Professional Documents
Culture Documents
2003 - A New Approach To Parametric Identification of A Single-Link Flexible-Joint Manipulator
2003 - A New Approach To Parametric Identification of A Single-Link Flexible-Joint Manipulator
273
© 2003 Kluwer Academic Publishers. Printed in the Netherlands.
1. Introduction
The growing demands for robot systems to perform tasks of high-accuracy track-
ing have attracted lots of attention to the effect of the joint flexibility recently.
Theoretical control methods aiming at tackling this issue are versatile [2]. These
include singular perturbation, feedback linearization [16], robust control [5, 18],
and adaptive control techniques [7]. Exact knowledge of the physical parameters
of the robots are prerequisite or at least beneficial to these sophisticated control
methodologies. In practice, these parameters are not known exactly and need to
be identified. In the case of a rigid robot, standard identification methods accom-
panying some optimal exciting trajectories are generally used [1, 12]. However,
implementation of these schemes are not easy in terms of both the seeking time
of those trajectories and the acceleration signals required [6, 14]. They are even
more difficult for flexible-joint robots due to the elastic degrees of freedom be-
ing uncontrollable [4, 14]. Therefore, more efficient identification schemes are in
demand.
Regarding this, a new method, based on the works in [8], is proposed for the
parametric identification of a single-link flexible-joint manipulator in this paper.
Given a priori the motor inertia, the proposed scheme identifies all the rest para-
meters asymptotically provided two criteria are fulfilled, i.e., the parameter errors
finally converge to some constants and the basis functions are linearly independent
274 J. T. HUANG
in terms of the steady-state trajectories. The control input is then selected to sustain
a steady-state motion in which the criterion is fulfilled. It is not hard to find such
a control algorithm in general. To demonstrate this, simulation of the proposed
scheme incorporating with a specific robust control method, which ensures the
practical stability of the tracking error dynamics, is presented [18]. The results
verified the effectiveness of the proposed design.
The remainder of the paper is organized as follows. Preliminaries for the design
and the tackled issue are given in Section 2. The proposed identification design is
introduced in Section 3. Simulation of the proposed scheme incorporating with a
specific robust control is undertaken in Section 4 to demonstrate the validity of the
proposed design. Discussions are finally given in Section 5.
2. Preliminaries
To provide a concrete framework for the later identification design, let’s consider
first a little more general four-dimensional nonlinear system
ẋ1 = x2 ,
ẋ2 = θ T f (x),
(1)
ẋ3 = x4 ,
ẋ4 = α T φ(x) + β T ψ(x)u(t)
where x = [x1 x2 x3 x4 ]T ∈ R4 is the system state, θ ∈ Rl , α ∈ Rm , β ∈ Rs are the
system parameters, f (x) ∈ Rl , φ(x) ∈ Rm ∈ Rr , ψ(x) ∈ Rs are the corresponding
known basis functions, and u(t) ∈ R is the control input. The linear-in-control
parameter vector β is assumed to be known a priori while the other parameters
are left to be estimated. To that end, most existing identification schemes require
the acceleration signals and persistent excitation, both are unfortunately hard to
get [9, 12]. In contrast, a closed-loop parametric identification design, modified
from that in [8], is proposed in this section to achieve the objectives without re-
quiring the acceleration signals. Moreover, the proposed design guarantees the
asymptotic stability of the parameter equilibrium provided the basis functions are
linearly independent in terms of the steady-state system trajectories. It can then
be incorporated with any a control scheme ensuring the asymptotic tracking of
a reference trajectory, along which the criterion is fulfilled. However, due to the
unknown nonlinearity θ T f (x) being not in the span of the control input, such a
task is not easy at all. The backstepping procedures are required in general under
such circumstances [2, 11]. The major and immediate application of such a design
is to the single-link flexible-joint manipulator and will be discussed in detail in the
next section.
Before introducing the main results, the following assumptions are required.
(A1) The system states x(t) in (1) are measurable.
(A2) The basis functions f (x) and φ(x) are continuously differentiable in x and
are bounded for bounded x.
