Section 7. PID Control

Section 7.
PID Control
Process Control and

Instrumentation
Process Control and Instrumentation PID Control
Section 7
PID Control
7.1 Introduction
The term PID refers to the three adjustments that are to be found on the controller:
Proportional, Integral and Derivative. In reality there is another term, the filter time, which
is conveniently ignored by most text books on the subject. PID control has been around since
about 1940 and has its roots embedded in pneumatics – with several of the terms dating back
to this period.
Indeed, one of the problems related to this technology is that there is no standardisation and
there are, consequently, a wide variety of different terms used to describe the same entity.
7.2 On/Off Control
Probably the most frequently encountered form of control is the On/Off temperature
controller typified by a simple thermostat found in virtually every home (oven, geyser,
refrigerator, etc.).
When applied to industrial process control the On/Off controller may be illustrated by the
basic batch process system shown in Figure 7.1 in which steam is used to heat a raw material
to a predefined temperature. The temperature of the product is measured using a
thermocouple that inputs to a simple On/Off type controller. The temperature to which the
product must be heated is set on the controller and is called the setpoint (SP). The Output of
the controller controls a solenoid valve supplying the heating steam.
Steam in
V1
SP Figure 7.1. Example of
simple process control
PV ERR On/off Output Actuator
A Σ controller
V2 system using ON/OFF
controller
Product in
Thermocouple Steam
out
Product
out
©Mick Crabtree 2007 Page 7.2

The on/off controller compares the process temperature (the Process Variable (PV)) with the
setpoint (SP) and produces an output that maintains the valve fully open as long as the process
temperature remains lower than the setpoint. As soon as the process temperature reaches the
setpoint, the controller fully closes the valve.
Assume the main valve (V1) is open. If the initial process temperature is 20°C and the
setpoint is 150°C then, under the influence of the controller, the control valve (V2) is also
open, steam will flow and the product starts to heat up.
The heat energy from the steam is supplied virtually instantaneously in what is termed a ‘step
change’. However, the actual process response shows a lagging characteristic because of a
combination of delays and retardations. These delays include the thermal capacities of the
product and the tank, the thermal transfer characteristics of the steam heating system, and the
insulating properties of the tank. This combination is called the process time lag.
When the process temperature reaches the setpoint temperature and the controller turns off the
heating steam, the product temperature will still continue to rise for some little time
afterwards because of the thermal capacity of the system (Figure 7.2). Eventually, however, it
will start to fall and, when the temperature reaches the setpoint, the valve reopens and re-
applies steam heating energy. Although re-applied immediately, the process time lag delays
the effects of the applied heat on the process medium. Thus, the temperature continues to fall
some little way below the setpoint until it starts to rise again.
The result of these process time lags is to impart a cyclic action to the process response —
with the temperature oscillating slowly above and below the setpoint (Figure 7.2).
Open Open Open
Valve
Closed Closed Closed
160°C
+10°C
Temperature (°C)
Off Off Off

150°C Setpoint
On On
-10°C
140°C
20°C
Time
Figure 7.2. Illustration of On/Off control showing both the valve and the
process reaction.

This action is illustrated in the static characteristic (Figure 7.3) which shows that the
controller Output can have only two values: Output (Max) and Output (Min) – dependant on
whether the error (ERR) is positive or negative.
Output (Max)
Figure 7.3. Static characteristic shows that the

controller Output can have only two values:
Output (Max) and Output (Min) – dependant on
whether the error (ERR) is positive or negative.
Output (Min)
_ 0 +
Error (ERR)
Deadband
Although, in this example, the excursion is ± 10°C, this will vary considerably from process
to process. Further, in many processes, the minimum and maximum excursions may fall
within acceptable limits and such a process control system might seem more than acceptable.
In the response shown in Figure 7.2 the On/Off action is always initiated when the
temperature equals the setpoint. In reality, all such devices exhibit a deadband — defined as
“the range through which an input can be varied without initiating a response”.
Figure 7.4 assumes a deadband of 6°C and illustrates that the action on the rising temperature,
from the application of steam heat, is similar to that described previously. In other words, the
heat is turned off when the temperature reaches the setpoint. However, the difference is now
that the controller does not restore heating when the temperature falls back to the setpoint but
only when it falls below the deadband.
Open Open Open
Valve
Closed Closed Closed
160°C
+10°C
Temperature (°C)
Off Off Off

150°C Setpoint
6°C Deadband
144°C
On On
-10°C
134°C
20°C
Time
Figure 7.4. Illustrating the effect of ON/OFF control with deadband.

One of the effects will be that the temperature will now have an even larger negative
excursion than before that will increase the on-time. Indeed, in this example, the effect of the
deadband is to give a total swing of between + 10 and - 16°C above and below the setpoint
— compared with the previous ± 10°C swing.
It should be noted that in many control systems, deadband is deliberately introduced in order
to prevent `chatter' or hunting and that some systems even have an adjustable deadband
control.
7.3 Proportional control
The On/Off system illustrated in Figure 7.1 may, or may not, prove to be satisfactory
dependent on how closely the temperature needs to be controlled. Whilst the deadband can
certainly be reduced, little can normally be done to reduce the cyclic excursions since they are
determined by the process time lag.
In On/Off control there are only two possible controller outputs, no matter the value of the
error signal, and thus the process is forced to oscillate. In order to avoid these forced
oscillations it is thus necessary to reduce the gain for small values of ERR. This is achieved
through the use of proportional control as shown in Figure 7.5. Here, the procedure is simple:
measure the process variable (PV); compare the PV with the target value or setpoint (SP) and
use the difference (the error signal, usually abbreviated to ERR) to determine the output (OP)
of the controller
SP
ERR = SP - PV Controller Reverse action

Σ K
OP Figure 7.5.
PV Block
Process PD
fluid in
diagram of
a closed
loop system
incorporatin
Steam in g error
difference
control.
Thermocouple
Process Steam out

fluid out
The error signal acts on the controller that, in the simplest proportional control, comprises an
amplifier with gain K. The OP which, in turn, determines the process demand (PD) is thus the
error times the gain. Put another way:
OP = ERR * K

The static characteristic for proportional control is shown in Figure 7.6.
Figure 7.6. The static characteristic for

proportional control.
Output
(Max)
Output Several things should be noted. Firstly, when

(Min) a controller is part of the loop, it is more
convenient to refer to the OP rather than the
PD. Secondly, as shown, it is customary to
_ + make use of a summing circuit in order to
Error (ERR) derive the error signal. This said, then either
the SP or the PV must be negative. Again, the
general rule is that the SP is taken to be positive and the PV negative in order to derive:
ERR = SP - PV
Assuming there is no phase shift through the process, then in order to allow the PV to have a
negative action, the output of the controller must be reversed. This REVERSE ACTION, in
fact, it would be the default.
In proportional control, the controller moves the final control element (the valve) to a definite
position for each value of PV (the temperature of the process liquid). Thus, initially, when the
temperature is low the resultant large error signal ensures that the valve is fully open (Figure
7.7). As the temperature increases, the controller reduces the deviation signal, which
progressively closes the control valve. This action continues until, ideally, the process
temperature slowly and smoothly reaches the setpoint, whereupon the deviation signal is zero
and the control valve is fully closed.
Setpoint
150°C
Temperature
20°C
Time
Figure 7.7. As the process temperature
100% rises, the deviation signal gradually
reduces to close the control valve.
Error signal
0%
Time

