Mechatronics 52 (2018) 22–35

Contents lists available at ScienceDirect

Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics

Robust position and vibration control of an electrohydraulic series elastic T

manipulator against disturbance generated by a variable stiffness actuator☆
⁎
Minh Nhat Nguyena, Duc Thien Trana, Kyoung Kwan Ahnb,
a
Graduate School of Mechanical and Automotive engineering, University of Ulsan, Namgu, Ulsan 44610, Republic of Korea
b
School of Mechanical Engineering, University of Ulsan, Namgu, Ulsan 44610, Republic of Korea

A R T I C LE I N FO A B S T R A C T

Keywords: This paper presents an electrohydraulic series elastic manipulator (ESEM) system containing a novel variable
Electrohydraulic stiffness actuator (VSA) and a hybrid robust control. The ESEM includes a series elastic manipulator (SEM), an
Series elastic manipulator adjustable stiffness mechanism (ASM), and an electrohydraulic servo system (EHS). Thus, the ESEM can benefit
Backstepping adaptive fuzzy sliding mode from the advantages of the EHS, such as high-power density and a high torque-to-weight ratio. Besides, the ESEM
control
system can use the advantages of the VSA to give the system a suitable dynamic in unknown environments as
Input shaping
well as low energy consumption for cyclic tasks. The proposed VSA adjusts the stiffness by changing the position
of the springs along the ball screw. This system can provide fast stiffness regulation in a much broader range.
However, the variant characteristics of the VSA, and nonlinearities and uncertainties in the EHS, such as friction,
leakages, and dependence of bulk modulus on temperature, are major challenges for the control design. The new
design of a backstepping adaptive fuzzy sliding mode control (BAFSMC) is addressed in this study. It is devel-
oped via sliding mode control, backstepping technique, and an adaptive fuzzy scheme. The controller is sepa-
rated into two control loops for the mechanical dynamics and the hydraulic dynamics. The SMC is embedded for
each loop to reduce the system's order and to ensure that the system's state variables reach and stay on the sliding
surface. The adaptive fuzzy scheme is used to replace the robust term in the control effort of the conventional
SMC to get rid of the chattering phenomenon and to deal with uncertainties in the mechanical and hydraulic
subsystems. The Lyapunov approach and backstepping technique are used to prove the robustness and stability
of the controlled system and to derive the adaptive laws. On the other hand, a fuzzy input shaping (FIS) scheme
which combines an input shaping technique (IST) and a fuzzy logic system, was proposed to minimize the
residual vibration at the end effector robustly over the expected operating range of the VSA system. Numerical
experiments and comparisons with some existing algorithms are presented to illustrate the theoretical results
and show the efficiency of the proposed controller.

1. Introduction execution and safe operation.

Traditionally, robots are used to handle repetitive tasks precisely in
complex environments. They are normally characterized by rigid links
and powerful actuators which make them dangerous for human
workers. Thus, to improve workplace safety, robot-human interaction
becomes an interesting topic. Investigations focus on the flexible im-
pedance properties of actuators, such as adjusting the robot equili-
brium, the stiffness, and the damping. Various variable stiffness con-
cepts were developed to reach a compromise between accuracy of
Several prototypes of VSA have been developed, such as MACCEPA
2.0 [1], MGR [2], VSRJ [3], AwAS-II [4], and 1-DOF VSA [5]. Gen-
erally, these systems combine the elastic elements, such as the spring
and the flexible beam, with an adjusting mechanism. Variable stiffness
actuation is based on the nonlinear deflection characteristics of the
springs. MACCEPA 2.0 uses a spring and a profile disk to obtain a de-
sired torque-angle curve, MGR adjusts the gear teeth to change the
applied force on the leaf spring. AwAS-II changes the lever ratio by
moving the location of the pivot. VSRJ and 1-DOF VSA use springs

Abbreviations: ESEM, Electrohydraulic series elastic manipulator; VSA, Variable stiffness actuator; AFSMC, Adaptive fuzzy sliding mode control; FIS, Fuzzy input shaping; DOF, Degree
of freedom; IST, Input shaping technique; AIS, Adaptive input shaping; EHS, Electro-hydraulic servo; EHA, Electro-hydraulic actuator; SMC, Sliding mode control; BAFSMC, Backstepping
adaptive fuzzy sliding mode control; ASBM, Adjustable stiffness base mechanism; DC, Direct current; MFs, Membership functions; IS, Input Shaping; BSMC, Backstepping sliding mode
control; BPID, Backstepping proportional integral derivative; ZV, Zero vibration; ZVD, Zero vibration and derivative
☆
This paper was recommended for publication by Associate Editor Prof. Bob Koch.
⁎
Corresponding author.
E-mail address: kkahn@ulsan.ac.kr (K.K. Ahn).

https://doi.org/10.1016/j.mechatronics.2018.04.004
Received 12 September 2017; Received in revised form 21 March 2018; Accepted 12 April 2018
0957-4158/ © 2018 Elsevier Ltd. All rights reserved.
M.N. Nguyen et al.
between the output joint and the actuators to regulate the joint stiffness
in both directions of the rotational motion.
A variable stiffness series elastic actuator (VSSEA) [4] consists of
two actuators. One actuator allows for stiffness tunability and the other
for regulating the equilibrium position of the two-link manipulator.
This mechanism increases the mechanical complexity, size, weight,
cost, and integration of the system. However, [6] has proved that the
performance of a lightweight rigid arm is inevitably degraded by im-
posing increasingly high safety constraints. Assuming that total link
inertia is minimized, a possibility for performance enhancement is left
with the design of compliant joint which can decouple the inertia of the
actuator proper form the inertia of the last link. Another work of [7] has
also shown VSA systems offer the possibility to move the load faster and
more safety than other solutions based both on rigid or flexible joint.
When a series elastic manipulator (SEM) contains a VSA, it can be used
efficiently for lift assistance or industrial manipulators to increase
workplace safety for human workers as well as provide suitable dy-
namics in unknown and dynamic environments [8]. The lower reflected
inertia can provide safe interaction with humans [6, 7, 9]. Shocks can
be absorbed by springs to prevent damage to the transmission of the
actuator, which improves the safety of the robots. A stiffness controller
is easily applied for VSA system and does not require using force sen-
sors. Furthermore, the fast stiffness response of the VSA can overcome
the limits of software systems during fast impacts [9]. The springs can
also exchange energy with the environment to reduce the energy con-
sumption in work that requires a high burst of power (such as the VIA
joint prototype [10], a soccer-ball kicking leg [11], hopping [1] or
throwing [12]). In cyclic tasks, the energy can be stored in the spring
during negative work and then released when power generation is re-
quired. A study on the optimal control of a VSA hammer showed that
varying the stiffness during the execution of a hammering task im-
proves the final performance substantially [13]. Analytical results
showed that a SEM with a VSA performing a hammering task can in-
crease speed up to 30% compared to not using VSA. However, the test
bench had a low torque-to-weight ratio and low energy efficiency,
which can be considered as the main factors that limit the performance
of actuators driven by electric motors.
By clamping a linear adjusting spring base mechanism into the
second link, a new type of VSA for series elastic manipulators is pro-
posed. This system can provide fast stiffness regulation in a much
broader range. The main novelty of this proposed VSA system is using a
ball screw to position the springs base, which can handle large loads
and highly precise positioning. A DC motor with a high-resolution en-
coder is used to drive the ball screws to change the forces applied po-
sition along the second link, and the springs act as the nonlinear elastic
elements. Furthermore, the spring forces acting on the motor shaft
through a gearbox are not considerable, and the stiffness regulation is
independent of the manipulator angular position; so, the VSA does not
require a complicated stiffness controller. Besides the advantages of the
VSA, the proposed ESEM can overcome the problems of the existing
prototypes by replacing the electric motors with an electro-hydraulic
servo system as a primary torque generator that has high power density
and high energy efficiency.
Electro-hydraulic servo systems have been widely investigated for
many applications, such as manipulators and aircraft. Its advantages
include large force and torque output durability, reliability, and small
size-to-power ratio, which is a remarkable characteristic of EHSs. When
an EHS is used to control the equilibrium position, the load capacity for
the repetitive task of the ESEM is expanded. However, the EHS's dy-
namics are very complicated because of nonlinear behaviors and
modeling uncertainties, such as friction, internal leakage, external
leakage, and parametric uncertainties [14]. These are challenges in
developing control strategies. Various control methods have been used
for trajectory-tracking control of hydraulic cylinders. The proportional
integral derivative (PID) controllers [15–17] were investigated to
control the position and force tracking for the electro-hydraulic
23
Mechatronics 52 (2018) 22–35
actuator (EHA). The feedback linearization was used in some research
work [18], but these methods did not account for the nonlinear dy-
namics of the cylinder and uncertain fluid parameters. Therefore, sev-
eral kinds of sliding mode control (SMC) methods were adopted for the
EHS system [19,20]. However, the SMC requires the knowledge of
uncertainty bound which is difficult to obtain in a practical system. The
chattering phenomenon is also a disadvantage in the application of the
SMC [21]. Thus, adaptive control algorithms have been proposed to
deal with the mentioned problems, sliding mode adaptive control
[22,23], feedback linearization adaptive control [24]. Although these
nonlinear control schemes showed the chattering elimination, good
tracking performance and robust to uncertainties, the considered con-
trolled plant appears to be a hydraulic actuator, which has a simpler
mathematical model than the ESEM system.
To handle a complex nonlinear system, many researchers have come
up with a robust control scheme which combines SMC with the back-
stepping approach. The combination uses the advantages of both SMC
and the backstepping control, such as order reduction, the systematic
and recursive design. The backstepping sliding mode control has been
applied in a linear induction motor [25], quadrotor [26,27], and
aeroelastic system [28]. In these systems, the actuator dynamics are
typically excluded from the system behavior to simplify the control
design. The control scheme has been employed in these systems without
considering the actuator dynamics, and it includes a proportional
control for position control and SMC for speed control. The stability and
robustness of the controlled system are proved based on the back-
stepping technique and Lyapunov approach. In the ESEM system, the
actuator dynamics are very complicated, especially because of the high
nonlinearities of the cylinder friction and leakages. They can influent
the dynamic characteristics and stability. Therefore, the actuator dy-
namics are considered in the position control for the ESEM system.
Because of the characteristics of the SEM, the first link connects to
the second link through a flexible joint, the residual vibration at the end
effector occurs during the trajectory tracking of the manipulator. Input
shaping technique (IST), a pre-filtering reference input, was introduced
by Singer and Seering [29] to eliminate the system residual vibration.
IST has been used in practical systems, such as flexible manipulators
[30] and ship cranes [31]. IST is a feedforward scheme, which gen-
erates shaped command based on the estimated oscillation parameters.
Thus, IST is not applicable for the ESEM system containing the VSA
system, because the stiffness regulation of the manipulator changes the
natural frequency and the damping ratio of the vibration along with the
stiffness. To overcome this disadvantage, several adaptive input
shaping (AIS) schemes have been developed. E. Pereira et al. [32] used
the algebraic identification method to obtain the natural frequency of
the arm. J. Park et al. [33] proposed the learning input shaping tech-
nique to update the IS parameters based on magnitude and phase dif-
ference of the residual vibration. Nevertheless, these schemes require a
detailed dynamic model of the plant, and the convergence rate of
adaptive values is much lower than the stiffness regulation response.
Based on the above analyses, an ESEM system containing a novel
VSA is proposed and discussed. The first examination of the proposed
system is to control the equivalent position of the manipulator robustly
to the nonlinear disturbances and uncertainties without the oscillation
of the end effector. The major contributions of this study are: 1) The
detail description, the nonlinear dynamical mechanism model, and the
state-space model of the ESEM are introduced. 2) In the residual vi-
bration elimination strategy, a new vibration-control method based on
the concept of [26] and fuzzy technique is designed for the character-
istics of the ESEM system. In this scheme, a fuzzy logic engine is em-
bedded to tune the parameters of the IST according to the stiffness
regulation of the VSA system and the cylinder actions to maintain the
robustness of the residual vibration suppression. 3) In the positioning
strategy, an adaptive robust controller is developed via the sliding
mode, backstepping approach [34] and adaptive fuzzy scheme [35].
The controller is separated into two control loops for the mechanical
M.N. Nguyen et al. Mechatronics 52 (2018) 22–35

