equations
sity of the FTSM have time-varying characteristics which considering both transverse and longitudinal deformation
increase the difficulty of control. Moreover, the grasping can more accurately describe the dynamic characteristics
mass of the FTSM is different when performing different of flexible manipulators (Yin et al. 2014). The flexible

3646
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

deformation of the FTSM is a nonlinear factor. How to reduce the influence of time-varying dynamic parameters
accurately describe the deformation is a difficult part of on space manipulators. The sliding mode control strategy
dynamic modeling. Scholars usually use the AMM is proposed to achieve accurate control of systems with
(Zhang 2009) and finite element (FE) method (Korayem variable parameters (Kayastha et al. 2023). Zhang et al.
et al. 2008) to describe flexible deformation. Mehrez and (2021) used the sliding mode controller to efficiently
El-Badawy (2010) established the dynamic model of improve the spacecraft hovering control accuracy. To solve
flexible-link manipulators based on the AMM. My et al. the problem of unstable control law in the sliding mode
(2019) described the deformation of the multi-link flexible control strategy, some scholars introduced the disturbance
manipulator arm linkage through the FE method. The observer (Yin et al. 2019) or intelligent control strategies
FE method is often used to describe flexible structures with (Yao 2021a) into the sliding mode controller. Wu et al.
large deformations and complex boundary cases. The (2020) proposed an improved sliding mode control scheme
FTSM presented in this paper is a small deformation prob- based on a fixed time disturbance observer, thereby reduc-
lem that has a regular shape. Therefore, the AMM can be ing the spacecraft vibration with coupled disturbances. Jia
used to describe flexible deformation (Aghajari et al. 2021). and Shan (2019) proposed an adaptive terminal sliding
In addition, the nonlinear disturbance torques will affect mode controller based on neural networks to improve the
the FTSM during rotation (Ahmadizadeh et al. 2021). trajectory tracking accuracy of the space manipulator
Many scholars have made outstanding contributions to arm. The FTSM is a typical dynamic uncertain system, in
the dynamic modeling of space manipulators (Zhao et al. which the dynamic model uncertainty consists of time-
2011; Yang et al. 2015). Usually, the Lagrange equation varying parameters and external disturbance torque.
(Vakil et al. 2008) and the Hamilton equation (Korayem Therefore, the intelligent control strategies such as neural
et al. 2022) can be used to establish the dynamics equations networks can be used to compensate for the dynamic
of space manipulators. Zhan et al. (2022) used the uncertainty to improve the FTSM control accuracy.
Lagrange method to establish the dynamic model of the In this paper, the FTSM dynamics model is established,
flexible joint space manipulator. Hu et al. (2022) estab- taking into account a variety of nonlinear factors. Com-
lished the dynamic model of the free-floating space manip- pared with the dynamic model of Shang et al. (2022b), this
ulator under external disturbance based on the Hamilton paper considers the two-dimensional deformation and the
equation. Establishing an accurate dynamic model is help- grasping mass in dynamic modeling. In addition, the
ful to improve the FTSM tracking control accuracy. There- RBF neural network is used to identify dynamic uncertain-
fore, the coupling effects of flexible structures’ two- ties including the disturbance torque and time-varying
dimensional deformation and nonlinear disturbing torque parameters. In the design of control law, the tanh function
of the FTSM need to be considered in dynamic modeling. is used to reduce the oscillation of input torque. Simulation
The dynamic parameters of the FTSM have time- and ground prototype control experiments show the
varying characteristics due to the length and grasping mass advantages of neural network compensation control strat-
variation. The change of dynamic parameters will lead to egy. The main contributions are the mathematical equa-
an inaccurate FTSM trajectory. The control strategy tions of the FTSM are derived considering various
design of space manipulators also encounters the same nonlinear factors; the dynamic model simplification
problems. For the high-precision control of space manipu- method by removing nonlinear terms is proposed; the con-
lators, many scholars have studied and achieved lots of trol law of the sliding mode controller with neural network
achievements (Zhao et al. 2021). In the early stage of the compensation is designed; the neural networks are used to
space manipulator control strategy research, some scholars identify the dynamic uncertainties of the FTSM.
use the PID control method to control the space manipula- The rest of this paper is arranged as follows: In Section 2,
tors (Nair et al. 2016). Wang et al. (2017) proposed the the FTSM mathematical equations considering various
high-precision attitude control method of the flexible nonlinear factors are derived, and a model simplification
spacecraft based on the PD controller. With the deepening method ignoring nonlinear terms is proposed. In Section 3,
of research, the PID control strategy cannot solve the influ- a sliding mode control strategy based on the RBF neural
ence of external disturbance torque on space manipulators. network compensation is designed. In Section 4, simulation
Based on this, some scholars proposed some self-anti- and ground prototype control experiments illustrate the
disturbance control strategies such as the active distur- advantages of the proposed strategy. Section 5 states the
bance rejection control (ADRC) and the disturbance conclusions.
observer (DOB) (Chen et al. 2018; Yang et al. 2012).
Dong et al. (2019) proposed a switched ADRC law for 2. Problem formulation and preliminaries
the variable structure near space vehicles with external dis-
turbances. Qiao et al. (2019) proposed a nonlinear distur- 2.1. Dynamic modeling of FTSM
bance observer and realized the tracking control of space
manipulator motion under multiple disturbances. In this paper, the FTSM working in orbit is carried out,
Although some self-anti-disturbance control strategies as shown in Fig. 1. Among them, Fig. 1(a) shows the imag-
can solve the problem of external disturbance, they cannot inary prototype of the FTSM. According to Fig. 1(a), the
3647
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig. 1. Schematic diagram of the FTSM; (a) Imaginary prototype of the FTSM; (b) Physical model of the FTSM.

space manipulator is installed on the space station and con- Therefore, the lateral deformation of the FTSM is
sists of flexible telescopic link 1, flexible link 2, end- (Shang et al. 2022b)
actuator, and servo motors. The two links are driven by lðtÞMn1
their respective servo motor. Fig. 1(b) shows the equivalent 0 ¼ cos ða1 lðtÞÞcoshða1 lðtÞÞ þ Aþ1 ð1aÞ
m
physical model of the FTSM, which consists of the flexible
telescopic link 1, a servo motor, and a tip payload. The A ¼ cos ða1 lðtÞÞsinhða1 lðtÞÞ
symbols’ physical significance in Fig. 1(b) is illustrated in  sin ða1 lðtÞÞcoshða1 lðtÞÞ ð1bÞ
the Nomenclature. It is assumed that flexible links in the
sinh ða1 lðtÞÞ þ sin ða1 lðtÞÞ
space rotate one after another to achieve operation tasks. e1 ¼ ð1cÞ
When the flexible telescopic link 1 moves, the flexible link coshða1 lðtÞÞ þ cos ða1 lðtÞÞ
2 is relatively stationary. Therefore, the influence of flexible d1 ðxÞ ¼ sin ða1 xÞ  sinh ða1 xÞ
link 2 can be ignored when studying the dynamics and con-
þ e1 ðcoshða1 xÞ  cos ða1 xÞÞ ð1dÞ
trol of the FTSM. In addition, we investigate the FTM
working in orbit in this paper. Therefore, the length of cx ðx; tÞ ¼ n1 ðtÞd1 ðxÞ ð1eÞ
the FTM is a function of time, which leads to the problem where a1 can be obtained by solving Eq. (1a).
of time-varying parameters in the dynamic modeling and The deformations of the beam in both lateral and longi-
control law design of the FTSM. At the same time, it can tudinal directions are coupled to each other. Then the lon-
be seen from Fig. 1(b) that the end-actuator is equivalent gitudinal deformation of the FTSM can be expressed as
to an ideal particle at the end of the FTSM. The variation (Abe 2009)
of tip payload will change the output characteristics of the rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
FTSM. Besides, the direction of the actuation force of the  2
@cy ðx;tÞ @cx ðx;tÞ
@x
¼ 1  @x
1
flexible telescopic link and the rotation direction of the ð2aÞ
 