Proportional gain and Proportional Band

Whilst many manufacturers refer to the proportional gain (K) of the controller, others refer to
the Proportional Band or PB. The Proportional Band is the change in transmitter signal
(expressed as percentage of its range) that is required to produce a 100% change in the
controller output.
If, for example, the temperature is measured over the range 0 to 180°C and the control valve,
from fully opened to fully closed, allows the temperature to be varied proportionally over this
temperature range, the Proportional Band is 100% (Figure 7.8).
Open
75%
Valve opening
Figure 7.8. When the

50%
control valve, from
fully opened to fully
25%
closed, allows the
temperature to be
Closed
varied proportionally
100% proportional band
over the measured
temperature range 0 to
20 40 60 80 100 120 140 160 180 180°C, the
Temperature (°C) Proportional Band is
100%.
However if, as shown in Figure 7.9, the temperature is measured over the range 0 to 180°C
and the control valve, from fully opened to fully closed, allows the temperature to be varied
only over the temperature range 20 to 150°C – a span of 130°C – the Proportional Band is:
130/180 = 72.2 %.
Open
75% Figure 7.9. When the

Valve opening
measured temperature range

50% is 0 to 180°C, and the control
valve, from fully opened to
25% fully closed, only allows the
temperature to be varied
Closed proportionally over the
72.2% proportional band temperature range 20 to
150°C, a span of 130°C, the
20 40 60 80 100 120 140 160 180 Proportional Band is only
Temperature (°C)
72.2%.
The first thing to notice is that proportional control starts to act as soon as the temperature of
the process exceeds 20°C. Consequently, the process reaction time could be speeded up by
reducing the temperature range over which proportional control is applied. In Figure 7.10, for
example, the steam valve is fully open until the process temperature reaches 100°C. Only then
does proportional control action start and throttling action take place.

Open
75%
Valve opening
50%
Figure 7.10. By
25% narrowing the
proportional band the
Closed 27.8%
gain is increased.
proportional
band
20 40 60 80 100 120 140 160 180

Temperature (°C)
In this case, the proportional band is reduced to 50/180 = 27,8 % and, consequently, it can be
seen that the Proportional Band is determined by the gain – the reciprocal of span:
Proportional Band = 100% /K
Proportional offset
So far, the complete process has been somewhat oversimplified. This is because it has been
assumed that once the setpoint temperature has been reached, no further heat need be applied
in order to maintain a temperature of 150°C. Clearly this is totally impracticable since, like it
or not, there will be heat losses.
Consider, for example, that the losses are fairly high and that in order to maintain the process
material temperature at 150°C the control valve needs to be about 60 % open. In other words,
quite a considerable amount of steam energy must be applied in order to just balance the heat
losses.
In order for the valve to assume, for example, a 50% open position, a steady-state deviation
signal is required. As shown in Figure 7.11, with a controller gain of K = 1, an OP of 50%
requires an error signal of 25°C. This, in turn, requires that there is a difference between the
process material temperature and the setpoint. Again, assume that, with the output valve set
to 50% the PV is 125°C. If the SP is set to 150°C the system is thus in equilibrium.
SP = 150°C
ERR = 25°C Controller OP = 50%

Σ K =1
Process PV = 125°C
fluid in Figure 7.11. With
the SP set to 150°C,
the system is in
equilibrium with a
Steam in
controller gain of K
= 1, an OP of 50%
and an error signal
Thermocouple of 25°C. This
requires that there is
a difference between
the process material
Process Steam out temperature and the
fluid out setpoint.

The consequence of this requirement is that the process temperature will, in fact, not reach the
setpoint temperature but instead will assume a lower value of 125°C in order to produce a
deviation signal of sufficient magnitude to maintain the valve in a 50 % open position. This
difference, due to the effect of the load, is called proportional offset. How big an error
signal is required will depend largely on the gain (K) of the controller.
One solution lies in increasing the controller gain K. This means that the error of 25°C will
now produce a larger OP. This, in turn will increase the PV and thus reduce the error signal.
In this manner it is possible to reduce the proportional offset as shown in Figure 7.12 – but
only at the expense of increased noise and instability.
Reducing proportional
offset
Setpoint
150°C
K=8 Figure 7.12. As the
gain K is increased,
Process variable
K=4
K=1 the proportional offset
is reduced – but only at
the expense of
increased noise and
instability.
20°C
Time
Proportional control can be used by itself with stable 1st order systems if small constant steady
state errors can be tolerated. Typically this might arise where the sensor inaccuracy is large or
even unimportant as might happen in liquid level control where only an approximate level
value is needed for correct operation. In addition, an error can often be tolerated in the inner
loop of a cascaded control system.

7.4 Manual bias
Another approach to compensating for the proportional offset error is to adjust the setpoint to
a value some 20 % higher than that required — for example, 180°C. On this basis, the
temperature will rise to a value of approximately 20 % below this — e.g. to the required value
of 150°C (Figure 7.13).
New set
180°C
point
Proportional
offset
Set point Figure 7.13. The
150°C 150°C
offset error (a) can
Proportional be compensated for
offset
125°C
by setting the
setpoint to a value
above that which is
Time Time really required (b).
(a) (b)
Clearly this is not a satisfactory arrangement and one solution is to employ a manual bias or
manual reset control. The block diagram of the control loop with manual reset (Figure 7.14)
shows that the manual reset control acts on the final control element in the form of a d.c. bias
or offset voltage. In this manner the reset control is adjusted to offset the output to a point
where the error signal is zero i.e. until the valve is 50 % open. Under stable conditions, this is
the setting required to maintain the temperature at the setpoint of 150°C.
Manual bias
SP
control
Figure 7.14. Block diagram of a
ERR Controller
Σ K =1
controller incorporating
manual reset.
PV
Process OP
fluid in The amount of offset required
Error
does not have to be actually
calculated. It only requires that
Steam in the operator adjust the manual
reset control slowly, allowing the
process loop to adapt to the
changing conditions, until the
Thermocouple deviation signal, as indicated on
the meter, is zero.
One of the drawbacks of using

Process Steam out
fluid out manual reset is that it cannot
adapt to changing conditions and
thus requires the intervention of
the operator whenever the setpoint is changed. In other words, whenever the setpoint is
changed, the manual reset control must be used to offset the final control element signal until
the error signal is again zero.

