You are on page 1of 13

HIGH PRECISION CONTROL SYSTEM FOR HYDRAULIC CYLINDER

R. Amirante, A. Lippolis, V. Marzullo

Politecnico di Bari, D.I.M.eG. – Sez. Macchine ed Energetica– via Re David, 200 Bari, Italy

Abstract

The aim of present study is to develop a precision driving system actuated by a hydraulic cylinder driven by a proportional
four way valve. In order to optimise the design parameters and to study the interaction between the proportional valve and the
cylinder, numerical simulation analysis as well as experimental test are performed. A comparison between numerical results
and experimental data is presented. The experimental data have been taken on a test rig located at the Politecnico di Bari.
First the frequency response of "proportional valve-cylinder" system is experimentally obtained. The attention of this analysis
is also focused on the ratio between the cylinder natural frequency and valve one.
The achieved results match reasonably well the literature data as well as the numerical results by means of a simulation code.
Typically the commercial valves are characterized by a critical dampening-factor so that they do not show any resonance
frequency, while the cylinders show relevant resonance phenomena. When combining the two components, it is crucial to
know the ratio of their natural frequencies. In particular, if the valve is characterized by a lower natural frequency, its behaviour
prevails so that no resonance peak can be observed. On the contrary, some resonance phenomena can be observed in all the
other cases. It is very important that all these phenomena are taken into account to realise a precise and accurate control of a
hydraulic axle.
Subsequently the attention is focused on the control technique. Three different control methods are presented:
1. a PID technique;
2. an expert control technique based on the valve-cylinder system characteristics;
3. an adaptive “One-Step-Ahead” control technique.
The final goal is to control by means of a hydraulic control system a horizontal machine center.

1 Introduction

The practice of hydraulic axes has became very wide-spread in many applications. Usually hydraulic axes are
more flexible than classical axes with ball-screw electrically-controlled, and also they can produce strong
accelerations. The main difficulty is the physical system control, just because the hydraulic systems are very much
not linear.
The present paper focalises its attention on a closed loop control on hydraulic system characterized by a cylinder
controlled by a four way valve; the cylinder position is measured by means of an optical linear encoder.
Usually a rapid dynamics of regulation valve means an high quality control for a closed loop system
hydraulically driven, but this opinion can be deceptive. The importance to analyse the whole system (the
regulation valve and controlled hydraulic cylinder) is underlined in this paper.
After a theoretical control analysis, the simplest classical control algorithm, the Proportional-Integral-
Differential (PID) technique, is tested on the hydraulic system. It is also proposed an adaptive control algorithm,
the One-Step-Ahead and an expert control technique, based on the knowledge of valve-cylinder system
characteristics, to control piston displacement and velocity. The One-Step-Ahead adaptive algorithm is very
interesting because does not require any previous knowledge of the controlled system, i.e. of the transfer function.
Therefore, this algorithm is very useful and suitable for experimental applications in which the system is a black
box controlled by a closed loop. This technique has shown very good in controlling of hydraulic transmission
tested in a pervious paper (Amirante R. et al., 1998).

2 Theoretical analysis.

System model, theoretical aspect.


In the present paper, the behaviour of an hydraulic system is analysed, this one consists of a differential cylinder
driven by a four way valve (fig.1).

