Professional Documents
Culture Documents
Abstract— A nonlinear position tracking controller with a controller [7]–[11]. Robust control methods under parameter
disturbance observer (DOB) is proposed to track the desired uncertainties have been proposed [12], [13]. Recently, the
position in the presence of the disturbance for electrohydraulic controller design in the presence of friction and internal
actuators (EHAs). The DOB is designed in the form of a second-
order high-pass filter in order to estimate the disturbance. The leakage was designed [14]. This method requires the knowl-
nonlinear controller is designed for position tracking as a near edge of the friction. All these methods improve the position
input–output linearizing inner-loop load pressure controller and tracking performance without considering their disturbances or
a backstepping outer-loop position controller. Variable structure the friction and load torque. However, when the disturbance
control is implemented in order to compensate for the error in significantly affects the position tracking performance, the
disturbance estimation. The desired load pressure is designed
to generate the pressure using the differential flatness property tracking is degraded since the force or torque available for
of the EHA’s mechanical subsystem. The disturbance within the systems is diminished. Therefore, a method for compen-
the bandwidth of the DOB can be cancelled by the proposed sating for the disturbance is needed in order to improve the
method. The performance of the proposed method is validated EHA performance. In practice, since direct measurements of
via simulations and experiments. the disturbance are not possible, it is necessary to estimate
Index Terms— Disturbance observer (DOB), electrohydraulic the disturbance using adaptation or estimation algorithms
actuator (EHA), position tracking. [15]–[19].
Several disturbance estimation methods for EHA have been
I. I NTRODUCTION proposed [16]–[19]. A constant friction estimator was pro-
posed in [16], and this method was found to be effective for
m ẍ p = −kx p − b ẋ p + A PL (5)
where m is the mass of the piston [kg], k is the load spring con-
stant [N/m], and b is viscous damping coefficient [N/(m/s)].
The goal is to make the piston position track the desired
position with disturbance compensation. In the problem to
be studied as depicted in Fig. 1, the load torque is in the
form of a sinusoidal signal. Generally, the main friction is the
Fig. 2. EHA structure. Coulomb friction in EHA [18]. Therefore, the disturbance d
in our system is in the form of a biased sinusoidal signal,
given by
parameter uncertainties, within the bandwidth of the DOB can
be canceled by the proposed method. The performance of the d = asgn(x 2 ) + m L sin(r x 1 ) (6)
proposed method is validated via simulations and experiments.
where a is the constant friction coefficient, m L is the amplitude
II. M ODELING of the load torque, and r is the rate of angle of the pendulum.
We assume that a, m L , and r are unknown but the upper
An EHA designed by the Korea Institute of Industrial Tech-
bound of m L , m L max , is known. The upper bound of the load
nology is shown in Fig. 2. In many EHA applications, the valve
torque frequency is also known. The assumption that the upper
dynamics is much faster than the dynamics of the remaining
bound of the load torque frequency is known is reasonable
parts of the system such that valve dynamics can be neglected
since the frequency of the load torque is determined by the
without significant reduction of control performance [4]–[6].
desired position profile.
Therefore, for simplicity, we use the following approximation:
Combining (1)–(6), the dynamics of the EHA can be
x v = kv i (1) formulated as the following state-space representation:
where x v is the spool position of the servo valve [m], i is the Mechanical ẋ 1 = x 2
input current of the torque motor [mA], and kv is the torque subsystem ẋ 2 = m1 (−kx 1 − bx 2 + Ax 3 − d)
motor gain [m/mA]. The control flow equation of the hydraulic
valve for the load flow rate can be written as Hydraulic
ẋ 3 = −αx 2 − βx 3 + γ Ps − sgn(u)x 3 u (7)
subsystem
1
