Professional Documents
Culture Documents
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1
H YBRID robot has drawn much attention for the past few
years due to the growing demand for the fast and high-
precision forming of large-scale complex component such as
freedoms, Neumann proposed an Exechon hybrid robot which
is constructed with a 2UPR&1SPR parallel structure and a 2R
serial structure [11]-[13]. With improved performance, the
aircraft structure drilling [1], [2]. Compared with traditional Exechon hybrid robot has been commercialized in machine-tool
machine tools, hybrid robot is more reconfigurable which manufacturers and integrators [14]. Besides, some other hybrid
enables it to deal with large and complex structures in a fast robots are also proposed and demonstrated through primary
way. To meet specific machining demand, hybrid robot can be experiments in recent years, such as TriVariant hybrid robot
mounted at any necessary base (fixed base or moving base) with [15], Georg V hybrid robot [16] and 5-Axis hybrid machine tool
any attitude [2]. The hybrid robot is usually constructed by [17]. In spite of many available hybrid structures, it is still a
integrating a parallel mechanism and a serial mechanism. challenging work to machine large-scale complex components
Parallel mechanism is used to provide a high accuracy and high especially the high-hardness structural parts. Motivated by this
dynamics motion, while serial mechanism is designed to demand, a 5-DOF hybrid robot, named TriMule, has been
cooperate with parallel mechanism to allow additional motions. invented, which comprises a 3-DOF parallel mechanism and a
Manuscript received November 7, 2017; revised February 21, 2018, May 19, xu_yang@tju.edu.cn, liuhaitao_tju@126.com, tianjinxjl@163.com,
2018; accepted July 4, 2018. This work is supported by Research Project of qi_liu@tju.edu.cn, tianhuang@tju.edu.cn).
State Key Laboratory of Mechanical System and Vibration under Grant W. L. Zhu is with the Department of Micro-engineering, Kyoto University,
MSV201806, National Natural Science Foundation of China under Grant Kyoto, Japan (e-mail: wule5033@gmail.com)
51622508, Grant 91648202. G. F. Gong is with the State Key Laboratory of Fluid Power and Mechatronic
X. Yang, H. T. Liu, J. L. Xiao, Q. Liu, and T. Huang are with the Key Systems, Zhejiang University, Hangzhou 310027, China (e-mail:
Laboratory of Mechanism Theory and Equipment Design of Ministry of gfgong@zju.edu.cn).
Education, Tianjin University, Tianjin 300072, China (e-mail:
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2
2-DOF wrist [18]. Compared to the established Tricept and II. LAYOUT OF TRIMULE HYBRID ROBOT
Exechon hybrid robots, it has a less number of revolute joints Fig. 1. shows one modularized TriMule robot, which can be
which means that higher stiffness can be potentially achieved, mounted on either moving platform or fixed base according to
leading to a better machining accuracy for difficult-to-cut specific machining demand. As can be seen, a 3-DOF parallel
materials. Besides the structure scheme, the motion mechanism is designed to execute translational motions with
performance of active joints can also influence its machining three active UPS limbs and one passive UP limb. Here, U, P
quality. Therefore, the high-precision motion control of active and S represent universal joint, prismatic joint and spherical
joints is of great importance to TriMule hybrid robot. joint, respectively. Servomotors with lead-screws are employed
Regarding the control strategy for hybrid robot, some to drive the active prismatic joints. A 2-DOF wrist is designed
research work has been reported in literatures. Neumann et al to achieve two rotational motions with two active revolute
proposed to control Tricept robot by combining a position joints. Servomotors with reducers are used to drive the active
control loop and a feedback control strategy [19]. The position revolute joints. At the terminal of serial rotating head, special
control loop is designed to position the terminal head in device can be mounted such as milling tool, support unit.
cartesian coordinate system, while the feedback control strategy For measurement and control, all the parallel limbs motors
is employed to correct the location error caused by force, and serial joints motors are equipped with absolute rotary
temperature, etc. For managing the interaction between Tricept encoders. Based on the encoder data, the position and
and environment, Fabrizio Caccavale et al proposed a model- orientation of terminal device can be real-timely calculated with
based impedance control strategy [20]. To solve the direct direct kinematic model. Through forcing the motors to track
kinematics of the Tricept along a desired trajectory, Siciliano inverse kinematic solution, desired position and orientation of
designed a jacobian-based closed-loop direct kinematics end-effector can be achieved.
algorithm [21]. Even a number of work has been done, little
effort has been made to improve the motion accuracy of the
active joints of hybrid robot, which provides the basis for the
high-precision position and attitude of terminal head. As the
active joints suffer from discontinuous friction, gravity
variation, inertia variation and other negative factors, this paper
focuses on developing an effective robust control method for it.