dynamics and the hydraulic dynamics. The SMC is embedded for each
loop to reduce the system's order and to ensure that the system's state
variables reach and stay on the sliding surface. An adaptive fuzzy
scheme is used to replace the robust term in the control effort of the
conventional SMC to get rid of the chattering phenomenon and to deal
with uncertainties in the mechanical and hydraulic subsystems. The
Lyapunov approach and backstepping technique are used to prove the
robustness and stability of the proposed controller and to derive the
adaptive laws. 4) Several experiments and comparisons of BAFSMC
with proportional integral derivative (PID) and backstepping sliding
mode control (BSMC); FIS with zero-vibration (ZV) and zero-vibration-
derivative (ZVD) are presented to illustrate the theoretical results and
show the efficiency of the proposed controller. Fig. 2. Structure of the variable stiffness actuator system.
The structure of the rest of this paper is as follow. Section 2 in-
troduces the mechanical structure and mathematical model of the An idea for the design of this VSA system is a combination of SEM
ESEM system. The robust controller and the FIS are designed in and an adjustable spring base mechanism (ASBM) as shown in Fig. 2.
Section 3. Section 4 shows the experimental results and the analyses. The ASBM uses a ball screw as a linear motion actuator which can
Last, some conclusions are summarized in Section 5. The Appendixes handle large loads with highly precise positioning. The coupling torque
present the definitions of the matrices, vectors, functions, and the time acting on the second link has different values for every position of the
derivative equations of Lyapunov function. nut. Not only the moment arm but also the tension of the springs is
adjusted by the nut position. Because of the requirements of working
condition, the DC motor drives the ball screws to provide the desired
2. Hardware design of the ESEM system with the VSA system stiffness. The VSA can change its stiffness in a much broader range and
high dynamic performance. Because the primary torque generator and
2.1. System analysis the VSA are two separate units, they do not affect each other during
operation. In addition, the disturbance torque acting on the DC motor
Hydraulic power systems are always a reasonable choice for the caused by the two springs is not considerable due to the high trans-
power source of heavy industrial manipulators because they can pro- mission ratio. Thus, the VSA system does not require a complicated
vide significant force or torque and fast response. A new generation of controller.
proportional servo valve with an integrated digital electronic is used in
this ESEM system to get rid of some disadvantages of the conventional
hydraulic system, such as pressure drop through the valve, low band- 2.2. The ESEM system setup
width, and low step response time.
As shown in Fig. 1, the EHS system combines a gear pump, an AC Fig. 1 is a schematic diagram of the proposed ESEM. The system
motor, a reservoir, a proportional servo valve, and a single hydraulic hardware includes three main parts: the hydraulic subsystem, the me-
cylinder. The AC motor operates at a constant speed to keep the sup- chanical subsystem, and the VSA system. Two high-resolution encoders
plied pressure always at the desired working value. The servo valve is are used to measure the angular displacement of the two links. Two
controlled to supply pressured oil in both directions to result in the pressure sensors with a pressure range of 0–160 Mbar (Kobold) are used
desired torque/position at the cylinder. to obtain the pressure of the two chambers, and these feedback signals

Fig. 1. Schematic diagram of the electro-hydraulic series elastic manipulator system.

24
M.N. Nguyen et al. Mechatronics 52 (2018) 22–35

Fig. 3. The experimental apparatus.

are used to estimate the feedback torque. In Fig. 2, the stiffness of the Table 2
system depends on the position of the ball screw nut. Therefore, the Setting parameters for the ESEM system.
desired stiffness can be obtained by regulating the angular displace- Components Parameters Specification
ment of the DC motor. Another encoder and a motor driver (MD03) are
used to carry out this task. The setting parameters for the ESEM system Hydraulic Pump Displacement 3.6 [cc/rev]
are as shown in Table 2. Rated rotation speed 1730 [rpm]
Relief pressure 150 [bar]
The proposed controller is implemented on an Advantech Industrial
Servo valve Model MOOG – D633
computer (Core i5 3.5 GHz) within Simulink environment combined Rated flow 10 [l/min]
and Simulink Desktop Real-Time Toolbox of MATLAB. An encoder Hydraulic cylinder Tube diameter – D 46 [mm]
Quad-04 card from Measure Computing Corporation collects data from Rod diameter – d 22 [mm]
Length of stroke – lcyl 35 [mm]
three encoders, and two multifunction data acquisition Advantech
Hydraulic oil Effective bulk modulus 1.5 × 109 [Pa]
cards, PCI171 and PCI6220, are installed in the PCI slots of the PC to Specific gravity 0.87
perform the peripheral interfaces. The experimental apparatus is shown Load cell Capacity 100 [kgF]
in Fig. 3. Rated output 2.052 [mV/V]
Remark 1. The parameters of the ESEM shown in Table 2 which were Encoder Model E60H
Resolution 8192 [p/rev]
obtained from, Autodesk Inventor, and datasheets of the devices.
Pressure sensor Capacity 160 [bar]
Rated output 16 [bar/V]
DC Motor Model IG – 36PGM
2.3. Mathematical modeling of the EHS system Gear head reduction ratio 1/5
Encoder resolution 26 [p/rev]
Using the principles of a hydraulic system, the governing nonlinear
equations describing the fluid flow distribution in the valve are written
as [36] (Table 1 The continuity equations for oil flow through the cylinder, are