FTSM do not interfere with each other. Therefore, the 1 @cx ðx;tÞ
  2 @x
influence of the actuation force of the flexible telescopic
link can be ignored in the process of the FTSM’s dynamics Rx  2
modeling. cy ðx; tÞ ¼  12 0 @cx@sðs;tÞ ds
R  2 ð2bÞ
The deformation of FTSM can be referred to as the 1 x 0
¼  2 0 n1 ðtÞd1 ðsÞ ds
deformation of flexible manipulators. As for the deforma-
tion of the flexible manipulator, it is mostly mathematically
derived from the vibration theory of beams. The deforma- Remark 1. For simplification, the symbols defined below
tion of the beam model during the motion is generally two- are all effective.
dimensional. The lateral deformation and longitudinal
lðtÞ ¼ l; cx ðx; tÞ ¼ cx ; cy ðx; tÞ ¼ cy ; cx ðl; tÞ ¼ cxl ;
deformation constitute the total deformation of the beam
cy ðl; tÞ ¼ cyl ;_cx ðx; tÞ ¼ c_ x ;_cy ðx; tÞ ¼ c_ y ;_cx ðl; tÞ ¼ c_ xl ;-
system. As shown in Fig. 1(b), we consider the two-
dimensional deformation of the FTSM. c_ y ðl; tÞ ¼ c_ yl ;d1 ðxÞ ¼ d1 ;d1 ðlÞ ¼ d1l ;d01 ðxÞ ¼ d01 ;d01 ðlÞ ¼ d01l ;
The FTSM is an infinite-dimensional dynamical system, d0 ðsÞ ¼ d^0 1 ; d00 ðxÞ ¼ d00 ;d00 ðlÞ ¼ d00 ;n1 ðtÞ ¼ n1 ;n€1 ðtÞ ¼ n
1 1 1 1 1l
€1
which is difficult to solve. In this paper, we adopt the
In the FTSM, external disturbance torque is inevitable.
AMM to discretize the continuous system of FTSM. Since
Therefore, the influence of disturbance torque should be
the second and higher-order modes are not easily excited,
considered in the dynamic modeling of the FTSM. Friction
only the first-order mode is retained (Gu et al. 2008).
models are generally used to characterize the effect of exter-
3648
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

nal disturbing torque on the system modeling. Compared The FTM will generate small deformation during the
with other friction models, the LuGre friction model can motion because of the slender structure and lightweight.
describe the mechanical behavior of friction torque more So the potential energy of the FTSM is the elastic potential
clearly. Especially, it has advantages in describing the stage energy of the FTM. Thus, the potential energy of the
from static friction to dynamic friction. For the above rea- FTSM is
sons, the LuGre friction model is used to characterize the Z l Z l
1  00 2 1  00 2
disturbance torque in the FTSM. P ¼ EI cx dx ¼ EI n1 d1 dx ð9Þ
In general, the disturbance torque can be described by 2 0 2 0

the LuGre friction model (Astrom and Canudas-De-Wit, Based on the Lagrange principle, we can get
2008), which is shown as follows. 8  
>
< dtd @T  @T þ @P ¼ Tm  Tf
@ h_ @h @h
T f ¼ c0 z þ c1 z_ þ c2 h_ ð3aÞ   ð10Þ
    >
: dtd @T  @n@T1 ðtÞ þ @n@P1 ðtÞ ¼ 0
@ n_ ðtÞ
z_ ¼ h_  c z h_ =f h_ ð3bÞ
1
0
  h  Substituting the expressions of T and P into Eq. (10)
i
f h_ ¼ F c þ ðF s  F c Þ exp h=h
_ s 2 ð3cÞ yields the dynamic equations of the FTSM that considers
two-dimensional deformation and first-order mode, as
   
Inspection of Fig. 1(b) shows that the position vector R Tm  Tf ¼ € hv00 þ 14 € hn41 þ h_ n_ 1 n31 v11  € hn21 þ 2h_ n_ 1 n1 v12
of the FTM can be written as  
þ €hn21 þ 2h_ n_ 1 n1 v13 þ € n1 v14 þ 12 ð€ n1 n21 þ 2n_ 1 n1 Þv15
2
 
Rx x þ cy
R¼ ¼ Bð h Þ ð4aÞ ð11aÞ
Ry cx  
 0 ¼ €hv14 þ 1€ 2
þ €n1 v13 þ €n1 n2 2n_ 1 n1
2
cos h  sin h 2
hn1 v15 1 þ v11
Bð h Þ ¼ ð4bÞ
 12 h_ n31 v11 þ h_ n1 v12  h_ n1 v13 þ n1 v16
2 2 2
sin h cos h

where B(h) represents the rotational transformation ð11bÞ

matrix. where vi-j is the coefficient in the dynamic equations, which
The kinetic energy of the FTM can be calculated by the are given in Appendix A.
following equation.
Rl T 2.2. Dynamical model simplification
T l ¼ 12 q 0 R_ Rdx_
2  2  2  3
2 _ ð5Þ
R x þ c y þ c h þ _
c 2
þ _
c 2
Eqs. (11a) and (11b) is the dynamics equations of the
¼ 12 q 0 4 5dx
l x x y
    FTSM considering the two-dimensional deformation.
þ 2 x þ cy c_ x  2_cy cx h_ Observation of Eqs. (11a) and (11b) reveals that there are
numerous nonlinear terms in the FTSM’s dynamics equa-
Similarly, the position vector RM of the tip payload is
  tions. The large range of rotational motion and the small
RMx l þ cyl deformation generated during the motion are coupled with
RM ¼ ¼ Bð h Þ ð6Þ each other in the FTSM. Besides, the existence of these
RMy cxl
nonlinear terms makes the dynamics equations of FTSM
The kinetic energy of the tip payload can be given as very complicated, which is not conducive to the control
law design in Section 3. Therefore, the simplification of
T M ¼ 12 M R_ TM R_ M
2  2  2  3 Eqs. (11a) and (11b) is needed. The simplification models
l þ cyl þ c2xl h_ þ c_ 2xl þ c_ 2yl ð7Þ of Eqs. (11a) and (11b) are to consider the lateral deforma-
¼ 12 M 4    
5 tion and the first-order mode, named as 1D1M model, and
þ 2 l þ cyl c_ xl  2_cyl cxl h_ ignore the nonlinear terms, named as INTs model. The fol-
lowing are the simplified dynamics equations of the FTSM.
The kinetic energy of the FTSM consists of the kinetic
energy of the FTM and the tip payload. Then, the total
2.2.1. 1D1M simplification model
kinetic energy of the FTSM from Eqs. (5) and (7) is
Using the 1D1M modeling method, the lateral deforma-
T ¼ Tl þ TM tion of the FTM is
2  2  2  3
2 _ cx ðx; tÞ ¼ n1 ðtÞd1 ðxÞ ð12Þ
R x þ c y þ c h þ _
c 2
þ _
c 2
¼ 12 q 0 4 5dx
l x x y
    The position vector R1 and RM1 at any point on the
þ 2 x þ cy c_ x  2_cy cx h_ ð8Þ
2  2   3 FTM as
2 _2  
l þ c yl þ c h þ c_ 2
þ c_ 2
þ 12 M 4
xl xl yl
5 Rx1 x
    R1 ¼ ¼ Bð h Þ ð13Þ
þ 2 l þ cyl c_ xl  2_cyl cxl h _ Ry1 cx
3649
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

