Section 8.

Loop Tuning

Process Control and

Instrumentation
Process Control and Instrumentation Loop Tuning

Section 8
Loop Tuning
8.1 Introduction

In order secure all the economic benefits that can accrue as a result of implementing advanced
control, all the loops within the plant should, ideally, perform optimally. This, however, is not
economically feasible. As a result, every effort should be concentrated on key control loops
that directly affect product-measured qualities such as:
¾ yield;
¾ purity;
¾ cost to produce; and
¾ up time (reliability).

By now, it should be apparent that the field equipment must perform to certain minimum
requirements.

“No amount of advanced control that relies on the use of poor field instrumentation can be
expected to yield worthwhile benefits. Think of control as a hierarchy: everything must work
well at the lower levels for the higher levels above to work.”
M.J. Oglesby (ICI Engineering Technology)

So what is the minimum acceptable equipment performance?

Generally, the minimum performance requirements laid down by industry are:

Process gain > 0.5 and < 2

Hysteresis < 2% (1% with positioner)
Stiction < 0.5%
Noise < 2%
Positioner overshoot < 20%
Linearity (G max/G min) <2

If you are not meeting these requirements, any attempts to optimise your loop by ‘twiddling’
with the knobs on your controller are likely to be a complete waste of time.

Inasmuch that process PID controllers vary considerably from manufacturer to manufacture,
so to do the processes on which they are to perform. As a result, loop tuning is often referred
to as an 'art form'. And so it was – until John G. Ziegler and Nathaniel B. Nichols delivered
their controversial paper in December 1941 at the annual meeting of the ASME.

The result was, the now famous "’Ziegler-Nichols’ methods of tuning controllers – methods
that have withstood the test of time and work just as well now as many of the later more
sophisticated optimising forms on the great majority of process applications.

©Mick Crabtree 2007 Page 8.2

Process Control and Instrumentation Loop Tuning

That is not to say that the Ziegler Nicholls method is the only, or even the best, method. A
fairly comprehensive list of the different methods available is listed below which tabulates 75
different methods!

Process reaction 43. Tan et al.

1. Ziegler and Nichols process reaction 44. Symmetrical optimum principle - Voda and
2. Hazebroek and Van der Waerden Landau
3. Astrom and Hagglund ultimate cycle 45. Friman and Waller
Ziegler-Nichols equivalent 46. Voda and Landau
4. Chien et al. - regulator 47. Smith
5. Astrom and Hagglund - regulator 48. Modulus optimum principle - Cox et al.
6. Chien et al. - servo 49. Cluett and Wang
7. Cohen and Coon process reaction 50. Abbas
8. Two constraints criterion - Wolfe Robust
9. Two constraints criterion - Murrill 51. Brambilla et al.
10. McMillan 52. Rivera et al.
11. St. Clair 53. Rivera et al. - alternative
Regulator tuning 54. Chien
12. Minimum IAE - Murrill 55. Thomasson
13. Minimum IAE - Shinskey 56. Fruehauf et al.
14. Minimum IAE - Huang et al. 57. Ogawa - Closed loop peak gain = 1.4; 20%
15. Minimum ISE - Hazebroek and Van der uncertainty in process parameters
Waerden 58. Ogawa - Closed loop peak gain = 1.4; 33%
Regulator tuning uncertainty in process parameters
16. Minimum ISE - Murrill 59. Ogawa - Closed loop peak gain = 1.4; 50%
17. Minimum ISE - Zhuang and Atherton uncertainty in process parameters
18. Minimum ITAE - Murrill 60. Lee et al. interpreted from graphs
19. Minimum ISTSE - Zhuang and Atherton Ultimate cycle
20. Minimum ISTES - Zhuang and Atherton 61. McMillan
Servo tuning 62. Regulator - minimum IAE - Shinskey
21. Minimum IAE - Rovira et al. 63. Regulator - minimum IAE - Shinskey
22. Minimum IAE - Huang et al. 64. Regulator - nearly minimum IAE, ISE,
23. Minimum ISE - Zhuang and Atherton ITAE - Hwang
24. Minimum ISE - Khan and Lehman 65. Regulator - nearly minimum IAE, ISE,
25. Minimum ITAE - Rovira et al. ITAE - Hwang
26. Minimum ISTSE - Zhuang and Atherton 66. Regulator - nearly minimum IAE, ISE,
27. Minimum ISTES - Zhuang and Atherton ITAE - Hwang
Direct synthesis 67. Servo - small IAE - Hwang
28. Haalman [16] 68. Servo - minimum ISTSE - Zhuang and
29. Minimum IAE - regulator – Pemberton, Atherton
Smith and Corripio 69. Regulator - minimum ISTSE - Zhuang and
30. Minimum IAE - servo - Smith and Corripio Atherton
31. 5% overshoot - servo - Smith and Corripio 70. Servo - nearly minimum IAE and ITAE -
32. 5% overshoot - servo - Hang et al. Hwang and Fang
33. Regulator - Gorecki et al. 71. Regulator - nearly minimum IAE and ITAE
34. Regulator - Gorecki et al. - Hwang and Fang
35. Chiu et al. 72. Simultaneous Servo/regulator - Hwang and
36. Astrom et al. (Honeywell UDC 6000 Fang
controller) Maximum sensitivity
37. Schneider 73. Astrom and Hagglund - = 1.4 - dominant
38. McAnany pole design
39. Leva et al. 74. Astrom and Hagglund - = 2.0- dominant
40. Khan and Lehman pole design
41. Hang et al. 75. Astrom and Hagglund - modified Ziegler-
42. Gain and phase margin - Ho et al. Nichols

