Professional Documents
Culture Documents
Abstract
In this paper, a robust velocity control problem for hydraulic elevators is investigated. The analysis is divided into two parts,
mechanical and hydraulic. A detailed mathematical model for the mechanics is established for the purpose of simulation, but the
control system design is carried out with a simplified model reduced from the detailed one. The three important characteristics of a
hydraulic elevator, including cylinder friction, pump friction, and pump leakage, are modeled through experiments. The leakage
property is characterized as a function of temperature and pressure. A two-stage nonlinear robust controller using the Lyapunov
redesign method is established for velocity tracking control. At the first stage, a robust controller for the mechanical part is designed
to yield the desired cylinder pressure for reference velocity tracking. At the second stage, a robust controller for the hydraulic part is
designed to track the reference pressure generated from the first controller. Simulation results validate that the proposed method is
robust in the presence of nonlinearities and uncertainties.
r 2004 Elsevier Ltd. All rights reserved.
Keywords: Hydraulic elevator; Control system design; Lyapunov redesign; Friction; Leakage; Stability; Robustness
0967-0661/$ - see front matter r 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.conengprac.2004.09.003
ARTICLE IN PRESS
790 C.-S. Kim et al. / Control Engineering Practice 13 (2005) 789–803
the velocity tracking control step, and the destination Kokotovic, 1995), and the Lyapunov redesign method
position control step. The load pressure compensation (Kim, Lee, & Cho, 2002; Qu, 1998). Also, numerous
step reduces the jolt at the departure of the car that may applications of the robust control methods for nonlinear
produce an uncomfortable ride for passengers. The systems with uncertainty have been reported (Alleyne,
velocity tracking control makes the car follow a reference 1996; Sohl & Bobrow, 1999; Yanada & Shimahara,
velocity profile that helps the passengers feel comfortable 1997; Yao, Fanping, Reedy, & Chiu, 2000; Corless &
during the movement. The destination position control Leitmann, 1997; Venkatesh, Cho, & Kim, 2002;
reduces the position error, which may exist after the Michino, Mizumoto, Iwai, & Kumon, 2003; Kwon,
velocity tracking control, at the target position. Han, & Ahn, 2004). Alleyne (1996) proposed a variant
A hydraulic pump connected directly to an electric of the backstepping method and improved the force
motor generates fluid power and controls the velocity of tracking performance of an electro-hydraulic actuator.
the car. A smooth transition of car depends on the To compensate for the parameter uncertainty of a
smooth movement of the cylinder. However, the nonlinear cement mill model, Grognard, Jadot, Magni,
following problems occur: (1) The system characteristics Bastin, Sepulehre, and Wertz (2001) designeda Lyapu-
change due to the variations of the load and oil nov-function-based controller. Yao et al. (2000) sug-
temperature; (2) jolts occur due to unequal pressures gested a discontinuous projection-based adaptive robust
at the departure and arrival points; (3) an undesirable controller for a single-rod hydraulic cylinder with
descent occurs due to oil leakage; and (4) unknown constant unknown inertia load. Sha et al. (2002) used
nonlinearities in the friction and continuity equations a discrete adaptive sliding mode control for a hydraulic
appear. These problems will eventually degrade the elevator velocity tracking system.
safety and comfort of the passengers (Nakamura, 2000). To improve the efficiency of the elevator system, we
In this paper, the velocity control problem of propose a two-stage nonlinear robust controller, using
hydraulic elevators is investigated (Jianmin, 2000; Kang the Lyapunov redesign method, for the velocity control
& Kim, 2000; Li, 2001; Sha, Bajic, & Yang, 2002; of the hydraulic elevator system. At the first stage, a
Teramoto & Nakamura, 1997). Sha et al. (2002) has robust controller for the mechanics is synthesized to
introduced a new approximated linear model for a control the velocity of the car. The control input to the
hydraulic elevator and investigated a sliding mode mechanics is used as a reference to the subsequent
control for velocity tracking in the discrete domain. pressure tracking control. At the second stage, another
Teramoto and Nakamura (1997) has introduced a third- robust controller for the hydraulics is designed for the
order linearized model with fixed parameters, for which purpose of tracking the reference pressure generated from
their open-loop experiments showed that the effects of the first controller. The proposed method showed good
nonlinearities and uncertainties were too large to control performance in the presence of uncertainties.