Q L = Cd wx v Ps − sgn(x v )PL (2) where x 1 is the position of the piston [m], x 2 is the velocity
ρ
of the piston [m/s], x 3 is the load pressure [N/m2 ], d is the
where Q L is the load flow rate [m3 /s], Cd is the discharge disturbance, u is the current input [mA], α := (4βe A/Vt ),
√
coefficient, w is the area gradient of the servo-valve spool β := (4βe Ctl /Vt ), and γ := (4βe Cd wkv /Vt ρ), and all states
[m], Ps is the supply pressure of the pump [N/m2 ], PL is are measurable.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
Bode Diagram
III. C ONTROLLER D ESIGN
0
A. Disturbance Observer
The DOB is designed in the form of a second-order high- −20
pass filter in order to estimate a disturbance d that is biased on
a sinusoidal signal with an unknown frequency. The dynamics
Magnitude (dB)
−40
of x̂ 2 are defined as
1
x̂˙2 = − kx 1 − bx 2 + Ax 3 − d̂ (8) −60
m
where x̂ 2 and d̂ are the estimations of x 2 and d, and x̂ 2 (0) = 0
−80
and d̂(0) = 0. The d̂ is defined by
t
d̂ = −K P (x 2 − x̂ 2 ) − K I (x 2 − x̂ 2 )dt. (9) −100
0
Then, the estimation
t error is defined as d̃ = d − d̂ whereas d̃ I −120
is defined as 0 d̃dt with d̂ I (0) = 0. Next, we will summarize 10
−1
10
0
10
1
10
2
10
3
The proposed DOB is in the form of second order high pass- positive number in order to avoid the problem of dividing by
filter. Thus, disturbance within the bandwidth of the DOB can zero [7]. 3
be estimated by the DOB. 3 Remark 4: It is important to note that (27) cannot be
solved “as is,” since it contains the control variable u on both
B. Tracking and Flatness Controller sides of an equation involving the sgn function. However, u
in the right side is used for only the sgn function. We can
The controller is designed to track the desired profile x 1d . see that the signof u is determined by [αx 2 + βx 3 − K 3 (x 3 −
In EHA, the piston motion is obtained by the load pressure. x 3∗ )+ ẋ 3∗ ] since Ps − sgn(u)x 3 > 0. Therefore, for the digital
Therefore, the load pressure becomes the desired load pressure implementation of the control law (27), the modified control
by the current input, and then the piston position moves to law is written as
the desired position by the load pressure. In the mechanical
1
subsystem, we regard the position x 1 and the load pressure u= αx 2 + βx 3 − K 3 e3 + ẋ 3∗ (29)
x 3 as the output and the input, respectively. Then, since all γ Ps − sgn(u b )x 3
variables in the mechanical subsystem can be parameterized in where u b = αx 2 + βx 3 − K 3 e3 + ẋ 3∗ is used in practice. 3
terms of differential functions of the position and the velocity Substituting (27) in (7) yields
except the disturbance, the mechanical subsystem is clearly
differentially flat ẋ 3 = −K 3 (x 3 − x 3∗ ) + ẋ 3∗ . (30)
m ẍ 1 + b ẋ 1 + kx 1 = Ax 3 . (24) From (24)–(30), the tracking error dynamics is given by
⎡ ⎤ ⎡ ⎤⎡ ⎤
Thus the use of flatness determines the desired load pressure ė0 0 1 0 e0
x 3d such that ⎣ė1 ⎦ = ⎣ 0 −K 1v 1 ⎦ ⎣e1 ⎦
1 d ė2
−K 0 −K 1 −K 2
e2
x 3d = m ẍ 1 + b ẋ 1d + kx 1d . (25)
A em Am em
⎡ ⎤ ⎡ ⎤
The tracking errors are defined as 0 0
t +⎣ 0 ⎦ + ⎣ 0 ⎦ e3
e0 = (x 1 − x 1d )dt −d̃ − d̃max sgn( p0 e0 + p1e1 + p2e2 ) A
0
m
e1 = x 1 − x 1d Ad Bm
ė3 = −K 3 e3 . (31)
e2 = x 2 − x 2∗
e3 = x 3 − x 3∗ (26) Now we study the stability of the tracking error dynamics (31).
Theorem 2: Consider the EHA (7). Suppose that the
where x 2∗ = ẋ 1d − K 1v (x 1 − x 1d ), K 1v > 0 and x 3∗ is yet to be DOB (8), (9) and the controller (27), (28) are used, and
defined. ωc of H (s) is greater than the maximum frequency of the
Remark 2: If e2 is defined as e2 = x 2 − ẋ 1d , then controllers disturbance. If Am is Hurwitz, K 3 > 0, and p0 , p1 , and p2
were designed to make the position tracking error e1 and are chosen such that
the velocity tracking error e2 to become zero. However, the ⎡ ⎤
position tracking error e1 and the velocity tracking error e2 ∗ ∗ p0
cannot simultaneously converge to zero in physical systems. Pm = ⎣ ∗ ∗ p1 ⎦
Therefore, in order to improve the position tracking perfor- p0 p1 p2
mance, e2 = x 2 − x 2∗ is designed using backstepping instead where Q m is negative definite and Q m = Am TP + P A ,
m m m
of e2 = x 2 − ẋ 1d . 3 then T > 0 exists such that the tracking error is bounded for
