Mechanism and Machine Theory: Ying Wu, Kaiping Yu, Jian Jiao, Rui Zhao

Mechanism and Machine Theory 92 (2015) 407435
Contents lists available at ScienceDirect
Mechanism and Machine Theory

journal homepage: www.elsevier.com/locate/mechmt
Dynamic modeling and robust nonlinear control of a six-DOF

active micro-vibration isolation manipulator with
parameter uncertainties
Ying Wu, Kaiping Yu , Jian Jiao, Rui Zhao
Department of Astronautic Science and Mechanics, Harbin Institute of Technology, 92 West Da-Zhi Street, Nangang District, Harbin, China
a r t i c l e
i n f o
Article history:
Received 1 August 2014
Received in revised form 8 June 2015
Accepted 12 June 2015
Available online 28 June 2015
Keywords:
Active micro-vibration isolation
Kane's method
Robust nonlinear
Parameter uncertainties
Stewart platform
a b s t r a c t
Dynamic model and vibration control of the Stewart platform are sophisticate and signicant for
vibration isolation especially the case with the base excitation. However, in the literature, no
complete and accurate solution is presented. In this paper, a novel nonlinear dynamic model of
six-spherical-prismatic-spherical (SPS) Stewart platform with the base excitation has been
derived via Kane's method. Peculiarly, the dynamic model can be utilized for control algorithm
designation to attenuate the base excitation. The order of the nonlinear coefcients has been
evaluated to simplify the model. The uncertainties of stiffness, damping and mass center location
have been studied based on the practical situation. An improved robust nonlinear controller,
which is composed of the linear control part, nonlinear part and excitation compliment part,
has been proposed to satisfy the two requirements at low frequency. The uniformly ultimate
boundary of the state vector has been veried by theoretical proof accompanied with numerical
simulation. It is concluded that the controller is of ne adjustability and can isolate a wide range
of external disturbance.
2015 Elsevier Ltd. All rights reserved.
1. Introduction
Optical satellites must deal with various disturbances on orbit [1]. For sensitive payloads, not being able to attenuate the
disturbances to a sufciently low level can either degrade the performance or cause malfunction. In order to satisfy the increasingly
stringent requirements of optical payload on satellite, many manipulators that can be used for six Degree-of-Freedom (DOF) active
vibration control have been conducted based on Stewart platform [17], which is of high load carrying capacity, good dynamic
performance and precise positioning [8,9]. However, dynamic models of the platform are not accurate enough to meet the high
standards of vibration isolation at low frequency. For instance, parameter uncertainties have not been conceived let alone the
nonlinear properties of the platform.
The methods of dynamic modeling can be primarily divided into two categories: vector method and energy method, both of which
lead to scalar equations but have different starting points [10]. Vector method starts with vector equations, including NewtonEuler
method, momentum principles, D'Alembert principle and Kane's method. Energy method begins with scalar energy expressions
which use Lagrange's equations, Hamilton's canonical equations, the Gibbs equations, or the BoltzmannHamel equations [11].
Dynamic equations formulated via NewtonEuler method [1215] would suffer a heavy computational burden since some constraint
forces are needed to be known. But the method is relatively easier to be understood and more direct than others. The Lagrange's
equation [1620] can be employed to analyze the dynamic characteristics of the platform. The fundamental issue is the selection of
Corresponding author. Tel./fax: +86 0451 8641 4320.
E-mail addresses: wuyinghit@gmail.com (Y. Wu), yukp@hit.edu.cn (K. Yu).
http://dx.doi.org/10.1016/j.mechmachtheory.2015.06.008
0094-114X/ 2015 Elsevier Ltd. All rights reserved.
408
Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435
general coordinates and the expression of the kinematics. Once these are done, the rest is fairly straightforward. Although the forces of
the joints are avoided, partial differentials of the kinematics with respect to general coordinates would cause additional computational
complexity of the equations. Compared with these two aforementioned methods, Kane's method [10,11,2125] which appears to be
distinctly advantageous for multi-body problems, leads to concise and intuitive dynamic equations [11]. Another characteristic of
Kane's method is that it's quite straightforward for the forward dynamics problems [29]. There are additional methods such as
generalized momentum [2628], virtual work [29] and screw theory [30].
Dynamic model of the Stewart platform is sophisticate and important for vibration isolation especially the case with the base
excitation. However, in the literature, no complete and concise solution is presented. If the base excitation exists, the lower platform
can no longer be regarded as an inertial reference. Therefore, we have to set an independent coordinate and calculate the transformation
matrices from the local frames to the inertial one, which will bring additional complexity into the dynamic model. Besides, since the
payload will be mounted on the upper platform, the disturbances can be transmitted to the payload directly without isolation.
Consequently, the dynamic model with the base excitation is essential on which proper control algorithm design will be based. In
this paper, we systematically present the dynamic equations with the base excitation through Kane's method. After a profound investigation of the actuator embedded in the link of the platform, it can be modeled as a piecewise linear spring which will bring structural
nonlinearity into the equations. The new dynamic equation is concise and complete. Moreover, it has explicitly physical meaning.
Apart from the modeling methods, a suitable control strategy is also signicant to multi-DOF vibration control. Generally, the
control strategies [31] can be divided into passive control [32,33], active control and semi-active control [34]. Passive vibration
isolation, which is effective for high frequency, is simple, reliable and stable. But it is not suitable for attenuating low frequency
vibration because two trades-offs potentially exist in the passive vibration system [35]. Active vibration control can address these
two limitations. To meet the actual situation, parameter uncertainties must be taken into account. As for the linearized model, classical
robust control method [3641] such as H and -synthesis can be used to handle these problems [39]. However, neither H control
nor -synthesis can be applied to nonlinear dynamic systems. Robust nonlinear controllers based on Lyapunov function can be utilized
as an alternative, which has been designed for nonlinear tracking control and active vibration control in both task space and joint
space [4249]. These controllers are based on the lemma given in [42] and [43]. Even though some work has been conducted on
this topic, they are far from perfect. Taking an example of the tracking controllers proposed by Kim et al. [45,46], there is a room
for improvement when it is utilized for vibration attenuation, which will be illustrated in this paper. Another one is presented by
Yang et al. [47,48]. The main idea of the controller is to decrease the stiffness matrix as an identity matrix and increase the value of
the damping matrix as much as possible, which seems to be hard to realize in practice.
For the sake of satisfying the two basic requirements in the ultra-low frequency domain, we have developed an improved robust
nonlinear controller by introducing a linear controller and two weighting matrices for the purpose of vibration isolation. The proposed
controller is composed of linear control part, nonlinear part and excitation compliment part. The linear one can be determined by
linear control algorithms, such as PID, LQR, pole placement, etc. The nonlinear one compensates the nonlinear and the uncertain
components. As for the excitation part, it compensates for the effect of base excitation. That the closed loop system can achieve
uniformly ultimate boundary has been proved in this article. In addition, the performance conditions of the system can be achieved.
Likewise, numerical simulations in the time domain demonstrate the ne performance of the controller. We assume that all the six
links are identical throughout the paper.
2. Dynamic model with base excitation
In this section, the dynamic model of the platform with the base excitation will be formulated step by step. The platform has six
DOFs including three displacements and three rotations. To start with, the kinematics of one link is analyzed. Then the partial
velocities and the angular velocities are derived in the matrix form. Thirdly, dynamic equations with base excitation are derived.
Fourthly, the axial force generated by the actuator is modeled and analyzed. Finally, the derived dynamic model is simplied into a
concise form with the geometrical nonlinearities omitted.
2.1. Kinematics analysis of one link
In kinematics analysis, all the vectors are expressed in the inertial coordinate. Fig. 1 depicts the inertial frame O, local frame B and P
that are xed in the center of the lower and upper platform. The origins of local frames xed in the sliding and cylindrical links are also
shown in the gure. Vectors of the six links are denoted as li (i = 1,,6)
*
l1 b16 p12 ; l2 b23 p12 ; l3 b23 p34 ; l4 b45 p34 ; l5 b45 p56 ; l6 b16 p56 :
The transformation matrix from frame B to inertial frame O,

2
R B=O
cc
Rotx; Rot y; Rot z; 4 cs ssc
sscsc
cs
ccsss
sc css
3
s
sc 5
cc
where c and s denote cos and sin, respectively. , and are the Cardano angles of the lower platform.
409
Payload
p56
z
p34
Upper platform
x P
p12
p
b45
One link
xB
z
x
b16
y( b y)
rb
Lower platform
Dis
b23
tur
ban
ce
(a)
x( b x)
(b)
Fig. 1. (a) The schematic diagram of the platform. (b) The top view of the platform.
The relationship between angular velocities of the lower platform and the derivation of the Cardano angles
2
3 2
bx
cos cos
4
b by 5 4 cos sin
bz
sin
sin
cos
0
32 3
0
0 54 5:
1
According to the small magnitude of the vibration

h
b
iT
Likewise, we have
2
RP=O
3
s
sc 5
cc
cc
cs
4 cs ssc ccsss
sscsc sc css
h
p
iT
where R(P/O) denoted the transformation matrix from frame P to inertial frame O. , and are the Cardano angles of the upper
platform.
A link is shown in Fig. 2, two parallel frames ci and si are xed in the mass center of the cylindrical part and the sliding part. The
transformation matrix from ci to O
Tci 2
Rot z; i Roty; i Rotx; i
c i c i c i si si ci s i
4 c i s i c i ci s i s i si
si
ci si
3
s i si c i ci s i
ci s i si si c i 5
ci c i
where i denotes the angle which is about the axis of the link. The angle is independent. Another two angles i and i can be determined from
ui cos i cos i
sin i cos i
sin i :
Obviously,
Tsi Tci
410

Mass center
pik
The sliding part
si
The cylindrical part
rpi
rp
bij
rbi
pi|P
bsi
li
ci
bci
Upper coordinate
t|B
bi|B
rb
rc
B
Base coordinate
Inertial coordinate
Fig. 2. Computational model of a link.
where Tsi is the transformation matrix from frame si to O.