A NEW APPROACH TO PARAMETRIC IDENTIFICATION 275
(A3) A specific bounded control input u(x, t) is applied to result in bounded states
x(t) and a non-constant steady-state trajectory vector x̄(t), i.e., x(t) → x̄(t)
as t → ∞.
(A4) The component functions of f (x̄(t)) and those of φ(x̄(t)) are linearly inde-
pendent separately, i.e., the relations a T f (x̄(t)) = 0 and bT φ(x̄(t))
= 0, with a, b being some constant vectors, imply that a = 0 and b = 0,
respectively, [3].
The design starts from constructing the parametric observers for θ and α in (1),
whose outputs are [8]
x2 (t )
θ̂(t) = y(t) + c1 f (x1 , ξ, x3 , x4 ) dξ,
x4 (t )
0 (2)
α̂(t) = z(t) + c2 φ(x1 , x2 , x3 , ξ ) dξ,
0
where c1 , c2 > 0, θ̂(t) and α̂(t) are estimates for the actual parameter θ and α,
respectively, ξ is the integration variable, y(t) and z(t) are the corresponding in-
termediate state vectors at disposal, however, they should be delicately selected
to render the whole closed-loop system asymptotically stable. To that end, we let
them be updated in the following ways [8]
x2 (t )
T ∂
ẏ(t) = −c1 θ̂ f (x) f (x) − ẋi f (x1 , ξ, x3 , x4 ) dξ ,
i=1,3,4
∂xi 0
ż(t) = −c2 α̂ T φ(x) + β̂ T ψ(x)u(x, t) φ(x) (3)
3
∂ x4 (t )
− ẋj φ(x1 , x2 , x3 , ξ ) dξ .
j =1
∂xj 0
Sustained the assumptions (A1)–(A4), it can be shown that the parametric ob-
servers in (2), with the dynamics of the intermediate states defined in (3), will
eventually identify the parameters θ and α in (1). The proof is similar to that in [8],
for brevity, it is omitted here. Nevertheless, a detailed analysis of this design for
the single-link flexible-joint manipulator will be given in the next section.
x (t )
Remark. The cancelling term ẋ4 (∂/∂x4 ) 0 2 f (x1 , ξ, x3 , x4 ) dξ appearing
in (3) calls for the measurement of the signal ẋ4 in general. However, in the case of
a system with f (x) being independent of the variable x4 , the cancelling term be-
comes zero and hence such a measurement is avoided. Similarly, the measurement
of the signals ẋ2 is not required if φ(x) is independent of x2 . These are the main
reasons for avoiding the measurement of the acceleration signals in the subsequent
designs.
276 J. T. HUANG
3. Identification Designs
The proposed scheme above will be applied to a single-link flexible-joint ma-
nipulator in this section. Given a priori the motor inertia, it identifies the rest
system parameters without requiring acceleration signals. Details are given in the
following.
The dynamics of a single-link flexible-joint manipulator, in state-space repre-
sentation, can be described by [7]
ẋ1 = x2 , (4.1)
mgl ke
ẋ2 = − sin x1 − (x1 − x3 ), (4.2)
Jl Jl
ẋ3 = x4 , (4.3)
ke µ 1
ẋ4 = (x1 − x3 ) − x4 + u(t) (4.4)
Jr Jr Jr
where x1 is the link position, x2 is the link angular velocity, x3 is the motor rotor
position, x4 is the motor rotor angular velocity, Jl is the link inertia, Jr is the motor
rotor inertia, ke is the joint elastic constant, m is the link mass, l is the link length,
g is the gravity constant, µ is the viscosity, and u(t) is the control input. The
dynamics described in (4) belongs to the class of systems in (1) and hence the
parametric observer constructed there can be directly applied. First, to unify the
notations, we let θ1 = mgl/Jl , θ2 = ke /Jl , α1 = ke /Jr , α2 = µ/Jr , β1 = 1/Jr .
In the sequel, it is assumed a closed-loop control input u(t) = u(x, t) is applied to
sustain (A3).