describe the system behavior sufficiently. The paper is structured as follows. In Section 2, the
To solve these problems, the hydraulic elevator system was derived mathematical models of the mechanical and
divided into two facets: a mechanical facet that includes the hydraulic facets of the hydraulic elevator system con-
car, rope, and pulleys, and a hydraulic facet that includes the sidered are presented. In Section 3, a cylinder friction
cylinder, logic valve, hydraulic power unit, and pipes. Then, model, a hydraulic pump friction model, and the leakage
the dynamic equations for each facet are derived. The characteristics of the pump, constructed through a series
mathematical model of the mechanics is a 14th-order linear of experiments, are explained. The controller design,
equation including all the pulleys and rope dynamics, whereas using a two-stage Lyapunov redesign method, is pre-
the mathematical model of the hydraulics is expressed as a set sented in Section 4. In Section 5, the simulation results of
of nonlinear equations with uncertainties, that is, flow the velocity control using the proposed models are
equations and continuity equations. Kim (2000) revealed that shown. Conclusions are given in Section 6.
the open-loop simulation results of the mathematical models
reported in the literature were inconsistent with open-loop
experimental results. Thus, a series of experiments have been 2. System modeling: dynamics
performed to obtain more accurate mathematical models for
the frictions in the hydraulic cylinder and pump (Armstrong- The equations of motion of the mechanics and
Helouvry, Dupont, & Wit, 1994; Bo & Pavelescu, 1982; hydraulics presented in this section are, respectively,
Karnopp, 1985). In addition, the leakage coefficient has been derived from classical Newtonian dynamics, the con-
modeled as a function of temperature and pressure in the tinuity equation, and the flow equation.
hydraulic equation.
Recently, many nonlinear robust control methods 2.1. Mechanics
have been proposed: sliding mode control (Bouri &
Thomasset, 2001; Wheeler, Su, & Stepanenko, 1998), The mechanics consist of a passenger car (cabin) in
the back-stepping method (Krstic, Kanellakopulos, & which the passengers ride, a hydraulic cylinder, ropes,
ARTICLE IN PRESS
C.-S. Kim et al. / Control Engineering Practice 13 (2005) 789–803 791
2.2. Hydraulics
Fig. 3 shows the hydraulic circuit of the elevator Fig. 1. Schematic of a hydraulic elevator and a detailed model of the
system considered, in which a logic valve, that is, the mechanics: Each rope segment is modeled as a spring–damper system.
main check valve (MCV), is used for the safety of the (a) Schematic of a hydraulic elevator. (b) A detailed model of the
system. Fig. 4 is a schematic of the MCV. Two solenoid mechanics.
valves associated with the MCV determine the up and
down movements of the car by switching the oil-flow the pump by increasing the input current to the motor,
direction. In Fig. 5, the control logic and operation the spool of the logic valve opens. At this moment, the
principle of this module are depicted. When both oil flows into the cylinder and the cylinder pressure
solenoid valves 1 and 2 are off, the hydraulic circuit increases. The increased cylinder pressure pushes up the
performs an up-movement. If we increase the flow from hydraulic jack and lifts up the car. If both solenoid
ARTICLE IN PRESS
792 C.-S. Kim et al. / Control Engineering Practice 13 (2005) 789–803
Table 1
Physical parameters of mechanics
A B C D
Pump velocity
Hydraulic jack Solenoid Solenoid
valve 1 valve 2 open
close Logic valve
on
off Solenoid valve 1
on
off Solenoid valve 2
Fig. 5. Control logic: A and B are the load pressure compensation step
and the velocity control step for an upward movement, respectively,
Pj while C and D are those for a downward movement.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Let iq be the input current to the pump; let T m be the 2jPp Pj j
output torque of the pump; let yp be the angular Qv ¼ C d Aðxv ÞsgnðPp Pj Þ ; (10)
r
displacement of the pump; let Pp be the pump pressure;
let Pj be the cylinder pressure; let Qp be the flow in the where
pump; let Qv be the flow in the logic valve; let xv be the
displacement of the logic valve spool; and let Aðxv Þ be 763 10 6
Aðxv Þ ¼ xv :
the opening area of the MCV that is proportional to the 9:2 10 3
spool displacement. The parameter values of the
Aðxv Þ is the opening area through which the oil flows
hydraulics are displayed in Table 2.
from area A to area B shown in Fig. 4. This value is
The relationship between the input current and the
normally obtained from the data-sheet of the logic valve
pump torque is given as
provided by the manufacturer. Because the response
T m ¼ K m iq : (4) time of the logic valve is very fast compared with other
dynamics, its dynamics is ignored in this context.