The position vector RM1 of the tip payload as 3.1. Design of sliding mode controller based on neural
  network compensation
RMx1 l
RM1 ¼ ¼ B ð hÞ ð14Þ
RMy1 cxl By simulation analysis in Section 4.1, the INTs model
has high model accuracy and is easy to calculate. So Eq.
The total kinetic energy of the FTSM is (17) is chosen for the design of the control law. Eq. (17)
is rewritten as follows.
T ¼ Tl þ TM  " #    
Rl v00 v14 €h 0 0 h Tf Tm
¼ RT1 R_ 1 dx þ 12 M R_ TM1 R_ M1
_ 1
q 0 þ þ ¼
 2
 2  2 v14 v13 €n1 0 v16 n1 0 0
Rl
¼ 12 q 0 x2 þ ðn1 d1 Þ h_ þ n_ 1 d1 þ 2xn_ 1 d1 h_ dx ð15Þ
2
ð18Þ
  2  2 Based on Eq. (18), the FTSM’s dynamics equations can
þ 1 M x2 þ ðb d1l Þ h_ þ n_ 1 d1l þ 2xn_ 1 d1l h_
2
2 1 be written in this way.
M ðqÞ€q þ C ðqÞq_ þ K ðqÞq þ D ¼ u ð19Þ
Similarly, the total kinetic energy expression of the
FTSM is shown as Eq. (9). where q, M(q), C(q), K(q), D and u are matrixes or vectors,
Combined with Eqs. (9), (10), and (15), the dynamic and the specific expressions are in Appendix B.
equations of the FTSM considering one-dimensional defor- Remark 2. For ease of writing, the following symbol
mation and first-order mode can be obtained as follows. expressions are defined.
 
€ € 2 þ 2h_ n_ 1 n1 v þ €
hv00 þ hn n1 v14 ¼ T m  T f ð16aÞ
1 13

_ ðqÞ ¼ M:
M ðqÞ ¼ M; C ðqÞ ¼ C; K ðqÞ ¼ K; M _
€
hv14 þ €
n1 v13  h_ n1 v13 þ n1 v16 ¼ 0
2
ð16bÞ
In this paper, the RBF is used to approximate the time-
varying terms in the FTSM’s dynamics model. Generally,
2.2.2. INTs simplification model the RBF neural network algorithm can be expressed as
Neglecting the nonlinear terms in Eqs. (11a) and (11b), (Shang et al. 2022c)
the dynamic equations of the FTSM can be rewritten as 8 h i
>
> x ¼ e T
_
e T
^
q T ^_ T ^€q T
q
€
hv00 þ €
n1 v14 ¼ T m  T f ð17aÞ >
>
>
>
>
> e ¼ ^q  q
€
hv14 þ €
n1 v13 þ n1 v16 ¼ 0 ð17bÞ >
>
>
> q ¼ ½ h n1 
>
>
In Section 4.1, we analyze the modeling accuracy of the >
>
< ^q ¼ ^h ^n1
two kinds of simplified models. Comparing and analyzing   ð20Þ
>
> hj ¼ exp ðxci 2t ÞT ðxci Þ
; i ¼ 1; 2;    ; n
the simulation results, it is found that the INTs model >
> 2
>
> i
has high model accuracy. Therefore, the INTs model is >
> T
>
> u ¼ ½ h h2 . . . hn 
used to design the control law. In addition, this paper uses >
>
1
>
> wT ¼ ½wk1 ; wk2 ;    ; wkn 
T
neural networks to compensate for the dynamic uncer- >
>
:
tainty caused by modeling errors and time-varying length, f^ ðxÞ ¼ wT u
thus improving the control accuracy of the FTSM.
The sliding mode function is defined as
3. Design of control law based on neural network v ¼ e_ þ Ze ð21Þ
compensation
where Z is the positive matrix in the sliding mode function.
According to Eqs. (19) and (21), we can get
In this paper, we study the FTSM, whose telescopic
structure causes the length of the FTSM to be time- M v_ ¼ M ð€e þ Z_eÞ
 
varying. It is impossible to obtain an accurate dynamics ¼ M ^€q þ Z_e  M€q
model of the FTSM because of the time-varying length     ð22Þ
of the FTM and the presence of disturbance torque. In Sec- ¼ M ^€q þ Z_e þ C ^q_  ðv  ZeÞ þ Kq þ D  u
tion 3.1, we design a sliding mode control strategy based on
the RBF neural network compensation. The neural net- ¼ u  Cv þ f
work is used to identify and compensate for the time- where f is the uncertainty terms vector in the FTSM’s
varying terms and disturbance torque in the FTSM dynam- dynamics equations. The detailed expression for f is
ics model. Then, in Section 3.2, we analyze the stability and    
robustness of the controller designed in this paper. f ¼ M ^€q þ Z_e þ C ^q_ þ Ze þ Kq þ D ð23Þ

3650
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Assumptions 1. (Shang et al. 2022c). In the ideal case, when The dynamical equations have the property that
the weight coefficients in the neural networks reach the _
vT(M2C)v = 0 (Yao 2021b).
optimal value w*, the RBF neural network output is close Thereby, Eq. (31) can be rewritten as follows.
to the uncertain terms vector. There is still an infinitesimal     
error vector d between the RBF neural network output and V_ 1 ¼ 12 vT M_  2C v  vT K v v þ vT d  dN tanh v
j
 _ 
the uncertain terms vector, as shown in Eq. (24). _T 1 _
þtrw c w þ v u T

   ð32Þ
¼ v K v v þ vT d  dN tanh jv
T
  _ 
_T _
f  f^ ðxÞ ¼ f  wT u ¼ d ð24Þ þtrw c1 w þ vT u
Following Eq. (24), we can obtain Design the adaptive law of the neural network as
_T
f  f^ ðxÞ ¼ d þ wT u  wT u ¼ d þ w u ð25Þ _
_
_
w ¼ cuvT þ kck v kw ð33Þ
where w is the weight coefficients error.
The Lyapunov function is defined as Eq. (26).
1 1 _T 1 _ Lemma 1. (Shang et al. 2022c) When the weight coefficients
V1 ¼ vT Mv þ tr w w ð26Þ are close to the optimal weight coefficients, the weight
2 2 c
By deviation of Eq. (26) in combination with Eq. (22), coefficients will stop changing. Thus, we can get.
we can gain
_T _
1 _ (
_ v_ þ tr w
V_ 1 ¼ vT M v_ þ 12 vT M w
c w_ ¼ 0
    ð27Þ ð34Þ
_ _T __ _
_ T
¼ vT u  Cv þ f þ 12 Mv þ tr w 1c w w_ ¼ w ¼ cuv  kck v kw
The sgn function is used in the traditional sliding mode Based on Eq. (34), we can rewrite Eq. (32) as follows.
function, which causes fluctuations in the input torque. In   
V_ 1 ¼ vT K v v þ vT d  dN tanh jv
order to solve the chattering phenomenon of flexible
_T
manipulators, the hyperbolic tangent function is used þtrw ðc1 ðcuvT þ kck v kwÞ þ uvT Þ
instead of the sgn function. The hyperbolic tangent func- _T
tion is demonstrated as ¼ vT K v v þ vT ðd  dN ðtanhðvÞÞÞ þ trw ðkk v kwÞ
 
 v  ejv  ejv _T
6 vT K v v þ kk v ktrw w  w
_ ð35aÞ
tanh ¼ v v ð28Þ  _ 
j e j þ e j _ 2
6 vT K v v þ kk v k k w kF k w kF  k w kF
where j is the coefficient. The value of j is too small, the  _
 _ 
input torque will fluctuate. If the value of j is too large, 6  K v min k v k2 þ kk v kk w kF k w kF  wmax
the tracking error becomes large. Therefore, the value of
j should be given moderately. k K v k P K v min ð35bÞ