7.5 Integral control
A method of overcoming the drawback of manual reset is through the use of automatic reset
in which the reset action uses an electronic integrator. Such control is thus more often known
as integral action — usually abbreviated to I. A PI controller thus incorporates both
proportional and integral action.
Automatic reset is similar to manual reset in that it offsets the signal to the final control
element, but uses the integral of the error signal as an input.
Electronic integration differs little from its mathematical concept — a measure of the area
under a curve. Thus, the integral of a steady-state input (the error signal) (Figure 7.15)
produces a steadily increasing signal — a measure, per unit time, of the area under the input
signal level curve. Should the input increase or decrease (Figure 7.16), the slope of the
integral will also increase (A) or decrease (B), in proportion to the area per unit time under the
signal curve. And if the input signal reduces to zero then the slope reduces to zero (C).
8 16
7 14
6 12
5 10
al
gr
te
Integral Figure 7.15. Basic

In
Input
4 8 (number of
squares per
electronic integration
unit time) where a steady state
3 6
input signal gives
Input
rise to a steadily
2 4
increasing output
1 2
signal.
0
8 0 1 2 3 4 5 6 7 8 16
Time
7 (C) 14
6 12
5 (B) 10 Figure 7.16.

al
gr
Reducing
te
Integral
In
Input
4 8 (number of the
squares per
unit time) integrator
3 6
(A) input to zero
Input only reduces
2 4
the output
slope to
1 2
zero.
0
0 1 2 3 4 5 6 7 8
Time

Again, in simplistic terms, the integral is proportional to the area per unit time and
(despite the fact at t = 7 s the input has fallen to zero) the measured area at t = 8 s is
still 15 square units! The only way in which the integral signal can be reduced,
would be for the input to go negative and thus reduce the total area per unit time
(Figure 7.17).
6 12
al
5 10
gr
te
In
Integral Figure
Input
4 8 (number of
squares per
7.17.
unit time) The
3 6
Input integrat
or
2 4
output
can only
1 2
be
reduced
0
0 1 2 3 4 5 6 7 8 if the
input is
Time
made
negative
.
Figure 7.18 shows a block diagram of PI controller incorporating automatic reset.
Integral
SP I
ERR Proportional
Σ K
PV
Process OP
fluid in
Error
Steam in
Figure 7.18. Block diagram

of PI controller
Thermocouple
incorporating automatic
reset.
Process Steam out

fluid out

Assume a step shift in the setpoint from 150°C to 160°C (Figure 7.19 (a)). In an open loop
situation this produces a step function error signal (Figure 7.19 (b)) that, after amplification,
acts on the final control element. This is the proportional component (Figure 7.19 (c)). The
step function error signal, however, is also applied to the integrator input, which produces a
steadily increasing output. This is the integral component (Figure 7.19 (d)) – the reset. This
is shown in the block diagram of Figure 7.20.
160°C
Set point (a)
150°C
Figure 7.19.
Error 10°C (b) A step change
signal in the setpoint
(a) produces
Proportional (c) a step change
component in the
deviation
signal (b) that
Integral produces both
component a
(d) proportional
component
(c) and an
10% integral integral
step component
Resultant (d) that are
valve 10% proportional
(e) step combined (e)
movement to act on the
2 minutes
final control
element.
Proportional output
Step error input

K
Σ Resultant output
Figure 7.20. Block

I diagram of
combined
Integral output
proportional and
integral action.
In this way, the controller applies automatic reset as a slowly increasing `d.c. offset' in much
the same manner as manual reset is applied. In a closed loop situation the integral output
would continue to rise, offsetting the output, until the deviation signal reduces to zero. At this
point the integral signal will be maintained at a constant value until the next disturbance.

The reset value is described in terms of its slope, or rate of increase (or decrease), for a given
input. Usually this rate is expressed as the integral time constant — the time taken for the
valve to move, under the influence of integral action, through the same distance as it had
under the influence of proportional action only. This is illustrated in Figure 7.19(e) which
shows how, for a 10% valve movement, the time taken for the integral action to increase the
valve movement by the same amount as the proportional action, is 2 min.
Another method of expressing the reset action is also in direct terms of its slope: how many
times, in one minute, does the integral value equal the proportion value. This is expressed in
`repeats per minute'. Figure 7.21 shows the valve position due to both proportional action and
to three different integral rates. In the first (curve A) the integral action rises at such a rate
that it equals the change due to proportional action after 0.67 min. In other words it would
`repeat' the proportional action 1½ times each minute.
1½ repeats/min
A 1 repeat/min
B ½ repeat/min
Valve position
10% integral Figure 7.21. Reset

change C
action may be
expressed in terms of
its slope — how
many times, in one
10% proportional
change
minute, does the
integral value equal
the proportional
value.
0.67 min 1 min 2 min

Time
In curve B the integral action is such that it equals the change due to proportional action after
one minute and will thus have a reset rate of 1 repeat/min. The rate at which curve C rises
shows that after one minute it only equals half of that due to proportional action and it thus
exhibits a reset rate of ½ repeats/min. In fact, the expression `repeats per minute' is the
reciprocal of the integral time and thus, in this example, the values of 1½, 1, and ½ repeats per
minute, would be expressed as integral times of 0.67, 1 and 2 minutes respectively.
These terms tend to be very confusing since both the integral time and the repeat time is
expressed in a very similar fashion — in either minutes or seconds (e.g. integral time:
minutes/repeat or seconds/repeat and repeat time: repeats/minute or repeats/second). To add
to the confusion, rarely will the controller indicate which parameter is being used and the
user, therefore, has to refer to the handbook.
The light at the end of the tunnel is that Foundation Fieldbus has decided to settle on integral
time expressed in seconds per repeat.
PI control is used for a wide range of process applications but is generally confined to fast
self-regulating systems exhibiting a first-order characteristic with little deadtime. Notably
these would be typified in the control of flow and low-capacity pressure systems.

7.6 Integral windup
Integral windup or, as it is often known, reset windup, is one of the most commonly
neglected and often undetected problems in feedback control.
If an error signal (ERR) persists for some time, the Integral term (and therefore the Output)
will continue to ramp up (or down) for as long as the error exists. In most cases feedback
action ensures that the error signal is, in fact, reduced. However, if a continuous error exists
which cannot be eliminated, then the output would continue to increase or decrease until the
final control element reaches its physical limit of fully opened or fully closed.
Because most actuators are limited in amplitude (and slew rate) then when the final control
element reaches one of its limits (i.e. saturates), and fails to reduce the error, the feedback
path is actually broken. In a pneumatic or electronic analog controller, the integrator output
will continue to ramp up or down until a physical limit such as air supply pressure or a supply
voltage, was reached. At this the integrator is 'saturated' or fully 'wound up'. However, in a
digital computer where there is, theoretically, virtually no upper or lower limit, the integrator
could carry on until the output reaches a value close to infinity.
Figure 7.22 shows a flow control loop in