The inertia tensors of the upper platform, the cylindrical part and the sliding part can be expressed as
T
Ip RP=O Ip=p RP=O

Ici
Isi
T
Tci Ici=ci Tci
T
Tsi Isi=si Tsi
10
where I(p/p), I(ci/ci) and I(si/si) are the inertia tensors relative to the center of the mass which are expressed in the local coordinates. Ip,
Ici and Isi denote the inertia tensors expressed in the inertial coordinate.
Link vector
li RP=O pijP RB=O tjB RB=O bijB pi tbi
11
where p(i|P), t(|B) and b(i|B) represents the vectors expressed in the corresponding local frames. pi, t and bi denote the vectors expressed
in the inertial frame.
Link length
Li li :
12
Unitary vector of one link

ui li =Li :
13
The absolute velocities and accelerations of bij and pik

rbi rb bi ; rpi rp pi
rbi rb b bi ; rpi rp p pi
rbi rb b bi b b bi ; rpi rp p pi p p pi
14
where p and b are the angular velocities of the upper and the lower platform.
Likewise, we have
rbi li i li rpi
15
rbi li i i li i li 2i li rpi
16
where i and i denote the angular velocity vector and the angular acceleration vector of one link, respectively. li and li denote the
sliding velocity and acceleration of one link. Take the cross product of Eq. (15) with ui and consider the condition i li = 0,

1

1
u rp p pi ui rb b bi :
Li i
Li
17
411
The sliding velocity of one link

T
T
li Li ui ui ui rp p pi ui ui rb b bi :
18
The angular and sliding accelerations of one link

i

1

1
ui rp p pi p p pi ui rb b bi b b bi
Li
Li

l L u u uT r p p u uT r b b l 2 :
i
i i
i i
p
p
i
p
p
i
i i
b
b
i
b
b
i
i
19
20
Additionally, Eqs. (18) and (20) can be rewritten as matrix form

h
Li uTi
~i
ui p
h
L uT
i
i
T T
~i
ui p
T T
i
h
rp
uTi
p
"
#
i r
h
p
uTi
p
T ~T
ui b
i
i
T ~T
ui b
i
rb
b
Jpi

rp
T
r
Jbi b
p
b
21
"
#

i
rp
r
rb
2
T
T
2
li Jpi
Jbi b Li
b
b
p
22
where the superscript tilde represents the skew matrix of a vector and

Jp
u1
~ 1 u1
p

u1
u6
;
J
b
~ u
~ 6 u6
p
b
1 1
u6
~ u
b
6 6

23
denote the Jacobi matrix and another ordinary matrix, respectively.

The velocities and the accelerations of the mass center of the cylindrical part and sliding part can be expressed as
24
25
rci rb b bi i bci ui
rsi rp p pi i bsi ui
aci

bci
T
Eui ui rp p pi p p pi
Li

b
T
rb b bi b b bi bci 2i ui
E ci Eui ui
Li

b
T
rp p pi p p pi
asi E si Eui ui
Li

b
T
2
si Eui ui rb b bi b b bi bsi i ui
Li
26
27
where E is the third order identity matrix, bci and bsi are as shown in the Fig. 2.
The position of the mass center of the payload can be expressed as
rco rp rc
28
in which rc = R(p|O)r(c|p) denotes the position vector of the payload expressed in the inertial coordinate. The velocity of the mass
center
rco rp p rc
29
and the acceleration

rco rp p rc p p rc :
30
412
2.2. The partial velocities and partial angular velocities

Since the kinematics of one link and the payload have been illustrated, the partial velocities and partial angular velocities can be
expressed. The general coordinates are selected to be the six DOFs of the upper platform
h
i
T
T
q x y z rp ; p :
31
The general velocities

h
q x y
i h
i
T
T
rp ; p :
32
The partial velocities and the partial angular velocities of the upper platform
rco h T i
E; ~rc
q j
vm; j
33
m; j
p
q j
0; E:
34
The subscript j = 1,2,, 6 denotes the six coordinates.

The partial angular velocities of one link
i; j
"
#
~ u
~p
~T
u
i
i; i i :
Li Li
q j
35
The partial velocities of the sliding velocities

vi; j
li
q j
h
i
T
T T
~i :
ui ui ; ui ui p
36
The partial velocities of the cylindrical part and the sliding part of one link
vci; j

b
rci
b
T
~ Ti u
~ip
~ Ti
ci Eui ui ; ci u
Li
Li
q j
37
vsi; j

rsi
b
b T
T
~i u
~ip
~ Ti :
~ Ti si u
E si Eui ui ; p
Li
Li
q j
38
2.3. The dynamic equations of the system

The axial forces exerted by the actuator, the spring and the damper of one link are denoted as Fi, then the general active forces can
be expressed as
F j mg vm; j
6
X
F i ui vi; j
i1
6

X
mci g vci; j msi g vsi; j :
39
i1
The general inertial forces

F j mam vm; j Im Im m; j
6

X
mci aci vci; j msi asi vsi; j
i1
6
6
X
X
Ici i i Ici i i; j Isi i i Isi i i; j
i1
40
i1
where m represents the sum of mass of the upper platform and the payload. msi denotes the mass of the sliding part. mci denotes the
mass of the cylindrical part. g = (0,0,g)T is the gravity acceleration.
413
The Kane's equation

F j F j 0:
41
Substitute the partial velocities, partial angular velocities and the accelerations into the Eq. (41) and omit the gravitational effects
since the satellite is in the microgravity environment. After a complicated derivation, we can obtain the nal form of the dynamic
model with the base excitation
p p Mb x
b b J p F
Mx
42
where
2
6
X
Q pi
6 mE
6
i1
M6
6
6
X
4 ~
~ i Q pi
p
mrc
2 6
X
Q bi
7
6
7
6
i1
7; Mb 6 i1
7
6X
6
6
X
4
T
~ i Q pi p
~ Ti 5
~ i Q bi
p
p
Im m~rc ~rc
T
m~rc
6
X
~ Ti
Q pi p
i1
i1
i1
3
6
X
~T 7
Q bi b
i
7
i1
7
7
6
X
T
~ 5
~Q b
p
i
bi i
i1
3
6
X
Q b bi 7
6

7
6 i1 bi b
ui
7; xp rp ; xb rb
6
;
b
7
6
6
~ 6 u6
p
p
b
5
4 X~
pi Q bi b b bi
2

Jp
u1
~ 1 u1
p
i1
3 2 6
3 2 6 T
3
6

X
Xu
X
~i
2
Q pi p p pi
msi bsi i ui 7 6
i Ici Isi i 7
7 6
6
7 6 i1
7 6 i1 Li
7
6
i1
76
76
7
p 6
7 6X
7 6X
7
6X
6
6
6 ~ ~T

pi ui
5
5
5
4
4
4
2
~ i Q pi p p pi p Im p
~ i ui
p
msi bsi i p
i Ici Isi i
L
i
i1
i1
i1
2
F F1
F2
F3
F4
F5
F6
in which
Q pi mci

2 1
b2ci
b
T
T
~ Ti Ici Isi u
~i
Eui ui msi E si Eui ui
2u
2
Li
Li
Li

1
b b
b b
T
T
~ Ti Ici Isi u
~i:
Q bi mci 1 ci ci Eui ui msi 1 si si Eui ui 2 u
Li Li
Li Li
Li
43
44
2.4. The axial force of the prismatic joint

In this part, the axial forces generated by the prismatic joints that consist of a spring, a damper and an actuator will be discussed.
According to the practical structure of the actuator, the device can be modeled as Fig. 3, in which k1 represents the stiffness when the
motion of the wedge is limited in the gap. Once the wedge contacts with other springs, the sum of stiffness of the spring combination is
denoted as k2. Assuming the gap range is [a0, a0], and then the force generated by the spring can be expressed as
f ki
8
i ba0
< i k i a0
a0 i a0
k1 i
:
i k i a0
i Na0
45
where i = li l0 and k = k2/k1 1. Therefore, the completely axial force can be written as
f i f ki f ci f ai
46
414
a0
a0
=li l0
k1
k2
Fig. 3. The computational model of the actuator.

in which, f ci ci i denotes the damping force and fai represents the active force generated by the voice coil. The joint force of all the
six links can be written as
F K1 k K1 gCl Fa
47
where K1, Cl are diagonal matrices. Each term of vector g() is

8
< i a0
g i 0
:
i a0
i ba0
a0 i a0 :
i Na0
48
Although the piecewise function is accurate to describe the stiffness, it is not convenient for dynamic analysis or for controller
design. As a compromise, a cubic function is utilized to t the piecewise curve. According to the practical actuator, a0 = 3 103 m
and the maximal range is 5 103 m. Thus, the tting function can be determined as
0
g i i a0 i i a0 i i
49
where = 2.5 104 m2 and = 0.225.

