com/12555
DOI 10.1007/s12555-011-0524-5
Keywords: Actuator saturation, anti-windup, conditional integration, modified internal model control,
real-time control.
For practical situations, the range of motor input voltage quency gain vectors, and r is a new control input, respec-
V is constrained to some certain voltages, called motor tively. On applying (6) to (9), yields
saturation limits. This occurs usually between output of
the motor controller and the PWM module. x = ( A − BK ) x + BK 0 r + ℑ. (10)
In order to design our proposed algorithm, we For the development of robust controller, suppose that
assumed that the relation between the actual actuator’s
input (V) and control signal produced by the AW xd = ( A − BK ) xd + BK 0 rd , (11)
compensator (u) is given by yd = Cxd (12)
V = β tanh(u / β ), (2) represent the desired performance of the closed-loop
where β > 0 is the maximum allowable voltage deter- system we wish for the manipulator. In last equation, the
mined by the actuator. Now, let us denote the actuator’s elements rd and yd are the reference trajectory and de-
non-implemented control signal by sired output in the joint space respectively. It is important
to note that, the value of the K controller is setup by trial-
φ (u ) = β tanh(u / β ) − u. (3) and-play based on transient and static performance from
simulation results for (11). Therefore, this part of the
Considering (1) to (3), we define a second-order non- design procedure is quite straightforward and is carrying
linear differential equation of integrated actuator and out independently. Now, for the problem of tracking the
manipulator, here called “available model” as desired trajectory, we define the following variables
q = u +℘( q, q , q, φ (u ) ) , (4) e = x − xd , ξ = y − yd , ℵ = r − rd . (13)
where Considering (13) and combining with (7), (10)-(12),
the error dynamics may then be written as
℘( q, q , q, φ (u ) ) = α −1
[(α I − M (q))q − Vm (q, q )q
(5)
− H (q, q ) + αφ (u )] . e = ( A − BK )e + BK 0ℵ + ℑ, (14)
ξ = Ce. (15)
Such that α is a non-singular diagonal matrix. The moti-
vation behind our definition is that, the effects of para- The control problem is now reduced to designing an
meter variations, external disturbances, and the common auxiliary control input ℵ to provide asymptotic stability
nonlinearities arise from using actuator in practical con- of the controlled robotic system, under the assumption of
trol applications can be lumped in a single term and con- unknown information of the actuator and robot dynamic
sidered as a perturbation denoted by ℘. Therefore, the (i.e., we have not any knowledge of parameters plate or
data sheet, which supply usually by the manufacturer).
control problem summarizes into the problem of stabiliz-
For this purpose, the following quantities are defined
ing the available model based on Internal Model Prin-
ciple. p
Ξ = e ( ) − ∑ b j e( ) ,
p p− j
(16)
4. DESCRIPTION OF CONTROL STRUCTURE j =1
p
Φ = ℵ( ) − ∑ b jℵ( ) ,
p p− j
In this section, we consider actuator saturation (17)
compensation to achieve optimal tracking and j =1
disturbance rejection for the motion control of robots as where bj’s are zero or any positive real scalars designed
an extended form of [16]. For this purpose, suppose (4) by engineer designer. The core idea of these defenitions
is rewritten in the following state and output equation is that perturbation can be approximated by a pth order
form as ordinary differential equation (ODE) as (18) [16,17].
x = Ax + Bu + ℑ, (6) p
ℑ( ) = ∑ b j ℑ( ) .
p p− j
y = Cx, (7) (18)
j =1
T 2n
where x = [q q ] ∈ ℜ is the state of the system,
y ∈ ℜn is measured output and only consists of the joint By taking time derivative of both sides of (14) we
arrive at
positions and finally
Ξ = ( A − BK ) Ξ + BK 0 Φ. (19)
0 I 0 0
A= , B = , C =[I 0], ℑ= . (8)
0 0 I
℘(q, q, q, φ ) By the last result, the design procedure is now to
search for a controller Ф as
Given the plant (6), (7), in the first stage we imple-
ment a feedback control scheme of the following form p
Φ = −∑ µ jξ (
p− j )
− µ0 Ξ (20)
u = − Kx + K 0 r , (9) j =1
where K and K0 are positive state-feedback and zero fre- such that the closed-loop system, defined by
1008 Alireza Izadbakhsh, Ali Akbarzadeh Kalat, Mohammad Mehdi Fateh, and Mohammad Reza Rafiei
p such that
Ξ = ( A − BK − BK 0 µ0 ) Ξ − BK 0C (∑ µ j e( ) )
p− j
(21)
j =1 p
but also it would cause the control system to avoid satu- Using some mathematical computation, it can easily
ration for a long time. This can be gained by changing be shown, for any arbitrary σ ∈ ℜ; ℑ1 (t ) + ℑ2 (t ),
the auxiliary control input ℵ injected to the inner con-
ℑ1 (t ).ℑ2 (t ) and σ .ℑ(t ) are belong to Z and hence Z is
trol loop. As can be seen from (15), ξ is a function of
tracking error e, therefore ℵ depends on the tracking an algebra. Therefore, the first condition, (24), is true for
Z. We show that Z separates points on the T set. To do
error only and it ensures that tracking error e(ξ) ap-
this, we choose the parameters of the ℑ(t ) in the form
proaches zero asymptotically as time increases.
