Professional Documents
Culture Documents
f = f2 + f1 T = (f2 − f1 )B/2. (3) where “∧ ” is used to indicate the estimated model parameters.
The sliding-mode control law is normally derived by subtract-
ing a sign function from the nominal control. However, there are
III. S LIDING M ODE C ONTROL L AW several disadvantages in using a discontinuous control law, such
In the sliding-mode control approach [24], [25], we define as chattering and the inability of the propellers to produce a
asymptotically stable surfaces (S) such that all system trajec- discontinuous control action. Hence, we define an approximate
tories converge to these surfaces in finite time referred to as the sliding-mode control law using a high-slope saturation function
reach time tr and slide along them until they reach the desired as in [26]
destination at their intersection. The reaching conditions are
normally established by defining (1/2)S T S as the Lyapunov f = fˆ − k1 sat(S1 /φ1 ) (11)
function and ensuring that its time derivative is negative. In the
S1 /φ1 , if |S1 | ≤ φ1
case of underactuated surface vessels, we define two surfaces to sat(S1 /φ1 ) = (12)
sgn(S1 ), if |S1 | > φ1
determine the two control inputs. Hence, the reaching condition
for each surface i may be defined as where φ1 is a positive constant which defines a small boundary
layer around the surface.
Si Ṡi ≤ −ηi |Si |, ηi > 0; i = 1, 2 (4)
In order to determine k1 , let us define the following bounds
where the value of constant ηi determines how fast the trajec- for the model parameters in (2):
tory will reach surface i, i.e., tr ≤ |si |/ηi .
|mii − m
ii | ≤ Mii |di − dˆi | ≤ Di , i = 1, 2, 3. (13)
A. Surge Control Law
We assume no uncertainty in our estimates of exponents αi
The first sliding surface is a first-order exponentially stable for simplicity. Note that our experiments have shown less than
surface defined in terms of the vessel’s surge motion tracking 1.5% uncertainty for these exponents as presented in Section V.
4006 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 11, NOVEMBER 2008
Define a Lyapunov candidate function similar to (4) that where “∧ ” indicates the estimated model and
guarantees reaching the set {|S1 | ≤ φ1 } in finite time and re-
main inside it thereafter b = m22 vxd − m11 vx (24)
α3
1 h = b (md vx vy + d3 sgn(ω)|ω| )
V1 = m11 S12 . (14)
2 − (m11 vx ω + d2 sgn(vy )|vy |α2 ) α2 d2 sgn(vy )|vy |α2 −1
The time derivative of (14) can be derived using (2), (9), + m33 ω (f − d1 vxα1 + 2λ2 m11 vx + m22 vy )
and (11)
+ 2λ2 m33 d2 sgn(vy )|vy |α2
f + m22 vy ω − d1 vxα1
V̇ = m11 S1 Ṡ1 = m11 S1 + Ṡr1 + m22 m33 vr + 2λ2 v̇yd − λ22 ṽy . (25)
m11
We define the yaw moment sliding-mode control law as
= S1 (m22 − m 22 )vy ω + (dˆ1 − d1 )vxα1
T = ĥ − k2 sat(S2 /φ2 ) /b̂. (26)
+ (m11 − m
11 )Ṡr1 − k1 sat(S1 /φ1 ) . (15)
It is interesting to note that the lateral motion control depends
Note that the saturation function is equal to the sign function not only on the yaw moment but also on the surge force f
for the set {|S1 | > φ1 } according to (12). Hence, the following as one would expect. It must also be noted that the control
reaching condition is achieved: law becomes singular if b̂ = 0. Since the control law is valid
only when surge motion is required (nonzero surge velocities),
V̇1 = m11 S1 Ṡ1 ≤ −m
11 η1 |S1 | (16) singularity could only occur when the tracking surge-velocity
error is more than 20%. This is due to the fact that m22 is
if k1 is selected as
normally more than 20% larger than m11 [23] (see, e.g., data
k1 = M22 |vy ω| + D1 vxα1 + M11 |Ṡr1 | + m
11 η1 . (17) in Section V). Hence, we assume that b > 0.
