Sunteți pe pagina 1din 42

Labor Regelung Einer Nichtlinearen

Hydraulischen Strecke

Regelungstechnischer Versuch 4

Backstepping-based Nonlinear Control of


a Hydraulic System

Version: 24.10.2013

Prof. Dr.-Ing. S. Liu

Lehrstuhl für Regelungssysteme

Fachbereich Elektrotechnik und Informationstechnik


Technische Universität Kaiserslautern

I
Contents
1 Introduction 5

2 System Modeling 7
2.1 Servo-pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Modeling of servo-pump . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Hydraulic cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 One-cylinder model . . . . . . . . . . . . . . . . . . . . . . 11
2.3.2 Two-cylinder model . . . . . . . . . . . . . . . . . . . . . . 13

3 Overall Model 14

4 Backstepping Control Theory 16


4.1 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.2 Application to Hydraulic System . . . . . . . . . . . . . . . . . . 23

5 Tasks 26
5.1 Tasks at home . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.1.1 Design a Numerical Derivative and Low-pass Filter Block . 30
5.1.2 Complementing Backstepping Algorithm . . . . . . . . . . 30
5.2 Tasks in Lab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2.1 IO-test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.2.2 Testing Encoder and Actuator Signal . . . . . . . . . . . . 35
5.2.3 Measurements and Estimations of Dv and fc . . . . . . . . 37
5.2.4 Realizing Backstepping Algorithm . . . . . . . . . . . . . . 37

Literaturverzeichnis 40

1
2
List of Symbols
A1 area of cylinder piston . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
A2 area of cylinder piston ring side . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
As area of pump control cylinder piston . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
CH hydraulic capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
dv damping coefficient of valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
EA bulk modulus in chamber A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13
EB bulk modulus in chamber B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Ee effective bulk modulus of oil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
FeL external load of pump control cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Ff friction of cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
FL external load of cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Fsf friction of pump control cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
kair air/oil ration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
ksL leakage coefficient of pump control cylinder . . . . . . . . . . . . . . . . . . . . . . . 11
ksp pump gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
kp parameter of P controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
kpu pump angle-voltage gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
kvu spool position-voltage gain of valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
KQ pump flow coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
kqb1 , kqb2 flow coefficient of chamber B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
kQv valve flow coefficient of servo pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
ms piston mass of pump control cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
n motor speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
PA pressure in chamber A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
PB pressure in chamber B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
PsL pressure in control cylinder of pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Po atmospheric pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Q fluid volume flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
QA volume flow of chamber A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
QB volume flow of chamber B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
QLex external leakage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
QLin internal leakage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Qp volume flow of pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Qv volume flow of valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
xc stroke of cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Tv time constant of valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
uv input voltage of valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
V10 initial volume in chamber A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
V20 initial volume in chamber B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13
vc velocity of cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3
Vgmax maximal displacement of pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
yv valve spool stroke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
α swshplate angle of pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
αr command value of swshplate angle of pump . . . . . . . . . . . . . . . . . . . . . . . 11
ηv volumetric efficiency of pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
ωv natural frequency of valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10

4
1 Introduction
Hydraulic systems are widely used in present industry. Compared to electrical
drives they can generate large forces or torques very fast with simple structures.
Hydraulic systems can be broadly classified into two categories. One is industrial
hydraulics, which is typically applied in mechanical manufacture, like extrusion
molding machines, pressers, rolling mills and others [Fin06]. The other is mobile
hydraulics, which are the core components of steering, braking, suspension and
power-train system of vehicles [KZH04] and also popular in robot manipulators,
ship rudders, airplane landing gears and so on [Mün06]. When the electronics are
integrated into the hydraulic parts, for instance sensors and servovalves, so-called
electro-hydraulic systems can be coined. The inputs are usually electrical signals
and the relevant system variables are measured and transformed into electrical
signals too. Thus the hydraulic components can be precisely controlled with good
performance.

Figure 1: The panorama of the electro-hydraulic testbed

The testbed built by Mannesmann Rexroth1 as shown in Fig. 1 and 2 is used for
the research of fault diagnosis and tolerant controller testing. It consists of two
functionally identical differential cylinders, e.g. working cylinder and load cylin-
der. The working cylinder represents the pressure stem of an extrusion machine.
1
Now the company is named as Bosch Rexroth after the mergence in 2001.

5
  
 




  





    # !"


  !"



Figure 2: Block diagram of the electro-hydraulic testbed

Its fluid power is supplied by a variable displacement axial piston servo pump.
The load cylinder outputs the resistance force, which is generated the moment
the material is pressed. Therein a fixed displacement pump is connected so that
load pressure can be built up quickly. Furthermore with a proportional pressure
relief valve an arbitrary load can be achieved. The direction change of fluid flow is
realized through several 2-position directional valves. The operation pressure of
whole system is limited under 250 Bar [Wit97]. The dynamics and the properties
of these parts together construct the nonlinearity of the hydraulic system, which
offers us the reason to utilize a nonlinear controller—Backstepping. As for the
control unit, signal measurements and control strategies are implemented with
the help of Dspace RTI 1104. In Fig. 3, the input signals and output signals we
will use are illustrated.
The explicit information about the card can be reached on the Dspace company
website. Part of students’ work relates to the operation on Dspace because of the
wide and important application of Dspace in industrial field. Dspace interacts
with Simulink in Matlab so that some block-assembled manipulations can finish
the control design. By looking into the Dspace manual, blocks and pins are
supposed to be selected to bridge the controller block in simulink with physical
world.
Generally the hydraulic circuit of the testbed can be divided in two main parts:
servo-pump and cylinders. The dynamic of hydraulic fluid as pressing force trans-
mission conductor will not be in consideration. In the modeling description sec-
tion, detailed mathematical model descriptions are given. Consequently the over-

6
Figure 3: Inputs and outputs between the testbed and Dspace

all mathematical model of this system is formulated. Based on the model we can
design control strategies to stabilize the system and track reference signals such
as a positioning trajectory. For obtaining a high exactness, the position tracking
control must keep the error under the level of 10−4 m with the help of a position
sensor at the error level 20µm. In the controller design section, the backstepping
nonlinear controller theory is introduced. Then aiming at this hydraulic system,
a backstepping controller is conceived.

