You are on page 1of 15

432 THE HYDRAULIC HANDBOOK

typical electro-hydraulic closed loop system is shown in Figure 5.1.1a. This can simply
be considered as having a block or box to represent each of the major components in a
system, shown schematically in Figure 5.1.1 b. The valve could have been built integrally
with the cylinder and it would be possible to draw these as a single block which has exactly
the same meaning in control terms. In a similar way the feedback amplifier (or signal
conditioning) has been incorporated into a single feedback path block with the transducer
itself, although it could have been shown separately. The block diagram also shows a
summing junction, where the feedback signal is subtracted from a demanded signal to
produce the error signal. This error indicates the difference between the actual system
output and the required output and is then used to operate the valve to supply hydraulic
fluid to a cylinder to move the output and correct the error. If the error signal is just
multiplied by a constant value before passing to the valve then this is called aproportional
controller, with aproportional gain constant which is selected in design or on final test and
commissioning.

The lines between the blocks have significance as intermediate variables in the control
loop. For example the input to the valve block is shown as the current to the valve and
the output from the same block is the valve spool position x. The demand signal can
be considered as a voltage level which must be set external to the loop, perhaps by a
manually adjustable potentiometer or from a supervisory control computer or PLC.

The advantages
In general a closed loop system will give more accurate control of the output, a better
dynamic response and be less sensitive to changes in system parameters, e.g. frictional
coefficients. The accuracy of an open loop system relies on a complex link through a
number of components between demand and output (the forward path in the figure is an
example). The use of a closed loop gives the possibility for better performance and more
SERVOSYSTEMS 433

precise and accurate control of the output (within the accuracy limitations of the feedback
transducer). This means that if any drift in the output occurs, maybe through internal
leakage or temperature changes in the forward path components, this will be detected and
automatically corrected. It would also be possible for the effect of any external loads
applied to the output to be detected and a correction made but the error would not
necessarily be eliminated. This depends on the controller used as discussed below.

The system output will not always be the required demanded value in a simple 'propor-
tional control' system. In a situation where the output actuator has to generate some force
to overcome any external load (perhaps a gravity load) then there will have to be a pressure
difference across the actuator piston. A pressure difference can only be maintained if the
valve is partially open (if it is exactly centered then the pressures will equalise). The valve
can only be open if it receives an input and hence for a proportional controller there must
be a difference between demand and feedback, an error. Errors of this sort are usually
called It should be noted that in order to give a particular value of valve
opening, the error can be kept small if the proportional gain is large.
There is an aspect missing from the simple block diagram given above and that is the
lag in components giving delays in time between the input and output from each block. All
components will experience these delays, even the electrical amplifiers. However, it is the
slowest components within the loop which have the most significance, and this will
generally be the combination of load and actuator. A re-consideration of the events when
an error occurs will indicate the importance of this delay. If the correction for an output
disturbance is delayed through the loop then it may arrive too late to make the required
correction. Worse than this if the disturbance is fluctuating then by the time the correction
has propagated around the loop it may actually arrive back at the output to be in-step with
a component of disturbance in the opposite direction and add rather than subtract. A signal
like this can actually chase itself around the delays in the loop, gradually increasing in
magnitude. This condition is known as and is a property of the loop itself.
A larger forward path gain will increase the speed of correction within the loop but
makes the unstable condition more likely. Thus the main problem is the choice of forward
path gain to optimise between the requirements of steady state accuracy and avoiding an
unstable response. If an acceptable compromise is not found then design changes or more
sophisticated controllers are required.

The system transfer function


A more mathematical approach can quantify the characteristics described above as well
as describing the other response features of servosystems. One way to achieve this is to
add appropriate mathematical models for the various components into the block diagram
and this is shown in Figure 5.1.2 for the same system.
The usual representation used in such diagrams is a lowercase's' to be the equivalent
of differentiation with respect to time (dJdt). The component models are thus differential
equations obtained either from a suitable analysis to obtain an equation of motion, or from
testing and matching the results with standard transfer function responses. All of these
equations can be called a transfer function, which relates an output to the appropriate input
as the following example for the valve shows:

There are only two blocks shown with such dynamic models, the electrical blocks are
shown with a simple constant or gain value as a transfer function. These latter could be
described more fully by similar differential equations but the time lags produced by these
would be so quick that they could not be seen even in the most rapid of transients. This can
be demonstrated by a more detailed comparison of the valve and the load equations used.
As given above the valve is described by a first order equation often called afirst order
lag, there is only one s-term. this equation the speed of dynamic change is described by
the numerical value taken by the time constant i, the smaller it is the more rapid will be
the response. The cylinder and load have been shown as a third order equation (highest
power this combines the spring-mass second order effect, described in the chapter on
Actuator performance with the natural integration effect between flow and position (a
flow into a cylinder gives a velocity, and so must be integrated to give position). The speed
of response of this equation is governed by the natural frequency term (On but in this case
the higher the frequency the faster the response. It is therefore easier to make comparisons
if the reciprocal of the natural frequency is used. Ifa valve is chosen which has a time
SERVOSYSTEMS 435

constant 5 times quicker than the load i.e. 5x n then the valve will have virtually

completed its response before the load has moved. This is shown in Figure 5.1.3 where the
step response for the load and valve are compared. Note that this shows only the second
order part of the load response. In fact most analyses will neglect the valve response if it
is this much faster than the load, or rather it may be better to select a valve with a faster
response. The electrical circuits are likely to have time constants which are 100 times
faster again and so may legitimately be neglected from the outset. The only occasion this
may not be valid is with very low frequency electrical filtering. If a more accurate analysis
is required with this filtering and the valve dynamics included, then it would also be
preferable to use a second order model for the valve itself.
This approach confirms that the load itself is the limiting component in the dynamic
response of the system. There are two ways in which this information is useful- it can be
used to estimate the maximum response rate likely from the system and also it can be used
to help analyse the stability of the system as described in the next section on frequency
response. The minimum response time will ultimately be limited by the load natural
frequency. The best step response time possible is in the order of and this assumes
that there is sufficient power available for the acceleration.
It is also possible to simplify this block diagram to an equivalent form which gives just
one transfer function, the so called Closed Loop transfer function. This is in fact the best
mathematical form to predict the overall system response. The effect of the proportional
gain term K could then be examined for its influence on the dynamic system response.
The typical response curves are shown in Figure for increasing values ofK As can
be seen the response gets quicker but increasingly more oscillatory which indicates that
instability is being approached.

In fact it is the total gain obtained by multiplying the numerators of all the blocks in the
loop which is important. This is sometime called loop gain. The controller gain is the only
term which is likely to be adjustable, everything else will be selected for other criteria. It
is sometimes thought that the gain of the feedback transducer can be adjusted to modify
the loop gain. Although it plays its part, it is better selected on the basis of a sensible output
voltage (usually 2V but up to 10V) corresponding to the full range of output.
SERVOSYSTEMS 437

The significance of frequency response


An alternative method which is used, very effectively, to describe dynamic response of
components and systems is in terms of their frequency response. Frequency response
describes the output from a component when it is excited from a sinusoidal input of
constant amplitude but over a wide range of frequency (which can be considered as a
gradually increasing frequency). The output waveform is compared to the input and two
measures of the relative response can be identified - the size of the output waveform
establishes the amplitude ratio; and the relative position in time establishes the phase
change (usually a phase lag). Figure 5.1.5 shows these two characteristics and the most
common way to represent them on a logarithmic frequency scale as a Bode diagram.
The higher the frequency at which the amplitude reduces and the phase lag occurs then
the better will be the system dynamic response. The frequency responses for the load and
valve described above can be compared as in Figure 5.1.6. This also shows the character-
istic difference between a first and third order transfer function.

This approach allows a more detailed explanation of the stability problem in terms of
sinewaves. As a reminder, the unstable condition occurs when the signal being fed around
the loop from the output adds to itself when it reaches the output again. This could be
formalised for a sinewave by saying that it is in-phase again when it has passed around the
loop. Since the signal passes through all the components in the system before returning to
the output, at any particular frequency the total phase change which occurs around the loop
can be summed. The summing junction itself contributes a 180° phase change (like
inverting the wave), and so the unstable condition can occur when the other components
contribute another 180°, and the signal has shifted a total of 360° and is in-phase again.
However this is not the only aspect, the signal must also be increasing in amplitude as it
passes around the loop for the output to increase in an unstable manner. A signal slightly
reducing in size around the loop i.e. an amplitude change just smaller than one, will be
stable but give such an oscillatory response that it could not be sensibly used. This
conforms to the Nyquist Criterion of stability, described in most control textbooks as an
438 THE HYDRAULIC HANDBOOK