Then, the equation of motion of the pump is
pffiffiffi
J mp y€ p þ C p ny_ p þ Dp Pp þ T f ¼ T m ; (5)
3. Friction and leakage models: experiment
where T f is the pump friction which will be estimated by
experiments (see Section 3.2). The flow equation and the
Fig. 6 shows a downward movement open-loop
continuity equation for the pump are given as
experiment, which depicts the relationship between the
Cl input current, cylinder pressure, and car velocity. As
Qp ¼ Dp y_ p pffiffiffi Pp (6)
n seen in Fig. 6, even if we decrease the input current by
5 A=s up to 5 s, the cylinder pressure and car velocity
and do not change. Specifically, when the input current
b drops to 34 A, the cylinder pressure begins to change
P_ p ¼ ðQ Qn Þ: (7) after 5 s, and the car begins to move down after 8.2 s.
Vp p
Hence, the region in which the cylinder pressure is not
The continuity equation between the logic valve and the affected by the change of current input is called the first
cylinder is given as dead zone, and the following region in which the car
velocity is not affected by the change of cylinder
b
P_ j ¼ ðQ Aj x_ j Þ: (8) pressure is called the secondary dead zone. In Fig. 6,
V j0 þ Aj xj n the first dead zone is from 2.4 to 5 s and the secondary
For the logic valve, the dynamic equation and the flow one is from 5 to 8.2 s.
equation are The first dead zone is known to occur due to pump
friction and the leakage characteristics. And the
M v x€ v ¼ Pj Aa þ Pp Ab Pps Ap C v x_ v K v xv ; (9) secondary dead zone is due to cylinder friction. These
Table 2
Physical parameters of hydraulics
54 45 30 Ff
The first dead zone The second dead zone
48 40 20
Fs
Cylinder pressure [bar]
10
Cylinder pressure Fc
36 30
0 V
30 Car velocity V0
25
-10
24 20
Fig. 7. Bo and Pavelescu’s friction model (Bo & Pavelescu, 1982).
Input current -20
18 15
12 10 -30
0 2 4 6 8 10
Time [s]
2000
Bo and Pavelescu compared several friction models in
0
the literature and proposed an exponential model of the
-2000
form
-4000
F f ¼ F c þ ðF s F c Þe av þ F v ðvÞ; (11)
-6000
where F c is the minimal level of Coulomb friction, F s is -8000
the level of static friction, F v is the viscous friction, v is -10000
friction + fluctuation
Table 3
Parameters in the cylinder friction model (13)
Leakage coefficient
High temperature fitting
pressure reaches the fixed relief valve pressure. Then, the 2.0x10-9
pump speed reaches a steady state, as y€ p ¼ 0: At this
time, if we close the relief valve, Qp ¼ 0 in (6) is 1.5x10-9
achieved. Physically, in this state, there is no output flow
of the pump. The pump is working only to compensate 1.0x10-9
the leak flow and maintain the relief valve pressure in
the pipe that connects the pump and valve block. 5.0x10-10
0 10 20 30 40 50
Therefore, the following equation holds: Pressure [bar]
Cl
Dp y_ p ¼ pffiffiffi Pp : (16) Fig. 14. Pump leakage experimental data and models.
n
By measuring the pump pressure, pump speed, and oil
the leakage function for high-temperature oil and
temperature, we established C l as a function of pressure
C low ðPp Þ is the leakage function for low-temperature
and temperature. The tests were repeated for various
oil. Since the logarithm of zero is undefined, an offset of
conditions of relief valve pressure and oil temperature.
0.0001 is applied to Pp in this context.
Fig. 14 shows the experimental data sampled from the
All of the friction and leakage models developed here
leakage tests, and, accordingly, the following leakage
were verified by comparing them with the experimental
model is proposed in logarithmic form as a function of
results under the same input and load conditions.
temperature and pressure:
Fig. 15(a) and (b) demonstrate the similar behavior of
C high ðPp Þ C low ðPp Þ the 4 states between the actual experimental results and
Cl ¼ ; (17)
40 15 the model-based simulation results using the same input
where C high ðPp Þ ¼ 0:1577e 8 log10 Pp þ 0:3373e 8 ; C low current (5 A/s) for an upward movement. The impulse
ðPp Þ ¼ 0:0687e 8 log10 Pp þ 0:1806e 8 ; and C high ðPp Þ is response of the mechanical facet, displayed in Fig. 15(c)
and (d), confirm the validity of the simulation model.
