Professional Documents
Culture Documents
Second-Order Sliding-Mode Control of Container Cranes: Brief Paper Giorgio Bartolini, Alessandro Pisano, Elio Usai
Second-Order Sliding-Mode Control of Container Cranes: Brief Paper Giorgio Bartolini, Alessandro Pisano, Elio Usai
www.elsevier.com/locate/automatica
Brief Paper
Second-order sliding-mode control of container cranes
Giorgio Bartolini ∗ , Alessandro Pisano, Elio Usai
Dipartimento di Ingegneria Elettrica ed Elettronica, Universita di Cagliari piazza d’Armi, I-09123 Cagliari, Italy
Received 24 November 2000; received in revised form 24 July 2001; accepted 2 May 2002
Abstract
Moving a suspended load along a pre-speci0ed path is not an easy task when strict speci0cations on the swing angle and on the transfer
time need to be satis0ed. Intuitively, minimizing the cycle time and the load swing are con5icting requirements, and their satisfaction
requires proper control actions, especially if some uncertainties in the system dynamics are present. In this paper we propose a simple
control scheme, based on second-order sliding modes, which guarantees a fast and precise load transfer and the swing suppression during
the load movement, despite of model uncertainties and unmodeled dynamic actuators. Such controller has been tested on a laboratory-size
model of the crane, and some experimental results are reported.
? 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Container cranes; Second-order sliding modes; Uncertain systems; Nonlinear systems
0005-1098/02/$ - see front matter ? 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 5 - 1 0 9 8 ( 0 2 ) 0 0 0 8 1 - X
1784 G. Bartolini et al. / Automatica 38 (2002) 1783–1790
by a fourth-order dynamics, which must be proved to enjoy Consider the auxiliary variable w = ’˙ + c’; where c is a
the desired properties of tracking and stability. By consid- positive arbitrary constant. In the new w–’ coordinates; (26)
ering (19) and (16), one obtains can be rewritten as
’˙ = −c’ + w;
ẋt = V0 + k’; (20) (27)
ẇ = −[b() − c]w + ["() − a()]’;
g k + 2l̇(t) where
’M = − ’− ’;
˙ (21)
l(t) l(t)
"() = c(b() − c): (28)
d
l̇ = l̇ − c2 (l − ld ): (22) Since the di>eomor0sm (’; ’) ˙ = T (’; w) preserves the
origin; the asymptotic stability of system (27) – (28) implies
The stabilizing e>ect the k parameter on the nonlinear that of system (26).
swing-angle dynamics (21) is shown to guarantee the De0ning the Lyapunov candidate function V (’; w) =
asymptotic vanishing of the load swing (Lemma 1). The 1 2
’ + 12 w2 , the corresponding derivative along the trajec-
2
rationale of the proof is the following: we 0rst observe that tories of system (27) – (28) is a quadratic form of the type
the rope-length zero-dynamics (22) is such that l and l̇
exponentially converge to the desired values in (49). Thus V̇ (·) = −[’; w]T M()[’; w]; (29)
the relationships
where
d
l(t) = ld + 1 e−c2 t; l̇(t) = l̇ + 2 e−c2 t (23) c − "()−a()+1
2
M() = : (30)
are satis0ed, for some constants 1 , 2 , at any t ¿ T ∗ , − "()−a()+1
2 b() − c
T ∗ being the time instant at which the sliding mode on
The matrix M() is positive de0nite if the following in-
s1 = s2 = 0 is permanently established. Then, by combining
equality holds:
(23), (49), (20), and (21), it is shown that the swing angle
asymptotically converges to the origin, provided that the 4"() ¿ ["() − a() + 1]2 ; || 6 XM : (31)
k coeScient is suitably chosen and that the trolley position
remains con0ned within a suitable compact domain. As a Note that the parameter c is absolutely 0ctitious, as it appears
0nal step, it is easily derived from (20) that, as soon as ’ only in the stability proof and does not a>ect neither the
tends to zero, the load moves with the desired constant structure of the sliding manifold nor that of the controller.
velocity V0 . After some algebraic manipulations, (31) can be rear-
ranged as follows:
Lemma 1. There exists open intervals K ≡ (km ; kM ) and
Xt ≡ (−XM ; XM ) such that; if k ∈ K and xt ∈ Xt ; the swing g1 () ¡ "() ¡ g2 (); (32)
dynamics (21) is locally asymptotically stable at the origin where
and the load trajectory tends to the prescribed one.
g1 () = a() + 1 − 2 a();
Proof. By (22); conditions (23) hold. Substituting (23); (33)
(49) and (20) into (21); and neglecting the terms containing g2 () = a() + 1 + 2 a():
’’˙ on the basis of the small oscillation assumption; it yields According to (33), (24) and (28), the limiting curves g1 ()
g k + 4V0 and g2 () depend only on the reference path, while the func-
’M = − ’− ’˙ + d(t)e−c2 t tion "() depends also on parameters c and k, therefore, the
l0 + 2 l0 + 2
actual crane parameters do not a>ect the stability condition
, −a()’ − b()’˙ + d(t)e−c2 t ; (24) (32) – (33).