The hyperbolic tangent function’s properties are pre- k w kF 6 wmax ð35cÞ
sented as follows (Aghababa and Akbari, 2012).
v   v    v  If V_ 1 6 0, the following inequalities hold.
     _ 
v tanh ¼ v tanh  ¼ jvjtanh P0 ð29Þ _  2
j j j K v min k v k þ kk w kF k w kF  wmax ¼ k k w kF  wmax =2
 2 ð36aÞ
The control law using the hyperbolic tangent function is k wmax =4 þ K vmin k v k P 0
designed as  2
4K v min k v k P k wmax ð36bÞ
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ffi
^
u ¼ f ðxÞþK v v þ dN tanh jv ð30Þ
  k w kF P wmax =2 þ wmax =4 ð36cÞ
¼ wT u þ K v v þ dN tanh jv
Provided that Eq. (36a)–(36c) is fulfilled, the stability of
where Kv is the parameter matrix; the parameter of the con- the FTSM can be guaranteed.
trol law satisfies dN P k d k. The control block diagram of the FTSM using the neu-
  _T _
_
V_ 1 ¼ vT u  Cv þ f þ 12 Mv_ þ tr w 1c w ral network compensation control strategy is represented in
 ! Fig. 2.
wT u  K v v  dN tanh jv  _T _ 
_
¼v T
þ tr w 1c w
_
Cv þ f þ 12 Mv ð31Þ 4. Results and discussion
_
¼ vT K v v  vT Cv þ 12 vT Mv In this section, simulation analysis and prototype exper-
     
_T __ iments are used to verify the effectiveness of the proposed
þvT d  dN tanh jv þ trw c1 w þ vT u
method of dynamic simplification and control strategies.
3651
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig. 2. Control block diagram of the FTSM using the neural network compensation control strategySubstituting Eq. (30) into Eq. (27) yields.

This section includes two subsections. The first subsection strated in this paper. This paper mainly studies the influ-
considers the effect of disturbing torque on the flexible ence of disturbance torque on the FTSM dynamic output
deformation and the accuracy comparison of simplified characteristic. These dynamic characteristics include flexi-
dynamic models. The other subsection includes the advan- ble deformation, rotation angle, rotation speed, and input
tage of neural network compensation control strategy torque. In this paper, the FTSM dynamic outputs are cal-
through simulation and physical prototype experiments. culated according to the single-link deduction algorithm
proposed by Shang et al. (2022d), as shown in Fig. 3.
4.1. Disturbance torque effect and modeling accuracy The object of this paper is the FTSM, which varies in
comparison length. Therefore, it is necessary to study the FTSM
dynamic output characteristics under different length
This section analyzes the influence of disturbance torque parameters. Three different length conditions are set in
on the dynamic response of the FTSM under different the simulation process. The simulation parameters are set
physical parameters, thereby drawing the conclusion that as EI is 200 Nm2; M is 2 kg; maximum speed is 1 rad/s;
disturbance torque cannot be ignored. More importantly, m is 10 kg. The flexible deformation of the FTSM at differ-
this section mainly carries out simulation experiments ent lengths is shown in Fig. 4.
based on the deformation accuracy comparison of three From Fig. 4(a), it can be observed that the amplitude of
FTSM’s dynamic models. Thus, it is found that the INTs the flexible deformation will increase marginally when the
simplified model has higher modeling accuracy and is con- disturbance torque is taken into account. More impor-
venient to design the control law. tantly, the flexible deformation will be oscillated close to
In order to investigate the effect of disturbance torque the steady state due to the existence of the disturbance tor-
on the FTSM dynamic model, the 2D1M model is demon- que. It can be seen from Fig. 4(b) that at the beginning of
3652
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig. 3. Calculation flow of the FTSM output characteristic.

the simulation time, the disturbance torque has no signifi- 2D1M model. Physical parameters of the FTSM will affect
cant effect on the rotation angle of the FTMS. After taking the flexible deformation. The FTSM parameters mainly
into account the disturbance torque, the disturbance include bending stiffness, tip payload mass, maximum
torque reduces the rotation angle amplitude. From Fig. 4 speed, and length. In order to make the proposed modeling
(c), it can be seen that the rotation speed amplitude de- method have high modeling accuracy under general physi-
creases due to disturbance torque. Besides, the disturbance cal parameters, it is necessary to study the FTSM flexible
torque will cause the rotation speed to fluctuate around the deformation under different physical parameters. Accord-
steady state. As can be seen in Fig. 4(d), the existence of ing to the FTSM flexible deformation simulation process
disturbance torque reduces the input torque amplitude, in Fig. 3, the flexible deformation curves of three dynamic
which is particularly obvious under the condition of small models with different physical parameters can be obtained,
length. However, with the increase in length, the influence as shown in Fig. 5. In the simulation process, the physical
of disturbance torque on input torque decreases. This is parameters of the FTSM are shown in Table 1.
because as the length increases, the inertia of the FTSM At the simulation time, the flexible deformation
increases, and the required driving torque amplitude decreases during 0–5 s and then increases during 5–10 s.
increases. However, the disturbance torque amplitude does The flexible deformation reaches the minimum value at
not change, which is only related to the rotation speed. The 2.5 s and the maximum value at 7.5 s. In the simulation
proportion of disturbance torque in input torque decreases time of 10–20 s, the flexible deformation gradually becomes
with the increase in length. Thus, the influence of distur- stable. As can be seen from Fig. 5 (a), as the bending stiff-
bance torque on the FTSM decreases with the increase in ness increases, the flexible deformation amplitude
length. When the FTSM reaches a steady state, the distur- decreases. On the contrary, it can be seen from Fig. 5(b),
bance torque makes the input torque fluctuate. Therefore, (c), and (d) that, as the tip payload mass, the maximum
it can be explained that the disturbance torque has an speed, and the length increase, the flexible deformation
increased impact on the FTSM dynamic output amplitude increases. According to all the local enlarge-
characteristics. ments in Fig. 5, under all the above physical parameters,
This paper mainly analyzes the modeling accuracy of the flexible deformation curves obtained by the INTs model
2D1M model, the 1D1M model, and the INTs model. are very close to that obtained by the 2D1M model. By
According to the flexible deformation accuracy analysis, comparing the three local magnification figures in Fig. 5
the modeling accuracy of the three simplified methods is (a), it can be seen that the three deformation curves in
compared. Generally, the nonlinear factor is an important the figure gradually coincide with the increase of the bend-
part of modeling accuracy, and the more nonlinear terms ing stiffness. It can be concluded that the modeling accu-
are considered, the modeling accuracy will improve accord- racy of the three modeling methods gradually approaches
ingly. However, the increase of kinds of nonlinear terms with the increase of the bending stiffness. By comparing
will bring a burden to the calculation process. The 2D1M the three local magnification graphs in Fig. 5(b), it can be
model considers most kinds of nonlinear factors. There- seen that with the increase of the tip payload mass, the
fore, the 2D1M model had higher modeling accuracy. In deformation curves obtained by the 1D1M model in the
order to improve the computational efficiency, the INTs figure gradually separate from that obtained by the
model and 1D1M model are proposed by simplifying the 2D1M model. It can be seen that with the increase of the
3653
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig. 4. Flexible deformation under different length considering disturbance torque.