SP
Σ P + I controller which an isolating valve is placed in
series with the main process stream.
PV Figure 7.22. Flow control loop (P + I)

in which an isolating valve is placed in
series with the main process stream.
Isolating valve Control valve
If the isolating valve closes then, as shown in Figure 7.23, the PV will fall to zero 1. This
will produce a continuous error signal 2 that acts on the integral term to ramp it up towards
infinity 4, whilst the output itself saturates at its upper limit – typically 20 mA 5.
2 PV falling to zero
results in continuous 3 Proportional output
positive error (ERR)
K
5 Output saturates at Figure 7.23.
SP Σ Σ high limit (e.g.20 mA) Action following
closure of the
isolating valve.
I 4Integral ramps
up towards
infinity
1 PV falls to zero
when isolating valve
closes

So, we’ve now arrived at a situation where the valve is fully opened and the output is
saturated. But, never forget that the integral term is still ramping up.
What happens when the isolating valve is opened? This is illustrated in Figure 7.24.
7 PV rising to maximum
results in continuous 8 Proportional output
negative error (ERR)
K
Output remains
SP Σ Σ saturated at its high
limit of, e.g.20 mA
Figure 7.24.
Action following
I opening of the
9Integral ramps down isolating valve.
from its previously high
6 When isolating valve value
opens PV rises to its
maximum – determined by
the control valve
The PV will, within a very short time, increase to a maximum determined by the valve
opening6. Since the setpoint has remained unchanged, the error signal (ERR) is now
reversed and the integral block now ramps down from whatever value it reached. Bearing in
mind that it ramps down at the same rate that it ramped up, it will take the same time to reach
its previous value, before the isolation valve was closed, as it did to get there. So, if the
isolation valve was closed for 25 minutes, it will take 25 minutes for the integral value to
reach the value that it was at previously.
Integral windup is often a problem in interrupted batching systems. Consider the batch
process in Figure 7.25 in which a tank is filling under the control of the level controller. If,
when the tank is not yet full, the main control valve is turned off, the PV will no longer
increase and a large error signal would thus be present. The error signal acting on the integral
will cause the output to ramp up until it saturates and the control valve is fully open. This is
called integral windup.
Main control
valve ON
Figure 7.25. If automatic
OP filling control is interrupted
and the PV no longer
increases, the error signal
SP acting on the integral will
cause the output to ramp up
until it saturates and the
control valve is fully open.
PV

If now the batching process is resumed and the main control valve is turned back on, the
output remains saturated and the tank will continue to fill above the set point. Bearing in
mind that, in order to decrease the integral action a negative input is required then the tank
level would have to exceed the setpoint before the polarity of the error signal could be
reversed.
As shown in Figure 7.26, this problem can be overcome by using what is called èxternal anti-
reset windup' (sometimes called ìntegral lockout' or ìntegral desaturation'). In ànti-reset
windup' the integral action is completely inhibited when the main control valve is closed.
Main control
valve OFF
OP Figure 7.26. Anti-reset

windup inhibits the
integral action when
SP
the main control valve
is closed.
Anti-windup
PV
You should also be aware that manufacturers also make use of a number of proprietary
solutions for dealing with windup. These include:
Conditional Integration.
If the controller output is saturated, and input and output are of the same sign, then set the
integrator input to zero.
Limited Integration.
If the integrator is saturated, and controller input is of the same sign, then set the integrator
input to zero.
Tracking Anti-Windup.
If the controller output is saturated, reduce the integrator input by some constant, times the
difference between unsaturated and saturated controller output.
Generally these measures work with only varying degrees of success and seldom eliminate the
problem. Consequently it is better to try and ensure that windup does not occur in the first
place.

7.7 Stability
Although, as we have seen, integral action reduces the proportional offset to zero, one of the
negative aspects of integration is that it has a destabilising effect on the loop.
Before we examine how integral action can increase the instability of a loop, we need to
examine the underlying causes or instability in general.
Consider the heat exchanger shows in Figure 7.27 in which the gain of the controller is
designated KC, and that of the process KP. This may be simplified as shown in Figure 7.28 in
which the total loop gain is: KL = KC * KP
SP
Controller Gain = Kc
ERR = SP - PV Controller
Σ Kc
OP
PV Figure 7.27. Schematic of
heat exchanger in which the
gain of the controller is
Process Gain = Kp
designated KC and that of
the process KP.
SP
+
- ERR
Σ Controller
Kc
Figure 7.28. Simplified block

PV diagram in which the total loop
Loop gain = Kc * Kp OP
gain is KL = KC * KP
Kp

Assume the values for KC and KP as shown in Figure 7.29 – in which case the total loop gain
KL is unity. If there is a positive PV disturbance c of +2 engineering units, this will result in
an error disturbance d of -2 to produce an OP disturbance e of -1. If, as shown, the process
gain is 2, with zero phase shift, the PV response f will be in direct opposition to that of the
disturbance.
ERR Disturbance
-2
+ 2
3
-
Σ Controller
Kc = 0.5 OP Disturbance -1
1 4
Kp = 2
PV Disturbance PV response
+2 -2
Figure 7.29. For a positive PV disturbance c of +2 engineering units, an
error disturbance d of -2 produces an OP disturbance e of -1. With a
process gain of 2 and zero phase shift, the PV response f will be in direct
opposition to that of the disturbance.
However if, as shown in Figure 7.30, the phase shift through the process is 180°, then the PV
response f will sustain the original disturbance and instability will undoubtedly occur.
ERR Disturbance
-2
+ 2
3
-
Σ Controller
Kc = 0.5 OP Disturbance -1
1 4 180° phase shift
Kp = 2
PV Disturbance PV response
+2 +2
Figure 7.30. If the phase shift through the process is 180°, then
the PV response f will sustain the original disturbance.

Referring to Figure 7.31, this can be shown in another manner:
ERR = SP - PV = SP - KP * OP
ERR = SP - KP * KC * ERR
ERR + KP * KC * ERR = SP
ERR (1 + KP * KC) = SP
ERR = SP/ (1 + KP * KC)
SP
ERR
Σ Controller
KC
Figure 7.31. Examination of

PV OP feedback equation.
KP
Examination of this last equation shows that if the loop gain KP * KC equals unity and there is
a phase shift of 180°, then the divisor is 0 and the error is infinite. Clearly there is instability.
On this basis, the conditions for instability are easy defined: the loop gain (KL) must be equal
to or greater than unity and there must be a phase shift of 180° through the process.
One important characteristic of linear systems is that when driven with a sinusoidal input at a
particular frequency, the response will be a sinusoidal function at the same frequency.
Consequently, the traditional test for stability has been conducted as shown in Figure 7.32 by
applying a sinusoidal waveform to the input of the process and comparing it with the
amplitude and phase of the output.
Input Output
Process
/X/ /Y/ θ Figure 7.32. Testing
stability by applying a
sinusoidal waveform to
the input and
comparing it with the
amplitude and phase
/Y/ of the output.
/X/ Input
Output
θ