The vector of axial force can be expressed as
F Kl k K1 Cl Fa
50
where Kl = (1 k)K1, and ().3 represents the cubic of each term instead of the whole one.
Reconsider Eq. (21), we have

Li Li0
r
T
T
r
Jpi p Jbi b
p
t
b
51
where t is a tiny time period.
Li Li0 Jpi

rp
T rb
Jbi
p
b
52
415
where rp = pt, p = pt; rb = bt, b = bt denote the displacements relative to the origin position. If initial conditions
are zero, then xp = xp and xb = xb, therefore

T
T
3
T
T
F Kl Jp xp Jb xb k K1 Cl Jp xp Jb xb Fa :
53
2.5. Dynamic model simplication

The dynamic equations shown in Eq. (42) are sophisticated because of b and p. However, consider the characteristics of
micro-vibration and the small magnitude of these terms, the geometrical nonlinearities can be omitted. Since and K1 are both of
large magnitude, the components which contain structural nonlinearity have to be reserved. The effects of the geometrical and the
structural nonlinearity will be illustrated in Section 5 Numerical simulation, where we will demonstrate the geometrical nonlinearity
can be omitted by qualitative analysis and numerical simulation.
Substitute Eq. (53) into Eq. (42), we can obtain the dynamic equations expressed in the task space
T
p Jp Cl Jp xp Jp Kl Jp xp k Jp K1 w Jp Fa
Mx
54
b Jp Cl JTb xb Jp Kl JTb xb represents the base excitation.

where w Mpb x
Reconsider Eq. (22) and eliminate the nonlinear terms, we have
T T
p JT
x
p Jp Jb xb :
55
Substitute Eqs. (50) and (55) into Eq. (42), we obtain dynamic equation expressed in the joint space
C K K w F
Ml
l
l
k
1
a
56
T T
T
where Ml = J1
and w0 J1
p MJp
p Mpb MJp Jb xb .
The dynamic model has the explicit expressions of the mass, damping and stiffness matrices, whose physical meaning is
transparent and form is more concise. Based on the model, the parameter uncertainties can be introduced and nonlinear vibration
controller can be designed consequently. The parameters of the platform are listed in Table 1. It costs only 0.148 s to compute the
system matrices by Matlab.
2.6. The validation of the dynamic model

It is necessary to validate the dynamic model for control purpose. In order to prove the dynamic model is correct, we have built a
dynamic model of the Stewart Platform in Adams/View (see Fig. 4). The corresponding parameters are shown in the Table 1. The
dynamic model is composed of the upper/lower platforms and six links (contains the sliding and the cylindrical parts). The two
platforms and the links are modeled as cylindrical structures. The lower platform is attached to the ground. One end of the sliding/
cylindrical part is modeled as the spherical joint; the other one is modeled as the sliding constraint. The two parts of the links are
connected by a spring-damper conguration.
Table 1
Parameters of the platform.
Parameters
Value
Location of the mass center

The radius of the upper platform
r(c|p) = [0, 0, 0.2] m

p
r p 0:1 6 m
p
r b 0:2 6 m
p
h 0:3 2 m
bci = 0.24 m
bsi = 0.18 m
mc = 0.6 kg
ms = 0.4 kg
m = 98 kg
I(ci|ci) = diag[6.32, 180, 180] 105 kg m2
I(si|si) = diag[8.71, 1100, 1100] 105 kg m2
I(m|p) = diag[1.552, 0.490, 1.552] kg m2
Cl = diag[10, 10, 10, 10, 10, 10] N s m1
K1 = diag[104, 104, 104, 104, 104, 104] N m1
K = diag[15, 15, 15, 15, 15, 15] N m1
C = diag[0.5, 0.5, 0.5, 0.5, 0.5, 0.5] N s m1
r = 2.5 %
The radius of the lower platform

The height of the platform
Distance between ci and bij
Distance between si and pik
Mass of the cylindrical link
Mass of the sliding link
Mass of the upper platform and the payload
Inertial tensor of the cylindrical part
Inertial tensor of the sliding part
Inertial matrices of upper platform
Damping matrix of the link
Stiffness matrix of the link
The perturbation of the stiffness
The perturbation of the damping
The perturbation of the mass center
416
Fig. 4. The dynamic model and results in Adams.
There are some essential tips for building the dynamic model in the software. In the theoretical model, the mass center is not at the
geometrical center. However, the location of the mass center in the model built by Adams is automatically set at the geometrical
center. To solve this problem, we have built a sphere with a very small radius in order to tune the mass center to be the same as
that in Table 1. We should keep in mind that the modal analysis for a nonlinear system is not regular although there exists nonlinear
modal analysis in the literature. For comparison and simplication, we have omitted nonlinear component in the dynamic model here.
We applied the vibration analysis module in Adams to carry out a modal analysis for the model. And the natural frequencies calculated
by the theoretical model can be identied in Fig. 6. The two results (see Table 2) are almost the same indicates that the analytical
model is veried.
3. Dynamic model with parameter uncertainties
A pivotal objective of control algorithms is robust to variations between the actual system and the nominal model on which the
control system design is based [10]. For active micro-vibration isolation systems, the primary uncertain parameters of interest are
the payload mass, stiffness, damping and mass center of the payload. But the mass matrix appeared in the system matrix as the
products with the stiffness and damping. Hence, mass uncertainties are not necessary to be considered independently. In this section,
the uncertain dynamic equations will be derived for parameter uncertainties in stiffness, damping and the mass center location of the
payload.
The position vector of mass center of the payload
rco rp ^rc
57
where ^rc rc 1 r , r is the installation error of the payload. Therefore

M M0 M
58
where M0 represents the nominal mass matrix. M = rMc and

"
Mc
0
m~rc
#
T
m~rc
T :
m~rc ~rc
59
Table 2
The natural frequencies calculated by Adams and theoretical method (unit:Hz).
Adams
Ours
1.4759
1.4682
1.5102
1.5024
2.7535
2.7541
4.8458
4.8661
5.2117
5.2050
7.7614
7.7748
417
In addition, take account of the uncertainties of the stiffness and the damping, the dynamic equations can be expressed as,
T
p Jp Cl Cl Jp xp Jp Kl Kl Jp xp k Jp K1 w Jp Fa :
M0 Mx
60
This is the dynamic model of the platform with parameter uncertainties. The base excitation also contains the uncertain
component; however, they can be included in the disturbance vector w.
Set
3
k Jp K1 ; Jp Fa
C0 Jp Cl Jp ; C Jp Cl Jp ; C C0 C
K0 Jp Kl Jp ; K Jp Kl Jp ; K K0 K:
We have
p C0 Cxp K0 Kxp w :
M0 Mx
61
According to matrix theory and omit the higher order parts

1
M0 M1

1
1
1
1
1
EM0 M1 M0 M0 r M0 Mc M0 :
62
Hence
1
C M0 C0 r M0 Mc M0 C0 M0 C
K M0 K0 r M0 Mc M0 K0 M0 K:
63
64
As a result
1
1
1
1
p M1
w M M
x
0 K0 xp M0 C0 xp M
65
where the uncertain part

1
1
1
1
1
1
r M0 Mc M0 K0 M0 K xp r M0 Mc M0 C0 M0 C xp :
Set x xTp
66
T T
xp as the state vector of the system, we can have

x Ax B B W x;
67
in which

A
0
1
M0 K0
1

0
0
E
;B
1
1 ; B
1
1 BH
M0
r M0 Mc M0
M0 C0
H r Mc M0 ; W Bw;
W Bw; x; Bn
1
1
1
n r Mc M0 K0 K xp r Mc M0 C0 C xp r Mc M0 w:
4. Robust nonlinear controller development

In this section, an improved robust nonlinear controller is developed based on the nonlinear dynamic model with parameter
uncertainties. By introducing the weighting matrices, the proposed controller can be tuned to isolate the micro-vibration at low
frequency perfectly. In this article, the norm of a vector and a matrix is dened as the Euclidean norm (|| ||), and the maximal
singularity of a matrix is dened as . Besides, the minimal singularity of a matrix is denoted as .
418
4.1. Control objectives and some assumptions

Based on the control objectives presented in previous studies, the target frequency range in this paper is focused on 0.7512 Hz.
There are two stringent requirements of the vibration control performance:
i. Keep the low frequency signal (0.7510 Hz) attenuation no less than 6 dB in all directions.
ii. For the high frequency signal (N10 Hz) and noise, the attenuation is not less than 11 dB.
To achieve these two requirements in a wide frequency range especially in the low frequency domain, an active control method
needs to be applied. For better understanding of the proof appearing in the next section, rst of all, we make a few assumptions in
this part.
Assumption i. Energy of the base excitation is bounded, that is
s
Z T
T
W Wdt W b w :
68
In some vibration control systems, the signals of external disturbances are difcult to be predicted accurately. However, the upper
bound of the disturbance can be estimated through effective identication algorithms or other equipment.
Assumption ii. The term n that contains nonlinearities and uncertainties of the system is bounded.
n :
69
This assumption seems impossible to be acted in practice since the upper bound cannot be precisely acknowledged. However, a
large value can be set in advance to cover the boundary, which is possible in real situations.
Assumption iii. Consider r is very small, therefore
H 1:
70
4.2. Robust nonlinear controller development

Subject to the assumption discussed in the previous section, the controller can be expressed as
Gx f 1 f 2
71
where G can be obtained through linear control algorithm, such as LQR, pole placement, H and so forth, which will ensure = A BG
as a Hurwitz matrix. It means that there exists a positive denite matrix P satisfying
T
A P PA Q
72
in which Q is also a positive denite matrix.