2
β
⎧ Cd wxs (p − p1 ) , xs ≥ 0 p˙ 1 = (Q1 − A1 x˙ p − QL)
⎪ ρ s V01 + A1 x p (3)
Q1 =
⎨C wx 2
(p − pa ) , xs ≤ 0
⎪ d s ρ 1 (1) β
⎩ p˙ 2 = (−Q2 + A2 x˙ p + QL)
V02 − A2 x p (4)
2
⎧Cd wxs (p − pa ) , xs ≥ 0
⎪ ρ 2 where xp is the hydraulic cylinder displacement (m), β is the effective
Q2 =
⎨ C wx 2 bulk modulus of the hydraulic fluid, V0i (i = 1, 2) are the initial volumes
⎪ d s (p − p2 ) , xs ≤ 0
⎩ ρ s (2) of fluid trapped at the sides of the actuator, Ai (i = 1, 2) are the areas of
the two sides of the actuator, and QL is the internal leakage.
where Q1 and Q2 represent fluid flows into and out of the valve, re-
Remark 2. Due to the high bandwidth of the servo valve, about 83 Hz
spectively. Cd is the orifice coefficient of discharge, ρ is the mass density
(from datasheet), the dynamics of the servo valve will be neglected.
of the fluid, ps is the pump pressure, pa is the return pressure,
Also, the small effect of the fluid inertial will be ignored in the dynamic
pi (i = 1, 2) are the pressures in the chambers, w is the area gradient of
model of the ESEM system. Hence, the spool displacement, xs, is
the servo valve, and xs is the spool displacement of the servo valve (m).
considered to be proportional to the input voltage, u.

Table 1 xs = ksp u (5)

System parameters of the test bench.
with ksp is the spool displacement gain (m/V), u is the input voltage of
Symbols Value Symbols Value the servo valve (V).
J1 0.45 Nm2 g 9.81 ms−2
J2 0.1 Nm2 Ps 150 bar
2.4. The dynamic model of the ESEM
m1 7 Kg Po 0 bar
m2 3 Kg β 1.25 × 109 Nm−2
l1 0.245 m A1 16.61 × 10−4 m2
The simple form of the ESEM system is shown in Fig. 4, with link-
d1 0.215 m A2 3.8 × 10−4 m2 frame and parameters assignment for each link. Jacobian matrices are
d2 0.245 m kt 2.16 × 10−8 m3/sVPa1/2 used to describe the kinematics and inverse kinematics of the end ef-
fector motion in the conventional Cartesian coordinate system [37].

25
M.N. Nguyen et al. Mechatronics 52 (2018) 22–35

l12 + l 02 + r02 − (r10 + x p)2

cos(θ0 + θ1) =
2l1 (l 02 + r02) (7)
where r1 = r10 + x p , and r10 is the length of the cylinder when x p = 0 .
In Fig. 5, the first excited torque, τ1, which is generated by the cy-
linder can be calculated by
τ1 = l1 Fcy = l1 sinαFc = l1 sinα (A1 p1 − A2 p2 ) (8)
l12 + r 12 − l 02 − r02
where cosα = 2l1 r 1
.
For the flexible joint analysis, springs k1 and k2 connect the first link
and the second link through a nut of a ball screw. The DC motor drives
the screws to change a3 to provide a suitable stiffness for the system.
The forces generated by the two springs are:
Fig. 4. System parameters assignment. Fs1 = Δx1 k1 (9)

Fs2 = Δx2 k2 (10)

The two-link manipulator dynamics are shown in (6), where Ti, Ӫi, di,
mi, li, Ji, are input torque, angular acceleration, the distance between where
joint and center of gravity point, weight, length, and moment of inertia
of link i, respectively. g is the acceleration of gravity. ⎧ Δx1 = a2 + a32 − 2aa3 cos (α1 + θ2) − a120
⎨ Δx = a2 + a32 − 2aa3 cos (α2 − θ2) − a220
⎩ 2 (11)
M (θ) θ¨ + V (θ , θ˙ ) + G (θ) = T − Td (6)
The second torque τ2 and the coupling torque τ12 can be calculated as
T T
where θ = [θ1 θ2]T , θ˙ = [θ˙ 1 θ˙ 2] , θ¨ = [θ¨1 θ¨2] , T = [τ1 τ2]T , Td = [τd1 τd2]T ,
τ2 = (Fs1 sinβ1 − Fs2 sinβ2) a3 (12)
M(θ) is the symmetric positive definite manipulator inertia matrix,
V (θ˙ , θ) is the vector of centripetal and Coriolis torques, G(θ) is the τ12 = (Fs1 sinβ1 − Fs2 sinβ2)(l1 + a3) (13)
vector of gravitational torques, which are shown in the Appendix A-1.
From Fig. 5.a, the relationship between first link angle, θ1, and where
cylinder displacement, xp, can be expressed as
⎧ a 2 + a12
2
− a2 ⎞
⎪ β1 = cos−1 ⎜⎛ 3 ⎟
⎪ ⎝ 2a3 a12 ⎠
⎨ a 2
+ a 2
− a2
⎪ β2 = cos−1 ⎜⎛ 3 22 ⎞
⎟
⎪ ⎝ 2a3 a22 ⎠ (14)
⎩
From Eqs. (9), (10) and (13) the disturbance torque vector is
achieved by:

⎡Td1 ⎤ = ⎡ τf + τ12 ⎤
⎢Td2 ⎥ ⎢ τd2 ⎥ (15)
⎣ ⎦ ⎣ ⎦
where τd2 is the external torques and the unknown function, the hy-
draulic friction torque, τf, a combination of the Coulomb friction and
the viscous friction [38] is considered as Eq. (16).
τf = l1sinαFf = l1sinα (FCoul sign(x˙ p) + Fυ x˙ p) (16)
The angular acceleration θ̈1can be obtained from Eq. (6) as

θ¨1 = B (θ2) τ1 + F (θ1, θ˙ 1, θ2, θ˙ 2) + d1 (t ) (17)

where B , F and d1(t) are known functions as shown in the Appendix A-
2, and d1(t) is the uncertainties in the mechanical subsystem of the
ESEM system. The uncertainties consist of the coupling torques, fric-
tions, and parametric uncertainties.
The time derivative of τ1 is calculated as:
τ˙1 = l1 sin(α )(A1 p˙ 1 − A2 p˙ 2 ) + l1 cos(α ) Fc
= K (x p , x˙ p , θ2) + uH (x p , p1 , p2 , θ2) + d2 (t ) (18)
where K , H , and d2(t) are known functions as shown in the
Appendix A-3.

(ps − p1 ) if u ≥ 0 (p2 − pa ) if u ≥ 0
ΔPp = ⎧ , ΔPr = ⎧ .
⎨
⎩ (p1 − pa ) if u < 0 ⎨
⎩ (ps − p2 ) if u < 0
and d2(t) is the uncertainties in the hydraulic subsystem: the leakages,
parametric uncertainties, the modeling error.
The state variables of the system are defined as
Fig. 5. Illustration of the EHS system and the VSA: a) the EHS system, b) the T
x = [x1 x2 x3]T = [θ1 θ˙ 1 τ1] . Then, the plant can be described by the
VSA system. following state space.

