See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/245449447

Adaptive Backstepping Control of Nonlinear Hydraulic-Mechanical System

Including Valve Dynamics

Article in Modeling, Identification and Control (MIC) · January 2010

DOI: 10.4173/mic.2010.1.3 · Source: DOAJ

CITATIONS READS

14 256

2 authors:

M. Choux Geir Hovland

Universitetet i Agder Universitetet i Agder
17 PUBLICATIONS 125 CITATIONS 97 PUBLICATIONS 1,301 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by M. Choux on 22 April 2014.

The user has requested enhancement of the downloaded file.

Modeling, Identification and Control, Vol. 31, No. 1, 2010, pp. 35–44, ISSN 1890–1328

Adaptive Backstepping Control of Nonlinear

Hydraulic-Mechanical System Including Valve
Dynamics
M. Choux G. Hovland

Mechatronics Group, Department of Engineering, University of Agder, N-4898 Grimstad, Norway. E-mail:
{martin.choux,geir.hovland}@uia.no

Abstract

The main contribution of the paper is the development of an adaptive backstepping controller for a nonlin-
ear hydraulic-mechanical system considering valve dynamics. The paper also compares the performance of
two variants of an adaptive backstepping tracking controller with a simple PI controller. The results show
that the backstepping controller considering valve dynamics achieves significantly better tracking perfor-
mance than the PI controller, while handling uncertain parameters related to internal leakage, friction,
the orifice equation and oil characteristics.

Keywords: adaptive observer backstepping, state feedback, nonlinear hydraulic-mechanical system, valve
dynamics

1. Introduction stepping controller was not included in the survey of

Bonchis et al. (2002). Hence, it would be of inter-
Control of nonlinear hydraulic-mechanical systems est to compare the backstepping and the PI controller
(NHMS) is challenging for several reasons: a) the sys- for an NHMS. In Zeng and Sepehri (2006, 2008) the
tem model is normally stiff with fast dynamics for the authors presented an adaptive controller to handle in-
hydraulics and relatively slow dynamics for the me- ternal leakage and unknown friction in a cylinder, un-
chanical parts, b) models usually contain strong non- known volumes in the orifice equation and temperature
linear elements such as the flow in orifices, friction, dependent oil characteristics.
valve overlap and input saturation, c) valves contain One physical phenomenon not considered in Zeng and
non-measurable states (position and velocity) and d) Sepehri (2006, 2008) is valve dynamics. Section 2.2
the oil characteristics depend on parameters such as shows that valve dynamics can be significant and
temperature and air content. should be included in the model-based controller. In
Bonchis et al. (2002) present an experimental evalua- addition to the valve dynamics, the adaptive controller
tion of ten different controller algorithms for an NHMS. developed in this paper also handles internal leakage
The results in the paper show that the simple PI con- and unknown friction in the cylinder, unknown vol-
troller performs reasonably well, and only a few of the umes in the orifice equation and temperature depen-
model-based controllers are able to improve the perfor- dent oil characteristics.
mance. Section 2 contains the model description including an
Adaptive backstepping is a model-based nonlinear con- experiment to determine the second order valve dy-
trol technique which has been recently applied to namics model, while sections 3 and 4 contain the con-
NHMS, see Zeng and Sepehri (2006, 2008). The back- trollers for two different scenarios: without and with

doi:10.4173/mic.2010.1.3 c 2010 Norwegian Society of Automatic Control

 Modeling, Identification and Control

valve dynamics while both controllers contain the non- 2.1. Linear Friction Model
linear orifice equation. Section 5 contains simulation
results with the two different backstepping controllers In this work the influence of valve dynamics is the main
compared with a PI controller. Finally, section 6 con- focus of the paper. Extra states added by considering
tains the conclusions. the dynamics of the friction model would complicate
the study of the valve dynamics. In this regard the
chosen friction model is linear:
2. Model Description
Ff ric = σ ẏ (1)
The tracking of the mass position y in the NHMS
shown in Fig. 1 is considered. The system in state space representation, with hy-
draulic units is:

k d+σ A
ÿ = − y− ẏ + pL (2)
M M 10M
4βA 4β
ṗL = − ẏ − cL pL
Vt Vt
√ r
400 10βCd wKv 1 √
+ p − p L xv (3)
Vt ρ
 
2D
x¨v = −ωv2 x˙v + xv + Kv u (4)
ωv

where u is the input current of the valve. If

the state variables [y, ẏ, pL , xv , x˙v ] are equal to
[x1 , x2 , 10M
A
x3
, x4 , x5 ], the system can be rewritten as:

ẋ1 = x2 (5)
T
ẋ2 = x3 + ϕ2 (x1 , x2 ) θ (6)
T
ẋ3 = b f (x3 )x4 + ϕ3 (x2 , x3 ) θ (7)
Figure 1: Translational hydraulic-mechanical system. ẋ4 = x5 (8)
T
ẋ5 = ϕ5 (x4 , x5 ) θ + u (9)
The parameters of the system are given in Table 1.
where θ is the vector of unknown parameters:
Table 1: Values of the system parameters with hy- 
−k −d + σ
T
draulic units. θ = [θ1 , · · · , θ6 ] =
, ,
M M
T
Par. Value Par. Value −2βA2 −4βcL 2
M = 41 kg Kv = 1 m/A , , −ωv , −2Dωv (10)
5M Vt Vt
k = 65000 N/m d = 500 Ns/m
A = 946 mm2 β = 12665 bar b is a known-scalar:
ρ = 900 kg/m3 Vt = 782 cm3
√
Cd = 0.65 w = 7 mm
r
40 10AβCd ωKv 1
p = 80 bar cL =1 b= (11)
M Vt ρ
σ0 = 5880 σ1 = 108
σ2 = 500 Fc = 100 and f is a nonlinear function:
Fs = 200 vs = 0.001
QL = QL l/min y =y m
p
f (pL ) = p − sign(xv )pL (12)
pL = pL bar xv = xv mm
ωv = 100 rad/s D =1 The vector functions ϕk (k ∈ {2, 3, 5}) are defined
as: ϕ2 (x1 , x2 ) = [x1 , x2 , 0, 0, 0, 0]T , ϕ3 (x2 , x3 ) =
[0, 0, x2 , x3 , 0, 0]T and ϕ5 (x4 , x5 ) = [0, 0, 0, 0, x4 , x5 ]T .

36
 Choux and Hovland, ”Adaptive Backstepping Controller”

−3 y
x 10

10

Position (m)
6
Figure 2: Experimental setup for NHMS.
4

2.2. Valve Dynamics

2
An experimental setup shown in Fig. 2 has been used
to determine the valve dynamics. A proportional and 0
a second order valve dynamics model are compared.
The proportional model is given by xv = Kv u while 8.6 8.7 8.8 8.9 9 9.1
the second order model is given by eq. (4). A step re- Time (sec)

sponse was generated by using a PI controller on the Figure 4: Transient response. Blue: measurements,
experimental setup in Fig. 2 and also on two simulated Black: second order model, Green: error
models with: a) proportional valve characteristics and second order model, Dashed: proportional
b) a second order valve dynamics model (eq. (4)). The model, Red: error proportional model.
results are shown in Figs. 3 and 4. Fig. 3 shows the en-
tire step response and the reverse step response, while
Position Proportional 2nd order Improvement
Fig. 4 is zoomed in at the transient response where the
RMS 7.16 · 10−4 3.41 · 10−4 52%
errors are the largest.
|MAX| 9.19 · 10−6 2.46 · 10−6 73%
−3 y
x 10
20 Pressure Proportional 2nd order Improvement
RMS 1.94 1.96 -1%
15
|MAX| 98.1 46.6 110%

Table 2: Comparison of RMS and MAX values against

10 experiments for a step response using a pro-
Position (m)

portional and a second order valve dynamics

model. Top: Position, Bottom: Pressure.
5

0
measured load pressure pL vs. the simulated load pres-
sures. The pressure RMS values for the proportional
and second order model are similar, while the MAX
−5 value shows a significant improvement for the second
8.5 9 9.5 10 10.5 11 11.5 12 12.5
Time (sec) order model. Hence, the effects of the valve dynamics
are important to consider in high-performance control
of NHMS.

Figure 3: Experimental and simulated results with two

different valve models. Blue: measurements, 3. Backstepping without Valve
Black: second order model, Green: error
second order model, Dashed: proportional Dynamics
model, Red: error proportional model.
In order to demonstrate the effects of introducing valve
Table 2 shows that the second order model represents a dynamics in the backstepping controller, a controller
significant improvement compared to the proportional without considering valve dynamics is developed first.
model. The step response (position) is improved 52% The complete controller including valve dynamics is
and 73%, respectively, for the RMS and MAX values. presented in section 4.
Table 2 also shows the RMS and MAX values for the The system without valve dynamics can be rewritten

37
 Modeling, Identification and Control

as: Stabilizing functions:

k d + σ2 A α1 (x1 , yr ) = ᾱ1 (27)