By substituting f (x) = [−sin x1 , −(x1 − x3 )]T and φ(x) = [(x1 − x3 ), −x4 ]T
into (2) and carrying out the integration there explicitly, the resulting estimated
parameter vector θ̂ and α̂ will be
ẏ1 (t) = −c1 sin x1 (θ̂1 sin(x1 ) + θ̂2 (x1 − x3 )) − (cos x1 )x22 ,
ẏ2 (t) = −c1 (x1 − x3 )(θ̂1 sin(x1 ) + θ̂2 (x1 − x3 )) − (x2 − x4 )x2 ,
ż1 (t) = −c2 (x1 − x3 )(α̂1 (x1 − x3 ) − α̂2 x4 + β1 u(x, t)) (6)
+ (x2 − x4 )x4 ,
ż2 (t) = c2 x4 (α̂1 (x1 − x3 ) − α̂2 x4 + β1 u(x, t)).
A NEW APPROACH TO PARAMETRIC IDENTIFICATION 277
It is noted that no measurement of the acceleration signals ẋ2 , ẋ4 are required to
implement the update laws (6), as explained earlier. Next, by directly differenti-
ating Equation (5) with time and taking Equation (6) into account, the resulting
parameter error dynamics can be easily obtained as
4. Simulation Results
To demonstrate the effectiveness of the proposed design, simulations of the pro-
posed design will be given in this section.
The reference trajectory for x1 to follow is
qd (t) = A sin(ωt), (10)
with the amplitude A and the angular velocity ω being at disposal. The robust
control method in [18] is adopted herein to track the trajectory qd in (10). For ease
of reference, the control law is rewritten in the following
u = uN + uF (11)
with
4
1 ∗ d qd
uN = Jr∗ q̈d + (mgl) ∗
q̈d cos qd − (mgl) ∗ 2
q̇ sin qd + J
ke∗ d l
dt 4
∗ ∗ ∗ ∗
+ µ w2N + (mgl) sin qd + Jl q̈d ,
uF = −k 3 JrM w
2
where
x4 − w2∗
2 =
w ,
k3
w2∗ = w2N
∗
+ w2F ∗
,
A NEW APPROACH TO PARAMETRIC IDENTIFICATION 279
in which the superscript ∗ denotes the nominal value, the subscripts M and m
denote the largest and the smallest admissible value of a parameter, and k is the
so-called attenuation factor at disposal. The control (11) ensures boundedness of
all closed-loop signals and, more importantly, arbitrarily small tracking errors by
using sufficiently large k values. Hence, assumptions (A2), (A3) sustain then. Ful-
fillment of (A4) can be verified by contradiction. First, if the basis functions sin x̄1
and (x̄1 − x̄3 ) are linearly dependent, then x̄3 = x̄1 −a sin x̄1 , a ∈ R. By substituting
them into (4.2) and integrating both sides explicitly, one obtains that
mgl 5.0 [N m]
mgl ∗ 4.5 [N m]
Jl 1.0 [kg m2 ]
Jl∗ 0.8 [kg m2 ]
JlM 1.4 [kg m2 ]
Jr 0.3 [kg m2 ]
Jr∗ 0.18 [kg m2 ]
JrM 0.42 [kg m2 ]
µ 0.1 [kg m2 /sec]
µ∗ 0.07 [kg m2 /sec]
ke 100.0 [N m]
ke∗ 75.0 [N m]
kem 60.0 [N m]
A 3.0 [rad]
ω 5.0 [rad/sec]
k 50.0 [–]
c1 1.0 [–]
c2 1.0 [–]
shown in Figure 1, while the parameter errors are depicted in Figures 2–5. It can be
seen that the parameter errors converge quickly to zeros within 30 seconds, which
demonstrate well the superiority of the proposed design over other methods.