In order to determine k2 , let us define a bound for the un-
certainty in h which, in turn, is defined based on the parameter
B. Lateral Motion Control Law uncertainties defined in (13)
We define the second sliding surface as a second-order expo- |h − ĥ| ≤ H. (27)
nentially stable surface in terms of the vessel’s lateral motion
tracking errors We also define the following bounds based on the geometric
mean of b [27]:
t
S2 = ṽ˙ y + 2λ2 ṽy + λ22 ṽy (τ ) dτ, λ2 > 0 (18)
1/β ≤ bb̂−1 ≤ β, b̂ = bmin bmax , β = bmax /bmin (28)
0
22 + M22 )vxd − (m
bmax = (m 11 − M11 )vx > 0
where ṽy = vy − vyd and ṽ˙ = v̇y − v̇yd . The estimates of these 22 − M22 )vxd − (m
bmin = (m 11 + M11 )vx > 0. (29)
error states are available through feedback and (1), (2), (7),
and (8). We define another Lyapunov candidate function that guaran-
We calculate a nominal yaw control law for zero dynamics tees reaching the set {|S2 | ≤ φ2 } in finite time
by taking the time derivative of the surface and setting it equal
to zero 1
V2 = m22 m33 S22 . (30)
2
Ṡ2 = v̈y + v̈yd + 2λ2 (v̇y − v̇yd ) + λ22 ṽy = 0 (19)
Noting that the saturation function is equal to the sign function
where the time derivatives of the second equations in (2) and for the set {|S2 | > φ2 }, the time derivative of (30) can be
(8) yield derived using (19)–(26) as
α2 d2 sgn(vy )|vy |α2 −1 v̇y + m11 (v̇x ω + vx ω̇)
v̈y = − (20) bT − h
m22 V̇2 = m22 m33 S2 Ṡ2 = m22 m33 S2
m22 m33
v̈yd = vr − vxd ω̇ (21) = S2 bb̂−1 ĥ − k2 sat(S2 /φ2 ) − ĥ
...
vr = (vyd ω − 2v̇xd )ω − sin θ x d + cos θ
...
yd . (22)
= S2 (bb̂−1 − 1)ĥ + ĥ − h − bb̂−1 k2 sat(S2 /φ2 ) . (31)
The nominal control is derived by substituting (2), (20)–(22)
into (19) Hence, the following reaching condition is achieved:
if k2 is selected as
22 m
k2 = β(H + m 33 η2 ) + (β − 1)|ĥ|. (33)
C. Stability Analysis
The surge force and yaw moment control laws in (11) and
(26) are derived based on the reaching conditions in (16)
and (32), respectively. Integration of these reaching conditions
verifies that the trajectory reaches the two corresponding sur-
faces in a finite time of less than (m11 /m 11 )(S1 (0)/η1 ) and
(m22 m33 /m 22 m
33 )(S2 (0)/η2 ), respectively. Furthermore, the
two surfaces in (5) and (18) are asymptotically stable. Thus, the
trajectory exponentially slides to the origin at the intersection
of the two surfaces. Thus,
t t
ṽx → 0, ṽx (τ ) dτ → 0, ṽy → 0, ṽy (τ ) dτ → 0 Fig. 2. Vessel with dc motors, receivers, gearboxes, and couplers.
0 0
(34)
and the kinematic relations in (1) guarantee trajectory tracking
in the inertial reference frame based on the real-time feedback
of θ and ω.
We also propose that ω is bounded-input–bounded-output
stable. Let us define the Lyapunov candidate function
1
V3 = m33 ω 2 . (35)
2
By using (2), the time derivative of V3 may be written as
V̇3 = ω [T − md vx vy − d3 sgn(ω)|ω|α3 ]
= ω(T − md vx vy ) − d3 |ω|1+α3 (36)
V̇3 < 0 if d3 |ω|α3 > |T − md vx vy |. (37)
m33 ω̇ + d3 sgn(ω)|ω|α3 = T − md vx vy
0, linear motion
T − m d vx vy = Fig. 3. Diagram of the experimental setup.
constant, circular motion.
Hence, after the initial transient period, the steady-state solution are only controllable in a high-speed range, we have reduced
will be zero and constant for straight-line and circular trajec- their speed at 1 : 16 ratio with custom-designed planetary gear
tories, respectively. trains. The motors were originally controlled using a joystick
and wireless receivers. However, we have modified the joystick
such that computed control signals can be transmitted through
IV. E XPERIMENTAL S ETUP
a data-acquisition card.