ÿ = − y− ẏ + pL (13)
M M 10M α2 (x1 , x2 , θ̂, yr , ẏr ) = ᾱ2 (28)
A 2βA2 4β A
ṗL = − ẏ − cL pL λ̂
10M 5M Vt Vt 10M α3 (x̄3 , θ̂, ȳr(2) , λ̂) = ᾱ3 (29)
√ r f (x3 )
40 10AβCd wKv 10M A
+ √ p − sign(u) pL u
M Vt ρ A 10M
(14) ᾱ1 = −L1 z1 (30)
∂α1
or, alternatively ᾱ2 = −z1 − L2 z2 − ω2T θ̂ + x2
∂x1
∂α1
ẋ1 = x2 (15) + ẏr (31)
T
∂yr
ẋ2 = x3 + ϕ2 (x1 , x2 ) θ (16) ∂α2
ẋ3 = b f (x3 )u + ϕ3 (x2 , x3 )T θ (17) ᾱ3 = −z2 − L3 z3 − ω3T θ̂ + Γ τ3
∂ θ̂ !
2
X ∂α2 ∂α2 (k)
where θ and ϕ are reduced by two states, ie. + xk+1 + (k−1) yr (32)
h
−(d+σ) −2βA2 −4βcL
iT ∂xk ∂yr
θ = [θ1 , · · · , θ4 ]T = −kM , M , 5M Vt , Vt , k=1

[x1 , x2 , 0, 0]T , ϕ3 (x2 , x3 ) = [0, 0, x2 , x3 ]T Adaptive control law:

ϕ2 (x1 , x2 ) = q
and f (x3 ) = p − sign(u) 10M A x3 . Following the tun-
λ̂
ing function design as in Krstić et al. (1995), the state u = α3 + yr(3) (33)
space system (15-17), which is in a strict feedback form f (x 3)

can be decomposed in sucessive subsystems for which

tuning functions and stabilizing functions are recur- Parameter update laws:
sively found, leading to the final adaptive control law ˙
u and the final update law for the uncertain parame- θ̂ = Γ τ3 (34)
ters θ and λ = 1b with estimated θ̂ and λ̂. Note that ˙
 
λ̂ = −γ sign(b) yr(3) + ᾱ3 z3 (35)
the symbol ˜ defines the estimation error, ie. θ̃ = θ − θ̂.
Error system: The design procedure (18-35) results
Coordinate Transformation in the following error system:

ż1 = −L1 z1 + z2 (36)

z1 = x1 − yr (18)
ż2 = −L2 z2 − z1 + z3 + ω2T θ̃ (37)
z2 = x2 − yr(1) − α1 (19)  
z3 = x3 − yr(2) − α2 (20) ż3 = −L3 z3 − z2 + ω3T θ̃ − b ᾱ3 + yr(3) λ̃ (38)

A Lyapunov function for this system is:

Regressor
1 T 1 |b| 2
V = z z + θ̃T Γ−1 θ̃ + λ̃ (39)
ω1 = 0 (21) 2 2 2γ
ω2 = φ2 (22) Its derivative along the solution of eqs. (34-35) and
∂α2 (36-38) is:
ω3 = φ3 − φ2 (23) 5
∂x2 X
V̇ = − Lk zk2 (40)
k=1

Tuning functions for θ̂: which proves from the Lasalle-Yoshizawa theorem that
global asymptotic tracking is achieved. The calcula-
τ1 = 0 (24) tions for the error system and the Lyapunov deriva-
tion are not shown in this paper, but are similar to the
τ2 = ω2 z 2 (25)
more complicated calculations for the controller includ-
τ3 = τ2 + ω3 z 3 (26) ing valve dynamics in Appendices A and B.

38
 Choux and Hovland, ”Adaptive Backstepping Controller”

4. Backstepping with Valve Tuning functions for b̂:

Dynamics π3 = z4 z3 (61)
(3)
In this subsection valve dynamics is included and it is ∂α3 λ̂ yr
assumed that all the states, i.e. position and velocity π4 = π3 − f (x3 ) x4 z4 + x4 z4 (62)
∂x3 2 f (x3 )2
of the mass, load pressure, position and velocity of the (4)
valve spool, are measured. The system rewritten as: ∂α4 λ̂ yr
π5 = π4 − f (x3 ) x4 z5 + x4 z5 (63)
∂x3 2 f (x3 )2
ẋ1 = x2 (41)
T
ẋ2 = x3 + ϕ2 (x1 , x2 ) θ (42) Stabilizing functions:
T
ẋ3 = b f (x3 )x4 + ϕ3 (x2 , x3 ) θ (43)
α1 (x1 , yr ) = ᾱ1 (64)
ẋ4 = x5 (44)
α2 (x1 , x2 , θ̂, yr , ẏr ) = ᾱ2 (65)
ẋ5 = u + ϕ5 (x4 , x5 )T θ (45)
λ̂
α3 (x̄3 , θ̂, ȳr(2) , λ̂) = ᾱ3 (66)
is in strict-feedback form but contains an unknown vir- f (x3 )
tual control coefficient bf (x3 ) which is not constant. α4 (x̄4 , θ̂, ȳr(3) , b̂, λ̂) = ᾱ4 (67)
A new extension of the tuning function design from
Krstić et al. (1995) is developed below in the special α5 (x̄5 , θ̂, ȳr(4) , b̂, λ̂) = ᾱ5 (68)
√
case where f (x) = p − x:

Coordinate Transformation
ᾱ1 = −L1 z1 (69)
z1 = x1 − yr (46) ᾱ = −z − L z − ω T θ̂ + ∂α1 x
2 1 2 2 2 2
∂x1
z2 = x2 − yr(1) − α1 (47)
∂α1
z3 = x3 − yr(2) − α2 (48) + ẏr (70)
∂yr
λ̂ ∂α2
z4 = x4 − y (3) − α3 (49) ᾱ3 = −z2 − L3 z3 − ω3T θ̂ + Γ τ3
f (x3 ) r ∂ θ̂
2
!
λ̂ X ∂α2 ∂α2
z5 = x5 − y (4) − α4 (50) + xk+1 + (k−1) yr (k)
(71)
f (x3 ) r ∂xk ∂yr
k=1
∂α3
Regressor ᾱ4 = −b̂ f (x3 ) z3 − L4 z4 − ω4T θ̂ + Γ τ4
∂ θ̂
2 3
ω1 = 0 (51) X ∂α3 ∂α3 X ∂α3
+ xk+1 + f (x3 ) b̂ x4 + y (k)
(k−1) r
ω2 = φ2 (52) ∂xk ∂x3 ∂yr
k=1 k=1
∂α2
! 3
(3)
ω3 = φ3 − φ2 (53) yr ∂α3 ˙ X ∂αk−1
∂x2 + + λ̂ + Γ ω4 zk
f (x3 ) ∂ λ̂ ∂ θ̂
(3) k=2
∂α3 ∂α3 λ̂ yr (3)
ω4 = − φ2 − φ3 + φ3 (54) λ̂ yr b̂
∂x2 ∂x3 2 f (x3 )3 − x4 (72)
(4) 2 f (x3 )2
∂α4 ∂α4 λ̂ yr
ω5 = φ5 − φ2 − φ3 + φ3 (55) ∂α4
∂x2 ∂x3 2 f (x3 )3 ᾱ5 = −z4 − L5 z5 − ω5T θ̂ + Γ τ5
∂ θ̂
4 4
X ∂α4 ∂α4 X ∂α4
Tuning functions for θ̂: + xk+1 + f (x3 ) b̂ x4 + y (k)
(k−1) r
∂xk ∂x3 ∂yr
k=1 k=1
k6=3
τ1 = 0 (56) !
(4) 4
τ2 = ω2 z 2 (57) yr ∂α4 ˙ X ∂αk−1
+ + λ̂ + Γ ω5 zk
τ3 = τ2 + ω3 z3 (58) f (x3 ) ∂ λ̂ ∂ θ̂
k=2
τ4 = τ3 + ω4 z4 (59) (4)
∂α4 λ̂ yr b̂
+ γ Π5 − x4 (73)
τ5 = τ4 + ω5 z5 (60) ∂ b̂ 2 f (x3 )2

39
 Modeling, Identification and Control

Adaptive control law: Cost Original Optimized

α1 1⊗ 2⊕ 1⊗ 2⊕
λ̂ α2 6⊗ 9⊕ 6⊗ 9⊕
u = α5 + y (5) (74)
f (x3 ) r α3 43⊗ 67⊕ 27⊗ 30⊕ 7 .
α4 707⊗ 1085⊕ 128⊗ 122⊕ 53 .
Parameter update laws: α5 29591⊗ 44486⊕ 699⊗ 513 ⊕ 222 .

˙
θ̂ = Γ τ5 (75) Table 3: Cost of calculation in number of multiplication
(⊗), number of additions (⊕) and number of
˙
b̂ = γ π5 (76) assignments (.) for each stabilizing function
˙ αi . Last column is the cost when the calcula-
 
λ̂ = −γ sign(b) yr(3) + ᾱ3 z3 (77)
tions are optimized.