Sliding mode controller (SMC) is a kind of variable structure
control method whose operation can be divided into reaching
phase and sliding phase [22]. In reaching phase, SMC drives the
system to a desired sliding surface. The reaching dynamic can
be shaped by designing a specific switching control [23]. Once
the system reaches the sliding surface, tracking error reduces to
be zero along a designed trajectory in spite of model uncertain.
With this good feature of robustness, SMC is firstly employed
to control TriMule hybrid robot. However, when a large model
uncertain appears, large switching control is needed to
guarantee a stable reaching phase, which can also cause system Fig. 1. Modularized TriMule hybrid robot
chattering [24]. To avoid this undesirable chattering and retain
the good robustness of SMC, a continuous friction feedforward III. DYNAMIC MODEL OF TRIMULE HYBRID ROBOT
sliding mode controller (CFSMC) is specially designed, which Dynamic model of TriMule hybrid robot reflects the
can converge TriMule robot to sliding surfaces with continuous relationship between active joints torque control input and
control effort. With a friction feedforward loop, CFSMC can active joints position. In consideration of the fast response of
also achieve a nice transient response while facing a torque control loop, the motors with servo drivers can be treated
discontinuous friction disturbance. Therefore, CFSMC is as proportional systems. Through incorporating the dynamics
expected to assure the TriMule hybrid robot an excellent of motors and rigid body, a full dynamic model expressed in
trajectory-tracking performance. joint-space coordinates is obtained as
The rest of this paper is organized as follows. Section II τ = A(q)q + Bq + F(q, q ) + V(q, q ) + G (q) + D (1)
presents the layout of TriMule hybrid robot. Section III
where 𝛕 is the active joints torque control vector defined by 𝛕 =
describes the model employed in this study. Section IV covers
[𝜏 𝜏 𝜏 𝜏 𝜏 ] (𝜏 , 𝜏 , 𝜏 , 𝜏 , 𝜏 are the torque control
the design procedures of proposed robust control methods
input respectively for 1th active joint, 2th active joint, 3th active
including SMC and CFSMC. Section V evaluates the robust
joint, 4th active joint and 5th active joint), 𝒒 is the active joints
controllers by experiments. Finally, the conclusions are drawn
position vector defined by 𝒒 = [𝑞 𝑞 𝑞 𝑞 𝑞 ]
in Section VI.
(𝑞 , 𝑞 , 𝑞 , 𝑞 , 𝑞 are respectively the position of 1th active joint,
2th active joint, 3th active joint, 4th active joint and 5th active
joint), 𝒒 is the active joints velocity vector, 𝒒 is the active joints
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3
acceleration vector, 𝐀(𝐪) ∈ 𝕽𝟓×𝟓 is the generalized inertia mode controller (SMC) is firstly designed and analyzed in this
matrix, 𝐁 ∈ 𝕽𝟓×𝟓 is the damping coefficient matrix defined by section.
𝐁 = diag(𝐵 , 𝐵 , 𝐵 , 𝐵 , 𝐵 ) (𝐵 , 𝐵 , 𝐵 , 𝐵 , 𝐵 are respectively Both SMC and CFSMC adopt the control framework
the equivalent damping coefficient of 1th active joint, 2th active illustrated in Fig. 2. As shown, inverse kinematics algorithm is
joint, 3th active joint, 4th active joint, 5th active used to calculate the command of five active joints (𝐪𝐫𝐞𝐟 =
joint), 𝐕(𝐪, 𝐪) ∈ 𝕽𝟓 is the Coriolis and centrifugal load [𝑞 ,𝑞 ,𝑞 ,𝑞 ,𝑞 ] ) according to the desired
vector, 𝐆(𝐪) ∈ 𝕽𝟓 is the gravitational load vector, 𝐅(𝐪, 𝐪) ∈ position (𝑥, 𝑦, z) and orientation (α, β) of terminal device. For
𝕽𝟓 is the active joints friction torque vector which can be each active joint, a position control loop (SMC or CFSMC) is
modelled as (2) [25] [26], 𝐃 = [𝑑 𝑑 𝑑 𝑑 𝑑 ] is the specially designed. Through forcing the active joints to follow
disturbing load vector arising from machining operation , etc. their command, the terminal device is desired to move along its
F (q, q ) = [ F1 F2 F3 F4 F5 ] reference trajectory.
ε 2
(2)
Fi = F C i (q ) ⋅ sgn( qi ) + ( F S i (q ) − F C i (q )) ⋅ e − ( qi / q i ) ⋅ sgn( qi )
where 𝐹 (𝐪) is the coulomb friction torque of ith active joint,
𝐹 (𝐪) is the stiction torque of ith active joint, 𝑞 is the
stribeck velocity of ith active joint.