amplitude ratio equal to unity at a phase change of 180 for what is called the Open Loop
transfer function (which is simply the product of all the blocks in the loop).
It helps explain why increasing the proportional gain makes the system unstable since
this causes a larger amplitude in the signal as it is passed around the loop. It also explains
why the valve should be chosen to have a faster response than the load. A slow valve will
have a lower frequency response and consequently give too much phase shift at too low
a frequency, although there are compromises possible as it also reduces the amplitude.
Frequency response analysis can be used as the basis for system design and the
application of more sophisticated controllers as described below. However, it has its
limitations, principally since it is a so called linear technique operating with system
equations which are linear (the differential equations in Figure 5.1.2 are linear). The
behaviour of hydraulic systems can be explained in these linear terms but they are in reality
non-linear and the coefficient values in these equations will change (for example the
natural frequency with piston position). Hence the design of systems is reliant on good
estimation of the coefficient values to use, and frequently it is better to use 'worst case'
values.

Characteristics of hydraulic components


Component selection can obviously influence system performance. Various parameters,
such as load mass and required stroke, are going to be fixed by the design specification.
The coefficients in the transfer function equations can relate the physical characteristics
of the system to its components for a double ended cylinder as follows:

The natural frequency was also discussed in the chapter on Actuator peiformance but
its importance here requires additional emphasis. An increase in the volume term reduces
the natural frequency and response. This volume comprises an inevitable volume in the
actuator itself plus any volume in the connection from the valve to the actuator body. For
this reason it is better to keep the valve as close as possible to the actuator and minimise
this volume component. If the connection volume can be considered negligible then a
useful simplification can be made, which will also be true generally. The minimum natural
frequency (the worst case) occurs when the piston is in the mid-position and hence the
relevant terms in area and volume become:
SERVOSYSTEMS 439

Where L total piston stroke


This emphasises the importance of actuator area in the dynamic response even though
it is within the square root, and explains why short stroke, large area actuators have a high
natural frequency. Area is normally chosen in order to give the required force either at stall
or dynamically. A larger area will have a consequence in an increase in the flow
requirement for the same maximum velocity and it is likely that this will also require a
correspondingly larger valve. Note also that an increased area also reduces the 'gain' of
the cylinder/load transfer function unless the valve size is also increased.
The conventional analysis for a linear actuator system is based, as here, on an equal area
cylinder. This is much the simplest case mathematically and is the choice for many
servosystems because the performance will be symmetrical. However, there are also many
systems with an unequal area actuator but there are some characteristics which may limit
the maximum performance. An unequal area cy linder operated from a symmetrical valve
(the same metering restriction on both ports) will have a higher velocity in the extend
direction than in the retract direction (unlike the case when operating from a constant flow
supply). In control terms this translates as a higher gain in the extend direction which will
provide the limiting case for stability. The ratio of extend to retract gains is the square root
of the ratio of the piston to annulus areas. The performance in the retract direction will
obviously be inferior unless some form of directional compensation can also be included.
The other limitation relates to the pressure levels necessary to balance the forces across
the piston when at rest. If used with a symmetrical valve there is no unique solution for the
individual pressures, a number of combinations will provide the force balance, and it is
likely that the valve will have to adopt an off-zero position depending on the history of
previous movement. Analytically, there is a unique solution for the completely symmetrical
case and also if the area ratio of valve opening matches the area ratio of the actuator. There
is a good high performance example in the main support actuators for a flight simulator
where this flow matching is employed. However, many lower performance systems work
quite adequately with the unsymmetrical configuration and a symmetrical valve.
The value for damping ratio is often the most difficult to predict. Friction in the actuator
and the load both contribute but there is also a 'fluid damping' component arising from
flow through the valve as a restriction. This latter term will be dependent on val ve opening,
and will fall to a minimum value when the valve is closed. Cylinders for servo use are
generally chosen to have low friction seals to reduce stiction and any hysteresis effects but
this combination is likely to give a 'worst case' estimate for 1;which is quite low, typically
1;~ 0.2 is likely unless there is significant load friction.
The normal rule for selecting valve sizing is based on a maximum power transfer
calculation. This shows that the pressure drop across the valve should be 113 of the supply
pressure when operating at the designed maximum power. This gives 213 of the supply
pressure available across the load actuator. This may never be an actual operating
condition but itis still a useful sizing rule of thumb. A valve sized too small will give good
440 THE HYDRAULIC HANDBOOK