26
M.N. Nguyen et al.
Fig. 6. Structure of the hybrid controller.
⎧ x˙1 = x2
x˙2 = B x3 + F + d1 (t )
⎨ 2
⎩ x˙ 3 = H u + K + d2 (t )
3. Control design
For vibration suppression and position control, the proposed control
includes two main components. The first is an inner loop control that,
produces the voltage for the servo valve to guarantee that the equiva-
lent position tracks the desired position. This control loop is also se-
parated into two controllers: a position control and a virtual torque
control. The second component, an outer loop control that, regenerates
the shaped position reference to suppress the vibration, is called “vi-
bration control”. The overall scheme of the proposed controller is
presented in Fig. 6.
3.1. Design of the backstepping adaptive fuzzy sliding mode controller
The backstepping adaptive fuzzy sliding mode controller is shown in
Fig. 7
The proposed BAFSMC system for the ESEM system is discussed in
two parts. Based on the chosen states of the system, the first part is
designed for position control; the second part is used for virtual torque
control to find the control law for the servo valve so that the manip-
ulator's angular displacement can track the desired commands. The
position is developed based on a sliding mode control and an adaptive
fuzzy technique. The adaptive fuzzy technique is used to replace the
sign function in the sliding mode control to reduce the chattering effect.
Besides, it also adjusts the robust gain to guarantee the variant dis-
turbances are always bounded. The virtual torque control is constructed
by a nonlinear controller and an adaptive fuzzy technique, which is
similar to the technique in the position control.
In this section, a backstepping fuzzy sliding mode control and a
backstepping adaptive fuzzy sliding mode control are designed.
3.1.1. Backstepping fuzzy sliding mode control
Step 1. Design the SMC to guarantee the tracking position error will be
as small as possible. This step will generate the virtual torque x3d.
Choosing the sliding surface as [39]
s1 = c1 e1 + e2
where c1 is a non-zero positive constant, the tracking error of first and
second state variable: e1 = x1 − x1d ; e2 = x2 − x2d .
Differentiating Eq. (20) with respect to time yields
Fig. 7. Proposed control scheme.
Mechatronics 52 (2018) 22–35
s˙1 = c1 e˙ 1 + e˙ 2 = c1 e˙ 1 + B x3 + F + d1 (t ) − x˙2d (21)
Define the torque tracking error as
s2 = x3 − x3d (22)
The equivalent control effort can be derived as the solution of
s˙1 (t ) = 0 without considering the uncertainty (d1(t) = 0)
x3eq = B −1 (x˙2d − c1 e˙ 1 − Π1s1 − F ) (23)
where Π1 is a positive constant.
A Lyapunov candidate function is chosen as follows:
1 2
V1 = s1
(24)
(19)
Then, its derivative is derived as
V˙1 = s1 s˙1 = s1 (c1 e˙ 1 + B x3 + F + d1 (t ) − x˙2d )
= s1 (c1 e˙ 1 + B (x3d + s2) + F + d1 (t ) − x˙2d ) (25)
In order to eliminate the chattering phenomenon in the conven-
tional SMC, a fuzzy logic inference mechanism is used to add a trans-
lation width, ɛ1* , into the virtual torque, referred to as EFSMC [35], with
the fuzzy linguistic rule base as follows.
Rule 1: If s1 is Positive, then x3d is x3eq – ɛ1*
Rule 2: If s1 is Zero, then x3d is x3eq
Rule 3: If s1 is Negative, then x3d is x3eq + ɛ1*
Then the control output of the position control loop can be ex-
pressed as
x3d = w11 (x3eq − ɛ1*) + w12 x3eq + w13 (x3eq + ɛ1*) (26)
where 0 ≤ w11 ≤ 1, 0 ≤ w12 ≤ 1, 0 ≤ w13 ≤ 1, are the firing strengths
of Rules 1–3, respectively; the relation w11 + w12 + w13 = 1 is valid
according to the special case of triangular membership functions; (26)
can be rewritten as
x3d = x3eq − ɛ1* (w11 − w13) (27)
By replacing Eqs. (23) and (27) into Eq. (25), the derivative of the
Lyapunov function is rewritten as follows:
V˙1 = −Π1s12 + s1 B s2 − s1 (B ɛ1* (w11 − w13) − d1 (t )) (28)
The ideal translation width, ɛ1* , can be obtained as Eq. (29) to match
the sliding condition [35].
d1
ɛ1* = + γ1
B w11 − w13 (29)
where γ1 is a small positive constant.
The value of ɛ1* is relative to the uncertainties of the mechanical
subsystem, d1(t). The virtual control output, xd3, is affected by the
translation width.
It is obvious that if the s2 is approximately equal to zero, the
tracking error s1 will converge to zero. Thus, the next step is to control
(20) the torque error is as small as possible.
Step 2. Design the virtual torque control loop to guarantee that the
tracking error s2 approaches zero. Taking the time derivative of Eq. (22)
s˙ 2 = x˙ 3 − x˙ 3d = K + uH + d2 (t ) − x˙ 3d (30)
Choosing the control signal ueq as the solution of s˙ 2 (t ) = 0 without
considering the uncertainties d2(t)
ueq = H −1 (−K + x˙ 3d − Π2 s2 − B s1) (31)
where Π2 is an arbitrary positive constant.
As the position control loop, a fuzzy logic inference mechanism is
used to add another translation width, ɛ *2 , into the control signal, re-
ferred as EFSMC [35], with the fuzzy linguistic rule base as follows
27
M.N. Nguyen et al. Mechatronics 52 (2018) 22–35

Rule 1, If s2 is Positive, then u is ueq – ɛ2* the Appendix B-1. The final form of the time derivative of V1 is obtained
Rule 2, If s2 is Zero, then u is ueq as
Rule 3, If s2 is Negative, then u is ueq + ɛ2* .
V˙1 = s1 (t ) s˙1 (t ) + α1 ɛ͠ 1 (t ) ^ɛ(t ) ≤ B s1 s2 − Π1s12 − s1 (t ) γ1 (w11 − w13)
1
The control signal of the virtual torque control loop can be ex- .

pressed as ⎡ ^ɛ (t ) (w11 − w13) ⎤

+ s1 (t )ɛ͠ 1 (t ) α1 ⎢ 1 − ⎥
⎢ s 1 (t ) α1 ⎥
u = w21 (ueq − ɛ *2) + w22 ueq + w23 (ueq + ɛ *2) (32) ⎣ ⎦ (41)

where 0 ≤ w21 ≤ 1, 0 ≤ w22 ≤ 1, 0 ≤ w23 ≤ 1, are the firing strengths Step 2. Apply an adaptive law for the translation width in the virtual
of rules 1)–3), respectively; the relation w21 + w22 + w23 = 1 is valid torque control loop
according to the special case of triangular membership functions Consider the Lyapunov function for the whole system as
The control signal u is rewritten as
s22 (t ) + α2 ɛ͠ 2 (t )
u = ueq − ɛ2 (w21 − w23) (33) V = V1 +
2 (42)
Consider the Lyapunov function for the whole system as where α2 is a small positive constant
s 2 (t ) The control signal u is rewritten as
V = V1 + 2
2 (34) u ̂ = ueq − ^ɛ2 (w21 − w23) (43)
The derivative of the Lyapunov function Eq. (34) is as follows:
The Lyapunov derivative Eq. (42) is given in the Appendix B-2. The
V˙ = V˙1 + s2 s˙ 2 (35) equation in final form is as follows:
.
Replacing Eqs. (28), (30), (31), and (33) into Eq. (35), the derivative
V˙ = V˙1 + s2 s˙ 2 + α2 ɛ͠ 2 ^ɛ
function Eq. (35) is represented as 2

≤ − Π1s12 − Π2 s22 − s1 (t ) γ1 (w11 − w13) − s2 (t ) γ2 H (w21 − w23)