2 System Modeling

2.1 Servo-pump
The pump, which is usually driven by an electrical machine, transforms the me-
chanical energy into hydraulic energy. The variable displacement pump allows
time varying displacement, which is defined the amount of fluid pumped per rev-
olution of the pump’s input shaft. The pump used in our experiment is actually
a servo-pump driven by an AC asynchronous motor. The actual look can be seen
in Fig. 4. The functional structure is interpreted in Fig. 5. As seen, it consists
of an axial piston pump with built-in proportional valve, a pressure transducer
and an external amplifier card for control. In the built-in proportional valve an
inductive position transducer is equipped for sensing the swivel angle and valve
position.
The pump’s pistons slide axially in a rotating cylinder block to give a pumping
action to the fluid. The piston shoes bear on a non-rotating swashplate. Vari-
ation of the swashplate angle changes the travel distance of the pistons, which
then changes the hydraulic displacement of the pump. The swashplate angle is

7
Figure 4: Actual view of servo-pump [Rex97]

determined by the offset cylinder and the control cylinder. Moreover the stroke
of the control cylinder is controlled by the proportional valve.
If the pump is off, the swashplate angle is held on 100% because of the preload
spring in offset cylinder side. With a rotating pump and de-energized proportional
valve the big control cylinder regulates to zero stroke as the valve spool is pushed
to the right by the spring and therefore the pump pressure is applied to the
control piston via valve port A. The balance between the pump pressure and the
spring force is kept between 8 and 12 bar. Increasing of the command value of
the swashplate angle makes the valve spool move from the central position to left
side until the swashplate angle (stroke of the control cylinder) reaches the set
value. The stroke of the control cylinder decreases because of the connection of
port A to the tank. Thus the swashplate angle will increase. On the contrary a
decrease in command value of the swashplate angle makes the valve spool move
from center to right. Meanwhile more fluid flows into the control cylinder and
increases the stroke. Finally the swashplate angle is reduced. The block diagram
of the servo-pump system is shown in Fig. 6.
Theoretically the volume flow rate can be calculated through the formula
Vgmax · n · tan(α) · ηv
Qp = , (1)
1000 · tan(αmax )

8
$ %&'%()*'+, - )./)/.+0/1'* 2'*2, 3 /44%,+ 56*017,. 8 5/1+./* 56*017,.
9 ).,*/'7,7 %).01: ; %)//* )/%0+0/1 +.'157<5,. = '1:*, >)/%0+0/1? +.'1%7<5,. @ 2'*2, %/*,1/07
A 2'*2, %)//* $B 2'*2, %).01:

Figure 5: Cross section view of servo-pump [Rex97]

where Vgmax is the maximal displacement, α stands for swashplate angle and αmax
is the maximum of it, n is the rotation speed of the pump and ηv denotes the
volumetric efficiency [Fin06]. Practically the swashplate angle varies in a small
range. It is reasonable to consider the approximation α ≈ tan(α) . Substitution
into Qp yields
Qp ≈ KQ · α (2)
where KQ is defined as pump flow coefficient, which is determined only by the
mechanical structure of the pump. Hence in light of this simplification the output
flow of the pump can be viewed as proportional to the swashplate angle. Despite
of the simplification, the later designed controller works also quite acceptably.

2.2 Modeling of servo-pump


This subsection will illustrate the interior structures of the servo-pump. Then
modeling based on the dynamics is established.
The servo-pump in fact is a closed loop control system. The swashplate angle α
is chosen as the state variable of the pump model [MJ96] [Pra01]. Fig.7 shows
the cascade structure of control loop. αr stands for the command value of the
swashplate angle and the α presents the stroke of the valve spool. Outer-loop
functions angle control. The angle error is amplified by the P controller and then
passed on to the inner control-loop as a command value. The valve spool stroke
is regulated by the inner PD controller. Thus the stroke of the control cylinder

9
Figure 6: Block diagram of servo-pump [Rex97]

αr 4Y α
SURSRUWLRQDO FRQWURO
3 3'
YDOYH F\OLQGHU

Figure 7: Closed control loop of servo pump

can be adjusted.
For precise modeling it is necessary to introduce the model of the proportional
valve and control cylinder. The fluid flow of the proportional valve Qv usually is
assumed to be proportional to the spool stroke yv . The controlled valve can be
approximated as a second order system [SS00] [Wat05]. In Laplace domain the
relation of the valve spool stroke yv and the input voltage uv can be written as

yv kvu ωv2
= 2 , (3)
uv s + 2dv ωv s + ωv2

where dv stands for the damping coefficient, ωv is the natural frequency and kvu
presents the spool position-voltage gain. Note that uv will be the important
controller input for the following backstepping algorithm.

10
By integrating the cylinder model the block scheme of the servo-pump can be
expanded as in Fig.8. It should be pointed out that the structure is partly sim-

As

yv Qv PsL 1
k vu ! v
2

# #
r 1
CH #
kp s 2 " 2 d v! v s " ! v
2 kQv As
ms

ksL Fsf
FeL

kQv flow coefficient of valve CH hydraulic capacity As piston area of control cylinder
PsL pressure in control cylinder ms mass of piston ksL leakage coefficient
Fsf friction of control cylinder FeL external load kp P controller

Figure 8: Nonlinear model of servo-pump

plified but it is still nonlinear because of the saturation of the valve spool stroke,
the friction in control cylinder, non-constant parameter CH and so on. Combin-
ing all the units gives a fifth-order system. Some of the parameters and state
variables are still unknown. Consequently the model in Fig.8 is not appropriate
for a real-time application.
More simplifications should be made without losing much accuracy of modeling.
• The valve is simplified as a first-order system.
• The cylinder is approximated as an integrator by ignoring the friction, ex-
ternal load and leakage2 .
Thus the servo-pump can be described by a second-order system as follows:
α kpu ksp
= , (4)
αr As Tv s2 + As s + ksp
where kpu is the angle-voltage gain, αr represents the command value of the
swashplate angle, ksp stands for the pump gain, which depends on the controller
parameters and valve flow coefficient, and Tv is the valve time constant.