Why so many? Well, much depends on the type of process itself, and what we are trying to
achieve. To start with there are vast differences in the requirements for a typical flow loop
and those on a surge level tank (Table 8.1).

©Mick Crabtree 2007 Page 8.3

Process Control and Instrumentation Loop Tuning

Table 8.1. Some guidelines for common control loops

Type of loop Characteristics Required control
Flow and liquid Fast response with no time delay PI controller with intermediate controller
pressure control Usually with small high-frequency noise gain
Liquid level control Noisy due to splashing and turbulence High gain PI controller for integrating
process.
Conservative setting for averaging
control when used for damping the
fluctuation of the inlet stream.
Gas pressure control Usually fast and self regulating PI controller with small integral action
(large reset time).
Temperature control Wide variety of the process nature Use PID controller to speed up the
– Usually slow response with time delay response.
Composition control Similar to temperature control usually Effectiveness of derivative action is
with larger noise and more time delay limited.

We then need to decide: are we trying to optimise on process disturbances or are we trying to
optimise on setpoint changes? Are we trying to minimise overshoot or, indeed, even
eliminate overshoot altogether (start-up on a pressure vessel)? And just how do we determine
performance?

8.2 Performance measures

No system can respond instantly to disturbances and changes in the setpoint. Consequently,
before we can assess how well a particular control scheme is likely to work, we must first
determine a set of performance criteria.

Assuming that the closed loop response is essentially a second order system, any one of the
responses shown in a Figure 8.1 could be used.

Setpoint change

Unstable:
Increasing
oscillation
Figure 8.1. Possible closed loop responses of
a second order system, to a step input change
Continuous in the setpoint.
oscillation (r = 0)

Under damped
(r > 0 < 1)

Critically damped
(r = 1)

Over damped
(r > 1)

©Mick Crabtree 2007 Page 8.4

Process Control and Instrumentation Loop Tuning

Theoretically, of course, the time taken for the system to settle completely after a change in
setpoint is infinite. However, in practice a ‘tolerance limit’ is defined which can be expressed
as a percentage of the setpoint change. A typical value of the tolerance limit is 5% and once
the plant is within this band it is considered to have achieved the setpoint.

One of the most commonly used performance criteria is what is termed ‘Quarter Amplitude
Damping’ (QAD) in which each cycle peak is one quarter of the amplitude of the previous
peak (Figure 8.2).

1A Figure 8.2. In the quarter amplitude damped

reaction, each successive swing is reduced by
1/4A
half so that the second excursion is a quarter
the amplitude of the first.

Although, for many applications, this is an adequate and easily achievable response, it is not
necessarily either a desirable response or even an optimal response.