Another two terms appeared in the controller can be expressed as
8
D1 1 x; t
>
>

<
x; t 1 x; t
1
f 1 x; t
>
>
: D1 1 x; t 1 x; t b
73
8
D2 2 x; t
>
>

<
x; t w 2 x; t
2
f 2 x; t
>
> D2 2 x; t
:
2 x; t b
w
74
where
T
1 x; t B x; t x V x; t B Px
75

T
2 x; t w B x; t x V x; t w B Px
419
76
and D1, D2 are the two weighting matrices and both of them are diagonal.
In addition, the two coefcients appeared in the controller satisfy the following restrictions

1

D1 H D1

w

1
w :
D2 H D2
77
78
The robust nonlinear controller is composed of the linear part, the nonlinear and uncertain part (Eq. (73)), the external excitation
part (Eq. (74)). And moreover, two weighting matrices are introduced into the algorithm to make the controller have ne adjustability. The controller proposed in this paper is an extension of the controller proposed by Kim et al. [44,45], where the controller was used
for tracking purpose. Their controller can be applied for vibration isolation by modifying the state space equations and tuning the control parameters. However, the two requirements cannot be satised by using Kim's controller. In Section 5.3, we will make a comparison between our controller and the reference's to show our modied controller is of good adaptability and adjustability.
4.3. Stability analysis and performance analysis
Theorem. With the three assumptions mentioned before, the state vector of the system will achieve uniformly ultimate boundary by applying the feedback control Eq. (71).
Proof. Dene a Lyapunov function as
V
1 T
x Px
2
79
where P is a positive denite matrix. The maximal and the minimal eigenvalues of P are max and min. Let
1
1
1
; :
2 min 2 2 max
80
Then
2
1 x V 2 x :
81
The time derivation of V along the trajectory of the dynamic model Eq. (67) is given as
V x P x x PAx B BH w Bn
1 T
T
T
x Q x x PBHGx x PB f 1 H f 1 n f 2 H f 2 w
2

1
2
T
min Q max PBHG x x PB f 1 Hf 1 n f 2 Hf 2 w
2
82
i. When 1(x, t) and 2(x, t) , thus
x PB f 1 H f 1 n f 2 H f 2 w
!
!
D1 1 x; t
D1 1 x; t
D2 2 x; t
D2 2 x; t
T
T

x PB
x; t H x; t n x PB w x; t w H x; t w
1
2
2

1

1 x; t D1 H D1
2 x; t D2 H D2 w
w
0
83
420
Take account of Eqs. (77) and (78), we can obtain

V 3 x
84
in which 3 12 min Q max PBHG:

ii. When 1(x, t) b and 2(x, t) b , thus

T
0
x PB f 1 Hf 1 n f 2 Hf 2 w
2 D1
H D1 1 x; t 1 x; t
1 x; t
w
2 D2
H D2 2 x; t 2 x; t
2 x; t
w
2 D2
2 D2
1 x; t
D1 1 x; t 2 x; t
D2 2 x; t
D1 D2
85
As a result
V 3 x
D1 D2
:
4
86
iii. When 1(x, t) and 2(x, t) b , thus

T
x PB f 1
H f 1 n f 2 H f 2 w
1 x; t D1 H D1
D2
2 D2
2 x; t
D2 2 x; t
Fig. 5. The computational procedure of P and the control block.
87
421
Then
V 3 x
D2
:
4
88
iv. When 1(x, t) b and 2(x, t) , thus

T
0
x PB f 1 Hf 1 n f 2 Hf 2 w

2 D1
1 x; t
D1 1 x; t 2 x; t D2 H D2 w
w
D1
:
89
Hence
V 3 x
D1
:
4
90
Consider all the four cases, we can conclude that: for arbitrary , we have
V 3 kxk
D1 D2
:
4
91
Combine Eqs. (81) and (91) and according to the lemma given by Bridge [43], we can derive the following inequality
consequently
kxk

1
2
2
D1 D2
2 3t
3t
kx0k e 2 2
1e 2
:
1
41 3
92
xb to xp
yb to yp
-50
-1
10
50
System: untitled1
Frequency (Hz): 1.5
Magnitude (dB): 35.8
10
Magnitude (dB)
Magnitude (dB)
50
System: untitled1
Frequency (Hz): 7.77
10
Frequency (Hz)
10
10
-50
-1
10
System: untitled1
10
10
Frequency (Hz)
10
10
zb to zp
-50
-1
10
50
Magnitude (dB)
Magnitude (dB)
50
System: untitled1
10
10
Frequency (Hz)
10
10
-100
-1
10
10
10
10
Frequency (Hz)
50
Magnitude (dB)
Magnitude (dB)
System: untitled1
-50
50
System: untitled1
Frequency (Hz): 1.5
System: untitled1
-50
-1
10
System: untitled1
10
10
Frequency (Hz)
10
-50
0
10
System: untitled1
Frequency (Hz): 5.2
10
10
Frequency (Hz)
Fig. 6. The transfer curves from the base excitation to the upper platform.
10
422
We can nd that if the time tends to innite, then
lim kxk
1
2
s
2 D1 D2
:
1 3
93
It demonstrates that uniformly ultimate boundary of the signal of the state vector is guaranteed. Therefore, by selecting ,
weighting matrices and P appropriately, we can achieve the two basic targets given before.
4.4. The computation of some matrices

The proposed robust nonlinear controller seems not straightforward to be applied because some parameter matrices, such as P,
cannot be selected properly. As for the weighting matrices, since they are diagonal, we will propose three principles to tune them
in order to satisfy the requirements. Assume matrix Q is given in advance, then Eq. (72)
T
ABG P PABG Q :
94
Obviously,

T
T
A P PA Q BG P PBG2kPBGk:
Structural nonlinearity only
95
Structural and geometrical nonlinearity
0.01
5
response (m or rad)
response (m or rad)
0.005
0
-0.005
-0.01
0
10
time (s)
15
-5
response (m or rad)
5
10
time (s)
15
20
0.01
0.005
x 10
0
-0.005
-0.01
0
-3
5
5
10
0time (s)
15
20
x 10
10
time (s)
15
10
time (s)
15
10
time (s)
15
-3
-5
-10
0
x 10
-3
2
0
-2
-4
0
-5
-10
1
20
response (m or rad)
response (m or rad)
response (m or rad)
-10
0
20
0.01
-0.01
0
x 10
without structural nonlinearity
-3
3
time (s)
Fig. 7. Responses of the platform (A = 5 mg, k = 0.8).
20
423
Hence,

1

T
A P PA Q kPBGk:
2
96
Assume
3
97
in which N 0. According to Eq. (84), we can obtain the following inequality

1
Q max kPBHGk:
2 min
98
Combine Eqs. (96) and (98), we have

(

T
kPBGk 1=2A P PA Q
:
max kPBHGk 1=2min Q
99
The computational process is illustrated in Fig. 5.

As for the two introduced matrices D1 and D2, we can select them depended on the following principles:
i. They had better be diagonal matrices for convenient.
ii. The rst three terms of D1 are equal, so do the last three terms.
iii. The rst three terms of D2 are equal, so do the last three terms.
structural nonlinearity only
structural and geometric nonlinearity
without structural nonlinearity

y
0.02
response (m or rad)
response (m or rad)
0.01
0
-0.01
-0.02
0
10
time (s)
15
20
0.01
0
-0.01
-0.02
0
10
time (s)
15
20
10
time (s)
15
20
10
time (s)
15
20
response (m or rad)
0.01
0
-0.01
-0.02
0
10
time (s)
15
20
0.02
response (m or rad)
response (m or rad)
response (m or rad)
0.02
0.01
0
0.01
-0.01
-0.02
0
010
time (s)
15
20
0.01
0
-0.01
-0.02
0
0.01
0.005
0
-0.005
-0.01
0
-0.01
-0.02
Fig. 8. Responses of the platform (A = 10 mg, k = 0.8).
424
0.012
0.035
0.040
0.050
0.100
0.150
Amplitude (m)
0.01
0.008
0.006
0.004
0.002
0
2.2
2.4
2.6
2.8
3
Frequency (Hz)
3.2
3.4
Fig. 9. Amplitude-frequency characteristic of x (k1 = 1 104 N m 1, k = 0.1).
Remark. The last two points mean that the weighting matrices can be selected as the following forms
D1
d1
d1
d1
d2
d2
d2
; D2
1 d1
1 d1
1 d1
2 d2
2 d2
2 d2
where d1, d2, 1 and 2 are real numbers.

In this article, the matrix Q is also selected as a diagonal matrix whose diagonal terms are equal. We set Q = 3E12 where E12
denotes the twelfth order identity matrix.
The LQR algorithm is utilized to calculate the feedback matrix G. First, two parameter matrices are chosen as

Q u 1
K0
0

0
; Ru 2 E6
M0
where v1 and v2 are two coefcients which are needed to be determined, E6 denotes the sixth order identity matrix. Since the matrix G
depends on v1/v2, we x v1 = 1 and make 2 [0, 0.1]. After some trials, we nd that when v2 = 0.05, the inequality Eq. (99) can be
satised.