The controller is designed as t ≤ T and converges to zero for t > T . 3
1 Proof: e3 exponentially converges to zero since K 3 > 0.
u= αx 2 + βx 3 − K 3 e3 + ẋ 3∗ (27)
γ Ps − sgn(u)x 3 For t ≤ T , since Ad is bounded, em converges to the bounded
ball. Then we study the stability for t > T . A Lyapunov
where candidate function V is defined as
1
x 3∗ = x 3d + − m K 0 e0 + (k − bK 1 + m K 12v − m K 1 )e1 V = em
T
Pm em (32)
A
+(b − m K 1v − m K 2 )e2 + d̂ where Pm is positive definite such that Q m = AmT P +P A ,
m m m
−d̃max sgn( p0 e0 + p1 e1 + p2 e2 ) . (28) which is negative definite. The derivative of V with respect to
time is given by
Note that the term −sgn( p0 e0 + p1 e1 + p2 e2 )d̃max is imple-
mented to compensate for the estimation error of the distur- V̇ = em
T
Q m em + 2em
T
Pm Ad (em ) + 2em
T
Pm Bm e3 . (33)
bance using VSC. Since d̃max ≥ |d̃| for t > T , em
T P A (e ) ≤ 0 for t > T .
m d m
Remark 3: In practice, Ps − sgn(u)x 3 is seldom zero, Thus
since
|x 3 | is seldom close
to Ps . In the rare case that
Ps − sgn(u)x 3 = 0, Ps − sgn(u)x 3 is set to a small V̇ ≤ em
T
Q m em + 2em
T
Pm Bm e3 . (34)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
10 d
x1
Position [mm]
5 x
1
−5
−10
0 1 2 3 4 5 6
Time [s]
(a)
Fig. 4. Block diagram of the controller structure. 1
T T
em 2Pm Bm e3 ≥ V̇ − em Q m em . (35) −0.5
−1
output input <0 0 1 2 3 4 5 6
Time [s]
Equation (35) shows that the relationship between em and (b)
2Pm Bm e3 is strictly output passive, and ėm = Am em + 1
Ad (e) is zero-state observable. Therefore, (31) is bounded-
0.5
input bounded-output stable. e3 exponentially converges to
Force [N]
zero. Therefore, em asymptotically converges to zero for 0
t > T. −0.5
Remark 5: In (27), the derivative of x 3∗ is used. The
−1
derivative of sgn(x) function is infinite at the zero point. 0 1 2 3 4 5 6
Time [s]
Since the sgn(x) function can be replaced with tanh(hx),
(c)
where h is a sufficiently large positive constant [21],
4
[22], d̃max tanh(h( p0 e0 + p1 e1 + p2 e2 )) is used instead of
d̃max sgn( p0 e0 + p1 e1 + p2 e2 ) for the digital implementation 2
Current [mA]
10 d 10 d
x1 x1
Position [mm]
Position [mm]
5 x 5 x
1 1
0 0
−5 −5
−10 −10
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time [s] Time [s]
(a) (a)
1 1
Position error [mm]
0 0
−0.5 −0.5
−1 −1
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time [s] Time [s]
(b) (b)
200 200
100 100
Force [N]
Force [N]
0 0
−100 −100
−200 −200
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time [s] Time [s]
(c) (c)
4 200
d
Estimated d
Disturbance [N]
2 100
Current [mA]
0 0
−2 −100
−4 −200
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time [s] Time [s]
(d) (d)
4
Fig. 6. Position tracking performance of only position tracking con-
troller without disturbance compensation when there is disturbance [Case 2]. 2
Current [mA]
−2
B. Case 2
The simulation results for Case 2 are shown in Fig. 6. −4
0 1 2 3 4 5 6
In Case 2, only the position tracking controller (37) in the Time [s]
proposed method was also used. Because of the load, which is (e)
heavier than the actuator mass, the force increased to overcome Fig. 7. Position tracking performance of the position tracking controller
the disturbance as compared with Case 1. Furthermore, the with Friedland algorithm when there is disturbance [Case 3]. (a) Position
force increased near zero velocity because of the friction. tracking performance. (b) Position tracking error (x1d − x1 ). (c) Force ( Ax3 ).
However, since the disturbance was not completely compen- (d) Estimated disturbance (d̂). (e) Control input (u).
sated for, the position tracking error was greater than that of
Case 1. Because of the Coulomb friction, the peak appeared method was designed to estimate constant friction. In (36),
in the control input. ż = 0 when x 2 = 0. Therefore, it was observed that the
estimation error appeared when the velocity was near zero.
C. Case 3 The estimation error resulted in position tracking error when
Fig. 7 shows the simulation results for Case 3. In [16], it the velocity was near zero. Thus, the peak and the oscillation
was shown that the Friedland method is effective to estimate a appeared near zero-velocity points in the position and the
time-varying disturbance when x 2 = 0 although the Friedland control input.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
10 d
x1
Position [mm]
5 x
1
−5
−10
0 1 2 3 4 5 6
Time [s]
(a)
1
Position error [mm]
0.5
0
Fig. 9. Photograph of the EHA test rig.