Power unit
Oil reservoir
4. Controller design
Motor
Fig. 15. Comparison between experimental data and simulation results: In (a) and (b), pump speed, pump pressure, cylinder pressure, and car
velocity can represent the validity of the motor–pump model, the pump-cylinder equations including pump friction and leak flow, the cylinder and
valve model, and the composite dynamics of the cylinder and the mechanics, respectively. In (c) and (d), the validity of the impulse response of the
mechanics is displayed for both the time domain and the frequency domain. (a) Pump speed and car velocity: upward movement. (b) Pump pressure
and cylinder pressure: upward movement. (c) Impulse response of the mechanics in the time domain: downward movement. (d) Impulse response of
the mechanics in the frequency domain: downward movement.
robust input, the control (20) has been shown to x_ 4 ¼ a3 ð2x4 x2 Þ a4 ð2x3 x1 Þ a5 x4
guarantee the ultimate boundedness of all possible þ a6 f c ðx3 ; x4 Þ þ a7 um ; ð22Þ
system responses within an arbitrarily small neighbor-
hood of the zero state (by letting e ! 0: Thus, a x6 ¼ a8 x4 ða9 þ Da9 Þx7 a10 x8 þ a11 x10 ; (23)
continuous control of (20) assures uniform asymptotic
stability. x_ 10 ¼ a12 x10 a13 x7 a14 T f ðx9 ; x10 Þ þ b1 ðuh þ ires Þ;
We investigated a two-stage robust controller for the
(24)
mechanical and hydraulic facets of the hydraulic
elevator. Fig. 16 shows a block diagram of the control where a1 ¼ 4C eq =ðDM c þ M c Þ; a2 ¼ 4K eq =ðDM c þ M c Þ;
algorithm. For the mechanics, we generated a reference a3 ¼ 8C eq =M j ; a4 ¼ 8K eq =M j ; a5 ¼ C j =M j ;paffiffi6ffi ¼ 1=M j ;
pressure of the cylinder, Pdj ; using the car velocity a7 ¼ Aj =M j ; a8 ¼ bAj =V j ; a9 ¼ ffiffiffi l =ðV j nÞ; a10 ¼
pbC
reference, x_ dc ; and the car velocity feedback, x_ c : Then, V p =V j ; a11 ¼ bDp =V j ; a12 ¼ C p n=J mp ;a13 ¼ Dp =J mp ;
for the hydraulics, using the cylinder pressure feedback, a14 ¼ 1=J mp ; b1 ¼ K m =J mp ; f c ðx3 ; x4 Þ ¼ f j ðx3 ; x4 Þ
Pj ; and the reference pressure, Pdj ; we generated the C j x4 ; and ires is the residual input current in the pressure
input current iq of the hydraulic pump. for the hydraulic compensation stage due to the switching of the control
part. mode from a pressure compensation problem to a
Defining the state variables as ½x1 x2 x3 x4 x5 x6 velocity tracking problem. As the mass of the car, M c ;
x7 x8 x9 x10 T ¼ ½xc x_ c xj x_ j Pj P_ j Pp P_ p yp y_ p T and um ¼ x5 ; varies from 0 to 1600 kg, we assumed that the nominal
(2)–(8) can be rewritten as value is 800 kg and DM c 2 ½ 800; 800: According to the
sensitivity analysis of the nonlinear model of the
x_ 2 ¼ a1 ðx2 2x4 Þ a2 ðx1 2x3 Þ; (21) hydraulic elevator system, the sensitive parameter
ARTICLE IN PRESS
798 C.-S. Kim et al. / Control Engineering Practice 13 (2005) 789–803
xcd xc
Load pressure
Reference velocity
compensation
generation Ppd PID-controller Load pressure
compensation Pp
d Pjd xc
xc
Reference velocity Reference pressure iq Pj
Hydrauic part Mechanical part
tracking tracking
dynamics dynamics
robust controller robust controller Velocity
tracking control xc
affecting the velocity control is defined as We make (29) stable by selecting suitable gains km1 and
M c ð54%Þ4C j ð31%Þ4bð7%Þ: As the effect of the oil km2 : And the system equation with uncertainties is
bulk modulus is relatively small, we considered the oil rewritten as
bulk modulus as a constant. The sensitivity analysis of a
ð4M c þ M j Þ€em þ ðC j þ km1 Þe_m þ km2 em
nonlinear system is well explained in Khalil (1992).