V˙ = −Π1s12 + s1 B s2 − s1 (B1 ɛ1* (w11 − w13) − d1 (t )) .
^ɛ (t ) (w 21 − w 23) ⎤
+ s2 (K + uH + d2 (t ) − x˙ 3d ) + s2 (t )ɛ͠ 2 (t ) α2 ⎡ s 2 (t ) −
⎢
⎣
2 α2
⎦ ⎥ (44)
2
= − ∑i = 1 Πi si2 − s1 (B1 ɛ1* (w11 − w13) − d1 (t ))
If the adaptive laws for two loops are designed as
− s2 (H ɛ *2 (w21 − w23) − d2 (t )) (36) .
^ɛ(t ) = s1 (t )(w11 − w13)
The ideal translation width, ɛ *2 , can be defined as Eq. (37) 1 α1 (45)
d2 .
ɛ2* = + γ2
w21 − w23 H (37) ^ɛ(t ) = s2 (t )(w21 − w23)
2 α2 (46)
where γ2 is a small positive constant. The translation width for the
virtual torque control loop, ɛ *2 , is profoundly affected by the un- Then Eq. (44) becomes
certainties in the hydraulic subsystem, d2(t). When the ideal translation V˙ ≤ −Π1s12 − Π2 s22 − s1 (t ) γ1 (w11 − w13) − s2 (t ) γ2 H (w21 − w23) (47)
widths are selected as Eq. (29), and Eq. (37), the derivative function
Eq. (36) is a semi-negative function. So, the whole system is stability. According to the inequality s1(w11−w13) ≥ 0, s2(w21-w23) ≥ 0, and
In practice, the uncertainties vary flexibly; the ideal translation the positive functions B , and H , we can obtain that V˙ (t ) ≤ 0 . V˙ (t ) is
widths, ɛi* (i = 1, 2) , are impossible to obtain exactly. If the translation negative semidefinite, that is, V(t) ≤ V(0). It implies s1(t), s2(t), ɛ͠ 1 (t ),and
width is too small, the error states will be away from the sliding surface. ɛ͠ 2 (t ) are bounded. Therefore, the proposed controller satisfies the con-
In case of using a large translation width, the error states reach the dition V˙ (t ) ≤ 0 or the closed-loop system is stable and lim e = 0 [39].
t →∞
sliding surface rapidly and may create large control efforts. Thus, Remark 3. The membership functions of the input are selected by
adaptive laws are proposed to estimate the ideal value of the translation analyzing the performance of the sliding variables, s1ands2.The
width. The proposed algorithm is presented in the next section. locations of the output membership functions are adjusted by the
adaptive laws (45), and (46).
3.1.2. Backstepping adaptive fuzzy sliding mode control
Step 1. Apply adaptive law for the translation width in the position
3.2. Design of the fuzzy input shaping
control loop
The estimated errors of the translation widths are defined as The proposed vibration control is developed by combining an input
shaping technique and a fuzzy logic. The fuzzy engine is used to adjust
ɛ͠ i (t ) = ^ɛi (t ) − ɛi * (i = 1, 2) (38) the parameters of the IST when the characteristics of the system are
where ^ɛi (i = 1, 2) are the estimated values of the ideal translation changed by the stiffness regulation. The conventional input shaping
widths. generates input command that completes a motion of the system
For the position control loop, the virtual torque can be represented without any residual vibration. Two types of input shaper, Zero-
as Vibration (ZV) and Zero-Vibration-Derivative (ZVD), are utilized in this
paper to suppress the oscillation of the end effector.
x3̂ d = x3eq − ^ɛ1 (w11 − w13) (39)
Thus, the Lyapunov candidate is reselected as 3.2.1. Input shapers
A Zero-Vibration (ZV) shaper was proposed to suppress the residual
s12 (t ) +α1 ɛ͠ 12 (t ) vibration in [29] can be expressed as the following general form:
V1 =
2 (40)
Rs (s ) = [A1 e−T1 s + A2 e−T2 s] R (s ) (48)
where α1 is a positive constant.
Differentiating Eq. (40) with respected to time, then substituting The reference signal, R(s), is shaped into the two-impulse sequence,
Eqs. (21)–(23), (38), and (39) into it, the whole equation is presented in Rs(s), by four parameters: the magnitudes of impulses A1 and A2, and

28
M.N. Nguyen et al. Mechatronics 52 (2018) 22–35

the delay time T1 and T2. To achieve zero vibration after the last im-
pulse, these parameters are obtained as
1 K
⎡ Ai ⎤ = ⎡ 1+K 1+K⎤
⎢ Ti ⎥ ⎢ 0 π ⎥
⎣ ⎦ ⎢ ⎣ ωd ⎥
⎦ (49)
where ωn is the system's natural frequency, ζ is the damping ratio, ωdis
the damped frequency of the system given byωd = ωn 1 − ζ 2 , and

K = exp ⎛⎜ ⎞⎟. The ZV results in an excellent vibration suppression.

−ζπ
1 − ζ2
⎝ ⎠
However, it is very sensitive to the modeling errors.
To improve the robustness of the IST under variations of the system
parameters, an additional constraint is used to add two more para-
meters to the input shaper as shown in Eq. (51). The input increases
from two to three impulses which refer to the Zero Vibration and De-
rivative (ZVD) shaper [29]. The ZVD can be presented as Eq. (50)

Rs (s ) = [A1 e−T1 s + A2 e−T2 s + A3 e−T3 s] R (s ) (50)

where the parameters of ZVD can be obtained by (51). Fig. 9. MFs for input/output of the proposed FIS with (a) MFs of the xVSA input;
(b) MFs of the ẋ p input; (c) MFs of the A1 output; (d) MFs of the A2 output; and
K2
⎡ 1 2K
⎤ (e) MFs of the T output.
⎡ Ai ⎤ = ⎢ (K + 1)2 (K + 1)2 (1 + K )2
⎥
⎢Ti ⎥ ⎢ 0 π 2π ⎥
⎣ ⎦
⎣ ωd ωd ⎦ (51)
Table 3
Rule table of fuzzy inference.
3.2.2. Fuzzy input shaping ẋp /xVSA P0 P1 P2 P3 P4 P5
In this research, the VSA adjusts the stiffness along with the system
parameters (ω and ζ) in a wide range. Even the ZVD shaper, which is P A1E A2E A1E A2E A1E A2E A1E A2E A1E A2E A1E A2E
more robust to the modeling errors than ZV, cannot perform well with T0E T1E T2E T3E T4E T5E
N A1R A2R A1R A2R A1R A2R A1R A2R A1R A2R A1R A2R
such a large deviation of actual frequency and modeling frequency. The T0R T1R T2R T3R T4R T5R
FIS is designed in this section to update the IS parameters during the
stiffness regulation of the manipulator to keep the estimated values
close to the actual values. The ZV, ZVD, FIS will be used to shape the position references in the
The fuzzy logic system consists of a fuzzifier, a defuzzifier, fuzzy FIS experiments in other to handle the residual vibration of the end
rules, and a fuzzy inference engine, as shown in Fig. 8, with two inputs effector.
and three outputs. The inputs are the position of the ball screws net,
xvsa, and cylinder velocity, ẋ p . Their membership functions of the po-
4. Experimental results
sition and the velocity are respectively triangular membership functions
(MFs) with fuzzy setsPi (i = 1, …, 5), and trapezoidal membership func-
In this section, several experiments were carried out to prove the
tions with fuzzy sets positive (P) and negative (N). The outputs are the
effectiveness of the proposed controller. The first one is an experiment
three parameters of the ZV, Ai (i = 1, 2) , and T2. They are triangular
for analyzing the dynamic characteristics of the proposed VSA. Next,
MFs with fuzzy sets extend (AiE), and retract (AiR) for the Ai (i = 1, 2)
two experiments are used to illustrate the effectiveness of the BAFSMC
output, and fuzzy sets Ti (i = 1, …, 10) for the T output. These MFs are
in tracking control and chattering elimination. Finally, some experi-
shown in Fig. 9. The fuzzifier is used to scales and converts the variables
ments substantiate the advantage of using FIS for vibration suppression
into the linguistic variables. The fuzzy rule table is finally designed and
over the expected operating range of the VSA. These algorithms were
shown in Table 3. The outputs of the fuzzy system are computed by
built by the combination of Simulink and the Simulink Desktop Real-
adopting the product inference and center-average defuzzifier.
Time Toolbox of MATLAB.
Remark 4. The parameters of these membership functions are specified
by using the magnitudes of impulses and the delay times. They are
calculated based on Eq. (49) with natural frequencies and damping 4.1. The variation performance of the VSA system
ratios obtained from analyzing oscillation of the end effector at several
positions of the VSA [40]. The results of the oscillation analysis are First of all, to investigate the dynamic characteristics of the pro-
showed in Table 5 posed VSA system, an experiment was set up using a PID controller with
KP1 = 2.5,KI1 = 0.4,KD1 = 0.05. The responses of the VSA to sinusoidal
The shaped reference generated by FIS is obtained by Eq. (52). y = 10sin(ωo t ) at the nut's starting position of 25 mm with ωofrom 0.1 to
1 e−T1̂ s + A
Rs (s ) = [A 2 e−T2̂ s] R (s ) 7 rad/sec in the frequency domain were collected to plot the Bode
(52)
diagram shown in Fig. 10. The PID controller was able to closely track
1 , A
where A 2 , T2 ̂ are estimated values. to the input signals that were lower than 3 rad/sec. The bandwidth of
the proposed VSA, in this case, is about 5.8 rad/sec. The analytical re-
sults show that the VSA system is fast enough to meet the application's
requirements. This PID controller will be used to control the position of
the ball screw nut, xVSA, for the next case studies.