For the simplicity of (1), the inertia matrix 𝐀(𝐪) is replaced
by a constant diagonal matrix 𝐀 , which means uncoupled
Fig. 2. Schematic diagram of CFSMC
constant gains inertia torques. Equation (1) can then be
rearranged as A. Sliding mode controller
=A
q −1 τ − A
−1Bq − A
−1F(q, q ) + E Considering the dynamics in (3), a sliding surfaces vector is
A =diag ( J J J J J ) defined as follows:
1 2 3 4 5
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4
From (10), following sufficient condition for the reachability diagonal matrix defined by 𝐋 = diag(𝐿 , 𝐿 , 𝐿 , 𝐿 , 𝐿 ).
of sliding surfaces can be obtained. Substituting (14) and (15) into (16) and assuming 𝝉𝑭 −
kΗ > F + AE (11) 𝐅(𝐪, 𝐪) = 𝐄 = [0, 0, 0, 0, 0] , differential of the feedback
max max
where 𝐇 = [1,1,1,1,1] , the comparation is operated between control vector needed for ensuring the exponential reaching law
the corresponding elements of vectors. in (16) can be driven as:
Equation (10) and equation (11) indicate that the switching τ B = A
q ref + Bq
+ A 2λ CFSMC (q
ref − q)
(17)
control vector 𝐤sgn(𝐬𝐒𝐌𝐂 ) plays a key role in the sufficient +A λ CFSMC CFSMC
λ ref
(q − q)
+ A Ws CFSMC sgn(sCFSMC )
+ AL
condition for system stability 𝐤𝐇 > 𝐅𝐦𝐚𝐱 + 𝐀𝐄𝐦𝐚𝐱 . As long as Through integral operation, we can further get the feedback
the switching gain 𝑘 is large than model uncertain 𝐹 +𝐽 ∙ control effort, which can be described as:
𝐸 , the ith active joint will converge to the sliding surface q 2 λ CFSMC (q ref − q)
τB = A ref + B q + A
and then accurately sliding to the desired state. Consequently,
L sgn(sCFSMC )dt
CFSMC CFSMC ref CFSMC
SMC can handle the model uncertain with a switching control +A λ λ (q − q) + A W s +A
vector. Yet large switching gain 𝑘 is needed if large 𝐹 +𝐽∙ (18)
𝐸 exists. Because the switching control signal is Accordingly, the control law of CFSMC can be obtained as
discontinuous, large switching gain 𝑘 can also cause q 2 λ CFSMC (q ref − q)
τ CFSMC = ProjτF (Γq ref )+A ref + Bq + A
a ’chattering’ problem. Therefore, it needs to make a choice
λ CFSMC λ CFSMC (qref − q) + A
between control accuracy and system stability while designing Ws L sgn(sCFSMC )dt
CFSMC
+A +A
the SMC.
(19)
B. Continuous friction feedforward sliding mode controller For rationalizing and establishing the stability of CFSMC,
According to the definition of stribeck friction torque in (2), following Lyapunov function candidates vector is defined.
𝟏
while the active joints are changing their motion direction, they 𝐕 𝑪𝑭𝑺𝑴𝑪 = 𝚽 𝑪𝑭𝑺𝑴𝑪 (𝒔𝑪𝑭𝑺𝑴𝑪 )
𝟐 (20)
will encounter discontinuous friction which could unavoidably 𝚽 𝑪𝑭𝑺𝑴𝑪 = diag(𝑠 ,𝑠 ,𝑠 ,𝑠 ,𝑠 )
affect their motion performance. To cope with this nonlinear where 𝑠 is the ith element of the sliding surface vector
friction load and assure smooth transient response, a friction 𝒔𝑪𝑭𝑺𝑴𝑪 .
feedforward control loop is specially designed as (12), which The first-order derivative of 𝐕 𝑪𝑭𝑺𝑴𝑪 is
can provide active friction compensation. CFSMC = Φ CFSMC s CFSMC
V
τ F = ProjτF (Γq ref ) (12) = Φ CFSMC (
e + 2λ CFSMC
e + λ CFSMC λ CFSMC e )
where 𝚪 is a positive constant diagonal matrix defined by 𝚪 = =Φ CFSMC
(ref
q − ( A τ −1 CFSMC
−A −1Bq
− A −1F (q, q ) + E
)
diag(𝛤 , 𝛤 , 𝛤 , 𝛤 , 𝛤 ) , the projection mapping Proj𝝉𝑭 (∎) = +2λ CSMC (q ) + λ CFSMC λ CFSMC (q ref − q ))
ref − q
[Proj𝝉𝑭 (∎ ), ⋯ , Proj𝝉𝑭 ∎ ] is defined as = Φ CFSMC ( − W sCFSMC − L sgn(sCFSMC ) − ( A −1 τ F − A −1F (q, q ) + E ))
= −Φ CFSMC
Ws CFSMC
−Φ CFSMC
(L sgn(s CFSMC −1F (q, q ) + E ))
) + ( A −1 τ F − A
τ F i max if ■i > τ F i max
(21)
Projτ F (■i ) = τ F i min if ■i < τ F i min (13)
i
■ From (21), following sufficient condition for the existence of
i otherwise the sliding motions can be obtained
Furthermore, to establish the stability of overall control −1Γq
LH > A −1F + E
ref + A (22)
max max
system and avoid undesirable system chattering, a continuous
where 𝐇 = [1,1,1,1,1] , the comparation is operated between
sliding mode feedback control loop is designed to cooperate
the corresponding elements of vectors.