tip payload mass, the modeling accuracy of the 2D1M In summary, according to the simulation analysis in
model and the INTs model is still very close. Similarly, Section 4.1, it can be seen that the disturbance torque
by comparing the local magnification graphs in Fig. 5(c) has a great influence on the FTSM dynamic characteris-
and Fig. 5(d), it can be seen that with the increase of the tics under the condition of small length. Therefore, the
maximum speed and the length, the deformation curves disturbance torque should be fully considered in dynamic
obtained by the 1D1M model and the 2D1M model grad- modeling and control law design. In addition, the INTs
ually separate. This directly indicates that the deformation simplified model has high deformation accuracy under
accuracy of the INTs model is close to that of the 2D1M. all physical parameters. Thus, the INTS model can be
Therefore, compared with the 1D1M model, the INTs used instead of the 2D1M model to design the FTSM
model has higher modeling accuracy. control law.
3654
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig. 5. FTSM flexible deformation under different physical factors.

4.2. Simulation and ground physical prototype control applied to the design of the neural network sliding mode
experiment control (NNSMC) strategy to solve this problem. To verify
the effectiveness of the NNSMC strategy, simulation anal-
The step function is used in the control law of the con- ysis, and ground physical control experiments are carried
ventional sliding mode control (SMC) strategy, which it is out. In the simulation analysis, this paper compares the
easy to cause chatter. In this paper, the tanh function is FTSM output characteristics under different control strate-

3655
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Table 1
Physical parameters of the FTSM.
Condition l (m) EI (Nm2) M (kg) h_ max (rad/s) m(kg)
Bending stiffness 1 (a-i) 2.5 80 2 0.5 10
Bending stiffness 2 (a-ii) 2.5 120 2 0.5 10
Bending stiffness 3 (a-iii) 2.5 160 2 0.5 10
Tip payload mass 1 (b-i) 2.5 200 1.5 1 10
Tip payload mass 2 (b-ii) 2.5 200 2.5 1 10
Tip payload mass 3 (b-iii) 2.5 200 3.5 1 10
Maximum speed 1 (c-i) 2.5 200 2 0.8 10
Maximum speed 2 (c-ii) 2.5 200 2 1.2 10
Maximum speed 3 (c-iii) 2.5 200 2 1.6 10
Length 1 (d-i) 1 200 2 1 10
Length 2 (d-ii) 2 200 2 1 10
Length 3 (d-iii) 3 200 2 1 10

gies. The effectiveness of the NNSMC strategy is illustrated rotational characteristics of the FTSM are shown in
by comparing the input torque, rotation angle, deforma- Fig. 6.(See Table 3.)
tion, and speed. The effectiveness of the neural network According to Fig. 6, the three control strategies can
compensation control strategy is verified by comparing achieve stable tracking of the rotation angle and speed of
the tracking error in the ground physical prototype control the FTSM in any condition. Compared with the SMC
experiment. The simulation and experimental results under strategy, the NNSMC(tanh) and the NNSMC(sgn) strate-
different control strategies are discussed in detail. gies can obtain smaller tracking errors of rotation angle
and speed. This is because RBF neural network can iden-
tify the disturbance torque. The NNSMC (tanh) has a
smaller tracking error compared with the NNSMC(sgn)
4.2.1. Simulation analysis
control strategy. This verifies that the control strategy
To illustrate the advantages of the NNSMC strategy
using the tanh function can reduce the torque ripple and
using the tanh function, this section compares the input
obtain smaller tracking errors. As the length of the FTSM
torque, rotation angle, and speed of the FTSM. More
decreases, the tracking errors of the three control strategies
importantly, the tracking error of rotation angle and speed,
gradually decrease. But the NNSMC(tanh) strategy can
deformation, and trajectory under different control strate-
still obtain smaller tracking errors. This proves that the
gies are compared to illustrate the advantages of the
NNSMC(tanh) strategy can obtain higher control
NNSMC strategy. In this paper, the SMC control strategy,
accuracy.
the NNSMC control strategy using the sgn function
To study the influence of different control strategies on
(NNSMC(sgn)), and the NNSMC control strategy using
deformation and input torque, the FTSM deformation
the tanh function (NNSMC(tanh)) are selected for simula-
and input torque simulation are carried out in this paper.
tion analysis. Two conditions are selected for simulation
Among them, the deformation simulation results are
analysis to illustrate the effects of different control strate-
shown in Fig. 7, and the input torque simulation results
gies under different conditions. The condition parameters
are shown in Fig. 8.
of the FTSM and controller used in the simulation are
According to Fig. 7, the deformation amplitude of the
shown in Table 2. The output characteristics of the FTSM
NNSMC strategy is smaller than that of the SMC strategy
can be obtained by the above three control strategies under
under any length. With the decrease of the FTSM length,
two conditions. In the simulation, the sine function is used
the deformation amplitude of the three control strategies
as the FTSM desired speed. The simulation results of the
decreases significantly. However, the NNSMC(tanh) strat-
egy can still maintain a small deformation amplitude. This
Table 2 verifies that the NNSMC(tanh) strategy based on neural
Parameters of the FTSM and controller in simulation. network compensation has the advantage of suppressing
Parameters Condition 1 Condition 2 deformation fluctuation.
Flexible link length of l (m) 2.5 1.5 According to Fig. 8, the input torque after neural net-
Flexible link mass m (kg) 3 3 work compensation is smaller than that without compensa-
Grasping Mass M (kg) 1 1.2 tion. This phenomenon is particularly obvious when the
Bending stiffness EI (Nm2) 100 100
direction of rotation speed changes. This is because the
Controller parameter Z 5I 5I
Controller parameter c diag [6,6] diag [6,6] control strategy of neural network compensation can
Controller parameter j 0.01 0.01 weaken the effect of disturbance torque. Therefore, the
Controller parameter Kv 15I 15I control strategy of neural network compensation can
Controller parameter dN 12 12 reduce the sudden change of input torque when the speed

3656
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig. 6. Rotation simulation results of the FTSM under different strategies.

direction changes. With the decrease of FTSM length, the short. The disturbance torque accounts for a smaller pro-
influence of disturbance torque on input torque is strength- portion of the input torque. This reflects that the input tor-
ened. This is because the inertia is small as the length is que curves obtained by the SMC strategy and the NNSMC

3657
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Table 3
Parameters of the ground control experimental platform.
Parameters Condition 1 Condition 2 Condition 3
FTSM length l (m) 0.6 0.7 0.8
Tip payload mass M (kg) 0.8 0.8 0.8
FTSM mass m (kg) 2.5 2.5 2.5
FTSM linear density q(kg/m) 4.17 3.57 3.13
FTSM thickness H (mm) 2 2 2
FTSM elastic modulus E (Gpa) 70 70 70
FTSM width W (mm) 200 200 200

Fig. 7. Simulation results of the FTSM deformation.

Fig. 8. Simulation results of the FTSM input torque.