2.3 Hydraulic cylinder


2.3.1 One-cylinder model

Cylinders are very common for linear actuation. The modeling of hydraulic cylin-
ders can be easily found in textbooks on hydraulics [Mer67, MR03, AGS06]. Thus
2
A detailed cylinder model can be found in chapter 2.3.

11
only a brief introduction will be made and the background of hydrodynamics will
be omitted.
xL AF

AE PD QGJK
FG
PC
FM QGHI

QC QD

Figure 9: Differential cylinder

In Fig.9 a hydraulic differential cylinder is shown. PA and PB represent the


pressure in the chamber A and B respectively. A1 and A2 denote the piston area
and ring-side area. QLin and QLex stand for the internal leakage flow and external
leakage flow. xc is the cylinder displacement, Ff the friction force and FL the
external load. The pressure dynamics is derived according to the rule of mass
conservation X X VH
Qin − Qout = V̇H + Ṗ , (5)
EH
where Q means liquid flow, VH the volume and EH is the bulk modulus that
indicates the compressibility of the fluid and can be formed as
∂PH
EH = −VH . (6)
∂VH
For mineral oil the value is around 1.4 ∼ 1.6 × 104 Bar. The bulk modulus
is affected by the system pressure, entrained air, mechanical compliance and
temperature. For engineering applications only the effective bulk modulus, which
is denoted as Ee , is utilized, which is usually expressed empirically. Considering
the significant influence of entrained air the following approximation is employed
[Wat05].
P
Po
+ kair
Ee = kair Eo P
, (7)
P
+ Po
where Po is the atmospheric pressure and the constant parameter kair is the air/oil
volume ratio.
Now the pressure dynamics in both chambers can be derived as

EA (PA )
ṖA = (−A1 vc + QA − QLin )
V10 + A1 xc
(8)
EB (PB )
ṖB = (A2 vc + QB + QLin − QLex ),
V20 − A2 xc

12
where V10 and V20 are the initial volume of chamber A and B respectively, vc
stands for the piston velocity and EA and EB can be calculated according to
(7). QA is the flow rate of chamber A, which is equal to the pump output flow
QA = KQ α because of the pump controlled structure of the system. QB is the
flow rate of chamber B. The pressure in chamber B is low. Therefore the effect
of the directional valves can not be ignored here. According to the characteristic
curve of the directional valve, QB can be approximated by
q
 k (P − P ) + k kqb1 (PB − Po ) + kqb2 > 0;
qb1 B o qb2
QB = (9)
0 k (P − P ) + k ≤ 0,
qb1 B o qb2

The parameters kqb1 and kqb2 can be identified through experiments. By ignoring
the leakages and combining Newton’s second law, the dynamic model of the
working cylinder can be described as:

ẋc = vc (10)
1
v̇c = (PA A1 − PB A2 − Ff − FL ) (11)
mc
EA (PA )
ṖA = (−A1 vc + KQ α) (12)
V10 + A1 xc
EB (PB )
ṖB = (A2 vc − QB ), (13)
V20 − A2 xc
where xc is the stroke of the cylinder, vc is the velocity of the cylinder piston, mc
means the mass of the piston and Ff and FL represent the friction and external
load respectively.

2.3.2 Two-cylinder model

According to the cylinder model derived in the last section the external load FL is
generated by the extrusion of two cylinder rods. This force can not be measured
with the existing hardware. Now a two-cylinder model will be derived, in which
the external load is the same as the pressure in the load cylinder chamber A,
which can be measured directly.
The load cylinder shown in Fig. 10 is identical to working cylinder. If the working
cylinder moves forwards, it moves backwards with the same velocity. Assuming
the friction in the load cylinder to be the same as in the working cylinder, the
piston motion can be written as:

ẋc = vc (14)
1
v̇c = (PBL A2 − PAL A1 − Ff + FL ), (15)
mc

13
where PAL and PBL represent the pressure in chamber A and B respectively.
The chamber B is in subpressure state and the pressure PBL is close to the tank
pressure, so PBL can be ignored. Now combine (15) with (11). The following
xc

PBL
FL PAL

Ff

Figure 10: Load cylinder

equation can be derived:


1
v̇c = (PA A1 − PB A2 − 2Ff − PAL A1 ) (16)
2mc
In this equation the external load is equal to the pressure in chamber A of the
load cylinder, which can be directly adjusted by the proportional pressure relief
valve. The two-cylinder model can be written as

ẋc = vc (17)
1
v̇c = (PA A1 − PB A2 − Ff ric − Fload ) (18)
m
EA (PA )
ṖA = (−A1 vc + KQ α) (19)
V10 + A1 xc
EB (PB )
ṖB = (A2 vc − QB ). (20)
V20 − A2 xc

3 Overall Model
In this chapter the whole system description is formulated, and the parameters
in the model are elaborated. However the process of the parameter identification

14
is omitted.