1
In order to simulate the dynamics of directional
proportional valve spool, the behaviour of the
valve is approximated by a second order system.
This approximation can be considered acceptable;
only the hypothesis that the damping effects are
proportional to velocity is not precise but
opportune in order to simplify the model. The
characteristic curves investigation, concerning
directional proportional valves, underlines that the
most of commercial valves have a near critical
damping coefficient value. Above reported
considerations concern the spool dynamics, but the
relationship between the flow rate and the
electrical signal is much more complex due to:
a) the relationship between the spool
displacement and opening area: in
Fig. 1: Hydraulic system, cylinder and valve
particular, a linear function (e.g.
rectangular opening area) simplifies the control system, while parabolic function or, in general, not linear
function allows a better control at low flow rates;
b) “the lap”: when the electrical signal is near to zero, the behaviour of the valve is strongly dependent on
this parameter. There are commercial valve with high dynamic performances, called “zero lap”, these
valves, usually, are characterised by a tolerance, between spool and sleeve edges, near to 0,5% in
relation to the total spool stroke. In this case, theoretically, there is not discontinuity between electrical
signal and flow rate near to zero; but, in reality, a discontinuity between these parameters always exists
because the tolerance above reported can not be preserved, due to pilot edge wear in time. It is important
to underline that these valves (zero lap) are in the minority, usually under lap valve are used for position
control, in order to obtain the best promptness, while overlap valves are used for velocity controls, these
valves are able to reduce internal leakage improving the hydraulic seal in normal position.
c) the relationship between the valve pressure drop and the flow rate is not linear.
Moreover, hysteresis, repeatability and inversion error make more complex the control.
The cylinder system is doubtless more complex than valve system. In this case the hydraulic fluid, with its
compressibility, is the elastic element of the system, while damper effect is due to different causes: friction
between piston and cylinder, internal leakage and, principally, the opening behaviour of the regulation valve. To
explain the last effect the free oscillations of cylinder piston can be analysed. For simplicity, a synchronous
cylinder with the same capacity volume V in both chambers is considered; the behaviour of a directional control
valve is simulate by a throttle on both chambers. With obvious meaning of symbols, the forces equilibrium of
cylinder can be written:
2x
( p1  p 2 ) A  M 2 (1)
t
the compressibility law of the hydraulic fluid, with reference to left and right chambers, is:
V V
p1  E 1 and p 2  E 2 (2)
V V
where E is equivalent (fluid and pipe) bulk module (Amirante R. et al., 1997). The time derivative of above written
equations is:
p1 E V1 E  x  p2 E V2 E x 
    A  Qt1  and      A  Qt 2  (3)
t V t V  t  t V t V  t 
where Qt1 and Qt2 represent the external leakage crossing the left and right throat orifice area. Usually flow
behaviour in the hydraulic valves is turbulent, but in order to simplify the calculations and to obtain a linear
equation, the flow behaviour is supposed laminar and then can be written
Qt1  K ( p1  p10 ) and Qt 2  K ( p2  p20 ) (4)
where K is an appropriate proportional constant, while p01 and p02 are pressure values outside the throttles.
Replacing eqs. (4) in eqs. (3) these expressions can be written:
p1 EA x E p 2 EA x E
  K ( p1  p10 )   K ( p 2  p 20 )
t V t V t V t V
(5) (6)
p1 E EA x E p 2 E EA x E
 K p1    K p10  K p2    K p 20
t V V t V t V V t V
Deriving (1) regards to time and adding it the same eqn. (1) multiplied by an appropriate constant, the following

2
equation is obtained:
 3 x E 2 x   p E   p E  2 EA 2 x EA
M  3  K 2   A 1  K p1    2  K p 2     K ( p10  p 20 ) (7)
 t V t   t V   t V  V t V
if the piston velocity c is introduced:
 2 c E c 2 EA 2 EA
 K  c K ( p10  p 20 ) (8)
t 2
V t MV MV
2 EA 2
The eqn. (8) is the classical equation of a second order dynamics system, in which  n  and a damper
MV
factor  is:
E E
K K
V V M E K
    (9)
2 n 2 EA 2 8V A
2
MV
In these formulas, to simplify the analytic treatment, some approximations have been considered: for example
right port volume equal to left port volume, the throat orifice areas are the same (and consequently K constants are
the same for both throttles), the flow rate trough throttles has been considered laminar.
The analytic result above reported can be physically explained: with closed throttles, if the piston was submitted
to a initial displacement from equilibrium condition and let it to oscillate free, the piston would transform
alternatively elastic energy (depending on compressibility of the hydraulic fluid) into kinetic energy of piston and
vice versa. On the contrary the elastic energy is dampened by the throttles presence.
The directional control valve acts like the throttles above described, in fact the volumetric flow (through the
same valve) can be ideally divided into two components: the first is the steady component and takes into account
of the valve opening conditions and valve supply pressure, while the second one takes into account the volumetric
flow rate oscillations due to pressure oscillations. This last one has a dampening effect, in fact when pressure level
rises (and consequently kinetic energy grows) this one induces a volumetric flow rate reduction trough the valve
similar to throttles leakage.
The cylinder system, in the most common practical applications, acts like an under dampened system, which has
strong resonance overshoots near to natural frequency n: for examples, the damper factor value calculated by eqn.
(9) for used experimental device is equal to 0,2. So the damper factor is function of valve operating conditions
(coefficient K in eqn. (9) ), then it is not a characteristic of cylinder but it is time dependent: the damper factor is
very near to zero when the valve is near the normal position, while it is increasing by the valve opening. Using
simple formulations and some approximations, it is demonstrable that dampening effects due to friction and
internal leakage in cylinder system are negligible in comparison to the dampening effects due to directional control
valve.
In order to understand the cylinder-valve system behaviour and to optimise the system control, it is very
important the knowledge of  parameter, defined as ratio between directional control valve natural frequencies and
cylinder natural frequencies. With reference to  parameter, it is possible individuate four different situations
(Sanelli M., 1997):
a)   1, it’s the most common case, in which the resonance peak of the cylinder is very reduced depending
on the high damper value of regulation valve near to the cylinder natural frequency. In this case it’s
suitable to use a PD control technique, to improve the regulation amplitude recovering the valve
dampening effect.
b)   1, in this case the valve natural frequency is smaller than cylinder natural frequency and so the valve
produces a little reduction of resonance amplitude due to the cylinder. In this case can be not necessary to
use the derivative component, while a proportional control, with a suitable amplification, is enough.
c)   1, the dampening power of the valve, near to the frequency in which the cylinder resonates, it is not
important, for this reason it is necessary choosing a control that allows a strong reduction of resonance
phenomena. It is necessary to use a PT1 control with limit frequency equal to 50% about the frequency of
the cylinder.