(sgn) strategy under Condition 2 are closer than under According to (ii) and (vi) in Fig. 9(a) and (b), the trajec-
Condition 1. This indirectly shows that the NNSMC strat- tory curves obtained by the two NNSMC strategies are clo-
egy can reduce the sudden change of input torque. More ser to the desired trajectory curves. It can be seen that the
importantly, compared with the NNSMC (sgn) strategy, two NNSMC strategies have better tracking accuracy than
the NNSMC(tanh) strategy effectively reduces the fluctua- the SMC strategies. In addition, the trajectory tracking
tion of input torque. This shows that the tanh function in curves obtained by the two NNSMC strategies basically
the control strategy can effectively reduce the fluctuation coincide. This view can also be reflected in (iv) and (viii)
of input torque. This verifies the effectiveness of the in Fig. 9 (a) and 9 (b). Based on this, according to Fig. 9,
NNSMC (tanh) control strategy. the trajectories obtained by the NNSMC(sgn) strategy
In space operations, the FTSM completes the task by and the NNSMC (tanh) strategy are similar in the X direc-
adjusting the end-actuator position. The tracking accuracy tion and Y direction under any condition, which is closer to
of the end-actuator directly affects the success of the task. the desired trajectory than the SMC strategy. This shows
Therefore, this paper analyzes the end-actuator trajectory that the NNSMC strategy can obtain a more accurate
in different control strategies. The trajectory simulation tracking trajectory. With the increase in the FTSM length,
results of the FTSM are shown in Fig. 9. the above phenomenon is more obvious. In order to show

3658
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig. 9. Trajectory simulation results of the FTSM.

the accuracy of the NNSMC(tanh) strategy, tracking tra- decreases, the tracking errors of the three control strategies
jectory error curves can be observed. According to the gradually decrease. However, the NNSMC(tanh) strategy
tracking error curves, the NNSMC (tanh) strategy has still has the smallest trajectory tracking error. This proves
the minimum tracking error. As the FTSM length that the NNSMC(tanh) strategy can effectively improve
3659
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig 9. (continued)

control accuracy. It is further demonstrated that the neural Therefore, through the simulation analysis of different
network compensation control strategy can effectively control strategies, it can be concluded that the NNSMC
improve the control accuracy of the FTSM. (tanh) strategy has high control accuracy.

3660
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig. 10. The FTSM’s ground control experiment platform; (a) Flexible telescopic link; (b) Composition of ground experimental platform.

Fig. 11. Experimental results of the FTSM’s ground control experiment platform.

3661
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Fig. 12. Evaluation index distribution maps of different control strategies.

4.2.2. Control experiment of the FTSM parameters under different conditions are shown in Tab.
To verify the effectiveness of the NNSMC(tanh) control 3. Similarly, the NNSMC(tanh) strategy, the NNSMC
strategy, the FTSM’s ground control experiment platform (sgn) strategy, and the SMC strategy are used to control
is built. To simulate the space environment, the flexible tele- the FTSM in different conditions. In ground control exper-
scopic link rotates horizontally in the experiment platform iments, the sine function is selected as the input signal of
to eliminate the influence of gravity. As shown in Fig. 10 the FTSM. In addition, the parameters of the controller
(b), the experimental platform consists of the ASME- are consistent with the simulation. The experimental results
MRB servo motor, cable-driven system, flexible telescopic can be obtained, as shown in Fig. 11.
link, NI controller, etc. The NI controller is divided into According to Fig. 11, the NNSMC(tanh) control strat-
three modules. Among them, the NI-Crio-9053 is the lower egy can obtain smaller rotation angle tracking errors and
computing module, which is responsible for compiling the acceleration amplitude in all conditions. With the increase
control program and realizing the rotation control of the of flexible telescopic link length, the rotation angle tracking
FTSM. The NI-9401 is a data acquisition module, which errors and speed fluctuation gradually increases. In general,
is responsible for measuring the rotation angle of FTSM the closer the slope of the input–output curve is to 1, the
by encoders and collecting it into the control system in better the speed control effect is. However, among the three
real-time. The NI-9264 is a motor drive module, which con- control strategies, the reason why the NNSMC (tanh)
trols the servo motor through voltage and transmits the strategy can achieve the smallest rotation angle tracking
control torque to the cable-driven system. errors is that the neural network compensates for the
To study the FTSM control effect, the experiment plat- uncertainty terms. It verifies that the NNSMC(tanh) con-
form simulates the FTSM under different conditions by trol strategy can obtain higher control accuracy. In addi-
changing the flexible telescopic link length. The FTSM tion, the flexibility of the cable-driven system between the

3662
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

servo motor and the flexible telescopic link causes an the FTSM flexible deformation fluctuate and the
increase in rotation angle tracking errors. The rotation amplitude of rotation angle and speed reduction.
angle tracking errors of the ground control experiment Therefore, the influence of the disturbance torque
are significantly larger than that of simulation control should be considered in the dynamic modeling and
experiments. controller design of the FTSM.
To compare the performance of different control strate- 2) In the dynamic modeling of the FTSM, the dynamic
gies in a quantitative way, four evaluation indexes are model considering the two-dimensional deformation
selected in this paper. The four evaluation indexes are the has high modeling accuracy. However, the 2D1M
average absolute error, standard error, average accelera- model is too complex to design control laws. The
tion, and acceleration standard deviation. The evaluation INTs model simplification method proposed can be
index distribution maps of different control strategies are used to design control laws instead of the 2D1M
obtained, as shown in Fig. 12. model. This is because the INTs model has high mod-
According to Fig. 12, the four indexes of the NNSMC(- eling accuracy and simple calculation.
tanh) strategy are the smallest compared with the NNSMC 3) The tanh function can be used instead of the sgn
(sgn) strategy. This shows that the control using the tanh function in the design of the control law to reduce
function has higher control accuracy. This indirectly shows the fluctuation of input torque. The control strategy
that the control law containing the tanh function can of neural network compensation can effectively
improve the chatter. Compared with the SMC control reduce the influence of the disturbance torque on
strategy, the average absolute error is reduced by 25.98% the FTSM. The NNSMC(tanh) strategy can effec-
in the NNSMC(tanh) strategy. Similarly, the standard tively improve the control accuracy and reduce the
deviation of error is reduced by 12.53%; the average accel- average value of the absolute value of tracking error
eration is decreased by 37.67%; the standard acceleration is by 25.98%.
reduced by 36.37%. This shows that the NNSMC(tanh)
control strategy has high control accuracy. This article mainly focuses on some innovative work on
In summary, the advantages of the NNSMC(tanh) con- the dynamics and control strategies of the FTSM. The
trol strategy are demonstrated by simulation and ground main innovations are summarized as follows: 1) A dynamic
control experiments of different control strategies. Com- model considering various nonlinear factors has been
pared with the SMC control strategy, the control strategy established. 2) A control strategy using neural networks
based on the neural network compensates for uncertain to compensate for dynamic uncertainty components
terms, which can effectively increase the control accuracy. including flexible nonlinear terms and external distur-
The control strategy of the neural network compensation bances is proposed. Neural network identification can be
can make the FTSM obtain higher control accuracy under used to improve the control accuracy of the FTSM. In
different flexible telescopic link lengths. addition, the combination of shock absorbers and optimal
control strategies can also improve the control accuracy of
5. Conclusion the FTSM. Shock absorbers can be installed in flexible
structures of the FTSM to reduce the impact of flexible
In this paper, the FTSM dynamic model considering the vibration. At the same time, the optimal controller param-
two-dimensional deformation and disturbance torque is eters are proposed to improve the accuracy of rotation
established. The influence of disturbance torque on the out- control.
put characteristics of the FTSM is emphatically analyzed.
The INTs model is proposed to simplify the dynamic equa-
tions of the FTSM by removing nonlinear terms. The sim- Declaration of Competing Interest
ulation analysis shows that the INTs simplified model has
high modeling accuracy, and the calculation is simple, The authors declare that they have no known competing
which is suitable for designing the control law of the financial interests or personal relationships that could have
FTSM. In order to reduce the influence of the dynamic appeared to influence the work reported in this paper.
model uncertainty composed of the disturbance torque
and the time-varying length on the FTSM control accu-
racy, the NNSMC(tanh) control strategy is proposed. Acknowledgments
The effectiveness and advantages of this control strategy
are proved by simulation analysis and ground control The project is supported by the Applied Basic Research
experiments. The specific research conclusions are as Program of Liaoning Province (2023JH2/101300159), the
follows: National Natural Science Foundation of China (No.
52275090), the Fundamental Research Funds for the Cen-
1) The disturbance torque has a great influence on the tral Universities (N2103025) and the National Key
output characteristics of the FTSM under the condi- Research and Development Program of China
tion of short length. The disturbance torque makes (2020YFB2007802).
3663
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Appendix A. Parameter expressions in the FTSM’s dynamic References