4.2. The equivalence position performances

Fig. 8. Configuration of a fuzzy logic system. The VSA system was adjusted to its lowest stiffness, xVSA = 0,

29
M.N. Nguyen et al.
Fig. 10. Bode plot of the VSA system with a PID controller.
Table 4
Parameters of the controllers.
Type of controller Parameters
GS-PID Control law 1: KP1 = 3.15 , KI1 = 45 , KD1 = 9 × 10−3 ,
Control law 2: KP2 = 2.7 , KI2 = 54 , KD2 = 45 × 10−3 .
Control law 3: KP3 = 3 , KI3 = 50 , KD3 = 9 × 10−3 .
Control law 4: KP4 = 2.8 , KI 4 = 57 , KD4 = 45 × 10−3 .
BSMC c1 = 60 , Π1 = 850 , η1 = 1, Π2 = 6.72 × 1010 , η2 = 6 × 108 .
BAFSMC c1 = 60 , Π1 = 900 , α1 = 0.01, Π2 = 6.72 × 1010 , α2 = 0.5 .
during the first ten seconds, then to its highest stiffness, xVSA = 50,
during the last ten seconds in the experimental setpoint tracking tests.
This experiment focuses on only the trajectory tracking response of the
controller. Thus, the IST is not used to generate the shaped reference
signals. In the first case, a pulse whose duty and cycle are 50% and 5 s is
used as an equivalence position reference. In the second case, the ESEM
system is run for a multi-steps trajectory. Since the ESEM system is a
low-bandwidth system, to enable the capability of vibration suppres-
sion of the IS, the reference signal must go through an appropriate low-
pass filter. To express the trajectory tracking performance and the ro-
bustness of the proposed controller, a gain-scheduled PID control, and a
Fig. 11. Comparison of the pulse position response of the ESEM system.
30
Mechatronics 52 (2018) 22–35
BSMC were set up for the comparisons. The GS-PID schedules the gains
based on stiffness states and kinematic position. The control laws 1–4
are used in the cases of extended cylinder at low stiffness, retracted
cylinder at low stiffness, extended cylinder at high stiffness, and re-
tracted cylinder at high stiffness, respectively. The BSMC uses two
conventional SMC systems for the two loops, where the robust term in
the control effort is uri = ηi sign(si ) (i = 1, 2) , ηi is the robust gain, and
the sign function is shown in the Appendix C-1. The system parameters
were adopted carefully based on the system design and experimental
processes.
Remark 5. In practice, because of the mechanical structure and
actuator capability, all the parameters of the above controllers are
chosen to achieve the best transient control performance. For the
proposed controller, the control parameters are selected to keep the
control effects, x3d, and, u, stay in their regions, and the derivative of
the Lyapunov function, as shown in Eq. (47), is a semi-negative
function. These parameters are shown in Table 4.
4.2.1. Case study 1: pulse reference
The performances, which are position responses, tracking errors,
torque responses, and control signals of the GS-PID, the BSMC, and the
proposed controller are shown in Figs. 11–13. As seen, the three con-
trollers can track the pulse reference, the BSMC and BAFSMC can both
obtain better control performance, but the GS-PID cannot perform well,
of which the error is too large to be accepted. The tracking errors are
shown in Fig. 11.b. The steady errors of the proposed control, the GS-
PID, and the BSMC are bounded by 0.02°, 0.04°, and 0.07°, respectively.
When the VSA system is regulated to high stiffness, the BAFSMC can
maintain the stability by using the adaptive law to estimate the trans-
lation width to confront the disturbances in the mechanical subsystem
online.
Similarly, in the virtual torque control loop, the BAFSMC has a
better torque tracking response than BSMC. Because the adaptive fuzzy
is used to deal with the uncertainties in the hydraulic subsystem, the
results in Fig. 12.b show that the feedback torque tracks the virtual
signal excellently. It is illuminated that the uncertainties in the hy-
draulic subsystem can significantly affect the precision and stability of
the controller.
Fig. 13 presents the control signals of the BSMC and the BAFSMC.
The results show that the BSMC contains oscillations in the control
signal because of the ‘overestimated’ robust gains in the robust terms.
Meanwhile, the chattering phenomenon is remedied in the control ef-
forts according to the online adjustment of the translation width.
M.N. Nguyen et al.
Fig. 12. Comparison of the pulse torque response of the ESEM system: (a)
BSMC; (b) BAFSMC.
Fig. 13. Comparison of the pulse control signal: (a) BSMC; (b) BAFSMC.
4.2.2. Case study 2: multi-step reference
As presented above, this case study discusses the multi-step tracking
ability of the proposed controller. The three controllers are then run for
the different positions shown in Fig. 14.a, which has a large distance
and a maximum angular displacement; both are near their physical
Fig. 14. Comparison of the multi-step position response of the ESEM system.
31
Mechatronics 52 (2018) 22–35
limits. The tracking errors are shown in Fig. 14.b. As seen, the results
show that the GS-PID cannot handle such a large movement well with
the tracking error around 0.9°. The proposed BAFSMC has a better
performance than the BSMC in terms of both position and virtual torque
control. The steady state error of the proposed BAFSMC during the
entire experiment is kept within [−0.02 0.02].
The torque performances are shown in Fig. 15. The results show that
the feedback torque of the proposed control keeps tracking the virtual
torque control effort better than the BSMC, as shown in Fig. 15c, which
means that all defined states are stable. Therefore, the robust tracking
performance is guaranteed even with parameter variations and external
load disturbance.
Fig. 16 shows the control signals of the BSMC and the proposed
control. The results substantiate the advantage of using adaptive fuzzy
to replace the robust term in conventional SMC. As shown in Fig. 16.b,
the control inputs do not exhibit chattering. Meanwhile, the chattering
still exists in the BSMC.
4.3. FIS experiments
In other to enable the effectiveness of input shaping algorithm, the
controlled plant has to track to the shaped reference. The GS-PID
controller does not satisfy this condition which reduces the effective-
ness of IS. The BSMC shows a good tracking response. However, the
chattering phenomenon of the control signal causes unexpected vibra-
tions at the end effector. Therefore, the GS-PID and BSMC will not be
investigated in this section. The BAFSMC substantiates its effectiveness
in controlling position and reducing the chattering effect. Thus, other
experiments using BAFSMC are carried out to demonstrate the vibration
suppressibility of the proposed FIS compared to the ZV and ZVD. The
reference signals of the BAFSMC will be shaped by ZV, ZVD, and FIS.
The parameters of the ZV and the ZVD are calculated by analyzing
the response of the second link to the pulse trajectory at the lowest
stiffness, xVSA = 0. The oscillation of the end effector was translated
into a peak in the periodogram, as shown in Fig. 17. According to the
result, the estimated value of the natural frequency is 8.9π rad/s. On
the other hand, the damping ratio obtained by [40] is 0.037.
To design the fuzzy input shaping, additional experiments are car-
ried out with different positions of the VSA based on BAFSMC. These
positions of the VSA are xVSA = 0, 10, 20, 30, 40, 50.The input mem-
bership functions are the positions of VSA as shown in Fig. 9.a. The
output membership functions are derived based on the magnitudes of
impulses and delay times which are computed based on Eq. (49) using
the natural frequencies and damping ratios as shown in Table 5.
M.N. Nguyen et al. Mechatronics 52 (2018) 22–35

Fig. 15. Comparison of the multi-step torque response of the ESEM system: (a)
BSMC; (b) BAFSMC.
Fig. 18.a presents the equivalent responses of the BAFSMC under the
shaped reference signal of ZV, and Fig. 18.d is the ball screw nut po-
sition response in the variable stiffness actuator. The results show that
the equivalent position can track the shaped pulse trajectory while the
VSA position keeps increasing to random values. The response of the
end effector corresponding to each position of the ball screw nut is
shown in Fig. 18.b and c. The black dotted line, the black line in
Fig. 18.c, and the black line in Fig. 18.b are the vibration of the end
effector without using the IS, with the ZV, and with the ZVD, respec-
tively. It is obvious that when the VSA position is at 0, the ZV and ZVD
suppress the vibration excellently because of the small error in the es-
timated values. In contrast, when the VSA position is adjusted to higher
values, the deviation of actual and estimated parameters increases, and
the residual vibration becomes larger. Although the ZVD is more robust
than the ZV [29], it cannot perform well in such a large range of
stiffness regulation. Besides, the effect of gravity also causes different
oscillations in both directions of the cylinder.
To overcome the problems mentioned above, the fuzzy logic engine
designed in Section 3 was used to update the ZV parameters online
corresponding to the VSA position and the cylinder actions. As the re-
sults presented in Fig. 19.b, the vibration vanished stably against the
stiffness regulation. The ZV parameters quickly change to the optimal
values, so the vibration suppression will not be affected by the adaptive
parameters process.
The FIS fuzzy output gains are shown in Fig. 20. The comparison of
vibration responses using different algorithms is zoomed in Fig. 21. In
this figure, the data is collected from 22.5 s to 32.5 s of the FIS ex-
periments corresponding to two positions of the VSA system, xVSA = 37
and xVSA = 45. The presence of the FIS does not influent the precise
trajectory tracking of the BAFSMC.