with the feedforward loop. Different from the traditional SMC,
Equation (21) and (22) indicate that the integral of switching
continuous sliding mode feedback control loop is designed
based on extended system [27] [28]. The nonlinear differential control vector 𝐀𝐋sgn(𝐬𝐂𝐅𝐒𝐌𝐂 )𝑑𝑡 plays a key role in the
equation governing TriMule extended system can be described sufficient condition for system stability 𝐋𝐇 > 𝐀 𝟏 𝚪𝒒𝐫𝐞𝐟 +
as 𝐀 𝟏 𝐅𝐦𝐚𝐱 + 𝐄𝐦𝐚𝐱 . As long as the gain 𝐿 is large than 𝐹 +
q =A −1 τ B − A
−1Bq + A −1 ( τ F − F (q, q )) + E
(14) 𝛤𝑞 +𝐸 , the ith active joint will reach its sliding
With regard to the system described by (14), a sliding
surface within the time (23) [31], and then its position tracking
surfaces vector is defined as [29]:
error will converge to be zero. Consequently, CFSMC can
sCFSMC = e + 2λ CFSMCe + λ CFSMC λ CFSMCe (15) handle the model uncertain with an integral of switching control
𝐂𝐅𝐒𝐌𝐂
where 𝛌 is a positive constant diagonal matrix defined by vector.
𝛌𝐂𝐅𝐒𝐌𝐂 = diag(𝜆 ,𝜆 ,𝜆 ,𝜆 ,𝜆 ). 1 1
For finite-time stability in reaching phase, following Wi sCFSMC i (0) + Li − Γi qref i max − Fi max − Ei max
1 Ji Ji
exponential reaching law is adopted [30]. ti = ln (23)
Wi 1 1
s CFSMC = − W s CFSMC − L sgn(sCFSMC ) (16) Li − Γi qref i max − Fi max − Ei max
Ji Ji
where 𝐖 is a positive constant diagonal matrix defined by
𝐖 = diag(𝑊 , 𝑊 , 𝑊 , 𝑊 , 𝑊 ) , 𝐋 is a positive constant
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5
V. EXPERIMENTAL VALIDATION Fig. 4. displays the control behavior of SMC. As can be seen
Fig. 3 shows the experimental mirror-milling machine, from Fig. 4 (a), the SMC can enable the closed-loop system to
which is set up with two 5-DoF TriMule hybrid robots. One effectively track the sinusoidal position reference. But Fig. 4(b)
robot is response for milling the thin-walled workpiece. The indicates that SMC also leads to a high frequency and large
other robot is used to support the workpiece, which can control amplitude speed oscillation. According to the sliding variable
the workpiece deformation caused by milling force. Through in Fig. 4 (c), this speed oscillation is caused by the system
the high-precision mirror position control of two robots, large chattering that occurs around the sliding surface. As the control
thin-walled workpiece can be well shaped. law of SMC is non-continuous, system chattering cannot be
In this setup, Panasonic minas-a5Ⅱmotors with servo drives avoided even with well-tuned SMC parameters. Fig. 4 (d)
reveals the discontinuous control signal of SMC. Therefore,
are used to drive the active joints. For terminal position and
SMC cannot assure a stable trajectory tracking.
orientation calculation, each motor is equipped with an absolute
rotary encoder, whose position resolution is 10000 pulses/rev. 0.06 0.20
PMAC PCI card is employed to perform the servo-control 0.05 0.15 Desired Velocity
CFSMC
algorithm with a cycle period of 0.442708ms. 0.04
0.10
Velocity(rad/s)
Position(rad)
0.05
0.03
0.00
0.02
-0.05
0.01
Desired Position -0.10
CFSMC
0.00 -0.15
(a) (b)
-0.01 -0.20
0.25 0.50 0.75 1.00 1.25 0.25 0.50 0.75 1.00 1.25
t(s)
t(s)
15000
8000 Total output
10000 Feedforward portion
6000
4000
5000
Slding variable
Control Signal
2000
0
0
-5000 -2000
0.03
0.05
of CFSMC is much smoother than the control signal of SMC,
0.02
0.00
as shown in Fig. 5 (d). To illustrate the role of friction
-0.05
0.01
feedforward control loop, the feedforward control input is also
-0.10
0.00
plotted in Fig. 5 (d). Therefore, chattering problem can be well
-0.15
(a) (b) avoided under the continuous control law of CFSMC.