¼ 2Aj ð km mm kf^m kÞ; ð30Þ
4.1. Mechanics: velocity tracking control
€
where kf^m k ¼ 2A1 j 4DM c x_ 2 M j d C _
j dþ2f c ðx3 ; x4 Þ :
max
Using a kinematic constraint such as x1 ¼ 2x3 þ d To design a robust input, let the Lyapunov function and
and parameters a1 a7 ; (21) and (22) can be rewritten pðe; tÞ be
as 1 1
V m ¼ ðem þ km Þ2 þ e_2m ; (31)
€ þ C j ðx2 dÞ
4ðDM c þ M c Þx_ 2 þ M j ðx_ 2 dÞ _ þ 2f ðx3 ; x4 Þ 2 2
c
¼ 2Aj um ; ð25Þ em þ k m
pðe; tÞ ¼ 0 2Aj rðÞ ¼ 2Aj e_m rðÞ; (32)
where d is a deformation value of the rope and um is the e_m
control input. The friction term f c ðx3 ; x4 Þ is the where rðÞ ¼ kf^m k:
uncertainty that we have to cope with. Then, the Since the stability of the control system in the case of
nominal system equation becomes kpðe; tÞk4e is well established in (Corless & Leitmann,
ð4M c þ M j Þx_ 2 þ C j x2 ¼ 2Aj um : (26) 1981; Khalil, 1992; Kim et al., 2002), we show the
stability for the case of kpðe; tÞkpe in (20). The
Let the position error be em ¼ xd1 x1 : Then, the error
derivative of the Lyapunov function (31) is given as
equation becomes
V_m ¼ ðem þ km Þ_em þ e_m e€m
ð4M c þ M j Þ€em þ C j e_m ¼ ð4M c þ M j Þx_ d2 þ C j xd2 2Aj um :
1
(27) ¼ ðem þ km Þ_em e_m fðC j þ km1 Þ_em
4M c þ M j
The control input to the mechanics of the hydraulic
þ km2 em þ 2Aj ðkf^m k þ km mm Þg
elevator system, um ; is defined as um ¼ jm þ km mm ;
where jm is a nominal control input and mm is a robust km2 C j þ km1 2
¼ 1 em e_m þ km e_m e_
control input. The nominal control input of the 4M c þ M j 4M c þ M j m
mechanical part is defined as 2Aj
ðkf^m k þ km mm Þe_m : ð33Þ
1 4M c þ M j
jm ¼ ð4M c þ M j Þx_ d2 þ C j xd2 þ km1 e_m þ km2 em ;
2Aj The substitution of mm ¼ pðe;tÞ ^
e kf m k into (33) yields
(28)
km2
where km1 is a derivative gain and km2 is a proportional V_ m ¼ 1 em e_m
4M c þ M j
gain. Using an linear variable displacement transformer
2Aj
(LVDT), we could measure the state variables x2 and x_ 2 : þ km kf^m k e_m
4M c þ M j
With the nominal control input um ¼ jm ; the error ( )
km 4Aj kf^m k
2 2
equation of the system is obtained as C j þ km1
þ e_2 : ð34Þ
ð4M c þ M j Þ€em þ ðC j þ km1 Þe_m þ km2 em ¼ 0: (29) 4M c þ M j e 4M c þ M j m
ARTICLE IN PRESS
C.-S. Kim et al. / Control Engineering Practice 13 (2005) 789–803 799
2Aj
If we choose km2 ¼ 4M c þ M j ; km ¼ 4M c þM j
kf^m k; kh1 40; kh2 ¼ 1; kh ¼ a11 b1 kf^h k; and e ¼ 0:035; then
km1 40; and e ¼ 0:02; then the derivative of V m becomes V h 40 and V_ h o0 are guaranteed. The robust control
( ) input for the hydraulics, mh ; can be obtained from (20)
km 4Aj kf^m k
2 2
_ C j þ km1 by setting pðe; tÞ ¼ a11 b1 e_h kf^h k:
Vm ¼ þ e_2 p0: (35)
4M c þ M j 4M c þ M j m
Finally, V m 40 and V_ m p0 were proved, and the control
input for the mechanics can be determined by the
5. Simulations
substitution of the computed gains into (20), (28), and
(32).