 ẋc = vc


 1
v̇c = (PA A1 − PB A2 − Dv vc − fc − Fload )


m



  
E (P )
 q
B B

 ṖB =

A2 vc − kqb1 (PB − Po ) + kqb2
Cylinder V20 − A2 xc (21)
EA (PA )


ṖA = (−A1 vc + KQ α)


V10 + A1 xc




·
+ kair


Po

E(·) = (· could be PA or PB )


 kair Eo

·
+ P·o
(
α̇ = vα
P ump (22)
v̇α = −p1 α − p2 vα + p1 kpu uin

Table 1: Parameters in cylinder part


m = 333.06 kg total mass of the working cylinder
and the load cylinder
Dv = 34.03 Bar ∗ m2 linear damping
fc = 3.596 kN Coulomb friction
Fload external load (time-varying)
A1 = 0.0314 m2 the area of the piston
A2 = 0.016 m2 the area of the ring side
V10 = 0.0035 m3 the initial volume in side A
V20 = 0.00801 m3 the initial volume in side B
kqb1 = 61.275 flow coefficient
kqb2 = −63.478 flow coefficient
Po = 1 atmospheric pressure
KQ = 0.0235 flow-rate coefficient
Eo = 1.6e4 Bar bulk modulus of oil
kair = 0.005 air coefficient

Through observation, parts of Eq. (21) can be simplified. Define FHY = PA A1 −


PB A2 . Then the second, third and fourth equation can be rewritten as
1
v̇c = (PA A1 − PB A2 − Dv vc − fc − Fload ) (23)
m
EA (PA )A1
ḞHY = (−A1 vc + KQ α)
V10 + A1 xc
 
EB (PB )A2
q
− A2 vc − kqb1 (PB − Po ) + kqb2 (24)
V20 − A2 xc

15
Table 2: Parameters in pump part
p1 = −44444 system parameter
p2 = −333.3333 system parameter
kpu = 0.0017 relation between input voltage
and set piston position

We also can write Eq. (24) for convenience of later usage in Backstepping algo-
rithm.
ḞHY = f (xc , vc ) + g(xc )α, (25)
in which EA (PA ), EB (PB ) are viewed as known parameters.

4 Backstepping Control Theory

4.1 Fundamentals
Backstepping control is a technique for designing stabilizing controls for a special
class of nonlinear dynamical systems. These systems are built from subsystems
that radiate out from an irreducible subsystem that can be stabilized using some
other method. Because of this recursive structure, the designer can start the
design process at the known-stable system and ”back out” new controllers that
progressively stabilize each outer subsystem. The process terminates when the
final external control is reached. Hence, this process is known as backstepping [3].
The method in this experiment is utilized to determine the controller. Especially
some key steps in the deduction of the controller as tasks should be finished by
the students. In the following, Integrator Backstepping Control theory will be
explained.
Suppose we have a system in feedback-linearizable form that is augmented by an
integrator, i.e. (
ẋ = f (x) + g(x)ξ, f (0) = 0
(26)
ξ˙ = u
where the variable ξ is referred to as a virtual control. Now suppose that the
feedback-linearizable system

ẋ = f (x) + g(x)u (27)

is stabilizable through the use of a feedback control u = α(x). This assumption is


critical to the concept of backstepping. In the following, it will become clear that

16
this indeed implies that the difference between every virtual control and every
corresponding state and real control u converges.
Now let’s define the difference between the virtual control and the ”desired value”
of the control as a virtual state variable, i.e.

z = ξ − α(x) (28)

Thus we can now rewrite the system taking into account the new state variable

 ẋ = f (x) + g(x)(α(x) + z),
(29)
 ż = ξ˙ − α̇ = u − ∂α(x) (f (x) + g(x)(α(x) + z))
∂x
Now due to our assumption that the system ẋ = f (x) + g(x)u is stabilizable,
we know there exists a control Lyapunov functional (clf) V (x) for the feedback
linearizable system that satisfies
dV
V̇ = (f (x) + g(x)α(x)) ≤ −W (x) ≤ 0.
dx
Next, let’s define a candidate augmented clf, Va (x, z), for the system (29)
1
Va (x, z) = V (x) + z 2 . (30)
2
Then take the derivative of the augmented clf:
 
∂α(x)
V̇a (x, z) = V̇ (x) + z ż = V̇ (x) + z u − (f (x) + g(x)(α(x) + z)) (31)
∂x

which can be rewritten as:


 
∂α
V̇a (x, z) = V̇ + z u − (f + g(α + z))
∂x
 
∂V ∂α
= (f + gα + gz) + z u − (f + g(α + z))
∂x ∂x
 
∂V ∂α ∂V
= (f + gα) + z u − (f + g(α + z)) + g
∂x ∂x ∂x
 
∂α ∂V
≤ −W (x) + z u − (f + g(α + z)) + g . (32)
∂x ∂x

Let’s observe the last term in Eq. (32). If we choose u appropriately, we can
ensure that the augmented clf is negative definite. One simple choice is
∂α ∂V
u = −kz + (f + g(α + z)) − g.
∂x ∂x

17
Eq. (32) turns out to be

V̇a (x, z) ≤ −W (x) − kz 2 .

Note that this control is by no means specific!


Another point worthy of being noted is that when a system has more than first
order, the control in the same principle should be a chain for every state. This
will be demonstrated in the following example.
Example 1. Design a backstepping controller for the following second order sys-
tem

ẋ1 = −x31 + x2 (33a)


ẋ2 = x22 + u (33b)

Step 1a: Define a virtual control for equation (33a), say α, and let z be a virtual
state variable representing the difference between the actual and virtual control of
(33a), i.e.
z = x2 − α
Thus in terms of the new state variable, we can rewrite (33a) as

ẋ1 = −x31 + z + α.

Step 1b: Define a candidate clf for this equation (arbitrarily choose a quadratic
lyapunov function)
1
V = x21 . (34)
2
Take the derivative:

V̇ = x1 ẋ1 = x1 (−x31 + α + z) = x1 (−x31 + α) + zx1 .

We can now select an appropriate virtual control α, which would make the first
order system stabilizable. In other words, the α provides a certain freedom to
parameterize the derivative of clf to satisfy the stability criterion. But the virtual
control α is not the actual state x2 in (33a). Therefore an instinctive thinking is
making the difference z between the virtual control α and actual state x2 converges
to zero.
Step 1c: We can choose

α = −k1 x1 k1 ≥ 0, (35)

which is NOT a unique choice. The derivative of the virtual control is given by

α̇ = −k1 ẋ1 = −k1 (−x31 + x2 ).

