Professional Documents
Culture Documents
Abstract: Servo-hydraulic control is regarded as a key technology in most mechanical testing systems for the
investigation of strength and fatigue properties of materials, mechanical components and structures. High
performance but robust control strategies are necessary as part of the test control system to ensure high test
quality and reproducible test results which are of special importance when considering multi-axial component
tests using high channel count test rigs with several active and synchronously operating servo-hydraulic test
cylinders. Such high performance multidimensional control algorithms demand sufficient information about
the dynamic characteristics of the used test cylinders, the test rig and – at least as a good estimate – of the
device under test. In this paper we introduce a special kind of identification strategy for deriving accurate
mathematical models for a class of commonly used servo-hydraulic test cylinders: low-cost symmetric test
cylinders with significant friction. In order to accurately describe the dynamics of the servo-hydraulic cylinder
system, we first build a mathematical model of the system which includes some still unknown parameters.
These unknown parameters are determined by utilizing a special identification scheme based on the black box
identification method to estimate initial values of the unknown parameters, followed by the grey box iden-
tification method for accurately estimating the complete cylinder system model. The developed black-grey
box identification method delivers accurate information on the cylinder dynamics. The comparison between
simulation results and measured data from real world experiments demonstrates the high accuracy of the
derived servo-hydraulic cylinder system model, and thus validates the effectiveness of the developed system
identification method.
Keywords: Electro-hydraulic servo system, servo-hydraulic test cylinder, fatigue test, friction, system mode-
ling, system identification, simulation
■■ 1 Introduction the higher demands on the industri- stems, severe coupling of the used
al testing system performance, their control axes can appear in combina-
Multi-axial testing systems with ser- application fields and ranges are tion with non-ideal mechanical link
vo-hydraulic test cylinders and de- expanding constantly. Consequen- elements. These effects have to be
dicated control systems are widely tly, the electro-hydraulic transmis- considered in the development and
used in experimental mechanics and sion and control systems become implementation of such test con-
related test applications because of more and more complex and the trol systems and control strategies.
their advantages, such as high ener- requirements on their control algo- It is further essential for any test
gy density, good linearity, little dead rithms become higher and higher. engineer to understand the dyna-
zone, high sensitivity, fine dynamic For instance, the duration and/or mic characteristics of the used test
performance, fast response, high the costs of mechanical tests should equipment and the plant in order to
precision and so on. With further remain low, but with increased test be able to setup such a test system
development of the electro-hydrau- precision and reproducibility it is with optimum dynamic response,
lic servo control technology due to necessary to ensure high suppressi- control precision and working reli-
on of external disturbances and the ability.
Lingjun Li, MSc.; Dr. Thomas ability to cope with non-linearities
Thurner, both Institute of Ligh- and time dependency in the actua- Throughout this paper, we descri-
tweight Design, Graz University tor dynamics and the overall plant. be a certain kind of system identi-
of Technology, Austria Furthermore, regarding multi-axis fication for building mathematical
electro-hydraulic servo control sy- models for a commonly used class
of servo-hydraulic test cylinder sy- and operation mechanism, however have several disadvantages, such
stems with significant friction. These some system information, such as as slow convergence speed. In [10],
test cylinder systems are composed the values of some physical and/or an adaptive algorithm to adjust the
of the cylinder itself, with a mecha- material parameters, is unknown. learning rate has been applied to
nically guided and sealed cylinder speed up convergence. In [11], a
rod, the servo-valve and the distan- In [3], we have built the mathemati- new parameter identification me-
ce and/or force measurement. Our cal model of a 250kN type hydraulic thod based on neural network has
proposed black-grey-box modeling cylinder system with a hydrostatic been proposed by using the upper
process, based on system identifi- bearing and thus with negligible and lower bounds of the parame-
cation experiments, can build and friction within the cylinder sub sy- ters of an electro-hydraulic load
parametrize an accurate model of stem. Based on a grey-box mode- simulator. In recent years, several
the dynamic properties of these ling approach, we have derived an researchers have used neural net-
cylinder systems to enable sufficient accurate system model which we works in combination with genetic
actuator knowledge as demanded reduced to a low order model wi- algorithms. They have realized an
by high performance but robust test thout significant loss in the quality optimal design of the topological
control algorithms in dynamic mul- of the system dynamics description structure and the weighting para-
ti-axial test rigs. for most of our applications. The re- meters of the neural networks [12].