The amplitude and phase may then be plotted separately as Bode plots (Figure 2.33). In the
first plot the gain (in decibels) is plotted against frequency whilst in the second plot the phase
shift (in degrees) if plotted against frequency. Both plots usually have the frequency in a
logarithmic scale from 0 to ∞.
4
2
Frequency (Hz)
0
Gain (dB)
-2 0.02 0.04 0.1 0.2 0.4 1 2 4 10
-3.2 dB
-4
-6
Figure 2.33. Bode
-8
plots in which the
-10 amplitude and
Frequency (Hz)
phases are plotted
0
0.02 0.04 0.1 0.2 0.4 1 2 4 10 separately.
40
80 800
Phase lag (θ)
120
160
1800
200
220
Examination of the plots shows that when the loop gain is 0 dB (unity gain) and higher, at a
frequency of 0.6 Hz, the phase shift is less than 80°. And indeed at just over 2 Hz, when the
phase shift is 180, the gain is only –3.2 dB (KP = 0.7). Clearly, the process is stable.
Another method of showing the results is in a Nyquist diagram (Figure 7.34) which plots the
gain and phase angle at different frequencies. On this it can be seen that at a frequency of 1.8
Hz, when the phase shift is 180°, the gain is only 0.45. Again, the process is clearly stable.
1.2
90°
1.0
0.8
0.6
3.6
0.4Hz
3.4 Hz Figure 7.34. Nyquist
2.5 Hz 0.2 diagram plots the gain
180° 1.8 Hz 0° and phase angle at
-1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.6 0.8 1.0 1.2 1.4 1.6 different frequencies.
-0.2
1.3 Hz
-0.4
1.1 Hz 0.06 Hz
-0.6
1.1 Hz
-0.8 0.18 Hz
0.97 Hz
-1.2
0.61 Hz 0.36 Hz
-1.0
270°

It is important to remember that so far we have only examined the stability of the plant and
have not taken into account the influence of the controller on the complete loop gain (KL).
Let us now re-examine the influence of the effect of integral action.
If we applied a sinusoidal waveform to the input of the two parallel function blocks, shown in
Figure 7.35, then the output of the proportional block should be in-phase whilst that of the
integral block will lag by 90°. The resultant of the two outputs will lag by a factor determined
by the amplitude of the proportional and integral components.
Proportional
output (in-phase)
Input
P
Σ
w Resultant output
I
Integral output
(lags by 90°)
Figure 7.35. With a sinusoidal waveform applied to the input of the two parallel
function blocks the output of the proportional block is in-phase whilst that of the
integral block will lag by 90°. The resultant will lag by a factor determined by the
amplitude of the two components.
This is shown using vectors in Figure 7.36 which again shows how the integral action lags the
proportional output by 90°. Thus, in considering the controller alone, the resultant magnitude
and lag will be determined by the relative magnitudes of the integral and proportional vectors.
And as the integral action increases, the magnitude and lag of the resultant also increases
(Figure 7.37).
Proportional output
Figure 7.36. With the

Integral action Resultant application of integral action
the resultant magnitude and lag
will be determined by the
relative magnitudes of the
integral and proportional
vectors.

Proportional output
Resultant
Integral action Figure 7.37. As the integral
action increases, the magnitude
and lag of the resultant also
increases.
Assume now that the process itself provides lag in excess of 90° as shown in Figure 7.38.
Now, with the application of both proportional and integral action the phase lag may well be
increased to a point where it meets the classical conditions for instability: a phase lag of 180°
and a gain of equal to or greater than 1 (Figure 7.39).
Figure 7.38. Process lag in excess

of 90°.
Process lag
Integral
action
Figure 7.39. With the

application of both
proportional and integral Resultant
action the phase lag may
be increased to a point
where it meets the Process lag
classical conditions for Proportional output
instability: a phase lag of
180° and a gain of equal
to or greater than 1.
From the foregoing it can be seen that in loops in which the phase lag is greater than 90°,
integral action may well lead to instability. It should also be noted that if the process lag is
less than 90° it is impossible to produce a total loop phase lag that equals 180°.

Figure 7.40 shows a total phase lag of 30°. As the integral action is increased it will tend
towards a maximum of 120° and can, therefore, never equal the conditions for instability.
Process
lag
Integral Proportional
output output Figure 7.40. Loops that exhibit small
process lags will tend to be stable. In
this example, where the process lag is
only 30°, increasing the integral
action will produce a total phase lag
that tends towards a maximum of
120° and can, therefore, never equal
the conditions for instability.
7.8 Derivative action
Many text-books will tell you that the primary function of derivative action is to speed up the
reaction time by generating a control action that is proportional to the rate of change of the
error signal – the time derivative. In this manner the control system generates a large
corrective effort immediately after a load change in order to begin eliminating the error as
quickly as possible.
This might be likened to the actions of a skilled operator on very slow process where the PV
moves very slowly after a setpoint change. In such cases the operator will often switch to
manual and fully open the final control element – allowing the PV to rise much faster. When
the PV reaches near the setpoint, the operator now closes the valve down to near the correct
value, or switches the controller back to Auto.
Whilst the foregoing discussion certainly provides good grounds for implementing derivative
action, the benefits described above are secondary. The primary reason for applying
derivative control is to improve stability.
We have seen that one of the problems of integral action is that in a loop tending towards
instability, integral action will aggravate the condition.
Derivative action can counter this condition. Again, if we apply a sinusoidal waveform to the
input of the two parallel function blocks, shown in Figure 7.41, then the output of the
proportional block should be in-phase whilst that of the derivative block will lead by 90°. The
resultant of the two outputs will lead by a factor determined by the amplitude of the
proportional and derivative components.

Proportional
output (in-phase)
Input
P
Σ
Resultant output
D
Derivative output
(leads by 90°)
Figure 7.41. With a sinusoidal waveform applied to the input of the two parallel function
blocks the output of the proportional block is in-phase whilst that of the derivative block
will lead by 90°. The resultant will lead by a factor determined by the amplitude of the two
components.
This is again shown in Figure 7.42, which shows derivative action leading the proportional
input by 90°. Once again, in considering the controller alone, the resultant magnitude and
lead will be determined by the relative magnitudes of the derivative and proportional vectors.
Resultant
Figure 7.42. Derivative action leads Derivative action
the proportional input by 90°.
Proportional output
If now derivative action is applied to the resultant shown in Figure 7. 37, due to a large
integral action, the overall lag of the controller output will be reduced (Figure 7.43).
Resultant due to
derivative action
Figure 7.43. If derivative
action is applied to the
resultant shown, due to a
large integral action, the Proportional output
overall lag of the controller
output will be reduced. Resultant due to
addition of integral
and derivative
Resultant due to actions
integral action

In the unstable condition shown in Figure 7.39, the effect of derivative action is, therefore, to
bring the overall loop into a stable condition (Figure 7.44).
Integral
action
Figure 7.44. Derivative

action may be used to bring
the overall loop into a stable
condition.
Process lag
Derivative action
Reduced resultant
due to derivative
action Proportional output
The block diagram is shown in Figure 7.45.
Integral
SP I
ERR Proportional
Σ P
OP
PV
Process Derivative
fluid in D
Steam in
Thermocouple
Process Steam out

fluid out
Figure 7.45. Full PID controller incorporating derivative action.