control but will waste energy. An oversized valve will not give as good control at low
openings nor will it waste energy but other valve performance features may become
significant including hysteresis and drift. There may of course be circumstance where a
better solution can be found by breaking the lhrd rule.
The valve also contributes to the non-linear aspects of the equations, mainly through its
behaviour as an orifice, since this shows that the flow rate is proportional to the square root
of pressure drop. Most electro-hydraulic valves also exhibit some amplitude dependency
in their dynamic response. The other feature of valve characteristics which is important
is the valve lap at null or zero opening. A zero lapped valve will give the most linear
characteristics with opening and is the most frequently used. A fully under-lapped valve
actually gives more linear performance in a mathematical sense but tends to be wasteful
of energy, it can also help improve system damping. Under-lapped valves have a higher
flow gain about the null but a reduced pressure gain and hence reduce the overall stiffness
of a closed loop system and affect its steady state accuracy.
The selection of the most appropriate val ve for a servosystem must be balanced with the
features and performance requirements of the system itself. In particular the dynamic
response of the valve must be sufficiently fast in relation to the load natural frequency or
time constants. A valve (or valve amplifier combination) which gives a linear relationship
between input and valve opening is generally easier to incorporate into a control loop.
However, dead-band or a large valve overlap is not a problem with most speed control
systems. A valve with low hysteresis is desirable since significant hysteresis can introduce
steady state errors with proportional control. Although these characteristics are most
frequently met by servo-valves it is equally appropriate to consider other valve designa-
tions for servosystem use if their characteristics are appropriate.

Pressure control
In systems where pressure is the controlled variable then there are some additional or
alternative considerations. Since pressure in a system changes very rapidly then more
attention needs to be paid to the valve dynamic response which should be specified in as
much detail as possible. It must either be selected sufficiently high relative to the pressure
response of the system as above or it can be selected to be the slowest responding
component and hence be the dominant feature of the dynamic response. It is also usual to
specify a characteristic of servo-valves which may be relevant and that is its pressure gain
characteristic. This usually shows that the full supply pressure change can be obtained
through a small proportion of the valve opening, as little as 5% total travel is usual.

Controller structures
The term controller structure refers to the way in which the control loop is built since
there are alternatives which allow modifications to improve the overall performance.
These may refer to the forward path of the loop alone or can include modification in
the feedback path. These modifications are sometimes called compensation in control
terms. The simplest controller to apply is the proportional controller but it has its
limitations in the compromise possible between good dynamic response, stability and
good steady state accuracy. The methods of compensation are therefore, alternative
SERVOSYSTEMS 441

ways to obtain the best performance from a hydraulic system design.


The most common alternative is the PID controller, standing for Proportional plus
Integral plus Derivative. This can be used in place of the simple proportional control in the
forward path. It means that in addition to operating the valve opening in direct proportion
to the error, additional components due to the derivative (rate of change of error) and
integral (summed total error) will be included.
The derivative term should improve the system response to sudden changes in demand,
which is not normally a problem with a hydraulic system. It could also help by improving
the damping ratio of the system but this will depend on the originalload/cy linder transfer
function (it is likely to help if this is second order dominant). Differentiation of any noise
on signals (particularly from feedback) can lead to problems and hence they may need to
be filtered.
The integral term is intended to improve accuracy by removing errors due to external
loads. The integrator will change the output until it sees a zero input, and the error is
'integrated to zero'. Unfortunately a position control system usually includes an integrator
implicit in the load/cylinder transfer function, and double integration leads to stability
problems. There is thus great difficulty in setting the integrator gain. Too low and it will
take a long time to remove the error and the output will apparently 'drift' towards the
required value. Too high a gain and the system will become oscillatory, higher still and
it will be unstable. Integral terms with even moderate gain may show a large, slow
overshoot which is evidence of what is called integrator wind-up. This occurs where the
output from the integrator is increasing whilst the error is large but unnecessarily since the
correction is already being made by the proportional term. There are standard solutions for
integrator wind-up.
Thus PID is not an ideal solution for hydraulic servosystems, although this will depend
on the exact form of the system transfer functions and most will benefit. Alternative
methods exist, both in the forward and feedback paths. These latter, sometimes called
feedback compensators, can be used principally to improve damping. A technique
effective for the normal position control transfer function is to use velocity and accelera-
tion feedback, sometimes called state feedback. However there are difficulties in deriving
the acceleration signal. The most useful alternative in hydraulic systems is to use an
additional feedback loop with a signal from a differential pressure transducer connected
across the load actuator. This acts in a similar way to the acceleration term when the mass
effects are significant. The signal should also be high pass filtered to prevent steady loads
causing additional errors in the steady state. These methods can be used with a propor-
tional plus integral forward path, which may be necessary since the integrator is the best
method to improve the steady state accuracy.