sults obtained from both simulation A new kind of genetic algorithm has
■■ 2 Related Work studies and real world experiments been applied in a parameter identi-
show that this method of mode- fication scheme of the digital elec-
Based on a brief discussion of the ling and simplification is effective tro-hydraulic governing system for
modeling and identification proces- for test cylinders with negligible a steam turbine [13].
ses for servo-hydraulic systems, the friction. This modeling process fails
work relevant to the content of the when significant friction phenome- All these identification algorithms
paper in hand will be addressed in na are appearing in a cylinder sy- can be used mainly in single-input
this section. stem, such as from mechanical gu- single-output (SISO) systems. With
iding elements and rubber seals. the increase of the number of con-
2.1 System Modeling trol axes and thus increased control
2.2 System Identification dimensions of the system, suitable
In general, to obtain a mathematical identification algorithms are much
System identification is the method more complex. Hence it is difficult
model of a certain dynamic system,
for estimating the structure and to apply identification algorithms
three kinds of modeling methods
parameters of a dynamic system in multi-input multi-output (MIMO)
are possible for deriving a proper
based on the information from the systems. Consequently, it is desi-
mathematical system description
input and output signals of the re- rable and in some cases necessa-
[1,2]:
lated systems. After 50 years of de- ry to develop simple and effective
velopment and research in the field identification algorithms which are
If we know all the relevant informa-
of system identification of electro- suitable for MIMO systems. Several
tion describing our system, such as
essential principles for the MIMO
its physical structure, the operation -hydraulic servo systems, several
system identification have been in-
mechanism, important material pa- powerful identification methods
troduced in [14]. Furthermore, the
rameters and so on, we can use the have been developed and success-
principles and methods of the sy-
white box modeling method based fully implemented [4,5,6].
stem identification are also impor-
on a purely analytical system de- tant for technicians and they can be
scription. In [7], Jang has applied adaptive ne-
found in [15].
twork algorithms to realize system
The second method is the so-called identification. His algorithm has a
black box modeling method, appli- self-learning function, neverthe-
■■ 3 Modeling and
ed when we have little or even no less, the complexity of the obtai- Identification of Servo-
knowledge about the system and ned overall control system grows Hydraulic Cylinder Systems
we can only use experiments with exponentially when increasing the
known input signals to estimate a number of model dimensions of the An electro-hydraulic servo system
mathematical system description system. Zhang has proposed a Fu- is a fully integrated system whi-
from the measured system respon- zzy-Tree algorithm [8,9] which has ch includes a digital control unit,
se. some advantages, such as fast cal- a hydraulic actuator and the assi-
culation and high precision. Many sting components, such as cabling,
The third method is the grey box research works on identification al- mechanical connections and so on.
modeling method: we have good gorithms based on neural networks When considering the control unit,
information about the system dyna- prefer BP feed forward neural net- we can in general distinguish bet-
mics, such as its physical structure works, however, these algorithms ween valve controlled systems and
QL = K q xv − K c pL (3)
Figure 1. Principal structure of a valve controlled symmetrically operating with K q being the flow gain and Kc
cylinder system with attached position and force sensor, sealing rings and
mechanical guidance elements. being the flow-pressure coefficient.