The curve "(), de0ned by (28) and (24), can be shaped
where = xt and d(t) is given by by properly setting k and c in order to make it lying within
the limiting curves g1 () and g2 () in a suitable domain of
d
1 g’ + [1 (k + 2l̇ ) − 22 ld ]’˙ interest X : || 6 XM .
d(t) = : (25) While ’ approaches zero, by (20), it is immediate to
ld (ld + 1 e−c2 t )
conclude that the actual trolley velocity becomes closer and
Since the “disturbing” term d(t)e−c2 t is exponentially closer to the desired value V0 . .
vanishing; in order to prove the lemma it suSces to guar-
antee the stability of the nominal part Remark. As a design example; consider the desired path
de0ned by l0 = 0:1 m; = 1 m−1 and XM = 3 m; and let the
’M = −a()’ − b()’:
˙ (26) reference horizontal speed be V0 = 0:1 m s−1 . A reasonable
G. Bartolini et al. / Automatica 38 (2002) 1783–1790 1787
120 where d (t) is bounded. Around the 0nal location, the swing
g2(ξ) dynamics is an LTI system a>ected by a vanishing distur-
100 bance; thus, choosing
γ (ξ)
g
80 ka ∈ 0; (38)
c1
60
g1(ξ)
the asymptotic stability is trivially guaranteed.
40
4. A multi-input second-order sliding mode controller
20
In this section it is shown that the problem of steer-
0
ing the system motion onto the desired manifold can be
-3 -2 -1 0 1 2 3
solved, both in the travel phase and in the arrival phase,
by means of a recently proposed multi-input second-order
Fig. 3. The curves g1 , g2 and " for l0 = 0:1 m, = 1 m−2 , c = 0:6 and sliding mode control strategy (Bartolini, Ferrara, Usai, &
k = 16:65. Utkin, 2000).
Di>erentiating twice (19) and (34), taking into account
(1) – (5), it yields
way for choosing c and k is to guarantee that the point Travel phase
"(0)=c(k=l0 −c) lies approximately at the medium between
sM = F2 (q; p) + 1−1 B(xt ; ’; p; )v;
g1 (0) and g2 (0). By continuity; (32) will remain satis0ed in
an open neighbour of = 0. In the actual case 12 (g2 (0) + Arrival phase
g1 (0)) ≈ 100; and choosing c = 0:6 and k = 16:65 one can
place "(0) as required; obtaining the shape in Fig. 3. Note sM = F2 (q; p) + 1−1 B(xt ; ’; p; 0)v; (39)
that the actual curve "() is entirely contained between the
limiting curves in the domain of interest || 6 3. If one wants where q =[xt ; ẋt ; l; l̇; ’; ’;˙ t1 ; t2 ], p ∈ P is the vector contain-
to enlarge the size of the stability domain Xt , a qualitative ing the system (crane and actuators) parameters, s =[s1 ; s2 ]T
criterion to adopt is that of reducing c while keeping constant and s = [s1 ; s2 ]T . The vector 0elds F2 (·) = [f21
(·); f22 (·)]
the value of "(0). and F2 (·)=[f21 (·); f22 (·)], are given in the appendix, while
the matrix B(·) is expressed as follows:
1 A2 −B1 sin(’)
B(xt ; ’; p; ) = : (40)
+(’) −B2 sin(’) − 2xt A2 A1 − B1 cos2 (’) + 2B1 xt sin(’)
ẋt = −c1 (xt − xf ) + ka ’; (35) Those facts, according to Bartolini et al. (2000), are
suScient to separate the multi-input control problem
g − c 1 ka ka in a set of almost-decoupled single-input control
’M = − ’− ’˙ + d (t)e−c2 t ; (36)
yf yf problems.