T
PBG 58:6623
max PBHG 5:9876
PBG 1=2A P PA Q
T
1=2A P PA Q 28:4166 1=2min Q 7:8933
max PBHG 1=2min Q
Consequently, the matrix P can be calculated. The specic value of G and P are listed in the Appendix A.
-3
1.5
x 10
response of x (m)
0.5
-0.5
-1
0
5
time (s)
Fig. 10. Response of x with initial condition changing (k1 = 1 104 N m 1, k = 0.8).
10
425
5. Numerical results and discussion

This section contains three parts: the effects of geometrical and structural nonlinearity, nonlinear vibration analysis and the
validation of the robust nonlinear controller. The rst one is essential to identify the effects of the geometrical nonlinearity is little
that we can omit them in the dynamic modeling process. The nonlinear dynamic analysis is based on the model expressed in the
link space since the excitation is only composed of acceleration. The other reason is the fact that we can judge the response of each
link whether it exceeds the maximal range of actuator or not. As for the validation of the controller, we just have to focus on the
dynamic model expressed in the task space since attenuating the responses of payload rather than the prismatic joint is our objective.
5.1. The effects of geometrical and structural nonlinearity
Even though the geometrical term p appeared in Eq. (42) is of the second order of the rotational velocities and the structural
nonlinearity k JpK13 is of the third order, we can still omit p and retain kJpK13 for several reasons. First of all, the magnitude
of Q pi, Q bi, pi(i = 1, 2, , 6) and li (i = 1, 2, , 6) are 101. The angular velocities of the six links depend on xp and xb. The maximal
value of the transfer functions from xb to xp is around 40 dB (see Fig. 6). The magnitude of the external disturbance is less than 103.
Consequently, the magnitude of the geometrical nonlinearity will be less than 104. However, the magnitude of the geometrical
nonlinearity is more than 103 due to the large value of the coefcients and K1.
To show the effects of geometrical nonlinearity is little and the structural nonlinearity cannot be omitted, two sinusoidal signals are
applied as the external disturbances to stimulate the system. The amplitudes of the acceleration are 0.05 ms2 and 0.1 ms2,
respectively; the frequency is 0.5 Hz.