equations
Abe, A., 2009. Trajectory planning for residual vibration suppression of a
The coefficients’ expressions in the FTSM’s dynamic two-link rigid-flexible manipulator considering large deformation.
Mech. Mach. Theory. 44 (9), 627–1639.
equations are shown below. Aghababa, M.-P., Akbari, M.E., 2012. A chattering-free robust adaptive
Z sliding mode controller for synchronization of two different chaotic
m l 2
v00 ¼ x dx þ Ml2 ðA1Þ system with unknown uncertainties and external disturbances. Appl.
l 0 Math. Comput. 218 (9), 5757–5768.
Aghajari, M., Dehkordi, S.-F., Korayem, M.-H., 2021. Nonlinear
Z l Z x 2 Z l 2
m ^ 02
^ 02 dynamic analysis of the extended telescopic joints manipulator with
v11 ¼ d1 ds dx þ M d1l ds ðA2Þ flexible links. Arabian J Sci Eng. 46, 7909–7928.
l 0 0 0 Ahmadizadeh, M., Shafei, A.-M., Jafari, R., 2021. Frictional impact-
Z l Z x Z l contacts in multiple flexible links. Int. J. Struct. Stab. Dyn. 21 (6),
m ^ 02
^ 02
v12 ¼ x d1 ds dx þ Ml d1l ds ðA3Þ 2150075.
l 0 0 0 Angelini, R., Costa, M., 2002. Multimedial maintenance manuals:
Z l
international space station applications for on-orbit maintenance
m support. Acta Astronaut. 51 (1–9), 415–425.
v13 ¼ d21 dx þ Md21l ðA4Þ
l 0
Astrom, K.-J., Canudas-De-Wit, C., 2008. Revisiting the LuGre friction
model stick-slip motion and rate dependence. IEEE Control Syst.
Z l
m Mag. 28 (6), 101–114.
v14 ¼ xd1 dx þ Mld1l ðA5Þ Chen, S., Xue, W.-C., Zhong, S., Huang, Y., 2018. On comparison of
l 0 modified ADRCs for nonlinear uncertain systems with time delay. Sci.
Z l Z x Z l China Inf. Sci. 61 (7), 1–15.
m ^d02 ds dx þ Md1l ^ 02 Ding, X.-L., Wang, Y.-C., Wang, Y.-B., Xu, K., 2020. A review of
v15 ¼ d1 d1l ds ðA6Þ
l 0 0
1
0 structures, verification, and calibration technologies of space robotic
systems for on-orbit servicing. Sci China Techno. Sci. 64 (3), 462–480.
Z l Dong, C.-Y., Liu, C., Wang, Q., Gong, L.-G., 2019. Switched adaptive
v16 ¼ EI d002
1 dx ðA7Þ active disturbance rejection control of variable structure near space
0 vehicles based on adaptive dynamic programming. Chin. J. Aeronaut.
32 (7), 1684–1694.
Ge, X.-X., Zhou, Q.-X., Liu, Z.-Q., 2020. Assessment of space station on-
Appendix B. Parameter value orbit maintenance task complexity. Reliab. Eng. Syst. Saf. 193 106661.
Gu, H.-C., Song, G.-B., Malki, H., 2008. Chattering-free fuzzy adaptive
robust sliding-mode vibration control of a smart flexible beam. Smart
The specific expression of each element in Eq. (19) are
Mater. Struct. 17 (3) 035007.
shown below. Hastings, D.-E., Putbrese, B.-L., La-Tour, P.-A., 2016. When will on-orbit
 servicing be part of the space enterprise? Acta Astronaut. 127, 655–666.
v00 v14
M ð qÞ ¼ ðB1Þ Hirzinger, G., Brunner, B., Landzettel, K., Schott, J., 1998. Preparing a
v14 v13 new generation of space robots - A survey of research at DLR. Rob
 Auton Syst. 23 (1–2), 99–106.
0 0 Hu, M.-L., Yang, X.-B., Dong, H.-L., 2022. Terminal sliding-mode
K ð qÞ ¼ ðB2Þ control for input-constrained free-float space manipulator via learning-
0 v16
based adaptive uncertainty rejection. Adv. Space Res. (Early Access)
Jia, S.-Y., Shan, J.-J., 2019. Finite-time trajectory tracking control of
F ¼ ½Tf 0
T
ðB3Þ
space manipulator under actuator saturation. IEEE Trans. Ind.
Electron. 67 (3), 2086–2096.
q ¼ ½ h n1 
T
ðB4Þ Jia, S.-Y., Shan, J.-J., 2020. Continuous integral sliding mode control for
space manipulator with actuator uncertainties. Aerosp Sci. Technol.
u ¼ ½Tm 0
T
ðB5Þ 106 106192.
Kaori, K., Yuki, I., Hideki, Y., et al., 2023. MADOCA: Japanese precise orbit
The damping of elastic joints is ignored, so the damping and clock determination tool for GNSS. Adv. Space Res. (Early Access)
matrix is 0. Kayastha, S., Katupitiya, J., Pearce, G., Rao, A., 2023. Comparative
study of post-impact motion control of a flexible arm space robot. Eur.
The values of the Gaussian function are as follows. J Control, 69 (Early Access).
8   Korayem, M.-H., Heidari, A., Nikoobin, A., 2008. Maximum allowable
>
> h ¼ exp ðxci ÞT ðxci Þ
>
>
j 2ð0:3Þ2 load of flexible mobile manipulators using finite element approach. Int.
>
>
>
> c1 ¼ ½ 1:2 1:2 1:2 1:2 1:2 T
J. Adv. Manuf. Technol. 36 (9–10), 1010–1021.
>
> Korayem, M.-H., Shafei, A.-M., Absalan, F., et al., 2014. Kinematic and
>
>
>
> c2 ¼ ½ 0:8 0:8 0:8 0:8 0:8 T dynamic modeling of viscoelastic robotic manipulators using
>
< timoshenko beam theory: theory and experiment. Int. J. Adv. Manuf.
T
c3 ¼ ½ 0:5 0:5 0:5 0:5 0:5  ðB6Þ Technol. 71 (5–8), 1005–1018.
>
> T
>
> c ¼ ½ 0 0 0 0 0  Korayem, M.-H., Shafei, A.-M., Doosthoseini, M., et al., 2016. Theoret-
>
>
4
ical and experimental investigation of viscoelastic serial robotic
>
> c ¼ ½ 0:5 0:5 0:5 0:5 0:5 
T
>
> 5 manipulators with motors at the joints using Timoshenko beam
>
>
>
> c6 ¼ ½ 0:8 0:8 0:8 0:8 0:8 
T theory and Gibbs-Appell formulation. Proc. Inst. Mech Eng Pt K-J
>
: Multi-Body Dyn. 230 (1), 37–51.
c7 ¼ ½ 1:2 1:2 1:2 1:2 1:2 T
3664
D. Shang et al. Advances in Space Research 72 (2023) 3645–3665