‘What’s wrong with QAD?’

There are several reasons why many process engineers are hostile towards quarter amplitude damping.

Firstly, they claim, it’s not safe enough – it doesn’t allow enough gain margin for process dynamic changes. This
could result in instability.

Secondly, as shown below, a quarter wave response consists of approximately 4 cycles. This means that the
valve has to reverse eight times every time there is a control error change.

Is this really desirable? Do you really want to work your valves so hard?

Thirdly, what happens if there is excessive deadband + hysterisis? On each reversal of the valve the controller
output has to ramp through the hysteresis band under the command of the controller's integrator. This slows
down the control, and increases variance. On slow processes such as temperature it can literally take hours
before the process finally settles out. This means that on many loops, QAD tuning results in much worse control
variance than even critical damping, where no valve reversals at all are made on a control error change.

At first glance it might be assumed that we want the process to follow a setpoint change as
fast as possible. In other words, the basic aim is to keep the error signal to a minimum. Thus,
in the response shown in Figure 8.3 (a) the shaded area is the integral of the error and could
be used as an index of performance: the smaller the area, the better the response.

©Mick Crabtree 2007 Page 8.5

Process Control and Instrumentation Loop Tuning

So, if we plotted the integral value on the same timescale, we'd end up with a curve that
attains a maximum value after all the oscillations have died down (Figure 8.3 (b)).
Process
Setpoint variable
+
+ + +
− − − −

Object: minimal area

Figure 8.3.(a) The basic goal is to
Time
(a) reduce the area (the integral) to a
Integral (area under

minimum.(b) The integral value

attains a maximum value after all
curve)

the oscillations have died down.

Total area of
the error signal

Time
(b)

Well, not really! Bearing in mind that the error signal is varying between a negative and a
positive value, we'd actually end up with an integral value that would also vary sinusoidally.
In order to overcome this problem we would thus have to take the integral value of the
absolute error – usually referred to as the Integrated Absolute Error (IAE).*
∞
Where: IAE = ∫ e (t ) dt
0
Please note that this performance criterion, and the ones that follow, merely defines a mathematical model
chosen for optimisation and which is used primarily for academic purposes, together with process
simulations, in the study of control algorithms. This has nothing to do with the integral action associated
with the controller.

However if, as occurs with a P only controller, the PV never attains the same value as the
setpoint, there is a standing offset (Figure 8.4 (a)) and the area under the curve (and thus the
integral value) will not converge to a final value but will increase, with time, without limit
(Figure 8.4 (b)).
Process
Set point variable
Proportional
offset

Figure 8.4. (a) The PV in a P only

controller never attains the same
value as the setpoint and there is
Time a standing offset. (b)
(a)
Integral (area under

Consequently, the integral value

will not converge to a final value
curve)

Increasing
area of the
but will increase without limit.
error signal

Time
(b)

©Mick Crabtree 2007 Page 8.6

Process Control and Instrumentation Loop Tuning

It must also be appreciated that merely reducing the integral is not necessarily the solution
that we’re looking for either. For example, in the response shown in Figure 8.5, the integral
(the shaded area) is definitely smaller. Unfortunately the peaks are also considerably larger.

Process
variable

Figure 8.5. Although the integral (the

shaded area) is smaller, the peaks are
considerably larger.

This may be overcome by taking the integral of the square of the error – known as the Integral
of the Square Error (ISE).
∞ 2

Where: ISE = ∫ [e (t )] dt
0

Now, large errors will contribute more and there will be a larger a penalty for a large
overshoot. Conversely small errors contribute less and there is thus a smaller penalty for
small persisting oscillations.

Although this solution minimises overshoot, it is at the expense of long settling time, and is
not directly related to economic performance.

Errors that are likely to occur later in time, rather than normal transients, can make use of
time-dependant functions such as the Integral of the Time-Weighted Absolute Error (ITAE)
or Integral of the Time Squared Error (ITSE)
∞

Where: ITAE = ∫ e (t ) ∗ t dt
0

∞ 2

and: ITSE = ∫ [e (t )] ∗t dt
0

Both these solutions emphasize the minimisation of error duration and are suitable for batch
operation or large set point changes. On the negative side there is the potential for
considerable overshoot. In addition they are quite removed from direct economic
performance.