The results of computer simulations using MATLAB/
SIMULINK are shown in Figs. 17–23. According to
4.2. Hydraulics: pressure tracking control
these results, the stability and velocity tracking perfor-
mance of the proposed controller are better than those
A robust controller for the hydraulics is designed to
of a m-controller used for the linearized system. The m-
generate an input current that can track the cylinder
controller was designed and analyzed by computing the
reference pressure originating in the mechanics con-
structured singular value m introduced by Doyle (1982),
troller. Assume that
which is a measure of robust performance. This means
a13 that the best controller is the one that gives the best
uh ¼ jh þ kh mh þ x7 ;
b1 worst-case response. The worst-case is the worst
where jh is the nominal control input, mh is the robust combination of possible disturbances, set point changes
control input, and ða13 =b1 Þx7 is the state feedback and model errors. For this simulation, the m-controller
control of the hydraulics. For the convenience of
calculation, define gðx4 ; x7 ; x8 Þ ¼ a10 x8 þ a9 x7 þ a8 x4 :
Here the states x4 ; x7 ; and x8 are observable, so that 60
Reference signal , µ control
gðx4 ; x7 ; x8 Þ is a known function. The time derivative of
50 Nonlinear robust control
(23) is given as
x_ 6 ¼ a11 x_ 10 gðx
_ 4 ; x7 ; x8 Þ Da9 x_ 7 : (36) 40
Car velocity [m/min]
As we know that xd6 is the derivative of the Fig. 17. Comparison of the car velocity with m-control and nonlinear
mechanics control input, um ; we determine that the robust control: upmovement.
nominal control input for the hydraulics is jh ; which is
defined as
40 250
1 Pp
jh ¼ _ 4 ; x7 ; x8 Þ þ x_ d6 þ a11 a13 x7 þ a11 a12 x10
gðx 35
a11 b1 Pj
200
þkh1 e_h þ kh2 eh g ires ; ð39Þ 30
Input current [A]
Pressure [bar]
25
where eh ¼ xd5 x5 : By applying uh ¼ jh to (37), the 150
10
where kf^h k ¼ a111b1 kða11 a14 T f þ Da9 x7 Þkmax : And the 50
Lyapunov function for the hydraulics is defined as 5
1 1 0 0
V h ¼ ðeh þ kh Þ2 þ e_2h : (41) 0 2 4 6 8 10 12
2 2 Time [s]
We can also prove the stability of this function in the Fig. 18. The input current, pump pressure, and cylinder pressure of
same manner as shown in Section 4.1. If we select the nonlinear robust control in Fig. 17.
ARTICLE IN PRESS
800 C.-S. Kim et al. / Control Engineering Practice 13 (2005) 789–803
8 4
µ control
6
Nonlinear robust control
0
2 -4
0
-8
-2
-4 µ control
-12
Nonlinear robust control
-6
-8
-16
0 2 4 6 8 10 12
0 2 4 6 8 10 12
Time [s]
Time [s]
Fig. 19. Comparison of the velocity tracking errors: up movement. Fig. 22. Comparison of the velocity tracking errors: down movement.
20 5
Reference signal , µ control Nominal (load 800kg, nominal friction)
10 4
Nonlinear robust control Load (1600kg) + nominal friction x 1.5
Load (0kg) + nominal friction x 0.5
0 3
Car velocity [m/min]
Car velocity [m/min]
-10 2
-20 1
-30 0
-40 -1
-50 -2
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [s] Time [s]
Fig. 20. Comparison of the car velocity of m-control and nonlinear Fig. 23. Robustness in the presence of load and friction variations.
robust control: down movement.
25
Pp 80
and the magnitude of the cylinder friction. The
20 robustness for the load and friction variations is shown
15
60 in Fig. 23. Given that the load variation is 800 kg and
40
the friction variation is fnominal frictiong 0:5
10
fnominal frictiong 1:5; and that the velocity tracking
Input current
5 20 error is comparatively smaller than the nominal control
0 0
error, the robustness of the proposed control scheme is
0 2 4 6 8 10
0 12 assured.
Time [s]
Fig. 21. The input current, pump pressure, and cylinder pressure of
the nonlinear robust control in Fig. 20.