of (23) as follow
Remark: As can be seen, (18) is a p order ODE. It can
c1 = 1, λ1 = − 1, ω1 =0, θ1 =0. (30)
be easily show that, the solution of this equation is a con-
tinuous function as: Since t1 ≠ t2 , then e−t1 ≠ e−t2 and therefore the sec-
p
ond condition is also verified. To show that Z vanishes at
ℑ(t ) = ∑ ci eλi t Cos (ωi t + θi ) . (23) no point of T, we simply observe that any system in the
i =1
form of (23) with ωi =0, θi =0 and ci > 0 has the prop-
It is interesting to investigate the capability of the last erty that
assumption, (18), from a function approximation capabil-
ity point of view. Herein, we will prove that (23) have ∀t ∈ T, ℑ(t ) > 0. (31)
the universal approximation capability. In the following,
Hence, Z vanishes at no point of T. Thus the three
we suppose that the input universe of discourse T is a
aforementioned conditions are satisfied. Therefore the
convex set in ℜn . First, we need the following useful result follows by Stone-Weierstrass Theorem.
Theorem.
5. EXPERIMENTAL RESULTS
Stone-Weierstrass Theorem [18]: Let Z be a set of
real continuous functions on a convex set T. If
The experimental setup using a 2-link elbow robot
1) Z is an algebra, that is, the set Z is closed under
manipulator is shown in Fig. 4. The first joint is driven
multiplication, addition, and scalar multiplication,
by geared permanent magnet DC motor characterized by
2) Z separates points of T, i.e.,
Barber-Colman Company operating within ±12 volt
∀t1 , t2 ∈ T , t1 ≠ t2 , ∃ ℑ(t ) ∈ Z : ℑ(t1 ) ≠ ℑ(t2 ), (24) input and second joint is driven by a vormo geared DC
motor under the same operating conditions. The first
3) Z vanishes at no point of T, that is, motor drives a shaft that carries a potentiometer, a tacho-
∀t ∈ T , ∃ ℑ(t ) ∈ Z : ℑ(t ) ≠ 0 (25) generator and a link, which is installed between the
motor and the potentiometer and the second motor is
then for any real continuous function ℑ(t ) on T and only equipped by a potentiometer. The measured input-
arbitrary ε > 0, there exists a function f (t) in Z such that output data are transferred to a computer (Pentium II 366
2
MHz) by a data acquisition card (ADVANTECH PCL- Without AW
1.8 Tracking AW
818L with up to 100 kHz sampling rate, 12 bit high Conditional Integration
speed A/D converter with a conversion rate of max 1.6 MIMC
RAWC
40KHz). The data acquisition card controls the practical 1.4
robot through user-defined programs in MATLAB/ )d 1.2
a
SIMULINK environment. A low pass filter with a cut off (rn
itoi 1
frequency of 5 Hz is also used for noise cancellation of so
p 0.8
the potentiometer. Now, to illustrate the windup
0.6
phenomenon, the performance of the proposed controller
is compared with three famous model-free AW methods 0.4
) 1.2
1 1 d
ra
p = diag 2 , 2 ,
(
n
io
ti
1
S + 10.95S + 30 S + 4.47 S + 5 so
p 0.8
Q1 = diag (400, 400), (33) 0.6
0
for MIMC. Moreover, using Ziegler-Nichols method and 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time(s)
manual tuning, we derived the controller gains for PID- (b) Joint 2.
controller as Td=diag(0.1,0.2), N=10, Ti=diag(1.11,1.25),
Tt=diag(5,5) and k'=diag(36,36). We indicate that the Fig. 5. Step responses of the manipulator system.
three mentioned famous AW control approaches cannot 15
Without AW
keep robustness against large external disturbances. To Tracking AW
that, when the system includes actuator saturation, the 10 Conditional Integration
MIMC
step response and control signal obtained as shown in RAWC
Figs. 5 to 6. These inferior responses are justifiable by 5
control signal saturation and accumulated error at the )lt
vo(
back of integral term. This poor response of PID con- eg 0
alt
trolled system that is large overshoot and settling time o
V
about 5 second has been compensated by using RAWC, -5
as shown in Fig. 5.