-0.01 -0.20
0.25 0.50 0.75 1.00 1.25 0.25 0.50 0.75 1.00 1.25
t(s)
t(s)
15
8000 Total Output
B. Tracking performance analysis
10 6000 On the 4-th joint of support robot, sinusoidal motion
5
4000 experiments with different frequency are carried out to test the
Slding variable
Control Signal
2000
performance of proposed CFSMC. For comparison, the
0
0
performance of traditional double closed-loop controller (TDC),
-2000
-5
SMC, as well as the CFSMC without feedforward control loop
-4000
-10
-6000
which is termed as CSMC, are also evaluated. TDC is a widely
-15
(c)
-8000
(d) used AC servomotor position controller which consists of an
0.25 0.50 0.75 1.00 1.25 0.25 0.50 0.75 1.00 1.25
t(s) t(s) outer position closed loop and an inner speed closed loop [32],
[33]. The position controller is designed with PD algorithm,
Fig. 4. Behavior of SMC for the sinusoidal position reference with T=1s and whereas the speed controller is constructed by PI equation.
A=π/119 (a) Position tracking curves; (b) Velocity tracking curves; (c) Sliding
mode variable 𝑠 curve; (d) Controller signal curve
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6
Fig. 6. displays the tracking results with respect to a A=π/119 is presented in Fig. 8. Hence, CFSMC can enable
sinusoidal position command with cycle T=10s and amplitude closed-loop system to well tack the sinusoidal position
A=π/119. As seen in Fig. 6 (a), four controllers can all enable command with cycle T=1s and amplitude A=π/119, the
closed-loop system to effectively track the command signal. sinusoidal position command with cycle T=0.5s and amplitude
However, the tracking error data in Fig. 6 (b) and Fig. 6 (c) A=π/119.
reveal that the SMC perform the worst with both large control Command Position TDC SMC
0.06 CSMC CFSMC (a)
error and severe chattering. According to (10), large switching 0.05
gain can help SMC to achieve small tracking error but also
Position(rad)
0.04
tends to give rise to system chattering. Different from SMC, the 0.03
0.02
CSMC and CFSMC do not have discontinuous control item and 0.01
hence can avoid the undesirable chattering, as shown in Fig. 6 0.00
(c). Further analysis shows, CSMC and CFSMC can assure a -0.01
2 4 6 8 10 12 14
similar tracking accuracy which is better than that of TDC and t(s)
0.0004 TDC SMC
SMC. As the motion speed is very low when the active joint is
0.04
0.00004 CSMC CFSMC (c)
Tracking Error (rad)
0.03
0.02 0.00002
0.01 0.00000
0.00
-0.00002
-0.01
5 10 15 20 25 30
-0.00004
t(s)
0.0002 2 4 6 8 10 12 14
SMC t(s)
Tracking Error (rad)
(b)
0.0001
Fig. 7. Tracking of a sinusoidal position signal with T=1s and A=π/119 (a)
0.0000 Position tracking curves; (b) Position tracking error curves (TDC & SMC); (c)
Position tracking error curves (CSMC & CFSMC)
-0.0001 Command Position TDC SMC
0.06 CSMC CFSMC (a)
-0.0002 0.05
5 10 15 20 25 30
Position(rad)
t(s) 0.04
0.00004 0.03
TDC CSMC CFSMC (c) 0.02
Tracking Error (rad)
0.00002 0.01
0.00
0.00000
-0.01
1 2 3 4 5
-0.00002 t(s)
0.0006
TDC SMC
Tracking Error (rad)
-0.00004
5 10 15 20 25 30
(b)
0.0003
t(s)
0.0000
Fig. 6. Tracking of a sinusoidal position reference with T=10s and A=π/119 (a)
Position tracking curves; (b) Position tracking error curves (SMC); (c) Position -0.0003
tracking error curves (TDC, CSMC & CFSMC)
-0.0006
Fig. 7. shows the tracking results with respect to a sinusoidal 1 2
t(s)
3 4 5
0.00003
facing the command signal. Tracking error data in Fig. 7 (b) and
Fig. 7 (c) indicate that TDC will lead to the largest tracking 0.00000
error and SMC will cause the most severe system chattering. In -0.00003
comparison, CSMC and CFSMC can achieve both good control
-0.00006
accuracy and good system stability. Further contrastive analysis 1 2 3 4 5
t(s)
between CSMC and CFSMC indicates that CFSMC can
effectively eliminate the overlarge tracking error (relative to Fig. 8. Tracking of a sinusoidal position signal with T=0.5s and A=π/119 (a)
regular tracking error) caused by the friction variation while Position tracking curves; (b) Position tracking error curves (TDC & SMC); (c)
Position tracking error curves (CSMC & CFSMC)
active joint is reversing. Similarly, tracking results for a Fig. 9 plots the tracking results of a sinusoidal position
sinusoidal position command with cycle T=0.5s and amplitude reference with cycle T=0.1s and amplitude A=π/119. Among
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7
the four effective controllers, TDC perform the worst with the of control system. To verify that proposed CFSMC has good
largest control error. As the position command is a fast- robustness to these load variations, further robustness test is
changing signal, the system under SMC does not show a carried out. Because the position of S-5th joint determines the
chattering problem. Compared to SMC, CSMC and CFSMC mass load distribution of S-4th joint, the inertia load of S-4th
can achieve smaller control error, which, however, are much joint changes while S-5th joint is working. In addition, Coriolis
greater than their control error in Fig. 6, Fig. 7 and Fig. 8. and centrifugal load, friction load of S-4th joint are also
Benefit from both fast-changing reference and continuous influenced by the position of S-5th joint. Hence, a multiple
control signal, CSMC and CFSMC successfully avoid the joints coupling motion experiment is carried out as shown in
undesirable system chattering. As the acceleration is very high Fig. 11. As can be seen from the figure, even while the S-5th
while active joint is reversing, the tracking error caused by the joint is moving along a triangular path, CFSMC can still
discontinuous friction is not obvious relative to the regular guarantee S-4th joint a satisfactory tracking accuracy, which is
tracking error. As a result, CFSMC does not show an advantage as good as the results in Fig. 10. Hence, CFSMC also has strong
compared with CSMC. Hence, both CSMC and CFSMC can robustness to inertia variation, friction load variation, Coriolis
enable closed-loop system to well tack the command signal and centrifugal load variation.