In order to fully appreciate the effect of derivative action on the control system it is important
to first appreciate the concept of differentiation. Differentiation is simply the action of
determining the rate of change of a parameter. So, if a step function is applied to the input of
a derivative block the theoretical output, (Figure 7.46) will be a sharp needle pulse whose
amplitude is determined by how rapidly the step function moves from its quiescent position
(A) to its new position (B). In practice, the output will not be a needle pulse but will decay in
such a manner as to stretch the pulse out. This is usually due to the influence of the capacitive
filtering system incorporated within the controller.
(B)
Error
signal
Figure 7.46. A step function applied to
(A) the input of a derivative block produces
Needle a sharp needle pulse whose amplitude
Derivative
pulse is determined by the rise time from (A)
output
to (B). The real response will decay and
(ideal)
stretch the pulse out.
Derivative Stretched
output pulse
(real)
Figure 7.47 shows the output of each individual function block, which combine to produce the
output shown.
Derivative output
D
Proportional output
Step error input
P Σ
Resultant output
I
Integral output
Figure 7.47. The output of each individual function block

combines to produce a composite output (OP).

It is important to recognize that that the derivative acts, not on the amplitude but on the rate of
change of the input – in this case the error signal. The result is that derivative action produces
a higher output for high frequency noise than low frequency noise. In general, derivative
action amplifies noise and, as a result, it is customary to incorporate a low pass adaptive filter
within the controller in order to get rid of any high frequency noise within the system.
Earlier, it was noted that if the process lag is small it is impossible to produce a total loop
phase lag that equals 180° and become unstable. In general, flow control loops have very
small process lags and would, therefore, never require the application of derivative action.
Note
Here is a statement taken straight from a so-called authoritative source:
“Derivative action can be added if desired. If the process value signal is not noisy and the
derivative is on the measured PV and not on the error, derivative action should stabilize
the controller. In general, if derivative action is added, the controller gain (KC) can be
increased and the integral term (Ti ) can be decreased. BEWARE: derivative action
often behaves poorly on industrial processes.”
Strictly, what they’ve said is true. But, it’s very misleading since it would appear to advocate
that we shouldn’t use it. Indeed several organizations have taken such statement so much to
heart that they actually prohibit the use of derivative action.
The fact is that derivative does act poorly on many industrial processes because of excessive
noise on the PV signal. The solution lies, not with the arbitrary prohibition of derivative
action but in applying good instrumentation practice to minimize the noise in the first place.
Derivative is normal expressed in time setting of minutes although the Fieldbus Foundation
has settled on time expressed in seconds.
7.9 Digital control
The increasing use of digital processing in process control has brought numerous benefits.
However, it must be remembered that the process variable e.g. temperature, pressure, flow,
etc. is invariably an analog signal. The conversion into a digital signal requires the analog
signal is sampled at a regular intervals in order to build up a representation of the original that
is as accurate as possible.
During each sample, separate calculations are made to determine the proportional, integral
and derivative values. These independent calculations are then used to increment or
decrement the existing output value. In other words, the calculations do not determine the
absolute value of the output but only changes that have occurred during the sample period.
This is called an Incremental Algorithm. This is illustrated in Figure 7.48.

OP2
Figure 7.48. In the Incremental

Algorithm, the calculations do not
OP1 determine the absolute value of the output
but only changes that have occurred
PV during the sample period.
Scan period
ERR2
ERR1
SP
T1 T2
At time T1 the difference between the PV and the SP will give rise to an error ERR1. This
ERR1 acting on the controller KC will thus give rise to an output:
KC x ERR1 = OP1.
And at T2, the difference between the PV and the SP will give rise to an error ERR2 that will
act on the controller KC to give an output:
KC x ERR2 = OP2.
In this case, OP2 is a totally independent calculation and is not dependent on the previous
value OP1.
In practice the application of the Incremental Algorithm is largely transparent to the user.
However, there are consequences that the user needs to be aware of.

7.10 Control algorithms
Having examined the function of each block (P, I and D) we now need to look at how they are
configured within the PID controller.
We have already seen how different manufacturers interchange the different nomenclatures
for the proportional block (proportional gain vs. proportional band) and integral
(minutes/repeat, just plain minutes, and repeats/minute). Equally, different manufacturers
organize the functional blocks into different arrangements, normally referred to as control
algorithms.
In essence there are three basic forms of the PID algorithm. These are:
¾ Parallel
¾ Ideal
¾ Series
Unfortunately, there is again a tremendous variation in the names used for these categories, by
different manufacturers and they often neglect to inform the user as to the category at all.
Indeed, often the only way to really tell which one you have is to examine the controller
equation – that is, if the manufacturer has provided one.
Many controllers are available with only one fixed arrangement but some PLC and DCS
systems allow a selection between different configurations.
Parallel algorithm
As the name implies, in the parallel algorithm the three function block are arranged in
parallel as illustrated in Figure 7.49. This is (as we’ve seen) the configuration that is most
often used as a teaching model and
is also often referred to as ‘ideal
P parallel’, non-interacting’,
‘independent’, or ‘gain
independent’.
Error input Output – Process

(ERR) Demand (PD) Figure 7.49. Parallel algorithm in
I Σ which the three function block are
arranged in parallel.
On the face of it this algorithm might appear ideal except, maybe, from the fact that each
function block has its own amplifier – adding to the overall cost. In reality, the independence
of each function block leads to a rather strange problem.

Consider the output of a parallel PI controller as illustrated in Figure 7.52
Proportional output
Step error input

K
Figure 7.52. Block
diagram of combined
Σ Resultant output proportional and
integral action.
I
Integral output
We saw earlier that if the contribution from the proportional block had a magnitude 10%, then
the integral time is given by the time it takes to raise the output a further 10%. In other
words, the contribution of the integral action has 'repeated' the proportional contribution in T
minutes.
In Figure 7.53 (a) the controller gain KP = 1 and the integral time is set such that it will repeat
the proportional action in T minutes.
Figure 7.53 (a) the controller gain

Output
10% integral
step KP = 1 and the integral time is set to
repeat the proportional action in T
10% proportional minutes. (b) If the gain is halved (KP
step = 0.5) the proportional action
produces an output step of 5% but
T minutes the integral action ramps up at the
same rate and the repeats in half the
(a) time (T/2 minutes).
Output
If the gain is halved (KP = 0.5) then

for a 10% step change in error, the
5% integral step proportional action would produce an
5% proportional step output step of 5% (Figure 7.53 (b).
However, in a parallel algorithm the
integral action would continue
T/2
(b) ramping up at the same rate such that
minutes
the proportional action is repeated in
half the time (T/2 minutes). In effect, therefore, when the gain is reduced, because the
integral action remains the same, it gets faster in relation to the proportional action and the
loop tends towards instability.