3
d)   1, in this case the natural frequency of the regulation valve is so greater than the resonance
depending on the cylinder. This one is not dampened and also a PT1 control is useless, it is suitable using
some controls able to produce artificial dampening effects, like that defined “State-Observer”.
In the present paper, after a complete
theoretical analysis, only the case a) will be
analysed experimentally in order to
demonstrate the good agreement between the
theoretical forecasts and the experimental
results.
In order to schedules the experimental
tests, a numerical simulation has been
performed a priori by a commercial code
named ITI-SIM. The code offers the
possibility to model engineering system and
in particular hydraulic plants. In fig. 2 the
block diagram used to simulate the
experimental plant is reported. The ITI-SIM
code uses the modelling by signal block
diagrams: the cylinder block has
Fig. 2: Numerical simulation block diagram
intrinsically the possibility to fit the
theoretical model above described. While the valve block do not permit to simulate the spool dynamics (spool
position is algebraically dependent to the valve input signal), but at every signal blocks some personalised macros
can be added to simulate particular systems. In the present case the macros added to the valve can simulate the
spool dynamics in according to second order system above illustrated. Numerical results have shown a good
agreement with the performed experimental test, but for conciseness reasons, only an example will be presented in
the following paragraphs.

Control theoretical aspects.


Before to illustrate the experimental results, in this paragraphs is theoretically analysed the system stability. With
reference to a cylinder position control, the block scheme equivalent to the experimental plant is presented in fig 3
(Marro G., 1999).
Kd
R + C
Gc(s)   2
 2

12v s  s  12c s  s 
-  2
v v   2
c c 
 

1
s
Fig. 3: Control block scheme - position control
where:
R(t) is the target position,
C(t) is actual piston velocity (controlled variable)
Gc(s) z-Transfer Function of controller
Ks is transfer function of valve-cylinder system; in agreement with the model
illustrated in previous section, both cylinder transfer function and valve transfer
 s s2  s s2  function is described by a second order system dynamics. Where c and v are
12v  2 12c  2 
 v v  c c  the natural frequencies of cylinder and valve, c and v are damper factors and Ks
is the proportional constant concerning valve-cylinder system. If the system were
linear, Ks would be the piston velocity connected with an input signal equal to 1
Volt. Since the system is not linear, Ks must be interpreted as the derivative of the
velocity as regards input signal, calculated with regard to the operative point.
According to the physical system, the controlled variable C(t) is the piston velocity; in fact the valve input signal
causes a valve spool displacement and so it fixes the volumetric flow rate across the valve, finally the piston
velocity is proportional to the volumetric flow rate. The feedback signal is measured by means of an optical linear
encoder, then the feedback variable is the piston position and the feedback transfer function is an integral type
function, for this reason the valve-cylinder system is an intrinsic integrator system and so the integral effect can be

4
neglected in the used control technique.
The previous block scheme (see fig. 3) is dedicated to the position control, but this one is, theoretically,
equivalent to the velocity control. In fact this scheme, using the reduction laws of block schemes, can be
transformed in an equivalent scheme (see fig. 4) with unitary feedback to represent velocity control scheme. In this
case, the input variable is the derivative of piston displacement, or better piston velocity, and so the control
transfer function concerning to velocity control must be the integral of the control transfer function concerning
position control.
Kd
sR + Gc ( s ) C
  2  2

s 12v s  s  12c s  s 
-  2
v v   c c2 
 

1
Fig. 4: Control block scheme – velocity control

As regards to linearized block scheme reported in fig. 3, it’s possible to analyse the stability of system using
Routh Criteria: supposing a proportional control Gc(s)=Kp, if  is the ratio between v/ c and replacing c with  ,
the transfer function equivalent to closed loop system can be written:
K s s  2 4
Ge ( s )  5 (10)
s  2 ( v    c ) s 4   2 ( 2  4 v  c  1) s 3  2 3 ( v    c  2 ) s 2   2 4 s   2 4 K s K p
imposing that there are not real positive roots, by Routh Criteria, after simple calculations it can be observed:
a) growing Kp constant, the valve-cylinder system becomes instable,
b) the most important parameter for stability control is the product of every constant on the direct and
feedback line, in the present case the product constant is Kp Ks . In particular it’s shown its ratio with
the frequency of cylinder ,
c) the stability depends on  ratio
In fig 5 it’s shown the stability parameter Kp Ks/  as function of  (assuming c = 0.05 and v = 0.9) for
proportional control and some others PD control. It is evident, for an assigned cylinder frequency, the stability
parameter, initially, grows like linear with the valve frequency ( parameter), as to as it arrives to a max value for
=0.7 and then it decreases with valve frequency.
The same analysis can be carried out for the PD control, in this case Gc(s) = Kp (1+d s)
K s K p s (1   d s ) 2 4
Ge ( s )  5 (11)
( s  2 ( v   c ) s   (  4 v  c  1) s 3  2 3 ( v    c 2 ) s 2   2 4 s)   2 4 K s K p (1   d s)
4 2 2