According to the “sub-optimal” sliding mode control al-
l̇ = −c2 (l − yf ); (37) gorithm (Bartolini, Ferrara, Levant, & Usai, 1999), during
1788 G. Bartolini et al. / Automatica 38 (2002) 1783–1790
where si is de0ned in (18) – (49), TM ki are the time in-
stants at which ṡi (i = 1; 2), is zero and Vi (i = 1; 2),
are proper constants computed on the basis of the un-
certainty upperbounds (Bartolini et al., 2000). The ad-
missible domains for the control e>ort is semi-in0nite,
and of the type Vi ∈ [ViL ; ∞), so that, in order to avoid
over-sizing of the controller, a trial-and-error proce-
dure, based on choosing V1 and V2 “suSciently large”
to ensure satisfactory performance, should be adopted in
Fig. 5. First test. The actual swing angle.
practice.
After a vicinity of the destination point has been attained
(e.g. |xt − xf | 6 ., . ¿ 0), the arrival phase is activated.
The control strategy is still of the type (42) but the sliding initial oscillation, intentionally introduced, is damped dur-
variables s1 ; s2 in (34), and new control amplitudes V1 and ing the transfer (see Fig. 6). Note that the actual trajec-
V2 de0ned by a trial-and-error procedure as before, must tory remains close to the desired one, and the oscillation
be used. disappears before the vertex of the actual trajectory is
reached.
100
40
Load vertical position [cm]
80
Load coordinates [cm]
30 x
60
20
40
10 20 y
0 0
0 20 40 60 80 0 5 10
Load horizontal position [cm] Time [sec]
40
0.2
20 0.1
10 0.0
0
-0.1
0 20 40 60 80 0 2 4 6 8 10
Horizontal load position Time [sec]
Fig. 6. Second test. The actual and reference load coordinates and the load swing.
Appendix A. The crane dynamics in explicit form Explicit expressions for xt(3) , l(3) and ’(3) , which can be
obtained by di>erentiating (43) and considering (44) – (46)
The dynamic equations (1) – (3) can be rewritten in matrix and (5), are given
form as
x(3)
xM
(3)
M l = -̇1 (·) + Ġ(l; ’)t + G(l; ’)t˙
l = -1 (·) + G(·)t; (43)
’M (3)
’
K1 (l’˙ 2 sin(’) + g cos(’) sin(’))
/1x (·)
1 K2 (l’˙ 2 + g cos(’))
-1 (·) = /1l (·) = ; (44)
+(’)
/1’ (·) K 1 K 3 K 0
− (l’˙ 2 sin(’) cos(’) − 2l̇’˙ sin2 (’)) − 2 l̇’˙ − g sin(’)
l l l
K0 = A1 A2 − B1 B2 ; K1 = A2 B1 − B1 B2 ; sM1 = xt(3) − k ’;
M
(3) d
K2 = A1 B2 − B1 B2 ; K3 = A1 A2 − A2 B1 : (46) sM2 = l(3) − ld + c2 (lM − lM ); (48)
1790 G. Bartolini et al. / Automatica 38 (2002) 1783–1790
The derivatives of ld can be expressed as Hong, K. -S., Park, B. -J., & Lee, M. -H. (2000). Two-stage control
d for container cranes. International Journal of Japan Society of
ld = l0 + xt2 ; lM = 2ẋ2t + 2xt xMt ; Mechanical Engineering (JSME Series C), 43(2), 273–282.
Levant, A. (1998). Robust exact di>erentiation via sliding mode technique.
d (3)
l̇ = 2xt ẋt ; ld = 6ẋt xMt + 2xt xt(3) : (49) Automatica, 34, 379–384.
Moustafa, K., & Ebeid, A. M. (1988). Nonlinear modeling and control of
By combining (48) – (49) with (43) and (47) the relevant overhead crane load sway. Journal of Dynamic Systems Measurement
vector 0elds turn out to be expressed as and Control, Transactions ASME, 110, 266–271.
Sakawa, A., & Shindo, Y. (1982). Optimal control of container cranes.
f21 (·) = /2x (·) − k(/1’ (·) + G31 (·)t1 + G32 (·)t2 ); Automatica, 18(3), 257–266.
Utkin, V. I. (1992). Sliding modes in control and optimization. Berlin:
Springer.
f22 (·) = /2l (·) − (6ẋt + 2c2 xt ) Virkkunen, J. (1990). Adaptive pole-placement control of a pilot crane.
Proceedings of the 11th IFAC world congress, Tallinn, Estonia (pp.
×[/1x (·) + G11 (·)t1 + G12 (·)t2 ] − 2xt /2x (·) + c2 313–318).