−0.5 10 Reference
Actual Position
Position [mm]
5
−1
0 1 2 3 4 5 6
Time [s] 0
(b) −5
200 −10
10 11 12 13 14 15 16
100
Time [second]
Force [N]
(a)
0
1
Position error [mm]
−100 0.5
−200 0
0 1 2 3 4 5 6
Time [s]
−0.5
(c)
200 −1
d 10 11 12 13 14 15 16
Estimated d Time [second]
Disturbance [N]
100
(b)
200
0
100
−100
Force
0
−200
0 1 2 3 4 5 6
Time [s] −100
(d)
−200
4 10 11 12 13 14 15 16
Time [second]
2 (c)
Current [mA]
−4
0 1 2 3 4 5 6 estimated accurately. Therefore, the position tracking error was
Time [s] improved when compared with that of Cases 2 and 3 and
(e) the position tracking performance was recovered. Note that
the estimated disturbance was similar to the force since the
Fig. 8. Position tracking performance of the proposed method with
disturbance compensation when there is disturbance [Case 4]. (a) Position required force to move the actuator was much smaller than
tracking performance. (b) Position tracking error (x1d − x1 ). (c) Force ( Ax3 ). that required to overcome the disturbance, but the disturbance
(d) Estimated disturbance (d̂). (e) Control input (u). is not the same as the force.
10 Reference the methods are shown in Figs. 10 and 11. The forces of
Position [mm] 5
Actual Position
both methods were asymmetric, since the zero angle of the
0 load and the zero position of the actuator were not identical.
−5 Furthermore, oscillations in the force appeared because of the
−10 structural vibration. Therefore, static friction was obscured
10 11 12 13 14 15 16 since the amplitude of the static friction was smaller than
Time [second] that of the oscillations. Fig. 11(d) shows the disturbance
(a) estimation result. The estimated disturbance includes biased
1 sinusoidal signal, effects of the assumptions for the modeling
Position error [mm]
[12] V. Pommier, R. Musset, P. Lanusse, and A. Oustaloup, “Study of two [17] Q. P. Ha, A. Bonchis, D. C. Rye, and H. F. Durrant-Whyte, “Variable
robust controls for an hydraulic actuator,” in Proc. Eur. Control Conf., structure systems approach to friction estimation and compensation,” in
Sep. 2003, pp. 1–6. Proc. IEEE Int. Conf. Robot. Autom., Apr. 2000, pp. 3543–3548.
[13] R. Nandakumar, G. D. Halikias, and A. Zolotas, “Robust control of a [18] A. Bonchis, P. I. Corke, D. C. Rye, and Q. P. Ha, “Variable structure
hydraulic actuator using the QFT method,” in Proc. Eur. Control Conf., methods in hydraulic servo systems control,” Automatica, vol. 37, no. 4,
2007, pp. 1–8. pp. 589–595, 2001.
[14] M. F. Rahmat, Zulfatman, A. R. Husain, K. Ishaque, Y. M. Sam, R. [19] C. S. Kim and C. O. Lee, “Speed control of an overcentered variable
Ghazali, and S. M. Rozali, “Modeling and controller design of an displacement hydraulic motor with a load torque observer,” Control Eng.
industrial hydraulic actuator system in the presence of friction and Pract., vol. 4, no. 11, pp. 1563–1570, 1996.
internal leakage,” Int. J. Phys. Sci., vol. 6, no. 14, pp. 3502–3517, [20] C. T. Chen, Linear System Theory and Design, 3rd ed. New York: Oxford
2011. Univ. Press, 1999.
[15] K.-S. Kim, K.-H. Rew, and S. Kim, “Disturbance observer for estimating [21] B. Yao, F. P. Bu, and G. T. C. Chiu, “Nonlinear adaptive robust control of
higher order disturbances in time series expansion,” IEEE Trans. Autom. electro-hydraulic servo system with discontinuous projection,” in Proc.
Control, vol. 55, no. 8, pp. 1905–1911, Aug. 2010. IEEE Conf. Dec. Control, Dec. 1998, pp. 2265–2270.
[16] B. Friedland and Y. J. Park, “On adaptive friction compensa- [22] K. D. Do and J. Pan, “Boundary control of transverse motion of marine
tions,” IEEE Trans. Autom. Control, vol. 37, no. 10, pp. 163–166, risers with actuator dynamics,” J. Sound Vibrat., vol. 318, nos. 4–5,
Oct. 1993. pp. 768–791, 2008.