the stability parameter Kp*Ks/ now is function of time constant d , or in particular of adimensional time constant
t=d  . In fig 5 there are four curves relative, respectively, to values 0.5, 1, 2, 3 of a-dimensional time constant t.
From the diagrams analysis, it can be observed how for low  (valve natural frequency lower than cylinder natural
frequency) the derivative component improves the dynamics of valve-cylinder system because it compensates the
dampening effect of the valve and also it is opportune to stabilize the system. Vice versa, when  grows, the
derivative contribute produces a negative effect on stability of the system because it reduces max value of the
stability parameter. In the next fig. 6 is reported the stability parameter as function of adimensional time constant
t=d  , for an assigned value of  = 0.16 like equal to the experimental plant  value; is evident the stabilising
effect of derivative component.

5
2 0.9

1.8 P control 0.8
PD Control t=0.5
1.6
PD Control t=1 0.7

1.4 PD Control t=2 0.6


1.2 PD Control t=3
0.5
KsKp/c

Ks K p/c
1
0.4
0.8
0.3
0.6
0.2
0.4
0.1
0.2

0 0
0 0.5 1 1.5 2 2.5
0 0.2 0.4 0.6 0.8 1 1.2 1.4
t

Fig. 5: Proportional and proportional derivative control: Fig. 6: Proportional and proportional derivative
stability limits on  variation control: stability limits for the 
experimental value (0,16)

3 The experimental plant

In order to validate theoretical analysis and perform experimental test, a test-rig is realized at the Department of
“Ingegneria Meccanica e Gestionale” of the Politecnico di Bari. The hydraulic scheme of circuit has been
reported in fig. 7, that is constituted by an open circuit and it is composed by:
- a 30 kW three-phase asynchronous electric motor. Its rotational speed can be regulated by means an electrical
inverter, within the range 03000 rpm in according with the pump speed range.
- a gear pump (63 cm3) which supplies hydraulic power. Its operation field is 7001500 rpm and its maximum
working pressure is equal to 350 bar.
- a directly operated pressure relief valve; by regulation of this
valve, the pressure value of 80 bar on the feed line is
established.
- a four-way valve, with a input signal range of -5  +5 Volt.
The valve has an overlap and when the input signal is in the
range 1 Volt the flow rate is zero.
- a differential cylinder, with a cylinder bore of 60 mm, a stroke
of 1 m and a piston rod diameter of 40 mm.
- flexible hoses (1 inch internal diameter) connect all the
components of the hydraulic circuit.
An automatic control system is used to electronically monitor
the behaviour of the hydraulic plant during the experimental tests,
such control system is managed by a personal computer equipped
by an Intel Pentium III processor and a 12-bit acquisition card,
which is characterized by a maximum sampling frequency of 1
MHz. Such a card manages 16 analogical acquisition channels,
two analogical output channel, two digital counter, 16 digital
Fig. 7: Layout of hydraulic system input/output channels and two counters TTL compatible. The
analogical input channels are been used to acquire pressure
signals, an analogical output channels has been used to send input
signal to four-way valve.
The cylinder piston displacement (the controlled variable) is measured by means of an optical linear encoder,
that produces two TTL compatible signals with 90 shift phase degrees, the pitch, corresponding to each time base
of square wave, is 8 m. The TTL compatible signals of optical linear encoder, normally, are acquired by the
counter; but sometimes the encoder signals are acquired on analogical channels, in this case piston displacement is
not been acquired in real time but the data are post processed.
The directional proportional valve displacement is varied by the analogical output signal of the data acquisition
card, thus allowing a rapid increase or decrease of internal orifices area. An electrical pressure transducer is linked
to the hydraulic circuit, to set up the direct operated pressure valve value.

6
4 Results and discussion

Regarding experimental tests, the first step has been to verify the dynamic performances of the proportional
valve: the experimental test-rig can not analyse the valve alone, but the whole valve-cylinder system. Taking into
account that the natural frequency of cylinder is near to 80 Hz, while the natural frequency of regulation valve is
near to 20 Hz, the weight of cylinder dynamics is negligible about the dynamics of the valve-cylinder system.
3
0.09
0.14

2.5 0.085
0.12

Proportional Signal (Volt)


2 0.08
0.1

velocity (m /s)
velocity (m/s)

1.5 0.075
0.08

1 0.07
0.06

Piston velocity 0.5 0.065


0.04
Proportional signal

0.06
0.02 0
100 200 300 400 500 600 700 800 900 1000 340 350 360 370 380 390 400 410 420 430 440
time (ms) time (ms)