Korayem, M.-H., Dehkordi, S.-F., Mehrjooee, O., 2022. Nonlinear Wang, Z.-H., Xu, M., Jia, Y.-H., Xu, S.-J., Tang, L., 2017. Vibration
analysis of open-chain flexible manipulator with time-dependent suppression-based attitude control for flexible spacecraft. Aerosp. Sci.
structure. Adv. Space Res. 69 (2), 1027–1049. Technol. 70, 487–496.
Lavı́n-Delgado, J.-E., Chávez-Vázquez, S., Gómez-Aguilar, J.-F., Alas- Wu, S.-X., Chen, L., Zhang, D.-X., Chen, J.-L., Shao, X.-W., 2020.
safi, M.-O., Alsaadi, F.-E., Ahmad, A.-M., 2022. Intelligent Neural Disturbance observer based fixed time sliding mode control for
Integral Sliding-mode Controller for a space robotic manipulator spacecraft proximity operations with coupled dynamics. Adv. Space
mounted on a free-floating satellite. Adv. Space Res. (Early Access) Res. 66 (9), 2179–2193.
Lu, Y.-B., Huang, P.-F., Meng, Z.-J., 2019. Adaptive prescribed perfor- Xiao, P.-F., Ju, H.-H., Li, Q.-D., Xu, H., Lu, C.-X., 2020. Task planning
mance control for the post-capture tethered combination via dynamic of space maintenance robot using modified clustering method. IEEE
surface technique. Aerosp Sci Technol. 94 105366. Access. 8, 2978122.
Mehrez, M.-W., El-Badawy, A.-A., 2010. Effect of the joint inertia on Yang, Z.-J., Fukushima, Y., Qin, P., 2012. Decentralized adaptive robust
selection of under-actuated control algorithm for flexible-link manip- control of robot manipulators using disturbance observers. IEEE
ulators. Mech Mach Theory. 45 (7), 967–980. Trans Control Syst. Technol. 20 (5), 1357–1365.
Misra, G., Bai, X.-L., 2017. Task-constrained trajectory planning of free- Yang, J.-Z., Peng, H.-J., Zhang, J., Wu, Z.-G., 2022. Dynamic modeling
floating space-robotic systems using convex optimization. J. Guid. and beating phenomenon analysis of space robots with continuum
Control Dyn. 40 (11), 2857–2870. manipulators. Chin J. Aeronaut 35 (9), 226–241.
My, C.-A., Bien, D.-X., Le, C.-H., Packianathe, M., 2019. An efficient Yang, T.-F., Yan, S.-Z., Han, Z.-Y., 2015. Nonlinear model of space
finite element formulation of dynamics for a flexible robot with manipulator joint considering time-variant stiffness and backlash. J
different type of joints. Mech Mach Theory. 134, 267–288. Sound Vib. 341, 246–259.
Nair, A.-P., Selvaganesan, N., Lalithambika, V.-R., 2016. Lyapunov Yao, Q.-J., 2021a. Adaptive trajectory tracking control of a free-flying
based PD/PID in model reference adaptive control for satellite launch space manipulator with guaranteed prescribed performance and
vehicle systems. Aerosp. Sci. Technol. 51, 70–77. actuator saturation. Acta Astronaut. 185, 283–298.
Qiao, J.-Z., Wu, H., Yu, X., 2019. High-Precision attitude tracking Yao, Q.-J., 2021b. Adaptive fuzzy neural network control for a space
control of space manipulator system under multiple disturbances. manipulator in the presence of output constraints and input nonlin-
IEEE Trans. Syst. Man Cybern.: Syst. 51 (7), 4274–4284. earities. Adv. Space Res. 67 (6), 1830–1843.
Shafei, A.-M., Shafei, H.-R., 2020. Considering link flexibility in the Yin, H.-B., Kobayashi, Y., Xu, J.-L., et al., 2014. Theoretical and
dynamic synthesis of closed-loop mechanisms: a general approach. J. experimental investigation on decomposed dynamic control for a
Vib. Acoust.-Trans. ASME. 142 (2) 021004. flexible manipulator based on nonlinearity. J. Vib. Control. 20 (11),
Shang, D.-L., Li, X.-P., Yin, M., Li, F.-J., 2022a. Dynamic modeling and 1718–1726.
fuzzy adaptive control strategy for space flexible robotic arm consid- Yin, X.-M., Wang, B., Liu, L., Wang, Y.-J., 2019. Disturbance observer-
ering joint flexibility based on improved sliding mode controller. Adv. based gain adaptation high-order sliding mode control of hypersonic
Space Res. 70 (11), 3520–3539. vehicles. Aerosp Sci Technol. 89, 19–30.
Shang, D.-Y., Li, X.-P., Yin, M., Li, F.-J., 2022b. Tracking control Zhan, B.-W., Jin, M.-H., Liu, J., 2022. Extended-state-observer-based
strategy for space flexible manipulator considering nonlinear friction adaptive control of flexible-joint space manipulators with system
torque based on adaptive fuzzy compensation sliding mode controller. uncertainties. Adv Space Res. 69 (8), 3088–3102.
Adv. Space Res. (Early Access) Zhang, D.-G., 2009. Recursive lagrangian dynamic modeling and simu-
Shang, D.-Y., Li, X.-P., Yin, M., Li, F.-J., 2022c. Vibration suppression lation of multi-link spatial flexible manipulator arms. Appl. Math
for two-inertia system with variable-length flexible load based on Mech. 30, 1283–1294.
neural network compensation sliding mode controller and angle- Zhang, L., Ge, P.-Q., 2021. High precision dynamic model and control
independent method. IEEE/ASME Trans. Mechatron (Early Access). considering J2 perturbation for spacecraft hovering in low orbit. Adv.
Shang, D.-Y., Li, X.-P., Yin, M., Li, F.-J., 2022d. Dynamic modeling and Space Res. 67 (7), 2185–2198.
fuzzy compensation sliding mode control for flexible manipulator Zhao, Y., Bai, Z.-F., 2011. Dynamics analysis of space robot manipulator
servo system. Appl. Math. Model. 107, 530–556. with joint clearance. Acta Astronaut. 68 (7–8), 1147–1155.
Sharma, G., Rout, B.-K., 2022. Energy efficient reactionless design of Zhao, Z.-J., Liu, Z.-J., 2021. Finite-Time convergence disturbance
multi-arm space robot for a cooperative handshake maneuver. Adv. rejection control for a flexible timoshenko manipulator. IEEE-CAA
Space Res. 71 (3), 1752–1768. J. Automatica Sin. 8 (1), 157–168.
Vakil, M., Fotouhi, R., Nikiforuk, P.-N., Salmasi, H., 2008. A con- Zhao, J.-D., Zhao, Z.-Y., Yang, X.-H., Zhao, L.-L., Yang, G.-C., Liu, H.,
strained lagrange formulation of multilink planar flexible manipulator. 2023. Inverse kinematics and workspace analysis of a novel SSRMS-
J Vib Acoust. 130 (3) 031007. type econfigurable space manipulator with two lockable passive
Wang, M.-M., Luo, J.-J., Zheng, L.-L., Yuan, J.-P., Walter, U., 2020. telescopic links. Mech Mach Theory. 180 105152.
Generate optimal grasping trajectories to the end-effector using an
improved genetic algorithm. Adv. Space Res. 66 (7), 1803–1817.

3665