with cycle T=0.1s and amplitude A=π/119. Command Position TDC SMC
Command Position TDC SMC 7 CSMC CFSMC (a)
0.06 CSMC CFSMC (a) 6
Position(rad)
0.05 5
Position(rad)
0.04 4
0.03 3
0.02 2
1
0.01
0
0.00
-1
-0.01 5 10 15 20 25 30
0.2 0.4 0.6 0.8 1.0 1.2 1.4 t(s)
t(s)
0.0006
TDC SMC TDC SMC
0.003 (b)
Tracking Error (rad)
Tracking Error (rad)
0.0004
0.002
(b)
0.0002
0.001
0.0000
0.000
-0.0002
-0.001
-0.0004
-0.002
-0.0006
-0.003 5 10 15 20 25 30
0.2 0.4 0.6 0.8 1.0 1.2 1.4 t(s)
t(s)
CSMC CFSMC (c)
0.00004
Tracking Error (rad)
0.0006 0.00002
0.0003 0.00000
0.0000
-0.00002
-0.0003
-0.00004
-0.0006
5 10 15 20 25 30
0.2 0.4 0.6 0.8 1.0 1.2 1.4 t(s)
t(s)
Fig. 10. Single active joint motion with sinusoidal position reference (T=12s
Fig. 9. Tracking of a sinusoidal position signal with T=0.1s and A=π/119 (a) and A=π) (a) Position tracking curves; (b) Position tracking error curves (TDC
Position tracking curves; (b) Position tracking error curves (TDC & SMC); (c) & SMC); (c) Position tracking error curves (CSMC & CFSMC)
Position tracking error curves (CSMC & CFSMC)
4 0.00002
Position(rad)
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8
D. Application for TriMule circular motion motion caused nonlinear dynamics. As a result, the position and
With the designed CFSMC, the support TriMule robot is orientation of support unit can be precisely controlled.
applied to perform a circular motion which needs a coordinated 900 0.02
X error (mm)
0.02
(radius:150mm, center: x=150mm, y=0mm, z=0mm, feed rate: 200 850
X (mm)
0.00 0.00
3000mm/min) in the XY plane is used to guide the support unit 100
-0.02 800
Desired X
(terminal device of support TriMule robot) position, as shown Actual X
-0.04
Desired S 1-th position
-0.01
zero value curves (α=0, β=0), which means the support unit is 200
t(s)
800 Desired S 2-th position 0.02
Y error (mm)
0.02 750
Y (mm)
0 0.00 0.00
-0.02 700
-100 -0.01
Desired Y
Actual Y -0.04 (g)
-200 650 -0.02
5 10 15 20 5 10 15 20
t(s) t(s)
1.0 750 0.02
Z error (mm)
0.5 0.02 700
Z (mm)
0.00 0.00
α error (rad)
0.05
Circular motion trajectory
α (rad)
-0.5
0 0
-1.0
0.00
-1.5
(d) (i)
-0.05 -2.0 -0.0002
5 10 15 20 -0.0001 5 10 15 20
t(s)
t(s)
0.10 0.0001 Desired S 5-th position 0.0002
0.2
0.0
0 0
0.00 -0.1
(e) -0.2
(j)
-0.05 -0.0001 -0.3 -0.0002
5 10 15 20 5 10 15 20
t(s) t(s)
Fig. 14. Circular motion of the support TriMule robot under CFSMC. (a)-(e)
motion of support unit, (f)-(j) motion of five active joints
0.00006
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 9
In this paper, the dynamical model of TriMule hybrid robot [21] B. Siciliano, “The Tricept robot: Inverse kinematics, manipulability
analysis and closed-loop direct kinematics algorithm” Robotica, vol. 17,
is simplified by approximating inertia matrix 𝐀(𝐪) with a no. 4, pp. 437-445, Jul. 1999.