18
Step 2a (real input step): We now find the virtual state equation
ż = ẋ2 − α̇ = x22 + u − (−k1 )(−x31 + x2 ).
Step 2b: The augmented clf is chosen to be
1
Va = V + z 2 (36)
2
The derivative of Va yields
V̇a = V̇ + z ż = (−x41 − kx21 + zx1 ) + z(x22 + u + k1 (−x31 + x2 ))
= −x41 − kx21 + z(u + x1 + x21 + k1 (−x31 + x2 )) (37)
Step 2c: We can now select an appropriate control u, which makes the second
order system stabilizable. Of the many possible choices let’s go with
u = −k2 z − (x1 + x22 + k1 (−x31 + x2 ))
= −k2 (x2 + k1 x1 ) − (x1 + x22 + k1 (−x31 + x2 )). (38)
Patently there is a lot of design freedom in the backstepping method, as at each
step backwards we only require the clf to be negative definite.

10
x1
9
x
2
8

6
x1 and x2

0
0 2 4 6 8 10
Time

Figure 11: The dynamic of system (33a), (33b) with backstepping controller

The initial values x(0) = [10, 5] for the system are assumed and k1 = 5, k2 = 1
are selected. Then the simulation for the dynamics under the backstepping control
is given in Fig. 11.

19
In Example 1, the backstepping control successfully drives the system to the ori-
gin. But during our experiment, we want to let the system track some predefined
signal. This attempt must be reflected in our design of the controller. And an-
other noteworthy point is that more free coefficients can be introduced in the
clf to facilitate the design, which sometimes is crucial. In the following, another
example is given to show these specifics.
Example 2. Consider a system containing a chain of integrators given by
ẋ1 = x21 + x2 (39a)
ẋ2 = x22 + x3 (39b)
ẋ3 = u (39c)
We want the x1 to track a signal, say, r(t). No doubt that we can supposedly see
x1 as a position variable and we expect this movement capable of following r(t)
exactly. So we have
z1 = x1 − r. (40)
The same procedure as in Example 1 is employed next.
Step 1a: Define a virtual control for equation (39a), say α1 , and let z2 be the
virtual state variable representing the difference between the actual and virtual
controls of (39a), i.e.
z 2 = x2 − α 1 .
Thus in terms of the new state variable, we can rewrite (39a) as
ẋ1 = x21 + z2 + α1 .

Step 1b: Define a candidate clf for this equation (again arbitrarily choose a
quadratic)
1
V = ρ1 z12 , (41)
2
where attentively the clf uses z1 instead of x1 and adds a new free coefficient ρ1 .
Later on, we will see these ρi will help a lot in determining a suitable control.
The derivative of V yields:
V̇ = ρ1 z1 ż1
= ρ1 (x1 − r)(x21 + z2 + α1 − ṙ)
= ρ1 (x1 − r)(x21 − ṙ + α1 ) + (x1 − r)z2 .

(42)
Step 1c: Select a virtual control for the first order system
α1 = ṙ − x21 − k1 (x1 − r) k1 ≥ 0
α̇1 = r̈ + k1 ṙ − (2x1 + k1 )ẋ1
= r̈ + k1 ṙ − (2x1 + k1 )(x21 + x2 ).

20
Hence, V̇ can be rewritten as

V̇ = −k1 ρ1 (x1 − r)2 + ρ1 (x1 − r)z2 .

Step 2a: Define a second virtual state, z3 = x3 − α2 . Thus

ż2 = ẋ2 − α̇1 = x22 + (z3 + α2 ) − (r̈ + k1 ṙ − (2x1 + k1 )(x21 + x2 )).

Step 2b: Augment the initial clf to reflect the presence of the new state variable
1
Va1 = V + ρ2 z22 .
2
The derivative of Va1 is

V̇a1 = V̇ + ρ2 z2 ż2
= −k1 ρ1 (x1 − r)2 + ρ1 (x1 − r)z2
+ ρ2 z2 x22 + (z3 + α2 ) − r̈ − k1 ṙ + (2x1 + k1 )(x21 + x2 )

(43)

Step 2c: Select a virtual control for this system to remove any potentially unde-
sired z2 , x1 and x2 terms (we can not remove z3 yet)
ρ1
α2 = − (x1 − r) − k2 z2 − x22 + r̈ + k1 ṙ − (2x1 + k1 )(x21 + x2 ). (44)
ρ2
So,
V̇a1 = −k1 ρ1 (x1 − r)2 − k2 ρ2 z22 + ρ2 z2 z3 .

And α̇2 = f (x, z) can be explicitly calculated. But here we skip this step. In our
practical application, some derivative steps are solved by a numerical derivative
method rather than using an analytic representation. One reason is the complexity
involved in the analytic representation, which consumes too much computational
resource even overrides the CPU. The other reason is that mostly the coefficients
of system are not exact based on approximation. Despite of getting through the
complex calculation, the value eventually obtained could be also not dominant.
Step 3a: Let’s recall z3 = x3 − α2 . The derivative of z3 is

ż3 = ẋ3 − α̇2 = u − f (x, z)

Step 3b: Augment the initial clf to reflect the presence of the new state variable
1
Va2 = Va1 + ρ3 z32
2

21
The derivative yields

V̇a2 = V̇a1 + ρ3 z3 ż3


= −k1 ρ1 (x1 − r)2 − k2 ρ2 z22 + ρ2 z2 z3 + ρ3 z3 (u − f (x, z))
ρ2
= −k1 ρ1 (x1 − r)2 − k2 ρ2 z22 + ρ3 z3 (u − f (x, z) + z2 )
ρ3

Step 3c: Select an actual control for this system


ρ2
u = −k3 z3 − z2 + f (x, z) k3 ≥ 0
ρ3
Then we get
V̇a2 = −k1 ρ1 (x1 − r)2 − k2 ρ2 z22 − ρ3 k3 z32 .
So far, we have now designed a feedback controller for stabilization of the 3rd
order system. Clearly, the backstepping method can be applied to a system of any
order, provided the system state equations are in the correct form.
For the simulation, we use that initial points x(0) = (1, 2, −5), and parameters
k1 = 100; k2 = 100; k3 = 100; ρ1 = 1e − 3; ρ2 = 1; ρ3 = 1e + 3. The reference signal
is a sine signal with amplitude 1 and frequency 3 rad/sec. The states dynamics
are shown in Fig. 12.