©Mick Crabtree 2007 Page 8.7

Process Control and Instrumentation Loop Tuning

8.3 Before we start

Before we embark on the actual process of tuning we need to appreciate that it involves
introducing some form of upset to the loop involved. We need, therefore, to be particularly
careful to ensure that we do not make things worse or even cause a shutdown. It is wise,
therefore, to take heed of some of the pre-test ‘back-stops’, proposed by David St. Clair and
Greg McMillan, that we need to put in place before we start tuning.

¾ You should have some idea of what you’re trying to do and what you can expect to do.
¾ You should have some idea of how fast and how far the process is going to react to
changes in the manipulated variable
¾ You should know and agree on how much change you will allow in the process and in the
manipulated variable.
¾ You should know and agree on who is going to make the actual adjustments to the
controller settings
o You should also be confident that they know how to do it physically
o You should also be confident that they are not likely to become confused if
something has to be done in a hurry
¾ You should know and agree on who is going to switch between manual and automatic
¾ You should know and agree on who is going to make the setpoint changes
¾ You should have knowledge of the safety interlocks and any other safety concerns for the
process
¾ You should pick a time to tune such that you will be available for a reasonable time
afterwards should problems arise
¾ You should record the existing settings and the controller output in the event you want to
return to them either in a hurry or simply to the system as you found it.
¾ You should make sure the changes are communicated to all operating person.
¾ If a log book is kept for this purpose, use it.

8.4 Basic tuning principles

We obviously don't have the space, the time, or the patience to cover all 75 tuning methods.
So, we will examine some of the more traditional methods that have stood the test of time.
These include those proposed by Ziegler-Nicholls, Cohen-Coon, and Pessen. We’ll also look
at Lambda tuning.

The basic procedure is simple: use Proportional Control as the principal method; add Integral
Control to remove offset; and then add Derivate Control (if necessary) to remove the
destabilising effect of Integral Control.

We stated earlier that the tuning process involves introducing some form of upset to the loop
involved.

©Mick Crabtree 2007 Page 8.8

 Process Control and Instrumentation Loop Tuning

8.5 Open loop reaction curve

In many regards the open loop reaction method is the most obvious solution. We introduce a
step change to the Process Demand (PD) and observe the Process Variable (PV). By
observing the process reaction we can immediately obtain information regarding the dynamics
of the process itself. This makes this method very useful for troubleshooting since we usually
have some idea as to what the response should look like. Consequently, if it doesn’t match up
to what we had expected, it gives us some clues as to the nature of the problem.

Critically, we can judge straight away if the process is self-regulating. Further, we can also
determine the process gain (KP) and the two critical time constants involved: the deadtime (θ)
and the response time (τ).

Armed with information regarding the dynamics of the process, we are now in a position to
adjust the controller parameters (Proportional, Integral and Derivative) to produce a ‘best-fit’
tuning solution.

Let’s look at the open loop reaction shown in Figure 8.6.

65.0

60.0
Percentage

55.0
∆PD
∆PV

50.0

45.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Time (seconds)

Figure 8.6. Open loop reaction showing the change in PV as a result of a step-change in
the process demand (PD).

The first thing to note is that this is a classical first order response plus dead-time (FOPDT).
The second point to note is that it is a self-regulating process in which the PV has evened out
to a new steady state value as a result of a step change applied to the process demand (PD).

We can also now calculate the process gain (Kp) which, as we saw earlier, is given by:
∆ PV
KP =
∆ PD
Examination shows that the change in output (∆PD) is:

60.0 – 50.0 = 10%

©Mick Crabtree 2007 Page 8.9

Process Control and Instrumentation Loop Tuning

And the change in process variable (PV) is:

62.0 – 46.0 = 16%

The process gain (Kp) is therefore:

∆ PV 16.0
KP = = =1.6
∆ PD 10.0
This conforms to the minimum performance requirement of the process gain lying between
0.5 and 2.