Error system: The design procedure (46-77) results

in the following error system (See Appendix A): Table 3 shows the cost of calculations for each stabi-
lizing function at each design step. The computation
of the final control law is optimized in order to reduce
ż1 = −L1 z1 + z2 (78) the cost of calculation and make real time application
ż2 = −L2 z2 − z1 + z3 + ω2T θ̃ (79) possible. Table 3 shows that significant reduction in
calculation time is possible by optimizing the code.
ż3 = −L3 z3 − z2 + b̂ f (x3 )z4 + σ34 z4 + σ35 z5
 
+ ω3T θ̃ − b ᾱ3 + yr(3) λ̃ + b̃ f (x3 )z4 (80)
5. Simulations

ż4 = −L4 z4 − σ34 z3 − b̂ f (x3 ) z3 + z5 + σ55 z5 5.1. Tracking Performance of

(3)
! Backstepping Controllers
λ̂ yr ∂α3
+ ω4T θ̃ + b̃ x 4 − f (x 3 ) x 4 (81)
2 f (x3 )2 ∂x3 In order to test the robustness of the controller, two
models of the plant are implemented. The first one,
described in section 3, is used to design the controller,
ż5 = −L5 z5 − σ35 z3 − σ45 z4 − z4 whereas a second model, more realistic is used to rep-
(4)
! resent the physical system. In this new model the dy-
λ̂ yr ∂α4 namics of the valve is represented by a second order
+ ω5T θ̃ + b̃ x 4 − f (x 3 ) x 4 (82)
2 f (x3 )2 ∂x3 transfer function, the friction in the cylinder is non-
linear and Stribeck and Coulomb effects are modeled.
where σik is defined as Moreover the compressibility of the fluid is not ne-
glected inside the load and thus can the cylinder accu-
∂α2 mulate fluid. Finally the uncertain parameters of the
σ34 = − Γ ω4 (83) new model differ from the ones used in the controller
∂ θ̂ design by up to +/- 20%. The simulation results are
∂α2
σ35 = − Γ ω5 (84) given in Fig. 5-10. Fig. 5 and 8 show the tracking for
∂ θ̂ sinusoidal and step references. Fig. 6 and 9 shows the
∂α3
σ45 = − Γ ω5 (85) tracking error, and Fig. 7 and 10 show the actuator
∂ θ̂ (valve) input. For the Figs. 5-7 the controller gains
equal [L1 , L2 , L3 ] = [180, 180, 180], while for Figs. 8-
A Lyapunov function for this system is:
10 the controller gains equal [L1 , L2 , L3 , L4 , L5 ] =
1 1 1 2 |b| 2 [180, 180, 180, 350, 350]. The reference position and the
V = z T z + θ̃T Γ−1 θ̃ + b̃ + λ̃ (86) tracking position are shown with dashed and plain
2 2 2γ 2γ
lines, respectively. The model used to develop both
Its derivative along the solutions of (78-85) is (See Ap- the backstepping controllers contain the following un-
pendix B): certainties: M ∗ = 0.9 M , A∗ = 1.1 A, k ∗ = 0.8 k,
X5 d∗ = 0.8 d, Cd∗ = 0.9 Cd , w∗ = 1.1 w, Vt∗ = 0.8 Vt ,
2
V̇ = − Lk zk (87) β ∗ = 0.8 β, ρ∗ = 0.9 ρ, p∗ = 0.9 p. The ∗-superscript
k=1 refers to the model used by the controller.

40
 Choux and Hovland, ”Adaptive Backstepping Controller”

Position Position
0.08 0.06

0.05
0.06

0.04

0.04
0.03
y (m)

y (m)
0.02 0.02

0.01
0

−0.02
−0.01

−0.04 −0.02
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
time (s) time (s)

Figure 5: Position tracking with the controller of sec- Figure 8: Position tracking with the controller of sec-
tion 3. tion 4.

Error −3
x 10 Error
0.025 5

0.02

0.015

0.01
0
0.005
y (m)

y (m)

−0.005
−5
−0.01

−0.015

−0.02

−0.025 −10
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
time (s) time (s)

Figure 6: Tracking error for Fig. 5. Figure 9: Tracking error for Fig. 8.

−3
x 10 Valve opening −3
x 10 Valve opening
6 4

5
3
4
2
3
1
2
xv (m)

xv (m)

1 0

0
−1
−1
−2
−2
−3
−3

−4 −4
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
time (s) time (s)

Figure 7: Input (valve opening) with the controller of Figure 10: Input (valve opening) with the controller of
section 3. section 4.