1
1
x

−1
0 1000 2000 3000 4000 5000 6000 7000 8000
50
x2

−50
0 1000 2000 3000 4000 5000 6000 7000 8000
5000

0
x3

−5000

−10000
0 1000 2000 3000 4000 5000 6000 7000 8000

Figure 12: The dynamic of system (39a), (39b) and (39c) with backstepping
controller

The the effect of tracking the reference signal can be seen from fig. 13

22
1.5
reference signal
x1

x1 and reference signal 1

0.5

−0.5

−1
0 1000 2000 3000 4000 5000 6000 7000 8000

Figure 13: The tracking results for system (39a), (39b) and (39c) with backstep-
ping controller

Note that configuring the appropriate gain is very important. Usually one can in
view of the physical systems give reasonable value.

4.2 Application to Hydraulic System


In the following, we will design the backstepping controller for our system. Our
system is a fifth-order system, which is elaborated in the modeling section, so that
the complexity is far more than the above two examples. Part of the key steps
will not be given here instead that they as tasks are supposed to be finished by
students so as to grasp the backstepping method. Let’s recall the system model

ẋ1 = x2 (44a)
Dv 1 1
ẋ2 = − x2 + x3 − Fl (44b)
m m m
ẋ3 = f1 (x1 , x2 ) + f2 (x3 ) + g(x1 )x4 (44c)
ẋ4 = x5 (44d)
ẋ4 = p1 x4 + p2 x5 + p3 u (44e)

Correspondingly, x1 = xc , x2 = vc , x3 = FHY , x4 = α and x5 = vα . The details


can refer to chapter 3.
Here, merely the necessary steps are listed. All the mathematical deductions are
omitted. As one of the most important homework, students can compare the

23
listed results with their ones and complement the missing steps. The symbols
used in the deduction are:

x1d : Tracked signal


x2d , · · · , x5d : virtual controller of x2 , · · · , x5 respectively
e1 , · · · , e5 : error or difference between x1 − x1d , · · · , x5 − x5d respectively
V1 , · · · , V5 : Lyapunov functionals
k1 , · · · , k5 : positive parameters appearing in the design of virtual controllers
ρ1 , · · · , ρ5 : positive parameters appearing in V1 , · · · , V5

e1 = x1 − x1d
e2 = x2 − x2d
e3 = x3 − x3d
e4 = x4 − x4d
e5 = x5 − x5d

ė1 = (e2 + x2d ) − ẋ1d


1 2
V1 = ρ1 e
2 1
x2d = ẋ1d − k1 e1
V̇1 = −ρ1 k1 e21 + ρ1 e1 e2 (45)

ẋ2d = ẍ1d − k1 (x2 − ẋ1d )


Dv 1 1
ė2 = − x2 + (e3 + x3d ) − Fl − ẋ2d
m m m
1 2
V 2 = V 1 + ρ2 e 2
2
x3d = ??
ρ2
V̇2 = −ρ1 k1 e21 − k2 ρ2 e22 + e2 e3 (46)
m
ẋ3d = numerical differential(x3d )
ė3 = f1 (x1 , x2 ) + f2 (x3 ) + g(x1 )(e4 + x4d ) − ẋ3d
1
V3 = V2 + ρ3 e23
2
1 ρ2
x4d = [−f1 (x1 , x2 ) − f2 (x3 ) + ẋ3d − e2 − k3 e3 ]
g(x1 ) ρ3 m
V̇3 = −ρ1 k1 e21 − k2 ρ2 e22 − k3 ρ3 e23 + g(x1 )ρ3 e3 e4 (47)

24
ẋ4d = numerical differential(x4d )
ė4 = e5 + x5d − ẋ4d
1
V4 = V3 + ρ4 e24
2
ρ3
x5d = ẋ4d − g(x1 )e3 − k4 e4
ρ4
V̇4 = −ρ1 k1 e21 − k2 ρ2 e22 − k3 ρ3 e23 − k4 ρ4 e24 + ρ4 e4 e5 (48)

ẋ5d = numerical differential(x5d )


ė5 = p1 x4 + p2 x5 + p3 u − ẋ5d
1
V5 = V4 + ρ5 e25
2 
1 ρ4
u = −p1 x4 − p2 x5 + ẋ5d − e4 − k5 e5
p3 ρ5
V̇4 = −ρ1 k1 e21 − k2 ρ2 e22 − k3 ρ3 e23 − k4 ρ4 e24 − k5 ρ5 e25 (49)

Finally, we get the control signal u. All the computations and deductions are
for it. The virtual controller x3d is missing. As exercise and task the x3d would
be derived by students. With the completion we can generate the control signal
correctly. If the simulation in matlab runs smoothly and successfully, the compi-
lation into C codes in real time workshop can be doable. Then downloading to
Dspace for real-time control sounds a reasonable and logical ensuing step.

25
5 Tasks

5.1 Tasks at home


The tasks in this section have to be finished at home.
Before the practical application of the backstepping algorithm in the testbed can
be done, it have to be tested in the given Simulink model. (For the principle of
backstepping refer to the above chapters.) The requirement is that the position
and velocity errors should be confined to 0.1 mm. The control input voltage
should be kept in relatively steady and minor level. The effect can be compared
with the one in Fig. 14.

Figure 14: The effect with adjusted backstepping controller

Next an explanation to the given simulink blocks is made.

26
27

Figure 15: simulink blocks overview


In Fig. 15 the azure block includes all the signal outputs, in which scope
Scope position and velocity errors can be used to observe the position and
velocity errors as well as input voltage. The red block is the backstepping algo-
rithm block need to be completed. The other blocks used to describe the model
and generate some activating signals should not be modified. In the backstepping
block (the red one) we can see Fig. 16. Wherein the orange block called x3 d and
the other identical light blue blocks named numerical diff should be designed.
Fload = 50 ∗ 1e5.