Our next task is to determine the deadtime (θ). In this case it is very easy to determine and, as
shown in Figure 8.7, is 10 seconds.
65.0

60.0
Percentage

55.0
∆PD
∆PV

50.0

45.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

θ Time (seconds)

Figure 8.7. The deadtime (θ) is easily determined as 10 seconds.

Our final task is to determine the response time (or time-constant (τ)) defined by the time
required for the variable to reach 63.2% of its final value (Figure 8.8).
65.0

60.0
Percentage

55.0
∆PD
∆PV
63.2%
50.0

45.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

θ τ Time (seconds)

Figure 8.8. The response time (τ) is defined by the time required for the variable to reach
63.2% of its final value.

©Mick Crabtree 2007 Page 8.10

Process Control and Instrumentation Loop Tuning

As we have seen, the change in process variable (PV) is:

62.0 – 46.0 = 16%

The value of 63.2% is thus:

(16 x 0.632) + 46.0 = 56.112%

and the response time (τ) (the time taken to reach this value from the end of the dead time
period) is:

45.2 – 25.0 = 20.2 seconds.

We now have the three values necessary to calculate our tuning parameters:

KP = 1.6
θ = 10 s
τ = 20.2 s

It is now only necessary to apply these parameters to the relevant formula as detailed in Table
8.3.

Table 8.3. Summary of open loop reaction curve formulae.

Mode of Control Zeigler-Nichols Cohen-Coon

P control τ τ ⎛ θ ⎞
Gain (KC) ⎜⎜1+ ⎟⎟
KP θ KP θ ⎝ 3 τ ⎠

PI control 0.9 τ 0 .9 τ ⎛ θ ⎞
Gain (KC) ⎜⎜1+ ⎟⎟
KP θ KP θ ⎝ 11 τ ⎠
⎛ τ + 0 .1 θ ⎞
Integral time (TI) 3.33 θ 3.33 θ ⎜⎜ ⎟⎟
⎝ τ + 2 .2 θ ⎠
PID control 1.2 τ 1.33 τ ⎛ 3θ ⎞
Gain (KC) ⎜⎜1+ ⎟⎟
KP θ KP θ ⎝ 16 τ ⎠
⎛ τ + 0 .2 θ ⎞
Integral time (TI) 2.0 θ 2.5 θ ⎜⎜ ⎟⎟
⎝ τ + 0 .6 θ ⎠
⎛ τ ⎞
Derivative time (TD) 0.5 θ 0.37 θ ⎜⎜ ⎟⎟
⎝ τ + 0 .2 θ ⎠

©Mick Crabtree 2007 Page 8.11

Process Control and Instrumentation Loop Tuning

Tuning method
1. Put the controller into MANUAL mode.
2. Make a small step change to the PD value.
The size of the step change will depend quite a lot of all local plant conditions. A large
change (10% or even greater) is desirable in order to clearly distinguish the change in PV
from measurement noise or other disturbances. However, in order to avoid detrimental
upsets to the process, the change should be as small as possible (less than 5%).
3. Record the response of the process (Figure 8.8).
4. Calculate P, I and D according to Table 8.3 above.

Lambda tuning
Lambda tuning is particularly effective for tuning relatively fast loops that are dead-time
dominant. Lambda tuning also helps to minimise interactions and suppress oscillations.

Lambda tuning provides an over-damped response that is extremely desirable for processes
operating at some critical limit. In the simplest sense, it attempts to achieve a non-oscillatory
closed-loop response having a predetermined time constant designated Lambda (λ). In effect,
therefore, we are trying to force the closed-loop response to conform to that shown in Figure
8.9 in which lambda (λ) is equal to the open-loop response time (τ).
Percentage

∆SP ∆PV
63.2%

θ λ Time (seconds)

Figure 8.9. Closed-loop response in which the closed-loop time constant (λ) is
modelled to equal the open-loop response time (τ).

The choice of the closed-loop time constant (λ) is determined by the degree of confidence in
the process model parameters KP, θ and τ. Typically it would vary from 1τ to 4τ.

Where there is a high degree of confidence use:

λ=τ
This will provide a relatively fast response to set point changes. If λ < τ the controller is
likely to be overly aggressive.
Where there is a low degree of confidence use a value:

λ = 2τ to 4τ.