Digital control
A closed loop controller may also be formed with the inclusion of a digital computer or
microprocessor within the loop. This requires the feedback signals to be in a suitable
digital form, either directly produced or through analogue to digital conversion. The
output from the digital control calculation will also be digital and will need analogue
conversion to drive the valve amplifier. The main advantages come from the possible
control calculations which can be implemented and these are sometimes referred to as the
control algorithm. The flexibility of control algorithms offer improvements for even
simple controllers, with the further opportunity of very much more sophisticated control-
lers which can maintain consistent system performance despite many parameter changes.
As simple examples the problem of integrator wind-up can be greatly alleviated by
disabling this part of the calculation when large errors exist from a sudden input change.
Another technique is known as gain scheduling, where different controller gains could be
used in different operating zones such as different parts of a piston stroke or for different
directions of travel for an unequal area cylinder. Another task made much easier with
digital control is the transfer from one control mode to another such as in the three way
transfer from position to speed to load control in different parts of the operating cycle for
some injection moulding machines.
Further discussion on aspects of both open and closed loop digital control is given in
the chapter on Computer control.

Position feedback in cylinders


There are many types of transducer which can be used with electro-hydraulic servosystems,
too many to be described here. However, there are some transducers which have been
produced specifically, or specially adapted, for use in linear actuators (hydraulic cylin-
ders) and information on these is less widely available. Measurement of speed, load and
pressure for feedback signals can be carried out with more conventional transducers.
Pressure transducers are discussed in the chapter on Instrumentation, but for details of
other devices consult an appropriate instrumentation handbook, bearing in mind the
requirements discussed above.
The reality ofincorporating a transducer into a cylinder as an integrated unit is relatively
recent butit is an eminently sensible idea operationally. There are potentially maintenance
complications if there is a transducer failure and access is required to replace it. However,
the added mechanical protection for the transducer should help to protect from damage and
improve reliability. In an external installation for the transducer it should be remembered
that the piston and rod are often free to rotate about the cylinder axis unless constrained
by the rod mounting, and this will complicate the transducer mounting.
The basic configuration used in the majority of designs is shown in Figure 5.1.7 for both
single ended and through rod cylinders. In both cases there is a transducer body and a rod-
like probe which are fixed relative to the cylinder body, and a modification to the standard
piston rod which includes the attachment of the moving part of the transducer. This
moving part varies with the operating principle of the transducer as described below. In
the single ended case the transducer probe will be immersed in the operating fluid and
subject to the operating pressures. The major features of the variants available are
summarised in Table 5.1.1 but it should be noted that these specifications are liable to
change as manufacturers develop their products.

Conventional transduction methods


There are variants of both hybrid track potentiometers and LVDTs (Linear Variable
Differential Transformer) which have been produced for mounting in hydraulic cylinders.
The principles of operation of both these devices are well established and hence have not
been detailed here. For the hybrid track potentiometer, a collar is attached to the piston rod
which makes contact with the resistive track itself, as the wiper, and connects through a
second sliding contact to a low resistance conductor to complete the return circuit. In the
case of the LVDT the conventional core of magnetically permeable material must be
attached to the piston rod and move within the transducer body.