constant diagonal matrix 𝐀 , and ignoring Coriolis and [22] W. Zhao, X. Ren and S. Wang, "Parameter Estimation-Based Time-
centrifugal load 𝐕(𝐪, 𝐪), gravitational load 𝐆(𝐪), disturbing Varying Sliding Mode Control for Multimotor Driving Servo Systems,"
IEEE/ASME Transactions on Mechatronics, vol. 22, no. 5, pp. 2330-2341,
load 𝐃. In our future work, we will reveal the exact value of Oct. 2017.
𝐀(𝐪), 𝐕(𝐪, 𝐪), 𝐆(𝐪) and 𝐃. And we will apply TriMule robot [23] C. L. Xia, G. K. Jiang, W. Chen, and T. N. Shi, "Switching-Gain
in the mirror milling of thin-walled component. Adaptation Current Control for Brushless DC Motors," IEEE
Transactions on Industrial Electronics, vol. 63, no. 4, pp. 2044-2052, Apr.
2016.
[24] Q. Xu, "Continuous Integral Terminal Third-Order Sliding Mode Motion
REFERENCES Control for Piezoelectric Nanopositioning System," IEEE/ASME
Transactions on Mechatronics, vol. 22, no. 4, pp. 1828-1838, Aug. 2017.
[1] J. L. Olazagoitia and S. Wyatt, "New PKM Tricept T9000 and its [25] X. C. Xi, A. N. Poo and G. S. Hong, "Tracking error-based static friction
application to flexible manufacturing at Aerospace Industry," SAE compensation for a bi-axial CNC machine," Precision Engineering, vol.
Technical Paper, Sept. 2007. 34, no. 3, pp. 480-488, Jul. 2010.
[2] M. Milutinović, N. Slavković and D. Milutinović, "Kinematic Modelling [26] C. C. de Wit, H. Olsson, K. J. Astrom and P. Lischinsky, "A New Model
of Hybrid Parallel-Serial Five-Axis Machine Tool," Fme Transactions, for Control of Systems with Friction," IEEE Transactions on Automatic
vol. 41, no. 1, pp. 1–10, Jan. 2013. Control, vol. 40, no. 3, pp. 419-425, Mar. 1995.
[3] Q. C. Li, L. M. Xu, Q. H. Chen and W. Ye, "New Family of RPR- [27] H. T. Yau, "Design of adaptive sliding mode controller for chaos
Equivalent Parallel Mechanisms: Design and Application," Chinese synchronization with uncertainties," Chaos, Solitons & Fractals, vol. 22,
Journal of Mechanical Engineering, vol.30, no. 2, pp. 217–221, Mar. no. 2, pp. 341-347, Oct. 2004.
2017. [28] N. S. Pai and H. T. Yau, "Generalized Projective Synchronization for the
[4] R. Clavel, "Device for the displacement and the positioning of an element Horizontal Platform Systems via an Integral-type Sliding Mode Control,"
in space," Swiss patent, no. 672089 A5, 1985. Journal of Vibration and Control, vol. 17, no. 1, pp. 11-17, Jan. 2011.
[5] A. Codourey, R. Clavel and C. W. Burckhardt, " Control Algorithm and [29] J. J. Slotine and W. P. Li, Applied nonlinear control, United States:
Controller for the Direct Drive Delta Robot," IFAC Proceedings Volumes, Prentice Hall, 1991.
vol. 24, no. 9, pp. 543-549, Sept. 1991. [30] F. Wang, C. Liang, Y. Tian, X. Zhao, and D. Zhang, “Design and control
[6] N. Liu and J. J. Wu, "Kinematics and Application of a Hybrid Industrial of a compliant microgripper with a large amplification ratio for high-
Robot-Delta-RST," Sensors & Transducers, vol. 169, no. 4, pp. 186-192, speed micro manipulation,” IEEE/ASME Transactions on Mechatronics,
Apr. 2014. vol. 21, no. 3, pp. 1262–1271, Jun. 2016.
[7] Y. Wu, Z. Fu, J. N. Xu, W. X. Yan, W. H. Liu and Y. Z. Zhao, "Dynamics [31] S. M. Mozayan, M. Saad, H. Vahedi, H. Fortin-Blanchette, and M. Soltani,
Analysis of Hybrid Parallel Robot," in International Conference on Fluid "Sliding Mode Control of PMSG Wind Turbine Based on Enhanced
Power and Mechatronics, pp. 1272-1275, Aug. 2015. Exponential Reaching Law," IEEE Transactions on Industrial Electronics,
[8] Y. Wu, Z. Fu, J. N. Xu, W. X. Yan, W. H. Liu and Y. Z. Zhao, " Kinematic vol. 63, no.10, pp. 6148-6159, Oct. 2016.