Another way of looking at this is to consider the vector diagram of Figure 7.54. This shows
that the phase lag of the resultant of Proportional and Integral action is still short of 180° and
is, therefore still stable. And, as we saw
Integral
action
earlier, increasing the Integral action
may well cause the resultant to meet the
classical conditions for instability: a
phase lag of and a gain of equal to or
greater than 1 (Figure 7.55).
Resultant
Figure 7.54. The phase lag of the
Process lag resultant of proportional and integral
Proportional output
action is still short of the conditions
for instability: a phase lag of 180° and
a gain of equal to or greater than 1.
Integral
action
Figure 7.55. Increasing the integral

action may cause the resultant to meet
Resultant the classical conditions for instability.
Proportional output Process lag But there is another factor that could
produce instability and that is a
reduction in the gain. This is shown in
Figure 7. 56. Here reducing the
proportional gain has brought the phase
shift to 180°. However, the amplitude of
Integral
action
the resultant is reduced and thus my not
result in an overall gain of equal to or
greater than 1.
Resultant
Figure 7.56. Increasing the
proportional action may cause the
resultant to meet the classical
Proportional output Process lag conditions for instability.
This illustrates that, despite first

examination it is, indeed, interactive.

Figure 7.57 shows the PV response of PI controller with a parallel algorithm to a 10% step
change in the Setpoint with the proportional gain (KP = 1) and the integral time set to 0.17
minutes. This shows a failrly rapid rsponse with a single overshoot.
65.0
10% Setpoint change
60.0
(engineering units)
Process Variable
PV response
55.0
KP = 1
50.0
I = 0. 17
45.0
0.0 30.0 60.0 90.0 120.0 150.0
Time (seconds)
Figures 4.57. Typical PV response of PI controller with a parallel algorithm

to a 10% step change in the Setpoint with the proportional gain (KP = 1) and
the integral time set to 0.17 minutes.
In Figure 7.58, the integral time remains the same but the gain has been reduced to 0.5.
Contrary to initial expectations the system has become much more cyclic – tending towards
instability.
65.0
60.0
(engineering units)
Process Variable
PV response
55.0
10% Setpoint change
KP = 0.25
50.0
I = 0. 17
45.0
0.0 30.0 60.0 90.0 120.0 150.0
Time (seconds)
Figures 4.58. Here, the integral time remains the same at 0.17 minutes but the
proportional gain has been reduced (KP = 0.5).

Ideal algorithm
The difference between this algorithm and the one described previously are both very small
and very confusing because this is also described as ‘ideal parallel’, ‘non-interacting’, and
‘ISA’. And to muddy the waters still further, at least one manufacturer refers to it as
‘interacting’.
The term ‘ISA’ is incorrect since the ISA has no association with this algorithm. It would
appear that this appellation arose because someone working on Fieldbus standard erroneously
thought that it would become the ‘Standard’ algorithm. It didn’t. In the ANSI/ISA-S51.1-
1979 (Rev. 1993) standard on Process Instrumentation Terminology this algorithm is used to
describe various examples. However, it not intended that this would be the ‘Standard’
standard.
The algorithm is illustrated in Figure 7.59 and shows that the P, I and D control blocks are
(more or less) independent of each other. The term ‘ideal non-interactive’ might better be
thought of as not so much as describing the algorithm as identifying it. Further, the term
‘ideal’ does not imply that it is an ideal system but rather that it is only suitable under ideal
conditions where an ideal, noise-free PV exists – not the typical input from a field sensor. A
classical teaching model, most frequently used in textbooks, this algorithm is also used for
computer optimisation and
process modelling.
P
SP
Figure 7.59. The
OP ‘ideal non-
Σ K I Σ
interactive’
algorithm in which
the P, I and D
PV control blocks are
independent of
D each other.
This particular arrangement has the advantage over the parallel algorithm in that only one
amplifier is used and that changes in the gain (K) actually affect both the integral and
derivative actions.
Surprisingly this makes it increasingly non-interactive as far as adjustments are concerned.
In Figure 7.60 (a) the controller gain is again KP = 1 and the integral is set to repeat the
proportional action in T minutes.
Now, if the gain is halved (KP = 0.5) then, as before the proportional action produces an
output step of 5%for a 10% step change in error (Figure 7.60 (b). However, in the ideal
algorithm, where KP is multiplied by the integral term the integral ramp rate is also reduced
by half such that the proportional action is repeated, as before in T minutes). In effect,
therefore, when the gain is reduced the integral action is also reduced and remains the same in
relation to the proportional action.

Figure 7.60 (a) the controller gain

Output
10% integral
step KP = 1 and the integral time is set
such that it will repeat the
10% proportional proportional action in T minutes.
step (b) If the gain is halved (KP = 0.5)
T minutes
the proportional action produces an
output step of 5% but the integral
(a) ramps up at half the rate and
remains the same in relation to the
proportional action.
Output
5% integral step
5% proportional step
T minutes
(b)
Again, we can look at this another way by considering the vector diagram of Figure 7.61.
Integral
action
Figure 7.61. As the proportional gain is
varied, the integral gain is also
changed. Although the magnitude of
the resultant changes, the phase angle
Resultant remains the same.
Process lag As the proportional gain is varied, in

Proportional this case reduced, the integral gain is
output also reduced. Thus, although the
magnitude of the resultant changes, the
phase angle remains the same.

Series algorithm
The origin of the series algorithm has its roots in pneumatic control where designers found it
difficult to build parallel structures using pneumatic components. With the introduction of
both analog and digital systems, many manufacturers introduced designs that were based on
the series algorithm.
The standard series algorithm is illustrated in Figure 7.62.
KI =1 KD =1
SP
ERR OP
Σ KP I Σ D Σ
PV
Figure 7.62. The standard series algorithm.
When we first examined Derivative action we saw that if a step change is applied to a
derivative block, the theoretical output, (Figure 7.63) will be a sharp needle pulse whose
amplitude is determined by how rapidly the step function moves from its quiescent position
(A) to its new position (B).
(B)
Error
signal
(A) Figure 7.63. A step function applied to
Derivative Needle the input of a derivative block produces
output pulse a sharp needle pulse whose amplitude
(ideal) is determined by the rise time from (A)
to (B). The real response will decay and
stretch the pulse out.
Derivative Stretched
output pulse
(real)
In older pneumatic and analog electronic controllers the response was limited by natural lags
inherent in the system and, in practice, the output will not be a needle pulse but will decay in
such a manner as to stretch the pulse out.
However, in a digital system the maximum output value is only be limited by the resolution of
the D/A converter used within the system. Moreover, since, as we saw earlier, a digital system
is likely to make use of an incremental algorithm, then even the smallest change from the
previous scan would be seen as a step input and the controller would increment or decrement
the existing output value. As a result, the use of a straight derivative block as shown in Figure
7.62 would be unusable, since the output would be continually jumping up and down between
limits.