pump controlled systems. Throug- tions are then combined to obtain Continuity equation of the fluid
hout this paper, we only consider a the full system model as the final
valve controlled servo hydraulic cylin- step in our modeling process. According to [16], the continuity
der system operated in position con- equation of the fluid is given by
trol, composed of the cylinder struc- 3.1.1 Modeling of the Cylinder
ture, the servo valve and the sensor dy V dpL
unit. These key components can be QL = A + Ct pL + t (4)
The cylinder model is built from dt 4 K dt
treated as individual sub-systems of three basic mathematical relation-
the overall cylinder system, and con-
ships: the pressure-flow equation
sequently, we derive the correspon- where A is the piston working area
of the control valve, the continuity
ding mathematical submodels and of the cylinder, y is the displace-
equation of the fluid and the pres-
then combine them into the overall ment of the piston rod, Ct is the
sure-load equation.
cylinder system model during the leakage coefficient of the cylinder,
proposed system modeling process. Vt is the whole volume of cylinder
Figure 1 shows the principal structu- Pressure-flow equation of the
chambers, and K is the bulk mo-
re of a typical symmetrically opera- control valve
dulus of the used oil. Note that we
ting servo-hydraulic cylinder system. do not consider the oil volume wi-
The flow through the valve orifices
thin the pipelines and the servo-val-
3.1 Modeling of Test is usually described by the orifice
ve since with very short pipelines
equation with the relationship bet-
Cylinder Systems
ween the valve spool position xv
this oil volume can be neglected
when compared to the oil volume
The first step to sufficiently descri- and the load pressure pL [16]:
inside the cylinder chambers.
be the dynamic characteristics of
an electro-hydraulic servo control QL = cv xv ps − pL (1)
Pressure-load equation
system is to obtain an accurate
mathematical model of the system where cv is the flow coefficient and The pressure-load equation can be
dynamics. Without a proper mathe- ps is the supply pressure. We assu- written as:
matical model it is difficult or some- me that ps is constant.
times even impossible to achieve d2y dy
an acceptable control performance QL has a nonlinear relationship with ApL = m 2
+B + Km y + F (5)
dt dt
for the overall rig control for ensu- xv and pL , and consequently, we
ring high test quality–especially for
have to linearize it by introducing where m is the load mass (including
dynamic test applications.
the first order Taylor series expansi- piston rod), B is the kinetic dam-
As the first step in the derivation on to Eq. 1 to obtain ping coefficient, K m is the spring
of the overall model of a position- stiffness of the elastic load and F is
-controlled servo-hydraulic cylinder Q = Q + ∆Q = Q + ∂QL ∂QL
L L0 L L0 xv 0 , p L 0 ∆xvthe
+ load force.
x , p ∆pl
system, the analytical models of the ∂xv ∂pl v0 L0
test cylinder, the servo valve and the (2) Applying the Laplace transform to
involved
∂QL ∂QL
QL = QLsensors are derived sepa-
0 + ∆QL = QL 0 + xv 0 , p L 0 ∆xv + x , p ∆pl Eq. 3, Eq. 4 and Eq. 5 leads to the
∂xv
rately. These sub-system descrip- ∂pl v 0 L 0 following equations
Figure 2. Schematic system diagram: The cylinder system (plant) consists of the servo valve and the test cylinder
structure (including its hydraulic-mechanic components).
Figure 3. Test cylinder system, including the cylinder unit with the included position sensor, the directly attached
servo valve and the digital cylinder controller unit.
ted as part of the real time cylinder digital cylinder controller. cutoff frequencies of fl=150Hz and
control system (SPH SmartControl). fh=450Hz.