6. Conclusions
was designed according to the assumption that m is 0.9
and that the linearized system order is reduced from 34 To establish accurate simulation models and to solve
to 9 using the balanced reduction theorem (Kang & the velocity tracking control problem of a hydraulic
Kim, 2000). According to the simulation result, the elevator system, experiment-based models for the
maximum velocity tracking error was reduced to 13–18 of hydraulic elevator were investigated. The dead zone
ARTICLE IN PRESS
C.-S. Kim et al. / Control Engineering Practice 13 (2005) 789–803 801
behaviors in hydraulic elevators were clarified through modeling of a hydraulic system based on experimenta-
experiments. The mathematical models for the cylinder tion were obtained.
and pump frictions were established by combining the
Karnopp’s friction model and Bo and Pavelescu’s
friction model. The leakage at the pump was mathema-
tically formulated by experiments resulting in a non- Acknowledgements
linear function of temperature and pressure. To improve
the performance of the velocity tracking control, a two- This work was supported by the Ministry of Science
stage robust control using the Lyapunov redesign and Technology of Korea under a program of the
method was investigated. The computer simulations National Research Laboratory, Grant number NRL
showed the stability and robustness of the proposed M1-0302-00-0039-03-J00-00-023-10. The authors would
controller in the presence of nonlinearities and un- like to thank Prof. Kyo-Il Lee at Seoul National
certainties. In this way, a practical methodology for University for his thoughtful guidance throughout the
designing a two-stage robust controller as well as system studies at their alma mater.
In the detailed model of the mechanics in (1), see Fig. 1(b), the system matrix Af is defined as
Af
2 3
0 1 0 0 0 0 0 0 0 0 0 0 0 0
6 7
6 a2;1 a2;2 a2;3 a2;4 a2;5 a2;6 0 0 a2;9 a2;10 a2;11 a2;12 a2;13 a2;14 7
6 7
6 7
6 0 0 0 1 0 0 0 0 0 0 0 0 0 0 7
6 7
6 7
6 a4;1 a4;2 a4;3 a4;4 a4;5 a4;6 a4;7 a4;8 a4;9 a4;10 0 0 0 0 7
6 7
6 7
6 0 0 0 0 0 1 0 0 0 0 0 0 0 0 7
6 7
6 7
6 a6;1 a6;2 a6;3 a6;4 a6;5 a6;6 a6;7 a6;8 0 0 0 0 0 0 7
6 7
6 7
6 0 0 0 0 0 0 0 1 0 0 0 0 0 0 7
6 7
¼6 7; ðA:1Þ
6 0 0 a8;3 a8;4 a8;5 a8;6 a8;7 a8;8 0 0 0 0 0 0 7
6 7
6 7
6 0 0 0 0 0 0 0 0 0 1 0 0 0 0 7
6 7
6 7
6 a10;1 a10;2 a10;3 a10;4 a10;5 a10;6 0 0 a10;9 a10;10 0 0 0 0 7
6 7
6 7
6 0 0 0 0 0 0 0 0 0 0 0 1 0 0 7
6 7
6 7
6 a12;1 a12;2 0 0 0 0 0 0 a12;9 a12;10 a12;11 a12;12 a12;13 a12;14 7
6 7
6 7
6 0 0 0 0 0 0 0 0 0 0 0 0 0 1 7
4 5
a14;1 a14;2 0 0 0 0 0 0 0 0 a14;11 a14;12 a14;13 a14;14
where
Lj1 ¼ 5:23 þ xj xc ; Lj2 ¼ 5:939 þ xj ; Lj3 ¼ Lj2 ; Lj4 ¼ 4:614 þ xj ; Lcb ¼ 1:65; and Lb ¼ 13:64 xc :
Wheeler, G., Su, C. Y., & Stepanenko, Y. (1998). A sliding mode Engineers Part I-Journal of Systems and Control Engineering,
controller with improved adaptation laws for the upper bounds on 211(6), 407–416.
the norm of uncertainties. Automatica, 34(12), 1657–1661. Yao, B., Fanping, B., Reedy, J., & Chiu, G. (2000). Adaptive robust
Yanada, H., & Shimahara, M. (1997). Sliding mode control of an motion control of single-rod hydraulic actuators: theory and
electrohydraulic servo motor using a gain scheduling type observer experiments. IEEE-ASME Transactions on Mechatronics, 5(1),
and controller. Proceedings of the Institution of Mechanical 79–91.