Eddy current device


The probe tube contains a single detector coil which is supplied with a carrier frequency
in the order of 10kHz. The moving part of the transducer is in fact a tube of special material
(Anticorodal B) which is inserted in the hollow piston rod. The eddy currents produced
in this tube interact with the coil where the two are overlapped, to change the impedance
of the coil. This impedance change can be detected by a number of methods to give either
a voltage or current output. The impedance is also very temperature sensitive and so active
temperature compensation is built in as part of the design (in fact in some variants, a virtue
is made of this and the temperature signal is available externally).

Variable resistive vector transducer (VRVT)


This device has a single detector coil within the probe tube and again operates with a carrier
frequency supply. The moving part is a high magnetic permeability core which moves
within the coil and changes its impedance. The coil is supplied with a constant current
drive and the in-phase potential difference across the coil is detected. This is demodulated
to give a voltage output linearly proportional to the core displacement. The core may be
SERVOSYSTEMS 445

attached directly to the piston rod or supplied within a sleeve fitting inside the rod and
outside the transducer probe.

Linear inductive position sensor (LIPS)


The probe contains two coils, one a sensing coil with a relatively coarse pitch which
determines linearity, and the second a balance compensating coil. These are supplied from
a high frequency oscillator (1 - 2 MHz) with careful control of the supplied voltage. The
coil impedance is again varied by the length of overlap with eddy current effects in a
surrounding tube of conducting material. Aluminium or stainless steel may be used but it
is possible that the piston rod material itself may give an adequate signal and no additional
sleeve is required. The detection of the output is also novel and involves a sampling
technique with values taken equi-spaced about the voltage peak. The high frequencies and
special signal conditioning techniques have led to a solution with all the electronics built
into the transducer housing giving a self contained package somewhat like a DCDC
LVDT.

Magnetostrictive
This transducer is digital in nature but unusually this is as a result of the discrete time
interval inherent in the process rather than quantisation in the position information. The
probe contains an outer protective tube, an inner tube or waveguide, and along the
centreline a conducting wire as shown in Figure 5.1.8. A permanent magnet is located
within the piston, moving with it and external to the probe. A current pulse is passed
through the circuit comprising the central conductor and the waveguide. The interaction
of the magnetic fields from the magnet and the pulse produce a torsional strain on the
waveguide as a result of an effect called magnetostriction (like piezoelectric but a
magnetic rather than an electrical field). This torsional pulse (produced instantaneously)
travels along the waveguide at a known speed and when it reaches the transducer head it
rotates the sensing coils to give an output. The positional information comes from the time
taken between the pulse occurring and the torsional wave reaching the sensing head, which
indicates the distance from the magnet to the head. This information is available as a time
interval but can easily be processed to give a proportional voltage.

Linear resolver
This is again inherently a digital type device but the signal is processed into analogue form.
The end of the probe houses four relatively short length coils, two act as supply and two
as detector coils. The moving section is a series of tubes, located end on end, inside the
piston rod, which each house a self contained length of measurement coil (relatively coarse
winding pitch). These coils are energised by the alternating field produced in the supply
coils. The signal in the detector coils then varies as its position changes relative to the
measurement coil windings. The winding pitch in both the detector and measurement coils
define the resolution limits and the measurement coil defines the basic linearity. The
segmenting of the measurement coil reduces the power requirements since only a
maximum of three adjacent lengths will ever be energised at one time.
This transducer only gives incremental information i.e. the increment of distance
446 THE HYDRAULIC HANDBOOK

moved from a starting point. To know the new absolute position the starting point must
also be known. A modified measurement coil can be incorporated at any position in the
stroke to provide a datum which can provide a check on absolute location. It is also
relatively easy to build in a back-up battery to maintain the absolute position information
in the event of a power failure.

Integrated digital sensing


This system uses a completely different mounting style and has been available from
several manufacturers in the past. It relies on the machining of small grooves in the piston
rod itself which are then coated to give a completely smooth finish (chromium plated or
ceramic coated). The position of the grooves can be detected by magnetic proximity
sensors built into the rod end housing. Two sensors are usually used carefully spaced
relative to the groove intervals to improve the resolution. The resolution is limited by the
closeness and machining accuracy for the grooves. This is a very compact system which
is robust, not exposed to hydraulic pressure and, depending on the detail design, the
electronics and sensors may be accessible for easy service. This system, however, only
gives incremental information as described above and requires a datum.

You might also like