41
 Modeling, Identification and Control

5.2. Comparison with PI Controller

−3 Absolute Positioning Accuracy (APA)
x 10
1.2
In Bonchis et al. (2002) the following comparison crite- R = sin, ts = 0
ria were defined: Mean Positioning Accuracy (MPA), R = sin, ts = tss
R = ptp, ts = 0
Absolute Positioning Accuracy (APA), Weighted Po- 1
R = ptp, ts = tss
sition Accuracy (WPA), Saturation Index (SAT), Ro-
bustness Index (RI) and Composite Index (CI). A PD 0.8
controller was compared with the other controllers on

APA (m)
all these criteria for the transient response (ts = 0) and 0.6
the steady-state performance (ts = 10), as well as for a
sinusoidal response and for a point-to-point response.
0.4
For the RI Bonchis et al. (2002) used a 50% reduction
in supply pressure.
0.2
In this paper, the criteria APA, MPA and WPA are
used to compare a PI controller with two backstepping
controllers BS1 and BS2 of sections 3 and 4, respec- 0
BS1 BS2 PI
tively. A PI controller is used instead of a PD, because
the spring in Fig. 1 makes the open-loop integrator dis- Figure 11: Comparison of Absolute Positioning
appear. The hydraulic system considered in Bonchis Accuracy.
et al. (2002) contained an open-loop integrator. More- Mean Positioning Accuracy (MPA)
0.35
over, the following three criteria are not considered in R = sin, ts = 0
this paper: SAT, RI and CI. Input saturation for the R = sin, ts = tss
0.3 R = ptp, ts = 0
system in Fig. 1 occurs when the valve opening reaches R = ptp, ts = tss
5mm. This saturation only occurs for the controller in 0.25
section 3. The criterion RI is not suited to benchmark-
ing when the nominal error is close to zero, which is 0.2
MPA (m)

the case in this paper. The CI makes use of the RI,

and hence is also not suited in our case. Nevertheless, 0.15
the robustness of the adaptive backstepping controller
can be seen for example in Fig. 8, where a 20% initial 0.1
error in model parameters are introduced.
Similar to the presentation in Bonchis et al. (2002), 0.05

Figs. 11-13 contain 4 bars, representing a) sinu-

soidal reference (entire response), b) sinusoidal refer- 0
BS1 BS2 PI
ence (steady-state response), c) point-to-point refer-
ence (entire response), d) point-to-point (steady-state). Figure 12: Comparison of Mean Positioning Accuracy.
Fig. 11 shows that both backstepping controllers BS1 Weighted Positioning Accuracy (WPA)
0.35
and BS2 perform better than the PI controller for the R = sin, ts = 0
APA criterion, where the BS2 controller performs sig- R = sin, ts = tss
0.3 R = ptp, ts = 0
nificantly better. For both the MPA and the WPA, the R = ptp, ts = tss
BS1 and the PI controllers give similar performance, 0.25
while the BS2 performs significantly better as seen in
Figs. 12-13. 0.2
WPA (m)

0.15
6. Conclusions
0.1
In this paper an adaptive backstepping controller con-
sidering valve dynamics for a nonlinear hydraulic- 0.05

mechanical system has been developed and the perfor-

mance has been compared with three different criteria 0
BS1 BS2 PI
to a simple PI controller. All three criteria show that
Figure 13: Comparison of Weighted Positioning
the adaptive backstepping controller taking valve dy-
Accuracy.
namics into account performs significantly better than
both the PI controller and a reduced version of the

42
 Choux and Hovland, ”Adaptive Backstepping Controller”

backstepping controller without taking valve dynamics ż3 = ẋ3 − yr(3) − α̇2
into account.
= b f (x3 )x4 + ϕT3 θ̂ + ϕT3 θ̃ − yr(3) − α̇2
Future research directions will focus on developing an !
output-feedback version of the backstepping controller λ̂
= b f (x3 ) z4 + y + α3 + ϕT3 θ̂ + ϕT3 θ̃
(3)
and the implementation of this controller on the exper- f (x3 ) r
imental setup shown in Fig. 2.
− yr(3) − α̇2
!
λ̂
= b f (x3 ) z4 + ᾱ3 − b λ̃ yr(3) + ϕT3 θ̂ + ϕT3 θ̃
References f (x3 )
− α̇2
Bonchis, A., Corke, P., and Rye, D. Experimen-
tal Evaluation of Position Control Methods for ∂α2
= b f (x3 )z4 − z2 − L3 z3 − ω3T θ̂ + Γ τ3
Hydraulic Systems. IEEE Transactions on Con- ! ∂ θ̂
2
trol Systems Technology, 2002. 10(6):876–882. X ∂α2 ∂α2
doi:10.1109/TCST.2002.804128. + xk+1 + (k−1) yr(k) − b λ̃ ᾱ3
∂xk ∂yr
k=1