The overall test system is thus com- 3.2.1 Input Signal Design for
posed of the PID controller, the ser- Identification We assume that the cylinder system
vo valve and the 40kN test cylinder can be considered linear, thus we
(with significant friction), as shown In the system identification pro- superimpose the low frequency si-
in Figure 2 for our single-channel cedures, two types of suitable co- gnal on the high frequency signal
control system (the sensor unit is lored noise signals are used as re- for identification and verification.
considered ideal, as discussed pre- ference setpoint signals: a) zero
viously). Figure 3 provides a pho- mean white Gaussian noise, 1st or- 3.2.2 Black-Grey Box Model Iden-
tograph of the used real world test der low-pass filtered with a cutoff tification
system – the test cylinder (as part of frequency of fc=10Hz, and b) zero
a multi-axis test rig) with the atta- mean white Gaussian noise, 6th order The grey box model identification
ched servo valve and the connected bandpass filtered with low and high method is not directly applicable
to our cylinder system composed To obtain an accurate mathemati- (the cylinder system) by evaluating
of the 40kN type cylinder and the cal description of our test cylinder the ratio of the obtained cross spec-
G761-3003 servo valve because the system, we combine the black box tral densities Pry = Pry / Pru . Hz
value of friction is still unknown. We model identification method and
When comparing the frequency re-
cannot estimate the initial values of the grey box model identification
sponse of the inaccurate white box
the unknown system parameters method into the so-called black-grey P P
from Eq. 13 accurately, but precise box model identification. model estimateryof the ruwhole system
values are needed for the grey box with the experimentally obtained
model identification (as demanded Black-Grey Box Model Identifica- frequency response
ω sv estimation from
K
the above mentioned process, we
sv
by identification routines, such as tion
the System Identification Toolbox in can tune the values of the uncertainm = 6.07kg
Matlab [19]). The principle of the proposed black- parameters ωh and ξ h from the
grey box model identification me- white box model until the two
Consequently, we apply a combined thod is shown as a process chart in K sv
frequency response curves roughly
black-grey box model identification Figure 4 and is described as follows: Y ( s) A
fit together on an amplitude = basis.
method. First, we build a white box firstly, based on the hydraulic the- I ( s ) s 2
2ξ
This enables us to obtain a proper s( 2 + h
s + 1)(
model of the whole system, i.e. we ory, we build the grey box model ωh ω
initial theoretical model, including ωh
roughly estimate the values of the of the whole system. According to
unknown parameters in Eq. 13 ac- valid estimates for a) the initial valu-
our practical expertise, we include
cording to our experience. Then, we es of some unknown parameters for
pre-estimates for the values of the
apply the black box model identi- the subsequent grey box model
unknown parameters as rough esti-
fication procedure which improves identification of the system and b)
mates (some of the values may have
the white box model by adjusting big errors) so as to achieve an ina- the initial values of the still unkown
some model parameter estimates. ccurate white box model. system parameters in the open loop
Using the obtained improved mo- Moreover, after designing the se- transfer function (Eq. 13). Then, the
del as a new initial model, we apply tpoint signal r , the signal is applied standard grey box model identifica-
the grey box model identification so tion method called prediction error
to the controlled servo-hydraulic
as to derive an overall system model method (PEM) is used to derive the
cylinder system, delivering the ou-
with satisfying accuracy. These steps final mathematical model of the
tput cylinder rod position signal y
are explained in more detail below. overall electro-hydraulic servo sy-
as the system response to our input
signal. Additionally, we also acquire stem.
Black Box Model Identification
the applied control variable u , i.e.
Based on the measured system in- the output signal of the PID control- ■■ 4 Results
put signal (position setpoint signal) ler. Then, the black box model iden-
and the measured output signal tification method (PSD estimation When utilizing the proposed black-
(cylinder rod position), we can cal- from measured system signals) is grey box modeling for identifying the
culate the power spectrum density used to obtain the frequency re- dynamic model of our cylinder system
P P =P Hz
Hz
/ Psystem in a non-para- with friction, 3
we/(s⋅obtain
mm3 /(s ⋅ A) very accurate
A)
ryry= Pryry/ Pruru
(PSD) of the plant. This procedure is sponse of the
a kind of a nonparametric black box metric representation: we calculate results, at least for the relevant ope-
model identification method since the cross power spectral densities rational bandwidth of such cylinder
types ranging from 0 to 100 Hz.