b w w w 0:5w 0:5w 0:5w T

x
w A sin2 f t
100
where A and f denotes the amplitude and the frequency of the excitation, respectively. In the following sections, the excitations have
the same form. Three cases are conducted:
Case i. Consider the structural nonlinearity only.
Case ii. Consider both the structural and geometrical nonlinearity.
Case iii. Consider the geometrical nonlinearity only.
x 10
Kim et al.
response(m or rad)
Wu et al.
x-displacement
-5
20
x 10
25
30
35
x 10
-5
20
25
-3
25
30
35
-0.01
20
response(m or rad)
response(m or rad)
30
35
30
35
25
time(s)
0.02
0.01
0
25
35
0.01
time(s)
-0.01
20
30
time(s)
-5
20
Yang et al.
y-displacement
-3
time(s)
z-displacement
response(m or rad)
response(m or rad)
response(m or rad)
Without control
-3
30
35
x 10
-3
-2
20
time(s)
25
time(s)
Fig. 11. Displacement responses of the platform (0.75 Hz).
426
The response curves of the platform under the excitations are shown in Figs. 7 and 8.
It is obvious to see that the response curves of Case i and Case ii are almost identical (the green and blue lines are overlapping)
which indicates the effect of the geometrical nonlinearity is little. However, if we get rid of the structural nonlinearity, the response
curves have changed a lot under the two kinds of excitations. It is concluded that the geometrical nonlinearity does a diminutive effect
to the dynamic system, but the structural nonlinearity must not be omitted especially when the amplitude of the excitation becomes
higher.
5.2. Nonlinear vibration analysis
The common phenomena of nonlinear vibration are jumping phenomenon, internal resonance, saturation phenomenon, diosmose
phenomenon, etc. We analyze the jumping phenomenon through sweeping frequency from 2.3 Hz to 3.3 Hz, and free vibration of the
system with initial conditions changes. Fig. 9 shows the jumping phenomenon appears when the amplitude of the acceleration
increases from 0.035 ms2 to 0.15 ms2.
To identify whether the system is sensitive to the initial condition, we set l1 range from 0.001 m to 0.0015 m and the rest maintain
zero. The result is shown in Fig. 10. In the gure, we can know that the nonlinear dynamic system is not sensitive to the initial
conditions since the amplitude changes little when the initial condition varies. Furthermore, that the period of the vibration does
not change implies the characters of the system remain unchanged.
5.3. Validation of robust nonlinear controller
Wu et al.
Kim et al.
response(m/s or rad/s)
Without control
x-velocity
0.1
-0.1
20
25
30
35
time(s)
z-velocity
0.2
-0.2
20
25
Yang et al.
y-velocity
0.1
-0.1
20
25
30
35
-0.2
20
25
35
30
35
0.1
z-velocity
0.04
0.03
0.02
30
time(s)
30
time(s)
0.5
25
35
0.2
time(s)
-0.5
20
30
time(s)
We have developed an improved controller by introducing two weighting matrices, computing several important matrices and
separating the coefcients into and w. As a result, the improved robust nonlinear controller is of ne adjustability, adaptability
and high efciency. In particular, the controller is quite suitable for vibration control at the low frequency range and the corresponding
control force is smaller than its counterparts. By applying various kinds of excitation, we will validate the effectiveness of the proposed
robust nonlinear controller and make comparisons among our controller, the one presented by Kim et al. and the controller developed
by Yang et al. [47,48] in this section. Our main attention is concentrated on the frequency ranging from 0.75 Hz to 12 Hz, since
35
0.01
-0.1
20
25
time(s)
-0.01
-0.02
-0.03
-0.04
20
20.5
21
21.5
22
22.5
time(s)
23
23.5
24
Fig. 12. Velocity responses of the platform (0.75 Hz).
24.5
25
Wu et al.
Kim et al.
Yang et al.
2
1
50
control force(N)
control force(N)
50
0
-50
-100
0
10
15
20
time(s)
3
25
30
control force(N)
control force(N)
5
10
15
20
time(s)
5
25
30
10
15
20
time(s)
4
25
30
35
10
15
20
time(s)
6
25
30
35
10
15
20
time(s)
25
30
35
-100
0
35
100
control force(N)
200
control force(N)
-50
100
100
0
-100
0
-100
0
35
100
-100
0
427
10
15
20
time(s)
25
30
50
0
-50
0
35
Fig. 13. The axial forces generated by the six actuators (0.75 Hz).
excitations with a higher frequency can be well isolated via passive approaches, which can be obtained directly from the Fig. 6. The
expressions of the external disturbances are as Eq. (100) shown and the maximal acceleration of the excitation is set at 0.05 ms2
(5 mg). The time of excitation lasts for 25 s. The rest 10 s is used for free vibration. The parameters of our controller are presented
in the Appendix. To clearly illustrate the responses in the following gures, we set the time interval as [20 s, 35 s].
The six gures appeared in the Figs. 11 and 12 are the displacement and velocity response curves of the platform under the
excitation of 0.75 Hz. The maximal and minimal attenuation of vibration is 30.8 dB and 8.7 dB when our method is applied to the
system. There is also some effect of vibration isolation except in the direction z by utilizing the controller presented by Kim et al.
Consequently, the displacements and velocities of the six directions cannot be isolated by using the controller at the same time.
That is the rst motivation why we introduce the weighting matrices and separate the controller into different parts. As for the
responses of the system with the controller developed by Yang et al., the effect is not obvious for the displacement isolation. However,
the amplitudes of the velocity responses of the system applied by Yang's controller have been magnied, which implies the controller
loses efcacy for velocity isolation. It can also be observed that once the external disturbance is removed, there still exists quite a long
time for the open-loop system to attenuate the vibration to small enough which is due to the slight damping ratio of the system. The
vibration of the closed-loop system of our and Yang's method will reach to zero in a very short time, which is noteworthy for
Table 3
The maximal displacement responses under various excitations.
Cases
Without control
Wu et al.
Kim et al.
Yang et al.
0.75 Hz
2 Hz and 0.75 Hz
12 Hz
Swept sinusoid
Displacement
(m)
Rotation
(rad)
Displacement
(m)
Rotation
(rad)
Displacement
(m)
Rotation
(rad)
Displacement
(m)
Rotation
(rad)
4.129E-3
2.533E-3
3.024E-3
5.025E-4
2.112E-4
1.106E-3
9.164E-4
4.574E-4
2.379E-3
2.354E-3
1.547E-3
3.845E-3
6.812E-3
1.011E-2
1.937E-2
6.234E-4
1.725E-3
2.322E-4
1.542E-3
2.832E-3
2.764E-5
7.464E-4
8.973E-4
2.955E-4
3.998E-3
2.402E-3
3.687E-3
5.025E-4
2.130E-4
1.125E-3
1.227E-3
6.725E-4
3.035E-3
3.343E-3
2.279E-3
5.823E-3
7.393E-3
1.148E-2
2.528E-3
6.358E-4
1.758E-3
2.234E-4
1.718E-3
3.756E-3
3.154E-5
1.269E-3
1.484E-3
4.052E-4
8.946E-6
4.078E-6
2.675E-6
1.463E-7
7.941E-8
5.104E-7
9.381E-7
6.540E-7
2.028E-7
3.061E-7
2.179E-7
3.882E-7
1.421E-5
4.301E-5
3.326E-6
1.869E-7
3.736E-7
6.098E-8
1.573E-6
2.834E-6
3.072E-8
2.708E-8
3.311E-8
1.311E-8
9.516E-5
6.638E-5
7.840E-5
1.668E-6
9.434E-7
6.159E-6
1.261E-5
7.703E-6
2.124E-6
7.915E-7
4.853E-7
1.148E-6
2.271E-4
4.117E-4
6.273E-5
2.380E-6
7.859E-6
9.376E-7
2.249E-5
4.934E-5
4.975E-7
1.215E-7
1.055E-7
3.451E-8
428
Table 4
The maximal control forces under various excitations (unit: N).
Cases
0.75 Hz
Wu et al.
Kim et al.
Yang et al.
2 Hz and 0.75 Hz
12 Hz
Swept sinusoid
Displacement
Rotation
Displacement
Rotation
Displacement
Rotation
Displacement
Rotation
25.76
5.651
23.42
55.28
76.80
72.12
36.21
15.44
13.60
47.57
36.38
10.54
62.55
105.1
83.05
50.60
42.35
17.63
30.56
5.820
26.25
39.44
46.96
82.89
51.66
25.51
13.65
55.63
42.82
11.78
69.48
80.02
62.65
68.98
58.54
25.95
0.102
0.013
0.101
0.094
0.108
0.066
0.108
0.057
0.081
0.182
0.147
0.040
0.065
0.101
0.074
0.194
0.165
0.043
0.789
0.539
1.447
2.740
2.899
1.385
2.268
2.355
1.244
1.129
0.787
1.266
0.834
1.715
1.776
0.649
1.421
1.486
the payload in the satellite. However, it still needs a long time for Kim's controller to reach zero. The control forces generated by the
three controllers are shown in Fig. 13. We can gure that the axial forces generated by the six actuators of our method are in the same
magnitude as Yang's method but smaller than Kim's method, which implies that our method is not very difcult to be realized in
practice. Specically, the control force of Kim's method is larger and more irregular than the axial forces generated by our controller,
which is the other motivation why we have to improve the controller. The maximal displacement responses and the control forces
under various excitations are listed in Tables 3 and 4.
The displacement and velocity responses of the platform under the excitation of 0.75 Hz and 2 Hz can be observed in Figs. 14 and
15. The maximal and minimal attenuation of vibration is 21.04 dB and 10.03 dB when our method is applied, which indicates that our
method is of appropriate adaptability. We can notice that the maximal vibration attenuations of the system applied by Kim's and
Yang's controller are 12.57 dB and 15.3 dB, respectively. But the control effect of Kim's controller is 1.67 dB in the direction z which
does not meet the requirement. For Yang's controller it loses efcacy in the direction z. It also can be observed in Fig 14 that the effect
of free vibration is still not good in some directions when Kim's controller is applied to the system. As for the velocity responses of
the system applied by Kim's controller, the velocity response of direction z is amplied. For Yang's controller, the velocity responses
of the three translational directions are all been magnied a lot, which implies the control has little effect on velocity attenuation
under the excitation. The axial forces of the three controllers are presented in Fig. 16, from which we can nd that the control force
of Kim's controller is more irregular and larger than ours and Yang's.
x 10
Kim et al.
response(m or rad)
Wu et al.
-5
20
x 10
25
30
35
x 10
-5
20
25
-3
25
30
35
-0.01
20
response(m or rad)
response(m or rad)
30
35
30
35
25
time(s)
0.02
25
35
0.01
time(s)
-0.02
20
30
time(s)
-5
20
Yang et al.
y
-3
time(s)
z
response(m or rad)
response(m or rad)
response(m or rad)
Without control
-3
30
35
x 10
-3
-5
20
time(s)
25
time(s)
Fig. 14. Displacement responses of the platform (0.75 Hz and 2 Hz).
Wu et al.
0.1
-0.1
20
25
30
35
time(s)
z-velocity
25
Yang et al.
y-velocity
0.1
-0.1
20
25
30
35
0.02
30
35
0.01
z-velocity
0.03
time(s)
30
35
30
35
-0.2
20
25
time(s)
0.5
25
35
0.2
time(s)
-0.5
20
30
time(s)
0.2
-0.2
20
429
Kim et al.
Without control
x-velocity
0.1
-0.1
20
25
time(s)
-0.01
-0.02
-0.03
20
20.5
21
21.5
22
22.5
time(s)
23
23.5
24
24.5
25
Fig. 15. Velocity responses of the platform (0.75 Hz and 2 Hz).
The maximal displacement response values under the excitation of 12 Hz are presented in Table 3 and the corresponding forces are
shown in Table 4. In such case, we keep the weighting matrix D2 to be the same as before. The attenuations of vibration of the three
controllers are obvious. The amplitudes of all the six directions are so small that we can regard the ultra-quiet environment
p can be
reached. Actually, it easy to learn that for the open-loop system, the maximal starting point of vibration isolation is 7:6 2b12 Hz
(see Fig. 6) as the maximal natural frequency of the system is 7.6 Hz. As a result, we can also gure out that the control forces of all
the three controllers are almost zero which indicates that the active controller is not necessary for high frequency excitations.
To validate whether our controller can attenuate the disturbance of varying frequency, a sine swept-frequency signal is used to
stimulate the system. The frequency varies from 0 Hz to 50 Hz linearly. The displacement response curves of the platform are
illustrated in the Fig. 15, from which we can observe that the system with different controllers barely vibrates under the excitation
of sine swept-frequency. It is clearly observed from the Fig. 17 or Table 3 that the responses of the open-loop reach to the peaks
because of the resonances. Consequently, the robust nonlinear controller proposed in this paper can deal with the situation of varying
frequency and providing an ultra-quiet environment for the payload. There exist some important phenomena which can be observed
in the bar graph of the control forces. One phenomenon is that the axial forces of our method and the other two methods are
near-symmetrically distributed on either side of zero point. As illustrated in Fig. 18, the maximal force generated by the actuators
for our method is 1.447 N. For the controllers proposed by Kim et al. and Yang et al., the corresponding forces lie in the same
magnitude as ours.
5.4. Discussion
Our results conrm that the dynamic model with the base excitation proposed in this paper is accurate and useful for vibration
control and that the improved robust nonlinear controller is efcient for controlling the low frequency excitations.
As for dynamic modeling of the Stewart Platform, there are many papers have been published before. However, for the case of
micro-vibration and the case with the base excitation, there is no complete or direct results can be applied to handle this problem.
It is the motivation that we derived the dynamic model with the base excitation for micro-vibration control purpose. There exist
several key points in the process of deriving the dynamic model. Firstly, the selection of the coordinate of the system is not an easy
job because the number of the components of the system is fairly large. Secondly, the derivation of the partial velocities is not obvious
since the effect of the base excitation has been considered. Thirdly, the formulation of the nal form of the dynamic model requires
430
Wu et al.
Kim et al.
Yang et al.
2
1
50
control force(N)
control force(N)
100
50
0
-50
0
10
15
20
time(s)
3
25
30
-50
0
35
10
15
20
time(s)
4
25
30
35
10
15
20
time(s)
6
25
30
35
10
15
20
time(s)
25
30
35
100
control force(N)
control force(N)
100
-100
0
10
15
20
time(s)
5
25
30
-100
0
35
100
control force(N)
100
control force(N)
-100
0
10
15
20
time(s)
25
30
50
0
-50
0
35
Fig. 16. The axial forces generated by the six actuators (2 Hz and 0.75 Hz).
1
0
-1
0
x 10
Kim et al.
response(m or rad)
Wu et al.
x
-4
10
15
-4
x 10
Yang et al.
y
-4
1
0
-1
0
10
15
x 10
-5
0
-4
-5
0
10
time(s)
10
15
10
15
5
time(s)
response(m or rad)
response(m or rad)
x 10
15
-4
time(s)
5
10
time(s)
-1
0
time(s)
z
response(m or rad)
response(m or rad)
response(m or rad)
Without control
x 10
15
x 10
-4
1
0
-1
0
5
time(s)
Fig. 17. The displacement responses of the six coordinates (swept sinusoid frequency).
431
The control force of Wu et al
numbers
4000
axis 1
axis 2
axis 3
axis 4
axis 5
axis 6
2000
0
-1.5
-1
-0.5
0
control force(N)
0.5
1.5
The control force of Kim et al
numbers
2000
axis 1
axis 2
axis 3
axis 4
axis 5
axis 6
1500
1000
500
0
-3
-2
-1
0
control force(N)
The control force of Yang et al
numbers
2000
axis 1
axis 2
axis 3
axis 4
axis 5
axis 6
1500
1000
500
0
-2.5
-2
-1.5
-1
-0.5
0
0.5
control force(N)
1.5
2.5
Fig. 18. The bar graph of the control force of the three methods (swept sinusoid frequency).
enough patients. Last but not the least, we have concerned the parametrical uncertainty and the structural nonlinearity, which are
important in the practice.
The improved robust nonlinear controller presented in this paper is based primarily on the lemma presented in the Refs. [42,43].
Kim et al. developed a controller for nonlinear tracking control, which is not well adjusted according to the references and our
numerical results. But the controller can be used for vibration control by modifying the state space equation and tuning the control
parameters. The control effect of displacement and velocity can be satised well except for the direction z under low frequency
excitations. However, the control forces of Kim's controller are the largest compared with ours and Yang's. And the control effects
of free vibration are still not satisfactory in some directions when Kim's controller is applied to the system. As for Yang's controller,
the control forces are smaller than Kim's and the control effects of displacements are also observable in some directions except for
the direction z. But the controller loses the efcacy to control the velocities of all the translational directions. In order to solve
nonlinear vibration isolation with parameter uncertainties and satisfy the basic requirements in all directions, we have improved
the controller by introducing a linear controller, two weighting matrices and separating the coefcients into and w . The proposed
controller contains the linear control part, nonlinear part and excitation compliment part. Theoretical proof accompanied with
numerical results conrms that the uniformly ultimate boundary of the state vector can be satised. We have also proposed the
computational process and the rule of some important matrices. It is concluded that the controller is of ne adjustability and can
meet the vibration attenuation requirements in all directions. In addition, the control force generated by the controller is smaller
and more regular than Kim's.
However, we have to acknowledge that there are some shortcomings of our method, such as that the magnitude of the excitation
has to be calculated or identied in order to tune the control parameters. There are still many interesting topics of the dynamic and
control of Stewart platforms, such as the dynamic isotropy, nonlinear vibration analysis of Stewart platform and so on.
6. Conclusion
This paper focuses on dynamic analysis and robust nonlinear control of a six-DOF active micro-vibration isolation manipulator
with parameter uncertainties. We can conclude that:
i. To design a suitable control algorithm to isolate the base excitation, a new nonlinear dynamic model that concerning the base
disturbance is presented via Kane's method in this paper, which is of explicitly physical meaning and concise form.
ii. The uncertain parameters, including stiffness, damping and the mass center location, are considered to satisfy the practical
situation. Qualitative analysis and numerical results both show that the effect of the geometrical nonlinearity is little, whereas
the structural nonlinearity cannot be omitted.
432
iii. An improved robust nonlinear controller is developed by introducing two weighting matrices, computing several important
matrices and separating the coefcients and w . The main characteristic of the controller is the combination of linear control
component, nonlinear component and uncertain component. The robust nonlinear controller is of good adjustability since the
two weighting matrices are introduced to the control algorithm.
iv. By applying this controller, the uniformly ultimate boundary of the state vector is guaranteed according to the theorem presented
in this article.
v. Numerical simulations demonstrate that the stringent quiescent environment of the payload in satellite can be achieved in a wide
frequency range by applying the nonlinear controller. The maximal and minimal attenuation of vibration is approximately 30 dB
and 8 dB, which satises the two requirements quite well.
Acknowledgments
The authors acknowledge the collaboration and the fruitful discussion with Dr. Yinhao Zhu and Dr. Haotian Zhou. The authors also
sincerely appreciate the anonymous referees for their valuable suggestions and questions.
Appendix A
The computation of some matrices
2
9:9835
0:0000
6 0:0000
9:9537
6
6 0:0000
0:0000
G6
6 0:0000
34:8292
6
4 19:1714 0:0000
0:0000
0:0000
0:0000 0:0000 1:1629 0:0000