28
29

Figure 16: backstepping algorithm construction


5.1.1 Design a Numerical Derivative and Low-pass Filter Block

Task 1: The designs of a numerical derivative and low-pass filter can refer to
corresponding book or articles (Abtast Reglung). Here no need of complicated
design. We can use the numerical derivative:
s(k) − s(k − 1)
ds = , (50)
sampling period
where the present state minus last state then divides the sampling period. How-
ever the derivative could be contaminated by noise. So a low-pass filter is neces-
sary. The following one could be used.
y[i] := α ∗ u[i] + (1 − α) ∗ y[i − 1], (51)
where u is the input and y is the output. α is defined as α = k∗sampling period,
in which k is a parameter we have to determine to achieve an acceptable result.
Note that if the parameter is too big, too much signal would be filtered, and if
too small, the filter has little effect. Choose the proper parameter to simulate in
Matlab. The sampling period is defined in the parameter file.

5.1.2 Complementing Backstepping Algorithm

Task 2: According to aforementioned backstepping theory, and on the basis of


our systems model finish the x3 d block in Simulink, which is exactly the x3d in
above equation description because some writing can be admitted in simulink.
Find proper parameter values for k1 , k2 after constructing the x3d . Before imple-

Table 3: k1 and k2 to be found in x3 d


k1 k2

menting the backstepping algorithm in reality, make sure by simulation that the
result is acceptable namely, the position error is less than 0.1mm.
Some explanations for the input signals to Backstepping Controller block can be
seen in table 4.
Note that because of the difference between simulation and practical implemen-
tation the parameters we get from simulation can not fit the practical imple-
mentation in Dspace so well. Therefore in the tasks-in-lab part, the parameters
actually still have to be tried and modified. But the simulation can ensure the
backstepping algorithm structure is right or not. Otherwise we can not expect a
set of parameters can be found to implement the backstepping control.

30
Table 4: Input signals for Backstepping Controller block

Parameter Value
z act x4 in eq. 44d
vz diff x5 in eq. 44e
x act x1 in eq. 44a
v act x2 in eq. 44b
PA is used to get FHY = x3 in eq. 44c
PB is used to get FHY = x3 in eq. 44c
xd reference signal, exact the x1d
vd the first derivative of reference signal, exact the ẋ1d
x d dot2 the second derivative of reference signal, exact the ẍ1d
x d dot3 the third derivative of reference signal (no need to use in the construction of x 3d)

5.2 Tasks in Lab


The realization of control in Dspace will utilize the Control Desk to observe and
control the testbed. Each signal you receive and send can be seen and adjusted
in Control Desk like Fig. 17

31
32

Figure 17: Control desk overview


5.2.1 IO-test

Task 3: For controlling the movement of cylinder besides the control of pump,
the valves’ states (open or close) are also needed to be put under consideration.
There are five different functions involving five different combinations of valves’
states.

Table 5: Directional Valves


Valve V0 V1 V2 V3 V4 V5 V6
Light index 7.1 7.2 7.3 7.4 23.1 23.2 4.0 left end right end
Pin Mapping Pin2 Pin3 Pin5 Pin6 Pin8 Pin9 Pin11 Pin12 Pin21
Channel IO1 IO3 IO5 IO7 IO9 IO11 IO13 IO12 IO21

The left end light and right end light in table 5 indicate if the working cylinder
has reached the farmost end. On the testbed this is signaled by two light near the
two ends. Meanwhile sensor would send the information out, which is as input
to IO12 and IO21.
Design a logic control with 5 different functional input. We name it ’working
mode selector’ that is a constant source with natural number from 1 to 5 output
corresponding to 5 different modes. This must be done by the utilization of
Simulink in matlab and control desk supplied by Dspace. The result should realize
that when one clicks one of the modes, the valves must behave correspondingly
in the predescribed way. Before accomplishing this task, do not try to implement
control algorithm. The lights can be seen (the red circles enclosed) in Fig. 18

33
34

Figure 18: Testbed lights


Table 6: Directional Valves
V0 -V6 Functionality
1000001 working cylinder forward
0000001 all stop
0001101 load cylinder forward
0100001 working cylinder backward
0010101 load cylinder backward

Table 7: Servo-pump: Mannesmann-Rexroth A10VSO 28 DFE1


Rotation Speed 1000 1/min
Volume Flow 28 L/min (max.)
Power 11kw
Swashplate Angle 17.7◦
Control Cylinder Stroke 17.72 mm (max.)

5.2.2 Testing Encoder and Actuator Signal

Task 4: This task could be divided into several parts:


• Design a manual controller in Control Desk. This step has two-fold signif-
icance. One is that the return of cylinders is in need of Load Cylinder
Forward. The other is that when the algorithm has to be interrupted or
terminated, switch to manual controller then to adjust the control output.
When the cylinder returns to the starting point, Be Sure to make the con-
trol output Zero. Implement a safety logic to switch the control output to
zero when the all stop button is clicked.
• Test Position Encoder. The Dspace supplies a digital incremental encoder
interface. In Simulink for using DS1104ENC POS C1 block, the DS1104
ENC SETUP block must be set up. Then utilize Control Desk to observe
if the input value makes sense of itself. If the value vindicates itself, then
consider an offset for that and explain why an offset value is necessary?
• Then under manual control through AD interface 5,6,7 to test Pressure in
Chamber A PA , Load Pressure PL and Swashplate angle zact .
According to the data sheets tab. (7,8, 9,10), the gain for each measurement
yields tab.(11) Consider those gains in the measurement and control output.