Krstić, M., Kanellakopoulos, I., and Kokotović, P. − b λ̃ yr(3) + ϕT3 θ̂ + ϕT3 θ̃ − α̇2
Nonlinear and Adaptive Control Design. Wiley, New ∂α2 T
York, 1995. = −L3 z3 − z2 + b̂ f (x3 )z4 + φ θ̂
∂x2 2
  ∂α2 T
Zeng, H. and Sepehri, N. Adaptive backstepping con- − b ᾱ3 + yr(3) λ̃ + ω3T θ̃ − φ θ̂ + b̃ f (x3 )z4
∂x2 2
trol of hydraulic manipulators with friction compen- ∂α2
sation using LuGre model. In Proc. American Con- + Γ (τ3 − τ5 )
trol Conference. pages 3164–3169, 2006. ∂ θ̂
= −L3 z3 − z2 + b̂ f (x3 )z4 + σ34 z4 + σ35 z5
 
Zeng, H. and Sepehri, N. Tracking Control of Hydraulic + ω3T θ̃ − b ᾱ3 + yr(3) λ̃ + b̃ f (x3 )z4 (90)
Actuators Using a LuGre Friction Model Compen-
sation. Journal of Dynamic Systems, Measure-
ment, and Control, 2008. 130(1):0145021–0145027.
doi:10.1115/1.2807181.

!
d λ̂
A. Calculations for the Error ż4 = ẋ4 − y (3) − α̇3
dt f (x3 ) r
System Including Valve ˙  
λ̂ d 1 λ̂
Dynamics = x5 −
f (x3 )
yr(3) − λ̂
dt f (x3 )
yr(3) − y (4)
f (x3 ) r
− α̇3
ż1 = ẋ1 − yr(1) λ̂ ∂α3
= z5 + yr(4) − b̂ f (x3 ) z3 − L4 z4 − ω4T θ̂ + Γ τ4
= x2 − yr(1) f (x3 ) ∂ θ̂
2 3
= z2 + α 1 X ∂α3 ∂α3 X ∂α3
+ xk+1 + f (x3 ) b̂ x4 + y (k)
(k−1) r
= −L1 z1 + z2 (88) ∂xk ∂x3 ∂yr
k=1 k=1
ż2 = ẋ2 − yr(2) − α̇1
! 3
(3)
yr ∂α3 ˙ X ∂αk−1
+ + λ̂ + Γ ω4 zk
= x3 + ϕ2 (x1 , x2 )T θ̂ + ϕ2 (x1 , x2 )T θ̃ − yr(2) − α̇1 f (x3 ) ∂ λ̂ ∂ θ̂
k=2
= z3 + ϕ2 (x1 , x2 )T θ̂ + ϕ2 (x1 , x2 )T θ̃ − α̇1 + α2 (3)
λ̂ yr b̂
T T − x4
= z3 + ϕ2 (x1 , x2 ) θ̂ + ϕ2 (x1 , x2 ) θ̃ − α̇1 2 f (x3 )2
∂α1 ∂α1 ˙  
− z1 − L2 z2 − ω2T θ̂ + x2 + ẏr λ̂ (3) d 1 λ̂
∂x1 ∂yr − yr − λ̂ yr(3) − y (4)
f (x3 ) dt f (x3 ) f (x3 ) r
= −L2 z2 − z1 + z3 + ω2T θ̃ (89) − α̇3