To explore RAWC ability in both set point and track- -10
1 1 -10
p = diag 2 , 2 ,
S + 10.95S + 30 S + 3S + 25
-15
Q1 = diag (400, 400), (35) 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time(s)
29.05S + 370 37 S + 375 (b) Joint 2.
Q2 = diag 2 , .
S + 10.95S + 30 S 2 + 3S + 25 Fig. 6. The joint-space control voltage histories.
1010 Alireza Izadbakhsh, Ali Akbarzadeh Kalat, Mohammad Mehdi Fateh, and Mohammad Reza Rafiei
1.5
Without AW For this purpose, two types of disturbances are
Tracking AW imposed to the second joint of robotic system. These
Conditional Integration
MIMC disturbances can be representative of additive
RAWC
uncertainties as well as usual output disturbances before
1
)d the saturation block.
ar
(n 1.5 Case 1 (Constant External Disturbance): The first
ioti
so
p
test includes a constant disturbance injected to the output
1
0.5 of inner-loop linear compensator u, as shown in Figs. 1
0.5 to 3. The performance of the four schemes are tested by
real-time experiments for set point control of joint 1 and
0
0 1 2 3 4 5 tracking a sinusoidal path by second joint given by
0
0 5 10
time(s)
15 20 25
xd (t ) = 1.256 − 0.628Cos (0.5t ) − 0.314Sin(t ). (36)
(a) Joint 1.
3 Figs. 7 and 8 show the time evolution for set-point and
tracking control, and tracking error of the joints,
2.5 respectively. Due to large number of Figures, the
actuator voltages has been omitted here, but they are of
2 acceptable behaviour. As can be seen, the tracking error
)d
a
(rn for RAWC is limited to about 0.015 rad for first joint and
o 1.5
itis 0.04 rad for second joint which is acceptable considering
o
p
the mechanical resolution limitations. These figures
1
Without AW
illustrate that all of these control methodologies provides
0.5
Tracking AW good responses in presence of constant external
Conditional Integration
MIMC disturbances except MIMC.
0
RAWC Case 2 (Large External Disturbance): To further test
0 5 10 15 20 25
time(s) the robustness of the proposed approach, a large external
(b) Joint 2. disturbances (in comparison to control signal), imposed
Fig. 7. The joint-space trajectory tracking in presence of to the second joint of robotic system. such a strong dis-
constant External disturbance. turbance is quite likely to move the actuator to the satu-
1 1
Without AW Without AW
0.8 Tracking AW 0.8 Tracking AW
Conditional Integration Conditional Integration
0.6 MIMC 0.6 MIMC
RAWC RAWC
)
0.4 )
0.4
d d
a a
r r
(
r
o
0.2 (
r
o
0.2
r r
r r
e e
n
o
i
0 n
o
i
0
t t
i
s 1 i
s
o
p -0.2 o
p -0.2
t
n
i
0.5 t
n
i
o
j
-0.4
o
j
-0.4
0
-0.6 -0.5 -0.6
-0.8 -1 -0.8
0 1 2 3 4 5
-1 -1
0 5 10 15 20 25 0 5 10 15 20 25
time(s) time(s)
(a) Joint 1. (a) Joint 1.
0.4 1.5
1
0.2
0.5
0
) )
d d
0
a a
r r
( (
r r
o
r
r -0.2 o
r
r
e e
n
o
i
n
o
i
-0.5
t t
i i
s
o
-0.4 s
o
p
t
p
t -1
n n
i i
o o
j j
-0.6
Without AW -1.5 Without AW
Tracking AW Tracking AW
-0.8 Conditional Integration -2 Conditional Integration
MIMC MIMC
RAWC RAWC
-1 -2.5
0 5 10 15 20 25 0 5 10 15 20 25
time(s) time(s)
(b) Joint 2. (b) Joint 2.
Fig. 8. The joint-space trajectory tracking errors in Fig. 9. The joint-space trajectory tracking errors in
presence of constant External disturbance. presence of large External disturbance.
A Robust Anti-Windup Control Design for Electrically Driven Robots - Theory and Experiment 1011
1.5
Without AW and avoids windup. Some experimental results were
Tracking AW shown to prove the superiority of the proposed approach.
Conditional Integration
MIMC The approach presented in this paper can easily be
RAWC extended to any robot of higher DOF.