5. Conclusions

This paper has presented a design of an electro-hydraulic series

Fig. 16. Comparison of the multi-step control signal: (a) BSMC; (b) BAFSMC.
Fig. 17. The periodogram of the end effector oscillation.
elastic manipulator containing a novel variable stiffness actuator. This
system takes advantages of the variable stiffness mechanism and the
high torque relative to size of the electro-hydraulic servo system to
provide a suitable dynamic and improve workplace safety for human
workers. The description, the nonlinear dynamical mechanism model,
and the state-space model of the ESEM were introduced. The first ex-
amination of the proposed system is to control the equivalent position
of the manipulator robustly to the nonlinear disturbances without the
oscillation of the end effector. Thus, the control strategy in this study
includes a positioning strategy for the EHS system and a residual vi-
bration elimination strategy for the end effector. Because this system is
a high-order and complex system with nonlinear elements, such as
friction, leakage, and parametric uncertainties, the proposed position
controller used a backstepping approach to create two loops, based on
system states: the position control loop and the virtual torque control
loop. Two adaptive fuzzy sliding mode control schemes were designed
for each system's state based on the theoretical model of the ESEM to
eliminate the chattering and guarantee the robustness in the presence of
the known uncertainties of both the mechanical and the hydraulic
subsystem. The stability analyses of the proposed control system were
derived in detail based on the Lyapunov approach. On the other hand,

Table 5
Frequencies and damping ratios of the oscillation at several positions of the VSA.
Natural frequency (rad/s)

xVSA = 0 xVSA = 10 xVSA = 20 xVSA = 30 xVSA = 40 xVSA = 50

Extend ωn = 9.2π ωn = 9.5π ωn = 9.9π ωn = 10.5π ωn = 10.75π ωn = 10.97π

Retract ωn = 8.9π ωn = 9.1π ωn = 9.7π ωn = 9.9π ωn = 10.4π ωn = 10.7π
Damping ratio
Extend ζ = 0.021
Retract ζ = 0.037

32
M.N. Nguyen et al. Mechatronics 52 (2018) 22–35

Fig. 18. Comparison of the vibration response of the end effector.

Fig. 21. The vibration response at two positions of the VSA.

Fig. 19. The vibration response of the end effector with the FIS.
the vibrations at the end effector vary flexibly along with the stiffness
regulation. The FIS was proposed to keep the robustness of the ZV
shaper in vibration suppression over the expected operating range of
the VSA. In the FIS, a fuzzy logic tunes the ZV shaper's parameters
corresponding to stiffness responses to ensure a small error of actual
and estimated parameters. Numerical experiments and comparisons
with some existing algorithms were presented to illustrate the theore-
tical results and show the efficiency of the proposed controller. The
analytical results showed that the ESEM with the hybrid controller
possesses the stable tracking control performance, free from chattering,
robust to uncertainties, and minimum residual vibration. It also de-
monstrated the credibility of the theoretical model of the ESEM system.
The proposed control scheme is promising for use in other nonlinear
systems that consider the actuator dynamics or systems that contain
nonlinear vibration response. The merits of the VSA compared with
fixed stiffness will be validated in our future work for a hydraulic ro-
botic hammer.

Acknowledgments

This work was supported by the Ministry of Trade, Industry &

Energy, Korea under Industrial Technology Innovation Program (No.
10067184).

Fig. 20. The adaptive parameters of the FIS.

33
M.N. Nguyen et al. Mechatronics 52 (2018) 22–35

Appendix A

1. The symmetric positive definite manipulator inertia matrix, the vector of centripetal and Coriolis torques, and the vector of gravitational torques
in Eq. (6) are shown as follows:
2
J + J2 + m2 l12 + 2m2 l1 d2cosθ2 J2 + m2 l1 d2cosθ2 ⎤ ⎡−m2 l1 d2sinθ2 (θ˙ 2 + 2θ˙ 1 θ˙ 2) ⎤
M (θ) = ⎡ 1 , V (θ , θ˙ ) = ⎢ ⎥,
⎢ J + m l d cosθ J ⎥ m2 l1 d2sinθ2 θ˙ 1
2
⎣ 2 2 1 2 2 2 ⎦ ⎣ ⎦
g (m d + m l )cos θ + gm d cos(θ + θ )
G (θ) = ⎡
1 1 2 1 2 2 2 1 2 ⎤
.
⎢ gm2 d2cos(θ1 + θ2) ⎥
⎣ ⎦

2. The known functions in the Eq. (17) are presented as:

J2
B = ,
J1 J2 + J2 l12 m2 − (d2 l1 m2cosθ2)2
2 2
(d 22 l1 m22cosθ2)(l1sinθ2 θ˙ 1 + g cos(θ1 + θ2)) J2 (d2 l1 m2sinθ2 (θ˙ 2 + θ˙ 1) − g cosθ2 (d1 m1 + l1 m2))
F = 2 2
+ 2
,
J1 J2 + J2 l1 m2 − (d2 l1 m2cosθ2) J1 J2 + J2 l1 m2 − (d2 l1 m2cosθ2)2
J2 (τd2 − τf − τ21 − τ2) + d2 l1 m2cos(θ2)(τd2 − τ2)
d1 (t ) = .
J1 J2 + J2 l12 m2 − (d2 l1 m2cosθ2)2

3. The known functions in the Eq. (18) are presented as:

A12 A22
K = −βl1sin(α ) x˙ p ⎜⎛ + ⎞
⎟,
⎝ V01 + A1 x p V02 − A2 x p ⎠

2 ⎛ A1 ΔPp A2 ΔPr ⎞
H = βCd wl1 kspsin(α ) + ,
ρ ⎜ V01 + A1 x p V02 − A2 x p ⎟
⎝ ⎠
A1 A2
d2 (t ) = −βl1 sin(α ) QL ⎜⎛ + ⎞
⎟ + l1 Fc cos(α ).
⎝ V01 + A1 x p V02 − A2 x p ⎠

Appendix B

1. The derivative of the Lyapunov function Eq. (41) is derived as follows:

.
V˙1 = s1 (t ) s˙1 (t ) + α1 ɛ͠ 1 (t ) ^ɛ(t )
1
.
= s1 (t )[B s2 − Π1s1 − ^ɛ1 (t )(w11 − w13) + d1 (t )] + α1 ɛ͠ 1 (t ) ^ɛ(t )
1
.
= B s1 s2 − Π1s12 + α1 ɛ͠ 1 (t ) ^ɛ(t ) − s1 (t )[ɛ1*(w11 − w13) − d1 (t ) + ^ɛ1 (t )(w11 − w13) − ɛ1*(w11 − w13)]
1
.
d1 (t ) d1 (t )
= B s1 s2 − Π1s12 − s1 (t )(w11 − w13) ⎡ + γ1 − ⎤ − s1 (t )(w11 − w13)ɛ͠ 1 (t ) + α1 ɛ͠ 1 (t ) ^ɛ(t )
⎢
⎣ w 11 − w 13 (w 11 − w13) ⎥
⎦ 1
.
⎡ ^ɛ (t ) (w11 − w13) ⎤
≤ B s1 s2 − Π1s12 − s1 (t ) γ1 (w11 − w13) + s1 (t )ɛ͠ 1 (t ) α1 ⎢ 1 − ⎥
⎢ s 1 (t ) α1 ⎥
⎣ ⎦

2. The derivative of the Lyapunov function Eq. (44) is expressed as follows:

.
V˙ = V˙1 + s2 s˙ 2 + α2 ɛ͠ 2 ^ɛ
2
.
= B s1 s2 − Π1s12 − s1 (t ) γ1 (w11 − w13) + α2 ɛ͠ 2 (t ) ^ɛ(t ) − s2 (t )[Π2 s2 + B s1 + H ^ɛ2 (t )(w21 − w23) − d2 (t )]
2
.
d (t )
= − Π1s12 − − s1 (t ) γ1 (w11 − w13) + α2 ɛ͠ 2 (t ) ^ɛ(t ) − H s2 (t ) ⎡ɛ2*(w21 − w23) − 2
Π2 s22 + ^ɛ2 (t )(w21 − w23) − ɛ *2 (w21 − w23) ⎤
2 ⎣ H ⎦
d (t ) d (t )
= − Π1s12 − Π2 s22 − s1 (t ) γ1 (w11 − w13) − H s2 (t )(w21 − w23) ⎡ 2
+ γ2− 2 ⎤
⎢ H w21 − w23 H (w21 − w23) ⎥
⎣ ⎦
.
− s2 (t )(w21 − w23)ɛ͠ 2 (t ) + α2 ɛ͠ 2 (t ) ^ɛ(t )
2
.
⎡ ^ɛ (t ) (w21 − w23) ⎤
≤ − Π1s12 − Π2 s22 − s1 (t ) γ1 (w11 − w13) − s2 (t ) γ2 H (w21 − w23) + s2 (t )ɛ͠ 2 (t ) α2 ⎢ 2 − ⎥
⎢ s 2 (t ) α2 ⎥
⎣ ⎦