35
Table 8: Hydraulic cylinder: Mannesmann-Rexroth CD H3 MF3/200/140/500
Piston ø 200 mm
Piston Rod ø 140 mm
Area Ratio 19.6
Piston Rod Length 1176 mm
Piston Rod Stroke 500 mm (max.)
Moving Mass 166 kg

Table 9: Absolute encoder: TR-Electronic CE-65-SSI


Encoder Capacity 24 bit (max.)
Steps Revolution 12 bit (4096)
Number of Revolution 12 bit (4096)
Resolution 20 µm

Table 10: Proportional valve: Mannesmann-Rexroth DBE 62-1X/315


Volume Flow 30 L/min (max.)
Operation Pressure 315 Bar (max.)

Table 11: Absolute encoder: TR-Electronic CE-65-SSI


Input and Output Gain
PA (input) 315
PL (input) 315
zact (input) 17.72
Desired Load (output) 1/315
Pump Maximum Pressure (output) 1
Control output (output) 1/10

36
5.2.3 Measurements and Estimations of Dv and fc

Task 5: Parameters Dv and fc are used to describe a relationship between friction


Ff ric and velocity v. The equation is approximately

Ff ric = Dv v + fc . (52)

And by making cylinders move in constant velocity, the friction can be calculated
through
Ff ric = PA A1 − PB A2 − Fload (53)
With PB A2 ≈ 0 and Fload = PL A1 . In view of these relationships, measure
’velocity vs friction’ data and calculate Dv and fc out by using polynomial fitting
technique.

Table 12: Velocity vs Friction


Friction
Velocity

Table 13: Dv and fc


Dv fc

5.2.4 Realizing Backstepping Algorithm

Task 6: We will supply another simulink diagram in this phase just as in Fig 19.

37
38

Figure 19: Simulink blocks in practical


Table 14: Parameters for Backstepping in practical
Parameter Value
k3 1e-5
k4 10
k5 100
beta1 100
beta2 10
beta3 1e-15
beta4 10

Table 15: k1 and k2 to be found in x3 d in pratical


k1 k2

Then students should do


• copy your IO test program into the block Digital I/O to implement the
valve control in practical
• copy your block x 3d into the block backstepping controller
• copy your three identical blocks numerical diff into the block backstepping
controller
• k1 and k2 have to be adjusted again until the requirement that position
error is kept less than 0.1mm is satisfied. The other parameters are given
in table 14. No need to care about them.

39
References
[AGS06] A. Akers, M. Gassman, and R. Smith. Hydraulic power system analysis.
Taylor & Francis, 2006.
[AL00] A. G. Alleyne and R. Liu. Systematic control of a class of nonlinear
systems with application to electrohydraulic cylinder pressure control. IEEE
Transactions on Control Systems Technology, 8:623–634, 2000.
[CLCL08] M. H. Chiang, L. W. Lee, C. C. Chen, and H. H. Liu. Velocity con-
trol for a variable displacement hydraulic servo system using adaptive fuzzy
sliding-mode control. Proceedings of the 17th World Congress IFAC, pages
13797–13802, 2008.
[CP99] J. Chen and R. J. Patton. Robust model-based fault diagnosis for dynamic
systems. 1999.
[Fin06] Dietmar Findeisen. Ölhydraulik. Springer, 2006.
[JK04] M. Jelali and A. Kroll. Hydraulic servo-systems:modelling, identification
and control. Springer, 2004.
[JMZ03] J. Jiang, M. M. Mahmoud, and Y. Zhang. Active fault tolerant control
systems: stochastic analysis and synthesis. Springer, 2003.
[KZH04] Wu K, Q. Zhang, and A. Hansen. Modeling and identification of a hy-
draulic transmission hardware-in-the-loop simulator. International Journal
of Vehicle Design, 34:52–64, 2004.
[Lju99] L. Ljung. System identification: theory for the user. Prentice Hall, 1999.
[Mer67] H. E. Merritt. Hydraulic control systems. John Wiley & Sons, Inc., 1967.
[MJ96] N. D. Manring and R. E. Johnson. Modeling and designing a variable
displacement open-loop pump. Journal of Dynamic Systems, Measurement
and Control, 118:267–271, 1996.
[Mün06] M. Münchhof. Model-Based Fault Detection for a Hydraulic Servo Axis.
PhD thesis, TU Darmstadt, 2006.
[MR03] H. J. Matthies and K. T. Renius. Eingführung in die Ölhydraulik. Teub-
ner, 2003.
[Pra01] E. A. Prasetiawan. Modeling, simulation, and control of an earthmoving
vehicle powertrain simulator. Master’s thesis, university of Illinois at Urbana-
Champaign, 2001.
[Rex97] Mannesmann Rexroth. Electrical closed loop control systems DFE1,
DFEE, and FE1 for the control of AA10SO variable displacement axial pis-
ton pumps., 1997.

40
[SS00] M. R. Sirouspour and S. E. Salcudean. On the nonlinear control of hy-
draulic servo-systems. Proceeding of the 2000 IEEE international conference
on robotics & automation, pages 1276–1282, 2000.
[TSL98] S. Tafazoli, C. W. Silva, and P. D. Lawrence. Tracking control of an
electrohydraulic manipulator in the presence of friction. IEEE Transactions
on Control Systems Technology, 6:401–411, 1998.
[Wat05] J. Watton. Modeling, monitoring and diagnosis techniques for fluid
power systems. Springer, 2005.
[Wit97] B. Wittmer. Inbetreibnahme und modellierung einer hydraulischen ver-
suchsanlage und implementierung einer geschwindigkeitsregelung. Master’s
thesis, TU Kaiserslautern, 1997.
[1] F. Franklin: Feedback Control of Dynamic Systems
Prentice-Hall, New Jersey, 4. Auflage, 2002
[2] Moeller, Fricke, Frohne, Vaske: Grundlagen der Elektrotechnik
Teubner-Verlag, Stuttgart, 1986
[3] H.K. Khalil, Nonlinear Systems (3rd ed.), Upper Sad-
dle River, NJ: Prentice Hall, ISBN 0-13-067389-7, (2002).
(http://www.egr.msu.edu/∼khalil/NonlinearSystems/).

41

S-ar putea să vă placă și