43
 Modeling, Identification and Control

∂α3
!
= z5 − b̂ z3 − L4 z4 − ω4T θ̂ + Γ τ4 b φT3 θ̂ φT3 θ̃
∂ θ̂ − λ̂ − 2
x4 − 3
− yr(4)
2 f (x3 ) 2 f (x3 ) 2 f (x3 )3
2 3
X ∂α3 ∂α3 X ∂α3
− α̇4
+ xk+1 + f (x3 ) b̂ x4 + y (k)
(k−1) r
∂xk ∂x3 ∂yr
k=1 k=1 = −L5 z5 − σ35 z3 − σ45 z4 − z4
3 (3) !
∂α3 ˙ X ∂αk−1 b̂ λ̂ yr (4)
λ̂ yr ∂α4
+ λ̂ + Γ ω4 zk − x
2 4 + ω5T θ̃ + b̃ x4 − f (x3 ) x4 (92)
∂ λ̂ ∂ θ̂ 2 f (x 3 ) 2 f (x3 )2 ∂x3
k=2
!
b φT3 θ̂ φT3 θ̃
− λ̂ − x4 − − yr(3) − α̇3
2 f (x3 )2 2 f (x3 )3 2 f (x3 )3 B. Lyapunov Derivative Including
= z5 − b̂ z3 − L4 z4 − ω4T θ̂ +
∂α3
Γ (τ4 − τ5 ) Valve Dynamics
∂ θ̂
3
∂α3 X ∂αk−1  
− f (x3 ) b̃ x4 + Γ ω4 zk V̇ = z1 (−L1 z1 + z2 ) + z2 −L2 z2 − z1 + z3 + ω2T θ̃
∂x3 ∂ θ̂
k=2 
! + z3 −L3 z3 − z2 + b̂ f (x3 )z4 + σ34 z4 + σ35 z5
φT3 θ̂ φT3 θ̃ ∂α3 T
− λ̂ − − yr(3) − φ θ̂
∂x3 3
  
2 f (x3 ) 3 2 f (x3 ) 3
+ω3T θ̃ − b ᾱ3 + yr(3) λ̃ + b̃ f (x3 )z4
(3) 
∂α3 T ∂α3 ∂α3 λ̂ yr + z4 −L4 z4 − σ34 z3 − b̂ f (x3 ) z3 + z5 + σ45 z5
− φ θ̃ − φ2 θ̂ − φ2 θ̃ + b̃ x4
∂x3 3 ∂x2 ∂x2 2 f (x3 )2 !!
(3)
= −L4 z4 − σ34 z3 − b̂ z3 + z5 + σ45 z5 λ̂ yr ∂α 3
+ω4T θ̃ + b̃ x4 − f (x3 ) x4
(3)
! 2 f (x3 )2 ∂x3
λ̂ yr ∂α3
+ ω4T θ̃ + b̃ x4 − f (x3 ) x4 (91) + z5 (−L5 z5 − σ35 z3 − σ45 z4 − z4
2 f (x3 )2 ∂x3 !!
(4)
λ̂ yr ∂α4
+ω5T θ̃ + b̃ 2
x4 − f (x3 ) x4
2 f (x3 ) ∂x3
!
d λ̂
y (4) ˙ b̃ ˙ |b| ˙
ż5 = ẋ5 − − α̇4 + θ̃T Γ−1 θ̃ + b̃ + λ̃λ̃
dt f (x3 ) r γ γ
5 (3)
˙   X λ̂ yr ∂α3
λ̂ d 1 =− Lk zk2 + x4 z4 − f (x3 ) x4 z4
=u+ ϕT5 θ̂ + ϕT5 θ̃ − (4)
y − λ̂ yr(4) 2 f (x3 )2 ∂x3
f (x3 ) r dt f (x3 ) k=1
˙
!
λ̂ (4)
− y (5) − α̇4 λ̂ yr ∂α4 b̂
f (x3 ) r + x 4 z5 − f (x 3 ) x 4 z5 + f (x 3 )z3 z4 − b̃
2 f (x3 )2 ∂x3 γ
˙ T
λ̂ λ̂ 
˙
= α5 + yr(5) + ϕT5 θ̂ + ϕT5 θ̃ − y (4) + z2 ω2 + z3 ω3 + z4 ω4 + z5 ω5 − Γ−1 θ̂ θ̃
f (x3 ) f (x3 ) r   
|b| ˙
! 
b φT3 θ̂ φT3 θ̃ − b ᾱ3 + yr (3)
z3 − λ̂ λ̃
− λ̂ − x4 − − yr(4) γ
2 f (x3 )2 2 f (x3 )3 2 f (x3 )3
5
X
λ̂ =− Lk zk2 (93)
− y (5) − α̇4
f (x3 ) r k=1

∂α4
= −z4 − L5 z5 − ω5T θ̂ + Γ τ5
∂ θ̂
4 4
X ∂α4 ∂α4 X ∂α4
+ xk+1 + f (x3 ) b̂ x4 + y (k)
(k−1) r
∂xk ∂x3 ∂yr
k=1 k=1
k6=3
! 4
(4)
yr ∂α4 ˙ X ∂αk−1
+ + λ̂ + Γ ω5 zk
f (x3 ) ∂ λ̂ ∂ θ̂
k=2
(4) ˙
∂α4 λ̂ yr b̂ T T λ̂
+ γ Π5 − x 4 + ϕ5 θ̂ + ϕ5 θ̃ − y (4)
∂ b̂ 2 f (x3 )2 f (x3 ) r

44 View publication stats