Fig. 8: Cylinder velocity time history and valve input signal Fig. 9: An enlargement of cylinder velocity. The natural
(2 Hz). cylinder frequency (about 80 Hz) is shown

1 10 frequency (Hz) 100


0.15 3
0.00 150

0.14 2.8

-1.00
2.6 120
0.13
amplitude catalog features -
2.4 dump. factor 0,7
0.12 -2.00
Proportional signal (Volt)

amplitude - experimental data


Amplitude (db)

2.2 90

Phase (deg.)
velocity (m/s)

0.11
-3.00 phase - experimental data
2
0.1
1.8 phase catalog features - dump. 60
factor 0,7
-4.00
0.09
1.6

0.08 30
1.4 -5.00

0.07 Piston velocity


1.2
Proprtional signal
-6.00 0
0.06 1
100 200 300 400 500 600 700 800 900 1000
time (ms)

Fig. 10: Cylinder velocity time history and valve input signal Fig. 11: Bode diagram, catalogue and experimental data
(16 Hz). referred to first only harmonica after a Fourier
analysis

It is clear that this system can be considered linear only in a very little range. According to technical catalogue
details, the experimental Bode-diagram of the valve has been built using an electrical input signal equal to 50% of
max input signal (5 Volt) added to a sinusoidal oscillation with a fixed amplitude equal to 5% referred to max
input signal. About the sinusoidal signal at frequency of 2 Hz, in fig. 8 it’s shown the time history of velocity
piston and electrical input signal. The piston velocity time history (see fig. 8, 10) shows strong oscillations around
a sinusoidal law. These oscillations are not due to electrical noise or to numerical causes but they are mechanical
oscillations depending on piston natural frequency. In fact the cylinder displacement is measured by digital signal
of a linear optical encoder, so the background noise is not present, the only possible error is depending on the
encoder resolution (8 μm). Moreover acquiring the two output encoder signals by analogical channels it is possible
to reach a 2 μm resolution, corresponding to 90 shift phase degrees, by a posterior data processing. The velocity
measurement is obtained as a mean value on a 2 ms time interval, consequently the final error is approximately 1
mm/s. On the contrary, the amplitude of velocity oscillations (see fig. 9) is approximately 15 mm/s, one order
greater than velocity error. This fact confirms the mechanical nature of high frequency phenomena above reported.
The same error analysis can be carried out with time measurement reference: a 100 kSample/s analogical

7
acquisition is performed on the two digital signal (TTL compatible) and, for this reason, the time measurement
error is about 10 μs on 2 ms time interval. Consequently, the velocity percentage error due to time measure is very
low and equal to 0,5%.
Figure 9 shows an enlargement of these oscillations. There are about 8 oscillations in a 100 ms time interval,
therefore the oscillation frequency is about 80 Hz. This experimental frequency value is in good agreement with
the cylinder natural frequency computed by eqn. (8). In reality this value can be theoretically evaluated only by an
exactly knowledge of fluid compressibility and connected volumes. In figure 1, two long hoses (about 1.5 m)
between cylinder and valve are evident, in opposition with an efficient industrial application. But the presence of
these big connected volumes is finalized to approach the cylinder frequency to valve frequency, in order to put in
evidence the phenomena plotted in fig. 9.
If the valve-cylinder system were linear, these oscillations would be dampened and the final output signal would
be characterised by the same frequency of input signal. Vice versa these oscillations persist due to no-linearity of
the system. Although the amplitude of oscillations due to natural frequency of cylinder is little in comparison with
the input signal amplitude, this effect introduces another noise to no-linearity of the system and in particular to the
goodness of control.
The time history of velocity piston and electrical
input signal for a frequency input signal of 16 Hz
16
is shown in fig. 10.
14 On piston velocity signal, a Fourier analysis is
valve input signal 16 Hz applied and the greatest amplitude value is found
12 valve input signal 10 Hz
valve input signal 12 Hz
on the first harmonica. However another amplitude
10
peak is visible near the natural cylinder frequency
amplitude (mm/s)

(see fig. 12). The experimental Bode-diagram, in


8 comparison with catalogue data, is shown in fig.
11.
6
After all, it is important to notice that the
4 velocity has the amplitude value of first harmonica
equal to 20 mm/s, while the sinusoidal component
2
amplitude of input signal is 0,25 Volt. Therefore
0
the Ks value, introduced in Control Theoretical
0 20 40 60 80 100 120 140 160 180 Aspects section, is 80 mm/sVolt.
The hydraulic cylinder natural frequency n is
frequency (Hz)

Fig. 12: Velocity amplitude spectral analysis at different valve approximately equivalent to 500 rad/s and the
signal input frequency. stability parameter Kp Ks/  (see fig, 6), for the
proportional control, is equal to 0.27. Therefore
the bigger Kp value to avoid the instability is 0.27 500/80 = 1.6 Volt/mm. This Kp value is near to the value
experimentally obtained.