In order to overcome this problem, it is common practice to include a low pass filter in front
of the derivative block as illustrated in Figure 7.64 that limits the excursion and stretches the
pulse out as shown.
KI =1 KD =1
SP
ERR Low pass OP

Σ KP I Σ filter
D Σ
PV
Figure 7.64. The ‘real interactive’ algorithm used for direct field interaction.
It is important to appreciate that this filter is not intended to be used as a noise filter and that if
excessive noise exists on the PV then use should be made of the proper noise filter within the
system that filters the incoming PV signal. This is frequently referred to as the ‘PV filter’.
A serious drawback of this arrangement is that each time the Derivative value (TD) is
changed, it would be necessary to change the filter time constant. Consequently, in practice,
the low pass filter response TLPF is configured so that it is directly geared to the value of the
derivative time, TD.
TLPF = αTD
where α is a constant and is usually fixed by the manufacturer at some value between ¼ and
1
/8.
⎡⎛ 1 ⎞ (1 + sTD ) ⎤
PD = K ⋅ ERR ⎢⎜⎜1 + ⎟⎟ ⋅ ⎥
⎣⎝ sTI ⎠ (1 + s αTD ) ⎦
With α fixed at 1/8 , for example, the derivative component would be 8 times more powerful
than the Low Pass filter.
Users should be aware that the value of α is sometimes user adjustable and arbitrarily set to a
nominal default by the manufacturer. This can lead to ridiculous situations. Michael Brown, a
well known practitioner in the field of loop optimisation, reported on a plant where the default
value of α had been set at unity. The derivative chain thus becomes:
⎡ (1 + sTD ) ⎤
⎢( ⎥
⎣ 1 + sTD ) ⎦
and the filter directly cancels out the derivative term. Totally pointless!
In another incident he reported on a case where the manufacturer’s default had been set to 2.5.
Now, the denominator is larger than the numerator and the complete block acts as large lag.

7.11 Alternative Control algorithms
In a conventional PID loop (Figure 7.65), control is based on the ERROR signal – the
difference between the SP and the PV. The ERROR signal itself cannot thus differentiate
between a disturbance of the PV and a change in the SP.
P
Figure 7.65. In a
SP conventional PID loop
(Type A equation) control
ERR OP is based on the ERROR
Σ K I Σ signal (SP – PV).
PV
D
In most processes, changes in the PV are normally smooth. However it is possible that the
operator can make a step change to the SP resulting in a step change in the ERROR signal –
with possible major damage occurring as a result. The above algorithm is usually referred to a
PID – inferring that the ERROR signal acts on all three components: the Proportional, the
Integral and the Derivative. This is also sometimes referred to as a ‘Type A equation’.
One solution is to implement the ‘Type C equation’ as shown in Figure 7.66.
SP
ERR OP
Σ I Σ Figure 7.66. In a ‘Type C’
equation the ERROR
signal acts only on the
PV Integral. Step changes
made to the set point SP
K P are only applied through
integral action. This
action is often referred to
as I-PD control.
D
In this algorithm, normal (relatively slow) fluctuations of the PV are applied directly to the
Proportional and Derivative components whilst the ERROR signal acts only on the Integral.
This, in turn, implies that step changes made to the set point SP are also only going to be
applied through integral action. This action is often referred to as I-PD control. The result is
that even if the operator should make excessive step changes, the OP will only ramp up or
down under the action of Integral action.

Although popular in the past, most systems now rather incorporate a built-in ramp generator
or ‘setpoint velocity limits’ to ensure that even if the operator should make step changes, to
the SP, the ERROR signal ramps up or down at a predetermined rate that is independent of
the Integral setting.
A third solution, the ‘Type B equation’, is shown in Figure 7.67.
P
Figure 7.67. In a
SP ‘Type B’ equation
the ERROR signal
ERR OP is not applied
Σ K I Σ through Derivative
but only through
Proportional and
PV Integral. This
action is often
D referred to as PI-D
control.
The major implication of this type of action is that the ERROR signal is not applied through
Derivative action. In any control loop where Derivative action is applied, large set point
changes can result in excessively large changes to the OP. This is simply the result of the
Derivative action acting on the rate of change of the ERROR step input.
In the ‘Type B equation’, only the relatively slow PV changes are applied to the Derivative
control action. This action is often referred to as PI-D control.
In the normal PID algorithm where derivative action is base on the ERROR signal, the system
provides a faster initial take-off and will suppress overshoot for a set point change. However,
in the PI-D algorithm, where derivative action is based on variations of the PV, it works
against a setpoint change, since it is unaware that the PV should be changing initially and that
the brakes should only be applied to the process as it approaches setpoint.
Again, many manufacturers rather make use of some form of setpoint velocity limiting to
prevent a large step change in error, from a manually entered setpoint change, resulting in a
large output change.
Largely unknown to users, this configuration is used as a default for the vast majority of
controllers on the market. Indeed, in some cases manufacturers only provide this one
configuration.

Section 7. PID Control

Uploaded by

Copyright:

Available Formats

Section 7. PID Control

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Section 7. PID Control

Uploaded by

Copyright:

Available Formats

Section 7.

Process Control and

7.2 On/Off Control

©Mick Crabtree 2007 Page 7.2

Open Open Open

Off Off Off

©Mick Crabtree 2007 Page 7.3

Figure 7.3. Static characteristic shows that the

Off Off Off

Figure 7.4. Illustrating the effect of ON/OFF control with deadband.

©Mick Crabtree 2007 Page 7.4

7.3 Proportional control

ERR = SP - PV Controller Reverse action

Process Steam out

©Mick Crabtree 2007 Page 7.5

The static characteristic for proportional control is shown in Figure 7.6.

Figure 7.6. The static characteristic for

Output Several things should be noted. Firstly, when

©Mick Crabtree 2007 Page 7.6

Proportional gain and Proportional Band

Figure 7.8. When the

75% Figure 7.9. When the

measured temperature range

©Mick Crabtree 2007 Page 7.7

20 40 60 80 100 120 140 160 180

Proportional Band = 100% /K

ERR = 25°C Controller OP = 50%

©Mick Crabtree 2007 Page 7.8

©Mick Crabtree 2007 Page 7.9

7.4 Manual bias

One of the drawbacks of using

©Mick Crabtree 2007 Page 7.10

7.5 Integral control

Integral Figure 7.15. Basic

5 (B) 10 Figure 7.16.

©Mick Crabtree 2007 Page 7.11

Figure 7.18 shows a block diagram of PI controller incorporating automatic reset.

Figure 7.18. Block diagram

Process Steam out

©Mick Crabtree 2007 Page 7.12

Step error input

Figure 7.20. Block

©Mick Crabtree 2007 Page 7.13

10% integral Figure 7.21. Reset

0.67 min 1 min 2 min

©Mick Crabtree 2007 Page 7.14

7.6 Integral windup

Figure 7.22 shows a flow control loop in

PV Figure 7.22. Flow control loop (P + I)

©Mick Crabtree 2007 Page 7.15

©Mick Crabtree 2007 Page 7.16

OP Figure 7.26. Anti-reset

©Mick Crabtree 2007 Page 7.17

Figure 7.28. Simplified block

©Mick Crabtree 2007 Page 7.18

1 4 180° phase shift

©Mick Crabtree 2007 Page 7.19

Referring to Figure 7.31, this can be shown in another manner:

Figure 7.31. Examination of