Analysis of 5-DOF Hybrid Parallel Robot," in International Conference [32] R. J. Wai, "Development of New Training Algorithms for Neuro-Wavelet
on Intelligent Robotics and Applications, pp. 153-163, 2015. Systems on the Robust Control of Induction Servo Motor Drive," IEEE
[9] M. Petko, G. Karpiel, K. Gac, G. Góra, K. Kobus, and J. Ochoński, Transactions on Industrial Electronics, vol. 49, no. 6, pp. 1323-1341, Dec.
"Trajectory tracking controller of the hybrid robot for milling," 2002.
Mechatronics, vol. 37, pp. 100-111, Aug. 2016. [33] J. Koivumaki, and J. Mattila, “Stability-guaranteed impedance control of
[10] K. E. Neumann, "Robot," United States Patent, no. 4732525, Mar. 1988. hydraulic robotic manipulators,” IEEE/ASME Transactions on
[11] K. E. Neumann, " Adaptive In-Jig High Load Exechon Machining Mechatronics, vol. 22, no.2, pp. 601–612, Apr. 2017.
Technology & Assembly," SAE Technical Paper, Sept. 2008. [34] R. Sato, “Generation Mechanism of Quadrant Glitches and Xompensation
[12] J. Zhang, Y. Q. Zhao and Y. Jin, "Elastodynamic Modeling and Analysis for it in Feed Drive Systems of NC Machine Tools,” International Journal
for an Exechon Parallel Kinematic Machine," Journal of Manufacturing of Automation Technology, vol.6, No.2, pp.154-162, Mar. 2012.
Science and Engineering, vol. 138, no. 3, 031011, Mar. 2016. [35] R. Sato, M. Maegawa, G. Tashiro, and K. Shirase, “Influence of Servo
[13] Y. Jin, Z. M. Bi, H. T. Liu, C. Higgins, M. Price, W. H. Chen et al., Characteristics on Motion Accuracy of Parallel Kinematic Mechanism,”
"Kinematic Analysis and Dimensional Synthesis of Exechon Parallel Key Engineering Materials, vol. 523-524, pp.762-767, Nov.2012.
Kinematic Machine for Large Volume Machining," Journal of
Mechanisms and Robotics, vol. 7, no. 4, 041004, Nov. 2015.
[14] Z. M. Bi and Y. Jin, "Kinematic modeling of Exechon parallel kinematic Xu Yang received B.S. degree in mechanical
machine," Robotics and Computer-Integrated Manufacturing, vol. 27, no. engineering and automation from China
1, pp. 186-193, Feb. 2011.
[15] T. Huang, Z. Y. Hong, J. P. Mei and D. G. Chetwynd, "Kinematic
University of Mining and Technology,
Calibration of the 3-DOF Module of a 5-DOF Reconfigurable Hybrid Xuzhou, China, in 2011, and Ph.D. degree in
Robot using a Double-Ball-Bar System," in Proceedings of the 2006 mechatronic engineering from Zhejiang
IEEE/RSJ International Conference on Intelligent Robots and Systems, University, Hangzhou, China, in 2017.
pp.508-512, Oct. 2006.
[16] H. K. Tonshoff, H. Grendel and R. Kaak, “Structure and characteristics of He is currently a postdoctoral fellow in the
the hybrid manipulator Georg V,” Parallel Kinematic Machines, pp. 365- school of mechanical engineering, Tianjin
376, 1999. University, China. His current research interests include
[17] X. J. Liu, X. Q. Tang and J. S. Wang, "A Novel 2-DOF Parallel
Mechanism Based Design of a New 5-Axis Hybrid Machine Tool," in adaptive robust control, robotics, electrohydraulic system.
Proceedings of the 2003 IEEE International Conference on Robotics &
Automation, Sept. 2003. Haitao Liu received B.S., M.S., and Ph.D. in
[18] T. Huang, C. L. Dong, H. T. Liu, X. D. Qin, J. P. Mei, Q. Liu et al., "a 5- mechanical engineering from Tianjin
DOF hybrid robot with multi-axis rotating support," Chinese Patent, no. University, Tianjin, China in 2004, 2007 and
201510401279.9, Oct. 2015.
[19] K. E. Neumann, " System and method for controlling a robot," United
2010, respectively. Currently he is a
States Patent, no. 6301525 B1, Oct. 2001. professor in Tianjin University. His research
[20] F. Caccavale, B. Siciliano and L. Villani, "The Tricept robot: dynamics interests include mechanisms and robotics.
and impedance control," IEEE/ASME Transactions on Mechatronics, vol.
8, no. 2, pp. 263-268, Jun. 2003.
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMECH.2018.2853764,
IEEE/ASME Transactions on Mechatronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 10
1083-4435 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.