Cylinder control
In all the experimental applications, the gear pump rotates to a constant velocity of 1200 rpm, to be sure that the
pressure relief valve is always open. The pressure on the proportional valve inlet port is 80 bar (fixed by pressure
relief valve) and the cylinder is in the outward stroke. The piston position is determined by a optical linear encoder
with a linear step of 8 m.

Cylinder position control


Initially, the position control tests are performed with many different value of regulator coefficients.

8
600 1

Piston position
0.9
500 control error
0.8

0.7

displacement error (mm)


400
displacement (mm)

0.6

300 0.5

0.4
200
0.3

0.2
100

0.1

0 0
0 500 1000 1500 2000 2500 1800 1900 2000 2100 2200 2300 2400 2500
time (ms) time (ms)

Fig. 13: Cylinder position control with the target position Fig. 14: Cylinder position control error with the target
500 mm. position 500 mm.

In figure 13 the classical piston time response is reported: after a dead time of about 85 ms, due to valve response
time (about 40 ms) and hydraulic fluid compressibility, the piston reaches a constant velocity compatible with the
valve characteristic at the maximum input signal. Only near the final position, the input valve signal goes down
and the regulator effect is evident at 1750 ms.
In figure 14 the absolute control error is reported near target band (1800 ms ÷ 2500 ms), to evaluate the goodness
of this kind of control. With the chosen control parameters, the displacement time history is absolutely monotonic
and the final absolute error is about 0.4 mm.
Growing the proportional coefficient, for example Kp=1.2, a little overshoot over the target value is evident (see
fig. 15). After this displacement overshoot, the cylinder changes its direction and then stops, with a final error
about 0.3 mm.

120 2 60 4

1.5 3
100 50

1 2

80 40
displacement (mm)

displacement (mm)

0.5 1
error (mm)

error (mm)
60 0 30 0

-0.5 -1
40 20

-1 -2

20 10
Piston position -1.5 Piston position -3
error (mm) error (mm)
0 -2 0 -4
0 500 1000 1500 2000 2500 0 200 400 600 800 1000 1200 1400 1600
time (ms) time (ms)

Fig. 15: Cylinder position control error, by P technique, Kp Fig. 16: Cylinder position control, by P technique, Kp = 1.6
= 1.2 and target position 100 mm. and target position 50 mm.

In figure 16, the same kind of control is tested, but using a proportional coefficient Kp=1.6. The time history
presents dampened oscillations around target value. In figure 17 the system instability is evident and the piston
oscillations have a frequency equal to 8 Hz, while the amplitude is approx 6 mm. Moreover, the proportional
algorithm does not present a considerable static error, because valve/cylinder system is an intrinsic integrator to
control piston position. Consequently, an integral technique is not necessary in this application.
To confirm the behaviour whole valve/cylinder system dynamics, a comparison between experimental data
obtained and numerical data is shown in fig. 18 with Kp=2. As told above, the numerical simulation is obtained by
a commercial code (ITI SIM r. 3.3) and the simulation scheme utilized is reported in fig. 2. The behaviour of
directional proportional valve is approximated by a second order system with n = 20 Hz and =0.7. The
comparison between experimental and numerical data is very good, the same frequency and amplitude values can
be observed.

9
60
120 8

6 50
100

4
40
80

Piston position (mm)


displacement (mm)

error (mm)
30
60 0

-2 20
40

-4
10 Experimental data
20 Numrical data
Piston position -6
error (mm)
0
0 -8
0 100 200 300 400 500 600 700 800 900 1000
0 500 1000 1500 2000
time (ms)
time (ms)

Fig. 17: Cylinder position control, by P technique only Fig. 18: Cylinder position control. A comparison
(Kp=2.5) and target position 100 mm. instability control is shown between numerical
plant simulation and experimental data (Kp=2).
In the following figure 19 the classical PD technique is applied on the hydraulic system. On the base of
theoretical consideration reported on previous paragraph, the use of differential parameter in the control technique
allows to increase the proportional coefficient without producing instability.
In this figure, the time history of piston displacement and control error is shown. The control coefficient are:
Kp=1.5 and Kd=0.05, while the target position is 500 mm. The output displacement (controlled variable)
approaches very well the target position.
600 0.2
Cylinder velocity control
It is possible to hypothesize an appropriate law of
0.15
500
piston velocity (see fig 20), at the same of the
0.1
position control. The velocity control tests can be
400
performed with many different kind of regulator
displacement (mm)

0.05
percentage error

techniques. The attention is focused on three different


300 0
control methods:
-0.05
1. a PID technique;
200 2. a close loop expert control technique;
Piston position -0.1 3. an adaptive “One-Step-Ahead” control
Percentage control error
100
-0.15
technique.
In the first time, a proportional integral and
0 -0.2 differential controls is used; in fig. 21 the theoretical
0 500 1000 1500 2000 2500 3000 3500 4000
time (ms)
and experimental time sequence of piston
Fig. 19: Cylinder position control, by PD technique (Kp=1.5, displacement, the displacement error are reported.
Kd=0.05) and target position 500 mm. Initially, the error grow very fast because of the dead
time, due to valve response time and hydraulic fluid
0.12
compressibility. Then the displacement error is
0.1
approximately constant (approximately 1 mm) with
exiguous oscillations. Finally, in the deceleration face,
0.08 the time delay of the valve produces a recovery of the
piston velocity (m/s)

initial piston displacement error.