The experiments are performed with a 0.45-m-long 1.614-kg We used a digital black and white 640 × 480-resolution
boat in a 1.9-m × 2.6-m indoor pool (Fig. 2). A diagram of camera for our feedback measurements. The camera rate is
the experimental setup is shown in Fig. 3. Two dc motors are 30 frames/s. The camera is installed 1.83 m above the center
mounted to propeller shafts using couplings built in-house. The of the pool and captures the image of the whole pool area.
propellers are 0.07 m apart (B = 0.07 m). Since the dc motors Since the camera image is distorted and in the form of an image
4008 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 11, NOVEMBER 2008
the surge and sway velocities are estimated using the estimated
velocities in the inertial reference frame and the inverse of the
transformation equations presented in (1).
V. R ESULTS
We performed a series of vessel surge, sway, and yaw motion
experiments by applying various known constant forces and
moments through a pulley system to estimate the parameters
of the model presented in (2). These experiments are not the
subject of this paper and are presented in [28]. The model data
and their uncertainty in SI units for the experimental vessel are
α1 = 1.510 ± 0.0075
Fig. 8 shows that the actual paths followed by the vessel are
α2 = 1.747 ± 0.013 very close to the desired straight-line paths in three different
experiments. The controller is very robust despite the error
α3 = 1.592 ± 0.0285.
in motor and camera calibrations. Although the path error
was very small, the vessel’s trajectory error in the x- and
y-directions was larger, as shown in Fig. 9. It seems that,
A. Straight-Line Trajectory
during the acceleration and cruise period, the boat falls behind
We conducted a series of straight-line trajectory control the trajectory. However, the vessel compensates during the
experiments. In each case, the vessel was commanded to follow deceleration phase, and the errors become smaller again. This
a straight line for 8 s starting from rest at one corner of the pool suggests that the interpolation process underestimates the input
and stopping near the opposite corner for approximately a 2-m voltage for the positive propeller motion, and hence, smaller
travel. We used a fifth-order polynomial to plan the trajectory in voltages result in slower vessel motion. Fig. 10 shows that the
the x-direction. The six boundary conditions used to determine heading angle during a typical experiment is not constant but
the polynomial coefficients were zero initial and final (after 8 s) remains bounded.
velocities and accelerations, and the initial and final position Fig. 11 shows the input voltages for the two motors during
based on the length of travel and initial heading angle. The a typical straight-line experiment and simulation. It can be
motion in the y-direction was defined as a linear function of the observed that the applied experimental voltages during the
x-direction based on the initial heading angle. The control law positive propeller motion are smaller. Another interesting phe-
parameters were selected as λ1 = λ2 = 0.5, η1 = η2 = 0.01, nomenon is that the vessel must maintain a constant heading
and φ1 = φ2 = 0.1. angle to go straight. However, after nearly 7 s, its heading
4010 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 11, NOVEMBER 2008
Fig. 13. Desired versus actual vessel trajectory during a circular experiment.
Fig. 11. Comparison of experimental and simulation control signals.
Kenneth R. Muske (M’05) received the B.S. Reza A. Soltan (M’08) received the B.S. degree in
and M.S. degrees from Northwestern University, mechanical engineering from K. N. T. University of
Evanston, IL, in 1981, and the Ph.D. degree from the Technology, Tehran, Iran, in 2006. He is currently
University of Texas, Austin, in 1994. working toward the M.S. degree in mechanical engi-
He was a Process Control System Consultant neering at Villanova University, Villanova, PA.
prior to his Ph.D. studies and was a Technical Staff He is also with the Center for Nonlinear Dynamics
Member with Los Alamos National Laboratory af- and Control, Villanova University. His current re-
ter receiving the Ph.D. degree. Since 1997, he has search interests are in the areas of dynamic systems,
been with the College of Engineering, Villanova nonlinear control, limit cycles, and stability theory.
University, Villanova, PA, where he is currently the
Mr. and Mrs. Robert F. Moritz Sr. Chair of Systems
Engineering and a Professor of chemical engineering. He is also with the Center
for Nonlinear Dynamics and Control, Villanova University. His current research
interests are in the areas of system modeling, control, and optimization with
application to automotive, chemical, and autonomous systems.