5. Discussions
In this paper we have proposed a parametric identification design for a single-
link flexible-joint manipulator. Given a priori the motor inertia, it guarantees the
bounded stability of the estimation error dynamics. Moreover, the asymptotic sta-
bility of the parametric equilibrium can be achieved on conditions that the basis
functions are linearly independent in terms of the steady-state trajectories and
the estimation errors converge to some constant vectors. It can incorporate with
any a control method capable of sustaining a steady-state motion within which
the criteria are fulfilled. The achievement does not require the measurement of
the acceleration signals. Simulation of this scheme with a specific robust control
design has verified the theory. For actual implementation of this design, however,
the requirement of all states being measurable may restrict its usefulness. There-
fore, further improvements are required for its practical applications and under our
investigation.
A NEW APPROACH TO PARAMETRIC IDENTIFICATION 281
References
1. Armstrong, B.: On finding exciting trajectories for identification experiments involving systems
with nonlinear dynamics, Internat. J. Robotic Res. 8(6) (1989), 28–48.
2. Bridges, M. M., Dawson, D. M., and Abdallah, C. T.: Control of rigid-link, flexible-joint robots:
A survey of backstepping approach, J. Robotic Systems 12(3) (1995), 199–216.
3. Chen, C. T.: Introduction to Linear System Theory, Holt, Rinehart and Winston, New York,
1986.
4. Dépincé, P.: Parameter identification of flexible robots, in: Proc. of the IEEE Conf. on Robotics
and Automation, Leuven, Belgium, 1998, pp. 1116–1121.
5. Elmaraghy, H. A., Lahdhiri, T., and Ciuca, F.: Robust linear control of flexible joint robot
systems, J. Intelligent Robotic Systems 34 (2002), 335–356.
6. Gautier, M. and Khalil, W.: Exciting trajectories for the identification of base inertial parameters
of robots, Internat. J. Robotic Res. 11(4) (1992), 362–375.
7. Ghorbel, F., Hung, J., and Spong, M. W.: Adaptive control of flexible-joint manipulators, IEEE
Control Systems Mag. 9 (1989), 9–13.
8. Huang, J. T.: An adaptive compensator for a class of linearly parameterized systems, IEEE
Trans. Automat. Control 37(3) (2002), 483–486.
9. Ioannou, P. A. and Sun, J.: Robust Adaptive Control, Prentice-Hall, Englewood Cliffs, NJ, 1996.
10. Kozlowski, K.: Modeling and Identification in Robotics, Springer, London, 1998.
11. Krstić, M., Kanellakopoulos, I., and Kokotović, P. V.: Nonlinear and Adaptive Control Design,
Wiley, New York, 1996.
12. Ljung, L.: System Identification: Theory for the User, Prentice-Hall, Englewood Cliffs, NJ,
1987.
13. Narendra, K. S. and Annaswamy, A. M.: Stable Adaptive Systems, Prentice-Hall, Englewood
Cliffs, NJ, 1989.
284 J. T. HUANG
14. Pham, M. T., Gautier, M., and Poignet, P.: Accelerometer based identification of mechanical
systems, in: Proc. of the IEEE Conf. on Robotics and Automation, Washingtion, DC, May 2002,
pp. 4293–4298.
15. Sira-Ramirez, H., Ahmad, S., and Zribi, M.: Dynamical feedback control of robot manipulators
with joint flexibility, IEEE Trans. Systems Man Cybernet. 22 (1992), 736–747.
16. Spong, M. W.: Modeling and control of elastic joint robots, ASME J. Dyn. Systems Measm.
Control 109 (1987), 310–319.
17. Swevers, J., Ganseman, C., Bilgin, D., De Schutter, J., and Van Brussel, H.: Optimal robot
excitation and identification, IEEE Trans. Robotics Automat. 13(5) (1997), 730–740.
18. Tomei, P.: Tracking control of flexible joint robots with uncertain parameters and disturbances,
IEEE Trans. Automat. Control 39 (1994), 1067–1072.
19. Tzes, A. P. and Yurkovich, S.: Application and comparison of on-line identification methods
for flexible manipulator control, Internat. J. Robotics Res. 10(5) (1991), 515–527.