0.06 In fig. 22 the time sequence of piston displacement
is shown, characterised by an expert control technique
0.04
in which an approximate law between piston velocity
and electrical input signal is assigned based on the
0.02
experimental valve-cylinder dynamics knowledge. To
0
delete the little error obtained on this experimental
0 100 200 300 400 500 600 700 800 900 1000 control law, a closed loop is however adopted to
time (ms)
improve the control quality by a PID technique. So
Fig. 20: Target Cylinder velocity law the control error is less than 0.2 mm
Also, an adaptive control technique has been used.
In particularly, the One-Step-Ahead adaptive control technique has been used, this technique is very powerful in
the control of hydraulic transmission, as demonstrated in a previous paper (Amirante R. et al., 1998).

10
80 1.4 80 1.4

1.2 Target displacement 1.2


70 70
Piston displacement 1
1
60 60 Displacement error

displacement error (mm)


displacement error (mm)
0.8 0.8

displacement (mm)
displacement (mm)

50 50
0.6 0.6

40 0.4 40 0.4

0.2 0.2
30
30
0
0
20
20 -0.2
Piston displacement -0.2
10
10 Target displacement -0.4
-0.4
Displacement error 0 -0.6
0 -0.6
0 200 400 600 800 1000 1200
0 200 400 600 800 1000 1200
time (ms)
time (ms)

Fig. 21: Cylinder velocity control by PID technique Fig. 22: Cylinder velocity control by expert control
(Kp=10, Ki =1,5 Kd=0,05) technique based on the knowledge of valve-
cylinder system characteristics (Kp=10, Ki =1,5
Kd=0,05)

80 1.4 In figure 24, the One-Step-Ahead adaptive control


1.2
technique combines a parameter estimation algorithm
70
1
with the control scheme: the Least Squares algorithm
60
0.8
will be used for an on-line estimation of the linearized
displacement error (mm)

model parameters on the basis of the previous


displacement (mm)

50
0.6
sequence of the hydraulic system inputs and outputs.
40 0.4
Then the control scheme will be applied to the
30
0.2
linearized estimated model of the controlled system.
20
0
The controlled system can be generally represented
Piston displacement -0.2 in the state variable notations as [13] (Dambrosio L.,
10 Target displacement
Displacement error
-0.4 1994) (Goodwin G.C. et al., 1984) (Goodwin G.C. et
0 -0.6 al., 1984) (Bryson A.E., 1975):
0 100 200 300 400 500 600 700 800 900
time (ms) 
X  f  X, U  , (12)
Fig. 23: Cylinder velocity control by “One Step Ahead”
technique where X and U represent the vectors of the controlled
variables and of the control variables , respectively:

 X1  U 1 
X  U  (13)
System X   , U   .
2 2

 ...   ... 
   
Xn U n 
Parameter
Estimator If the model is linearized and this one is approx by means of forward
finite differences the time derivative at the left hand side, the following
matrix relation can be obtained:
Design
Calculation X t  A t  t  X t   t  B t   t   U t  t , (14)

The classical control systems can evaluate A and B on the basis of


Control the controlled system mathematical or empirical models, e.g., transfer
law functions. The present technique does not require a previous knowledge of
Fig. 24: Block layout of the One- such relationships. On the contrary, it evaluates the values of A and B on
Step-Ahead control techni- the basis of the input and output data referring to an appropriate past time
que. window. The Least Squares Algorithm in recursive form is used as the on-
line parameter estimator of A and B because of its simplicity and very high
convergence rate.
In order to apply the Least Squares Algorithm, Eq. (14) is rewritten in the following form:
X t   t  t   t  t , (15)

11
where:

 X 
 t   t  A t   t B t   t , t   t   t   t  . (16)
  U t  t 

The Least Squares Algorithm considers Xt and Ut-t as known terms and give appropriate relations to estimate
the unknown terms Ut-t. The following recursive formulae [10] are used to obtain ’t-t, which represents a
reasonable estimation of t-t (in the following the apex will be omitted for the sake of simplicity):

tit  ti2t 
Pt i 3t  t 2t
1 tT2t  Pt i 3t  t 2t

 Xi,t t  ti2t  t 2t ,  (17)

where:

Pti 3t  t2t  tT2t  Pti 3t


Pti 2t  Pti 3t  . (18)
1 tT2t  Pti 3t  t2t
i i
In Eqs. (17), (18),  t-t denotes the i th raw of the matrix t-t, while P t-3t indicates the covariance matrix of 
i th
t-t, and Xi, t is the i state variable.
The One-Step-Ahead adaptive control technique considers the system linearized mathematical model given by:

X t  A t  t  X t   t  B t   t  U t   t , (19)

or, alternatively, by:

X t   t t   t t , (20)

where At-t and Bt-t, obtained by means of Eqs. (17), (18), represent an estimation of the corresponding matrices
in Eq. (14). Similarly, t-t represent an estimation of the corresponding matrix in Eq. (15).
The target of the One-Step-Ahead adaptive control is to minimize the error tracking defined as:

 t  X t  X *t , (21)

where Xt* denotes the desired output at time t. Taking into account Eqs. (20) and (21), the error tracking
pertaining to the time t+t can be expressed as:

tt  t t  X*tt . (22)

The control system has to annihilate the error tracking. Accordingly, the following relation can be obtained from
Eq. (22):

t   t  X *t t . (23)

Taking into account Eqs. (16), Eq. (23) can be expressed as:

X * t  t  A t  X t  B t  U t . (24)

The One-Step-Ahead adaptive control assumes that the parameter matrices At-t and Bt-t remain unchanged from
the time t -t to the time t, i.e.:
A t  A t t ,B t  B t  t . (25)

Since the matrices At-t and Bt-t are estimated by means of Eqs. (17) and (18), the only unknown in Eq. (24) is
represented by the vector of control variables Ut:


U t  B 1t  t  X * t  t  A t  t  X t .  (26)

Eq. (26) provides the control laws, i.e., the hydraulic system input variables required to warrant that the state
variables follow the desired time sequence X*. Such a requirement is exactly satisfied for linear systems. For the
present non linear system, the previous requirement is only approximately fulfilled, being the error tracking
dependent on the time step t. As a consequence the matrices At-t and Bt-t must be continuously updated at each

12
time step. Eq. (26) also allows to infer that the number of controlled system outputs (state variables) must be equal
to the number of controlled system inputs (control variables).

5 Conclusions

In the present paper a theoretical analysis of a precision driving system actuated by a hydraulic cylinder driven
by a proportional four way valve is developed. The interaction between the proportional valve and the cylinder to
optimise the design parameters are considered by numerical simulation analysis and experimental test. The
experimental data have been taken on a test rig located at the Politecnico di Bari.
At beginning it has been analysed the dynamics of cylinder-valve system and so the frequency response of this
system is experimentally obtained. This analysis is very important just because all these phenomena are taken into
account to realise a precise and accurate control of a hydraulic axle. In fact, regarding experimental tests above
reported, it is possible to observe that the piston velocity time history has shown strong oscillations around a
sinusoidal law and these oscillations are due to mechanical oscillations depending on piston natural frequency.
The experimental Bode-diagram, in comparison with catalogue data, is reported referring to first harmonica after
a Fourier analysis and same differences are evident.
Subsequently, the attention is focused on the position and velocity cylinder control technique. In particular three
different control methods are presented: a PID technique, an expert control technique based on the valve-cylinder
system characteristics and an adaptive “One-Step-Ahead” control technique. After a theoretical stability analysis
which allows to evaluate the best values of control coefficients, the different results are reported, to evaluate the
best performances. The precision obtained is very interesting also with the innovative One Step Ahead control
technique. In the following the final aim is to control by means of a hydraulic control system a horizontal machine
center.

References

Amirante R., A. Dadone, L. Dambrosio, A. Lippolis, 1998 - Experimental results of “One-Step-Ahead”


adaptive control applied to hydraulic transmission; The VI IEEE Mediterranean conference on control and
systems.
Amirante R. - Lippolis A. , 1997 - Analisi sperimentale di transitori su trasmissioni idrostatiche; Atti del VI
convegno Nazionale dell’AIIA, Ancona - vol. “Meccanica”.
Bryson, A.E., Ho Yu-Chi, 1975 - Applied Optimal Control - Optimization, Estimation, and Control; Hemisphere
Publishing Corporation, New York.
Dambrosio L., 1994 - Development and Application of One-Step-Ahead Adaptive Control System for
Helicopters; MS Thesis, Department of Aerospace Engineering, Mississippi State University,.
Goodwin G.C., and Sin, K.S., 1984 Adaptive, Filtering, Prediction and Control; Prentice-Hall Inc., Englewood
Cliff, New Jersey.
Mannesmann Rexroth , 1995 - Proportional valve and servo-valve; Mannesmann Rexroth GmbH.
Marro G. , 1999 – Conttrolli automatici, Zanichelli.
Sanelli M. , 1997 – Feasibility analysis, dynamic analysis of components, solutions with dedicated NC systems –
International workshop “Power transmissions”.

13

You might also like