we use a non-parametric PSD esti- PryPryandPP
ruru
with Welch’s method in
mation (periodogram averaging,
Welch method). The result of this Matlab, which enables us to derive In order to verify the identification
black box model identification al- the frequency response of the plant results, a suitable Simulink system
ωωsvsv KKsvsv
lows us to improve the white box
model with better estimates of the mm==66 .07kg
.07 kg
values of the unknown parameters,
such as ωh and ξ h .This step is es-
sential for the following grey box KKsv sv
model identification in order to ob- Y Y( s()s) A A
== 2 2
tain accurate identification results. I (Is()s ) s s 2ξ2hξ h ss
s (s ( 2 2+ + s + s+ 1)(1)( ))
ωωhh
ωω
hh ωω +1
sv sv+ 1
Grey Box Model Identification
Table 1. Comparison of the determined values of certain parameters of the cylinder system
Pry PryP Pruwhich are obtained by
ru
the developed system identification process
Systems
( Hz ) ξ h ωHz ω/(ssv ⋅K K sv ( m 3 /(s
Pry = Pry / Pru Pry =PryPry=ω
/PPry / Pru
h ru sv (3Hz
m )A)
sv A) ⋅) A)
m ⋅3 /(s
Cylinder system with 19 l/min servo valve m =m = 6kg
6.07 .07kg
343 0.9766 160 0.0114
(type Moog G761-3003)
Cylinder system with 38 l/min P
servo
ry Pru
valve Pry PryPru Pru
343 1.0012 160 0.0230
(type Moog G761-3004) K sv K sv
Y ( s) Y ( s) = A A
results for both cylinder systems, ω =
ωwesv
experiments sv
and Kidentification re-
ω
K sv with a differentsv sv K sv M.F. I and
( s ) I Zulfatman:
( s ) s 2 s 2
“Open-lo-
2ξ 2ξ h s + 1s)( s )
observe that only the servo-valve sults system valida- op and s ( 2s (+ 2 h+recursive
closed-loop s + 1)( )
gain parameter has been increased m = 6.07
te the universality of kg
m =m
our proposed
6.07
= 6kg
.07kg ωh ωhωh ωh ω sv +ω 1 sv + 1
identification of an electro-hy-
by a factor of 2, as we have expec- method. draulic actuator system,” 2010
ted. IEEE Conference on Robotics
Future research will K sv focus on theK sv K sv Automation and Mechatronics,
■■ 5 Conclusion development
Y ( s) YA ( s) Y ( s)
of identification and A A page 285-290, 2010.
= strategies = =2
control
I ( s) s 2 I
2ξ h ( for
s ) multi-axial
I ( s ) s s22 ξ h 2ξ[7] Jang
s s sJ.-S. R.: “ANFIS: Adaptive-
In this paper, we derive the mathe- s ( 2 + test
servo-hydraulic + 1)(s ( s2(+) 2 + s + h1)(s + 1)( )
sapplications. )
ωh ωh ωω ω
sv h+ 1ω h h ω h
ω sv-network-based
+ω1sv + 1 fuzzy inferen-
matical model of an electro-hydrau- ce system,” IEEE Trans. on Sy-
lic servo system utilizing a novel References stems, Man and Cybernetics,
black-grey box model identification Vol. 23, page 665-685, 1993.
method for the servo-hydraulic test [1] Khamesee M.B., Craig David G.: [8] Zhang J.G., Mao J.Q., Dai J.Y.
cylinder systems with significant “Controller design using white and Wei K.H.: “Fuzzy-tree mo-
friction. The identification method box and black box methods del and its application to com-
performs well for cylinder systems for a magnetically suspended plex system modeling,” IFAC,
with some unknown physical pa- robotic system,” 2005 IEEE In- page 1-12, 1999.