1
)d
a
r(
n
REFERENCES
o
iit
so
[1] V. Kapila and K. M. Grigoriadis, Actuator Satura-
p tion Control, Marcel Dekker, 2002.
0.5
[2] Y. Ting, S. Tosunoglu, R. Freeman, and D. Tesar,
“Saturation avoidance methods for serial robots op-
erating under a failure,” Journal of Robotic Systems,
vol. 16, no. 12, pp. 667-678, 1999.
0
0 5 10 15 20 25 [3] W. Gao and R-R. Selmic, “Neural network control
time(s)
(a) Joint 1. of a class of nonlinear systems with actuator satura-
3 tion,” IEEE Trans. on Neural Networks, vol. 17,
no.1, pp. 147-156, 2006.
2.5 [4] G. Liu and A. A. Goldenberg, “Comparative study
of robust saturation-based control of robot manipu-
2 lators: analysis and experiments,” International
)d Journal of Robotics Research, vol. 15, no. 5, pp.
ra
(n
tio
is
1.5 473-491, 1996.
o
p [5] W. Peng, Z. Lin, and J. Su, “Computed torque con-
1 trol-based composite nonlinear feedback controller
Without AW for robot manipulators with bounded torques,” IET
Tracking AW
0.5
Conditional Integration Control Theory and Applications, vol. 3, no. 6, pp.
MIMC 701-711, 2009.
RAWC
0 [6] E. A. Ruiz, A. Zavala-Rio, V. Santibanez, and F.
0 5 10 15 20 25
time(s) Reyes, “Global trajectory tracking through static
(b) Joint 2. feedback for robot manipulators with bounded in-
Fig. 10. The joint-space trajectory tracking in presence puts,” IEEE Trans. on Control Systems Technology,
of large External disturbance. vol. 17, no. 4, pp. 934-944, 2009.
[7] J.-Z. Xiao, H.-R. Wang, W. Zhang, and H.-R. Wei,
ration zone. The technical specifications such as tracking “Adaptive robotic control based on a filter function
error and trajectory tracking ability are illustrated in Figs. under the saturation of actuators,” Proc. of Interna-
9 to 10. Tracking error presented in Fig. 9 looks small, tional Conference on Machine Learning and Cy-
lower than %3.5, which is satisfactory for many practical bernetics, pp. 283-287, 2006.
applications. From Fig. 10, it can be seen that the joint [8] E. Zergeroglu, W. Dixon, A. Behal, and D. Dawson,
positions for RAWC reaches the reference signals in a “Adaptive set-point control of robotic manipulators
short time while having good performance in steady state. with amplitude-limited control inputs,” Robotica,
Experimental results show that, there is virtually a little vol. 18, no. 2, pp. 171-181, 2000.
change in the responses for the RAWC compared with [9] A. Z-Río and V. Santibanez, “Simple extensions of
other anti-windup method. Thus, the proposed robust the PD-with-gravity-compensation control law for
anti-windup control method demonstrates high capability robot manipulators with bounded inputs,” IEEE
of overcoming not only the model uncertainties, but also Trans. on Control Systems Technology, vol. 14, no.
the external disturbances in presence of saturation nonli- 5, pp. 958-965, 2006.
nearity. [10] Y.-S. Lu, “Integral variable-structure control with
variable-structure sliding dynamics for anti-reset
6. CONCLUSIONS windup,” Proc. of the Institution of Mechanical En-
gineers, Part I: Journal of Systems and Control
Many Anti-Windup control techniques may improve Engineering, pp. 209-216, 2008.
performance of the control system subject to some [11] F. Morabito, A. R. Tee1, and L. Zaccarian, “Nonli-
actuator saturation. However, such a design near anti windup applied to Euler-Lagrange sys-
modifications does not guarantee asymptotic closed-loop tems,” IEEE Trans. on Robotics and Automation,
stability in presence of external disturbance. External vol. 20, no. 3, pp. 526-537, 2004.
disturbance besides causing tracking error can itself [12] A. S. Morris, Measurement and Instrumentation
aggravate the actuator saturation problem. The use of a Principles, Butterworth-Heinemann, 2001.
robust control approach for trajectory tracking of robot [13] M. Kanamori, “Static anti-windup controller design
considering the actuator saturation model was developed for planar 2DOF robot manipulators with actuator
in this paper. The proposed approach guarantees stability, saturation,” Proc. of IEEE International Confe-
ensures tracking and disturbance rejection capabilities, rence on Robotics and Automation, pp. 3314-3319,
1012 Alireza Izadbakhsh, Ali Akbarzadeh Kalat, Mohammad Mehdi Fateh, and Mohammad Reza Rafiei