0:0000 2:1127 0:0000 0:0000
9:9982 0:0000 0:0000 0:0000
0:0000 9:9392 0:0000 0:0000
0:0000 0:0000 9:9410 0:0000
0:0000 0:0000 0:0000 9:9707
17:3445
0:0000
0:0000
17:4278
0:0000
0:0000
0:0000 41:6106
48:8599
0:0000
0:0000
0:0000
3
0:0000
0:0000
2:9872 0:0000
0:0000 2:5728 0:0000 0:0000 7
7
40:0496 0:0000
0:0000 0:0000 7
7
0:0000 16:9325 0:0000 0:0000 7
7
0:0000
0:0000 14:4058 0:0000 5
0:0000
0:0000
0:0000 6:5940
3
2:2370
0:0000 0:0000 0:0000 0:4080 0:0000 0:0196 0:0000 0:0000 0:0000 0:0046 0:0000
7
6 0:0000
1:6529
0:0000
0:2172
0:0000
0:0000
0:0000
0:0083
0:0000
0:0024
0:0000
0:0000
7
6
6 0:0000
0:0000 0:6086 0:0000
0:0000 0:0000 0:0000
0:0000 0:0000 0:0000
0:0000 0:0000 7
7
6
6 0:0000
0:0000 0:0000 7
0:2172 0:0000 0:0986
0:0000 0:0000 0:0000 0:0024 0:0000 0:0005
7
6
6 0:4080 0:0000 0:0000 0:0000
0:1559 0:0000 0:0058
0:0000 0:0000 0:0000
0:0012 0:0000 7
7
6
6 0:0000
0:0000 0:0000 0:0000
0:0000 0:3057 0:0000
0:0000 0:0000 0:0000
0:0000 0:0000 7
3
7
P 10 6
6 0:0196 0:0000 0:0000 0:0000
0:0058 0:0000 0:0062
0:0000 0:0000 0:0000
0:0010 0:0000 7
7
6
7
6 0:0000 0:0083 0:0000 0:0024 0:0000 0:0000 0:0000
0:0065
0:0000
0:0009
0:0000
0:0000
7
6
7
6 0:0000
0:0000
0:0000
0:0000
0:0000
0:0000
0:0000
0:0000
0:0022
0:0000
0:0000
0:0000
7
6
7
6 0:0000
0:0024
0:0000
0:0005
0:0000
0:0000
0:0000
0:0009
0:0000
0:0004
0:0000
0:0000
7
6
4 0:0046 0:0000 0:0000 0:0000
0:0012 0:0000 0:0010
0:0000 0:0000 0:0000
0:0003 0:0000 5
0:0000 0:0000 0:0000
0:0000 0:0000 0:0000
0:0000 0:0003
0:0000
0:0000 0:0000 0:0000
2
3
0:1647 0:0000 0:0000 0:0000 0:8004 0:0000
6 0:0000 0:0621 0:0000 0:3018 0:0000 0:0000 7
7
6
6 0:0000
0:0000 0:0000 0:0000 0:0000 0:0000 7
7
H6
6 0:0000
0:0082 0:0000 0:0647 0:0000 0:0000 7
7
6
4 0:0301 0:0000 0:0000 0:0000 0:1715 0:0000 5
0:0000
0:0000 0:0000 0:0000 0:0000 0:0000
2
H 0:8655 b 1
H 0:8355 b 1
5 10
Case 1. 0.75 Hz
s
Z T
T
W Wdt kWk 9:74
0
s
Z
T
n ndt n 1
Wu et al.

8

9 >
1
3

< 1
= >
w 10N9:74; 5 10
1
1

D1 diag1; 1; 1; 0:1; 0:1; 0:1 ;>

1
>
: w 10
w
D2 diag30; 30; 30; 1; 1; 1
D2 H D2
Kim et al.
w 10N9:74; 5 10
D diag1; 1; 1; 1; 1; 1

w 62
w
1H
Yang et al.

15; 1 75
2
1 1 10 ; w 10
Case 2. 0.75 Hz and 2 Hz

s
Z T
T
W Wdt W 10:74
0
s
Z
T
n ndt n 1
Wu et al.
3
w 11N10:74; 5 10
D1 diag1; 1; 1; 0:1; 0:1; 0:1
D2 diag30; 30; 30; 1; 1; 1

1

1
D1 H D1

1
>
>
: w 10
w
D2 H D2
8
>
>
<
Kim et al.
3
w 11N10:74; 5 10
D diag1; 1; 1; 1; 1; 1
Yang et al.
10; 1 75
2
1 1 10 ; w 10
Case 3. 12 Hz
s
Z T
T
W Wdt W 1:0
0
s
Z
T
n ndt n 1

w 67
w
1H
433
434
Wu et al.

8

9 >
1
3

< 1
= >
w 2N1; 5 10
D1 H D1

D1 diag1; 1; 1; 0:1; 0:1; 0:1 ;>

1
>
: w 10
w
D2 diag30; 30; 30; 1; 1; 1
D2 H D2
Kim et al.
3
w 2 N 1; 5 10
D diag1; 1; 1; 1; 1; 1
)
w 20
w
1H
Yang et al.
10; 1 75
2
1 1 10 ; w 10
Case 4. Swept sinusoid.