34
M.N. Nguyen et al.
Appendix C
1. The sign function used in BSMC is defined by
1 if si (t ) ≥ 0,
sign(si ) = ⎧
⎨
⎩ 1 if si (t ) < 0.
−
References
[1] Vanderborght B, Tsagarakis NG, Semini C, Ham RV, Caldwell DG. MACCEPA 2.0:
Adjustable compliant actuator with stiffening characteristic for energy efficient
hopping. 2009 IEEE international conference on robotics and automation. 2009. p.
544–9.
[2] Wang W, Fu X, Li Y, Yun C. Design of variable stiffness actuator based on modified
gear–rack mechanism. J Mech Robot 2016;8(6). 061008-061008-10.
[3] Migliore SA, Brown EA, DeWeerth SP. Novel nonlinear elastic actuators for pas-
sively controlling robotic joint compliance. J Mech Des 2006;129(4):406–12.
[4] Jafari A, Tsagarakis NG, Sardellitti I, Caldwell DG. A new actuator with adjustable
stiffness based on a variable ratio lever mechanism. IEEE/ASME Trans Mechatron
2014;19(1):55–63.
[5] Tonietti G, Schiavi R, Bicchi A. Design and control of a variable stiffness actuator for
safe and fast physical human/robot interaction. Proceedings of the 2005 IEEE in-
ternational conference on robotics and automation. 2005. p. 526–31.
[6] Bicchi A, Tonietti G, Bavaro M, Piccigallo M. Variable stiffness actuators for fast and
safe motion control. Robotics research. The eleventh international symposium.
Berlin, HeidelbergBerlin Heidelberg: Springer; 2005. p. 527–36.
[7] Bicchi A, Tonietti G. Fast and “soft-arm” tactics [robot arm design]. IEEE Robot
Autom Mag 2004;11(2):22–33.
[8] Giorgio G, et al. Variable stiffness actuators: the user's point of view. Int J Robot Res
2015;34(6):727–43. 2015/05/01.
[9] Tsagarakis NG, Laffranchi M, Vanderborght B, Caldwell DG. A compact soft actuator
unit for small scale human friendly robots. 2009 IEEE international conference on
robotics and automation. 2009. p. 4356–62.
[10] Haddadin S, Weis M, Wolf S, Albu-Schäffer A. Optimal control for maximizing link
velocity of robotic variable stiffness joints. IFAC Proc Vol 2011;44(1):6863–71. 1//.
[11] Schempf H, Kraeuter C, Blackwell M. Roboleg: a robotic soccer-ball kicking leg.
Proceedings of 1995 IEEE international conference on robotics and automation. 2.
1995, vol. 2. p. 1314–8.
[12] Kim BS, Song JB. Hybrid dual actuator unit: a design of a variable stiffness actuator
based on an adjustable moment arm mechanism. 2010 IEEE international con-
ference on robotics and automation. 2010. p. 1655–60.
[13] Garabini M, Passaglia A, Belo F, Salaris P, Bicchi A. Optimality principles in variable
stiffness control: the VSA hammer. 2011 IEEE/RSJ international conference on
intelligent robots and systems. 2011. p. 3770–5.
[14] Merritt HE. Hydraulic control systems. New York, NY, USA: Wiley; 1967.
[15] Truong DQ, Ahn KK. Parallel control for electro-hydraulic load simulator using
online self tuning fuzzy PID technique. Asian J Control 2011;13(4):522–41.
[16] Truong DQ, Ahn KK. Force control for hydraulic load simulator using self-tuning
grey predictor – fuzzy PID. Mechatronics 2009;19(2):233–46. 2009/03/01/.
[17] Truong DQ, Ahn KK. Force control for press machines using an online smart tuning
fuzzy PID based on a robust extended Kalman filter. Expert Syst Appl
2011;38(5):5879–94. 2011/05/01/.
[18] Re LD, Isidori A. Performance enhancement of nonlinear drives by feedback line-
arization of linear-bilinear cascade models. IEEE Trans Control Syst Technol
1995;3(3):299–308.
[19] Liu Y, Handroos H. Technical note Sliding mode control for a class of hydraulic
position servo. Mechatronics 1999;9(1):111–23. 1999/02/01/.
35
Mechatronics 52 (2018) 22–35
[20] Wu M-c, Shih M-c. Simulated and experimental study of hydraulic anti-lock braking
system using sliding-mode PWM control. Mechatronics 2003;13(4):331–51. 2003/
05/01/.
[21] Bandyopadhyay B, Janardhanan S, Spurgeon SK. Advances in sliding mode control:
concept, theory and implementation. Berlin, Germany: Springer-Verlag; 2013.
[22] Ho TH, Ahn KK. Speed control of a hydraulic pressure coupling drive using an
adaptive fuzzy sliding-mode control. IEEE/ASME Trans Mechatron
2012;17(5):976–86.
[23] Ahn KK, Nam DNC, Jin M. Adaptive backstepping control of an electrohydraulic
actuator. IEEE/ASME Trans Mechatron 2014;19(3):987–95.
[24] Garagic D, Srinivasan K. Application of nonlinear adaptive control techniques to an
electrohydraulic velocity servomechanism. IEEE Trans Control Syst Technol
2004;12(2):303–14.
[25] Lin F-J, Shen P-H, Hsu S-P. Adaptive backstepping sliding mode control for linear
induction motor drive. IEE Proc Electr Power Appl 2002;149(3):184–94.
[26] Bouabdallah S, Siegwart R. Backstepping and sliding-mode techniques applied to an
indoor micro quadrotor. Robotics and automation, 2005. ICRA 2005. Proceedings
of the 2005 IEEE international conference on. 2005. p. 2247–52. IEEE.
[27] Madani T, Benallegue A. Backstepping sliding mode control applied to a miniature
quadrotor flying robot. IEEE industrial electronics, IECON 2006-32nd annual con-
ference on. 2006. p. 700–5. IEEE.
[28] Chen C-L, Peng CC, Yau H-T. High-order sliding mode controller with backstepping
design for aeroelastic systems. Commun Nonlinear Sci Numer Simul
2012;17(4):1813–23. 2012/04/01/.
[29] Singer NC, Seering WP. Preshaping command inputs to reduce system vibration. J
Dyn Syst Meas Control 1990;112(1):76–82.
[30] Zain MZM, Tokhi MO, Mohamed Z. Hybrid learning control schemes with input
shaping of a flexible manipulator system. Mechatronics 2006;16(3):209–19. 2006/
04/01/.
[31] Agostini M, Parker GG, Groom K, Schaub H, Robinett RD. Command shaping and
closed-loop control interactions for a ship crane. Proceedings of the 2002 american
control conference (IEEE Cat. No.CH37301). 3. 2002. p. 2298–304. vol. 3.
[32] Pereira E, Trapero JR, Díaz IM, Feliu V. Adaptive input shaping for single-link
flexible manipulators using an algebraic identification. Control Eng Pract
2012;20(2):138–47. 2012/02/01/.
[33] Juyi P, Pyung-Hun C, Hyung-Soon P, Eunjeong L. Design of learning input shaping
technique for residual vibration suppression in an industrial robot. IEEE/ASME
Trans Mechatron 2006;11(1):55–65.
[34] Krstic M, Kanellakopoulos I, Kokotovic PV. Nonlinear and adaptive control design.
New York, USA: John Wiley & Sons; 1995.
[35] Rong-Jong W, Kuo-Ho S. Adaptive enhanced fuzzy sliding-mode control for elec-
trical servo drive. IEEE Trans Ind Electron 2006;53(2):569–80.
[36] Niksefat N, Sepehri N. Designing robust force control of hydraulic actuators despite
system and environmental uncertainties. IEEE Control Syst 2001;21(2):66–77.
[37] Oh S, Kong K. Two-Degree-of-freedom control of a two-link manipulator in the
rotating coordinate system. IEEE Trans Ind Electron 2015;62(9):5598–607.
[38] Olsson H, Åström KJ, Canudas de Wit C, Gäfvert M, Lischinsky P. Friction models
and friction compensation. Eur J Control 1998;4(3):176–95. 1998/01/01/.
[39] Slontine JJE, Li W. Applied nonlinear control. NJ: Prentice-Hall; 1991.
[40] Meirovitch L. Elements of vibration analysis. McGraw-Hill; 1986.