rameters that cannot be estimated ternational Conference on Me- [9] Zhang J.G., Mao J.Q., Xia T. and
accurately prior to the identification chatronics and Automation, Wei K.H.: “Fuzzy-tree model
process. Simulation results and their Vol. 3, page 1346-1350, 2005. and its applications to com-
comparison with measurement data [2] Zaker B., Gharehpetian G.B., plex system modeling,” ACTA
from real world experiments show Mirsalim M. and Moaddabi N.: AUTOMATICA SINICA, Vol. 26,
that the developed identification “PMU-based linear and non- page 378-381, 2000.
strategy is satisfying in terms of de- linear black-box modeling of [10] Pralos A.G., Fernandz B., Atiyz
livering an accurate model of the power systems,” 2013 21st Ira- A.F. and Tsai W.K.: “An accele-
cylinder system dynamics. Further nian Conference on Electrical rated learning algorithm for
multilayer perception net- [13] Dai Y.P., Deng R.G., Liu J., Sun K. [17] Sun Y.S., Jin B.Q. and Xiong
work,” IEEE Trans on NN, Vol. 5, and Wang Y.: “Study on para- X.Y.: “Research united simulati-
page 493-497, 1997. meter identification for steam on of electro-hydraulic servo-
[11] Zhang B., Zhao K.D. and Sun turbine DHE governing system -proportional value controlled
F.Y.: “Neural network parame- based on genetic algorithm,” cylinder servo position sy-
ter identification of electro-hy- Proceedings of the CSEE, Vol. stem,” Hydraulics Pneumatics
draulic load simulator,” ACTA 22, page 101-104, 2002. & Seals, Vol. 4, page 38-42,
AERONAUTICA ET ASTRONA- [14] Manring N.D.: Hydraulic Con- 2009.
UTICA SINICA, Vol. 30, page trol Systems, Hoboken, NJ, USA: [18] Moog, Inc.: G761 Series Ser-
374-379, 2009. John Wiley & Sons, Inc., 2005. vovalves, Technical reference,
[12] Ling S.H., Lam H.K. and Leung [15] Ljung L.: System Identificati- East Aurora, NY, USA: Moog,
F.H.F.: “A variable-parameter on: Theory for the User, 2nd ed. Inc., http://www.moog.com/li-
neural network trained by im- Upper Saddle River, NJ, USA: terature/ICD/g761seriesvalves.
proved genetic algorithm and Prentice Hall, 1999. pdf, Nov. 2013.
its application,” Proceedings of [16] Jelali M. and Kroll A.: Hydraulic [19] Ljung L.: System Identifica-
International Joint Conferen- Servo-system: Modeling, Iden- tion Toolbox User’s Guide,
ce on Neural Networks, page tification and Control, London, R2011b, Natick, MA, USA: The
1343-1348, 2005. Great Britain: Springer, 2003. MathWorks, Inc., Sep. 2011.
Uporaba servohidravličnih sistemov je ključnega pomena pri napravah, namenjenih preizkušanju trdnostnih
lastnosti in utrujenosti materiala tako strojnih delov kot konstrukcij. Da lahko zagotovimo visoko kakovost
testiranja in dobro ponovljivost rezultatov, je potrebno uporabljati robustne strategije vodenja takšnega
sistema. To je še posebej nujno v primeru večosnih testnih sistemov, kjer uporabljamo večje število sinhrono
delujočih servohidravličnih valjev. Za učinkovito uporabo primernih regulacijskih algoritmov so potrebne
natančne informacije o dinamičnih lastnostih aktuatorjev in celotne testne naprave.
Prihodnje raziskave bodo usmerjene v razvoj metode za identifikacijo in regulacijske strategije, primerne za
večosne servohidravlične sisteme.
Ključne besede: elektrohidravlični servosistem, servohidravlični valj, test utrujanja materiala, trenje, mode-
liranje, identifikacija, simulacija