Since the period of the excitation is time-varying, we cannot set the control parameters as before. In this case, we just select the
parameters the same as the Case 3.
References
[1] E.H. Anderson, J.P. Fumo, R.S. Erwin, Satellite ultraquiet isolation technology experiment (SUITE), Aerospace Conference Proceedings, March 1925, Montana,
2000.
[2] C. Liu, X. Jing, S. Daley, et al., Recent advances in micro-vibration isolation, Mech. Syst. Signal Process. 56 (2015) 5580.
[3] Z.J. Geng, L.S. Haynes, Six degree-of-freedom active vibration control using the Stewart platforms, IEEE Trans. Control Syst. Technol. 2 (1994) 4553.
[4] J. Spanos, Z. Rahman, G. Blackwood, A soft 6-axis active vibration isolator, Proceedings of the American Control Conference, June, Washington, 1995.
[5] R.G. Cobb, et al., Vibration isolation and suppression system for precision payloads in space, Smart Mater. Struct. 8 (1999) 798812.
[6] G.S. Hauge, M.E. Campbell, Sensors and control of a space-based six-axis vibration isolation system, J. Sound Vib. 269 (2004) 913931.
[7] A. Preumont, et al., A six-axis single-stage active vibration isolator based on Stewart platform, J. Sound Vib. 300 (2007) 644661.
[8] D. Stewart, A platform with six degrees of freedom, Proc. Inst. Mech. Eng. 180 (1965) 371386.
[9] B. Dasgupta, T.S. Mruthyunjaya, The Stewart platform manipulator: a review, Mech. Mach. Theory 35 (2000) 1540.
[10] Y.K. Kim, M.S. Whorton, Equations of Motion for the g-LIMIT Microgravity Vibration Isolation System, NASA/TM-2001-211301, Marshall Space Flight Center, AL,
2001.
[11] Y. Yun, Y. Li, A general dynamics and control model of a class of multi-DOF manipulators for active vibration control, Mech. Mach. Theory 46 (2011) 15491574.
[12] B. Dasgupta, T.S. Mruthyunjaya, A NewtonEuler formulation for the inverse dynamics of the Stewart platform manipulator, Mech. Mach. Theory 33 (1998)
11351152.
[13] B. Dasgupta, T.S. Mruthyunjaya, Closed-form dynamic equations of the general Stewart platform through the NewtonEuler approach, Mech. Mach. Theory 33
(1998) 9931012.
[14] R. Oftadeh, M.M. Aref, H.D. Taghirad, Explicit dynamics formulation of Stewart Gough platform: a NewtonEuler approach, IEEE/RSJ International Conference on
Intelligent Robots and Systems (IROS), October 1822, Taiwan, 2010.
[15] S. Pedrammehr, M. Mahboubkhah, N. Khani, Improved dynamic equations for the generally configured Stewart platform manipulator, J. Mech. Sci. Technol. 26
(2012) 711721.
[16] G. Lebret, K. Liu, F.L. Lewis, Dynamic analysis and control of a Stewart platform manipulator, J. Robot. Syst. 10 (1993) 629655.
[17] Z. Geng, et al., On the dynamic model and kinematic analysis of a class of Stewart platforms, Robot. Auton. Syst. 9 (1992) 237254.
[18] H. Guo, H. Li, Dynamic analysis and simulation of a six degree of freedom Stewart platform manipulator, Proc. Inst. Mech. Eng. C J. Mech. Eng. Sci. 220 (2006)
6172.
[19] H. Gattringer, R. Naderer, H. Bremer, Modeling and Control of a Pneumatically Driven Stewart Platform, Motion and Vibration Control, Springer, Netherlands,
2009. 93102.
[20] Y. Yun, Y. Li, Multi-degree of freedom vibration model for a 3-DOF hybrid robot, IEEE/ASME International Conference on Advanced Intelligent Mechatronics, July
1417, Singapore, 2009.
[21] M. Liu, C. Li, C. Li, Dynamics analysis of the GoughStewart platform manipulator, IEEE Trans. Robot. Autom. 16 (2000) 9498.
[22] Q. Meng, et al., Dynamic modeling of a 6-degree-of-freedom Stewart platform driven by a permanent magnet synchronous motor, J. Zhejiang Univ. Sci. C 11
(2010) 751761.
[23] C. Yang, et al., Dynamic modeling and computational efficiency analysis for a spatial 6-DOF parallel motion system, Nonlinear Dyn. 67 (2012) 10071022.
[24] P. Wu, H. Xiong, J. Kong, Dynamic analysis of 6-SPS parallel mechanism, Int. J. Mech. Mater. Des. 8 (2012) 121128.
[25] K. Yu, Y. Wu, Modeling and active vibration control of six-DOF manipulator through -synthesis with parameter uncertainties, International Symposium on
Precision Engineering Measurement and Instrumentation, SPIE, Aug 1011, Changsha, 2014.
[26] A.M. Lopes, Dynamic modeling of a Stewart platform using the generalized momentum approach, Commun. Nonlinear Sci. Numer. Simul. 14 (2009) 33893401.
[27] A.M. Lopes, F. Almeida, The generalized momentum approach to the dynamic modeling of a 6-dof parallel manipulator, Multibody Syst. Dyn. 21 (2009) 123146.
[28] A.M. Lopes, E.J.S. Pires, Complete Dynamic Modeling of a Stewart Platform Using the Generalized Momentum Approach, Nonlinear Science and Complexity,
Springer, Netherlands, 2011. 199210.
[29] Y. Zhao, F. Gao, Inverse dynamics of the 6-dof out-parallel manipulator by means of the principle of virtual work, Robotica 27 (2009) 259268.
[30] J. Gallardo, et al., Dynamics of parallel manipulators by means of screw theory, Mech. Mach. Theory 38 (2003) 11131131.
[31] A. Preumont, Vibration Control of Active Structures: An Introduction, Springer Science & Business Media, 2011.
[32] B.D. Marneffe, M. Avraam, A. Deraemaeker, et al., Vibration isolation of precision payloads: a six-axis electromagnetic relaxation isolator, J. Guid. Control. Dyn. 2
(2009) 395401.
[33] Y. Zhang, Jitter control for optical payload on satellites, J. Aerosp. Eng. 27 (2012) 04014005.

[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
435
W. Zhao, B. Li, P. Liu, et al., Semi-active control for a multi-dimensional vibration isolator with parallel mechanism, J. Vib. Control. 19 (2012) 879888.
C. Collette, S. Janssens, K. Artoos, Review of active vibration isolation strategies, Recent Patents Mech. Eng. 4 (2011) 212219.
K. Zhou, J.C. Doyle, K. Glover, Robust and Optimal Control, Prentice Hall, New Jersey, 1996.
G.E. Dullerud, F. Paganini, A course in Robust Control Theory, Springer, New York, 2000.
A. Packard, J.C. Doyle, The complex structured singular value, Automatica 29 (1993) 71109.
L. Liu, B. Wang, Multi objective robust active vibration control for flexure jointed struts of stewart platforms via H and synthesis, Chin. J. Aeronaut. 21 (2008)
125133.
L. Liu, K.K. Tan, Y. Guo, et al., Active vibration isolation based on model reference adaptive control, Int. J. Syst. Sci. 45 (2012) 97108.
Y. Xu, H. Liao, L. Liu, et al., Modeling and robust H-infinite control of a novel non-contact ultra-quiet Stewart spacecraft, Acta Astronaut. 107 (2015) 274289.
J.C. Martin, L. George, Continuous state feedback guaranteeing uniform ultimate boundedness for uncertain dynamic systems, IEEE Trans. Autom. Control 26
(1981) 11391144.
M.M. Bridges, et al., Robust control of rigid-link flexible-joint robots with redundant joint actuators, IEEE Trans. Syst. Man Cybern. 24 (1994) 961970.
D. Kim, J. Kang, K. Lee, Nonlinear robust control design for a 6-DOF parallel robot, KSME Int. J. 13 (1999) 557568.
D. Kim, J. Kang, K. Lee, Robust tracking control design for a 6-DOF parallel manipulator, J. Robot. Syst. 17 (2000) 527547.
H. Kim, Y. Cho, K. Lee, Robust nonlinear task space control for 6-DOF parallel manipulator, Automatica 41 (2005) 15911600.
T. Yang, et al., Nonlinear robust control method for active vibration isolation using a Stewart platform, IEEE International Conference on Robotics and
Biomimetics, February 2126, Bangkok, 2009.
T. Yang, et al., Robust back stepping control of active vibration isolation using a Stewart platform, IEEE International Conference on Robotics and Automation, May
1217, Kobe, 2009.
I.M. Buzurovic, D.L. Debeljkovic, Robust control for parallel robotic platforms, IEEE 16th International Conference on Intelligent Engineering Systems (INES), June
1315, Lisbon, 2012.

Mechanism and Machine Theory: Ying Wu, Kaiping Yu, Jian Jiao, Rui Zhao

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Mechanism and Machine Theory: Ying Wu, Kaiping Yu, Jian Jiao, Rui Zhao

Hochgeladen von

Copyright:

Verfügbare Formate

Mechanism and Machine Theory 92 (2015) 407435

Contents lists available at ScienceDirect

Mechanism and Machine Theory

Dynamic modeling and robust nonlinear control of a six-DOF

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

The transformation matrix from frame B to inertial frame O,

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

According to the small magnitude of the vibration

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

The sliding part

Fig. 2. Computational model of a link.

where Tsi is the transformation matrix from frame si to O.

Ip RP=O Ip=p RP=O

Unitary vector of one link

The absolute velocities and accelerations of bij and pik

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

The sliding velocity of one link

The angular and sliding accelerations of one link

Additionally, Eqs. (18) and (20) can be rewritten as matrix form

denote the Jacobi matrix and another ordinary matrix, respectively.

and the acceleration

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

2.2. The partial velocities and partial angular velocities

The general velocities

The subscript j = 1,2,, 6 denotes the six coordinates.

The partial velocities of the sliding velocities

2.3. The dynamic equations of the system

The general inertial forces

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

The Kane's equation

2.4. The axial force of the prismatic joint

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

Fig. 3. The computational model of the actuator.

where K1, Cl are diagonal matrices. Each term of vector g() is

where = 2.5 104 m2 and = 0.225.

where t is a tiny time period.

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

2.5. Dynamic model simplication

b Jp Cl JTb xb Jp Kl JTb xb represents the base excitation.

2.6. The validation of the dynamic model

Location of the mass center

r(c|p) = [0, 0, 0.2] m

The radius of the lower platform

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

Fig. 4. The dynamic model and results in Adams.

where ^rc rc 1 r , r is the installation error of the payload. Therefore

where M0 represents the nominal mass matrix. M = rMc and

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

According to matrix theory and omit the higher order parts

where the uncertain part

xp as the state vector of the system, we can have

4. Robust nonlinear controller development

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

4.1. Control objectives and some assumptions

4.2. Robust nonlinear controller development

in which Q is also a positive denite matrix.

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

i. When 1(x, t) and 2(x, t) , thus

Y. Wu et al. / Mechanism and Machine Theory 92 (2015) 407435

Take account of Eqs. (77) and (78), we can obtain

in which 3 12 min Q max PBHG:

iii. When 1(x, t) and 2(x, t) b , thus

Fig. 5. The computational procedure of P and the control block.