©Mick Crabtree 2007 Page 8.12

Process Control and Instrumentation Loop Tuning

Having decided on the value for λ, the tuning relations may be calculated as follows:
τ
KC =
K P (λ + θ )
and:
θ2
TI = τ +
(λ + θ )
The above calculations are taken from H.L. Wade. It should be noted that there are several
variations in the methods used in determining the time constants and the formulae. For
example, instead of the open loop response time (τ) being taken as the time taken to reach
63.2% of the final value, it is often calculated as the time taken to reach 98% of the final value
and is referred to as (T98). This would obviously modify the tuning calculations accordingly.

Lambda tuning is widely used in the pulp and paper industry where a strong connection exists
between paper uniformity and upstream control loop interactions. Unlike Ziegler Nicholls,
Lambda tuning allows users to set the speed of response for each individual loop. Thus,
higher priority control loops can be tuned faster and lower priority loops can be tuned slower.
This freedom to select the response speed of individual loops helps reduce unwanted loop
interactions that are detrimental to the overall process performance. A simple example of this
is a cascaded control loop where the inner loop should be tuned five to 10 times faster than
the outer control loop.

Approximate FOPDT
In reality, few processes will exhibit the true FOPDT response illustrated in Figures 5.6 to 8.
Rather, because most processes comprise a series of lags, they will take on the S-curve
response in which there is no clear demarcation between where the deadtime (θ) ends and the
response time (τ) starts (Figure 8.10).

Figure 8.10. Most

processes comprise a
series of lags that take on
the S-curve response in
which there is no clear
100%
demarcation between
of PV
where the deadtime (θ)
∆PD
ends and the response time
(τ) starts.

θ τ

Indeterminate area This problem may be

overcome by drawing a tangent to the steepest part of the curve as shown in Figure 8.11.
What is now, strictly speaking, the ‘apparent’ deadtime (θ) is calculated from the point at
which a step was applied to the PD, and the point at which the tangent crosses the horizontal

©Mick Crabtree 2007 Page 8.13

Process Control and Instrumentation Loop Tuning

line indicating the previous steady state value of the PV. This point also marks the starting
point from which the response time (τ) may be calculated (63.2% of the PV).

Tangent to point of
steepest rise

100%
of PV
63.2%
of PV

θ τ

∆PD

Figure 8.11. The ‘apparent’ deadtime (θ) is calculated from the point at which a step was
applied to the PD, and the point at which the tangent crosses the horizontal line indicating
the previous steady state value of the PV. This point also marks the starting point from
which the response time (τ) may be calculated (63.2% of the PV).

It is not always easy to precisely determine the steepest part of the curve and hence the point
at which the tangent crosses the line may not be exact. In such cases, where some uncertainty
exists, is it better to round the figures are up or down?

Here it is useful to examine the Ziegler Nicholls formulae given in Table 8.3 and repeated
below:
1.2 τ
Gain (KC) =
KP θ

Integral time (TI) = 2.0 θ

Derivative time (TD) = 0.5 θ

Clearly, if you under-estimate the deadtime (θ)you are estimating that the process is easier to
control than it really is. This results in a higher calculated value for the gain (KC) and a
smaller calculated value for the integral time (TI) – both tending to produce oscillation and
subsequent instability in the control loop. It follows that over-estimating the deadtime will

©Mick Crabtree 2007 Page 8.14

Process Control and Instrumentation Loop Tuning

move the calculated values towards a lower gain and longer integral time – both tending
towards better stability.

A similar argument can be applied when estimating the value of the response time (τ) where,
if you estimate the time constant too long, you are estimating that the
process is easier to control than it really is. This over-estimation results in a higher calculated
value for the gain (KC) – again tending towards instability
.
Consequently, if there is any uncertainty in estimating the process parameters, the basic rules
are:

Use a longer estimate for the deadtime (θ).

Use a shorter estimate for the response time (τ).

Integrating processes
In the tuning methodologies given so far, it has been assumed that the processes are self-
regulating – with the PV value starting from some point of equilibrium and steadying out at a
new stable value.

In some cases, however, the process may not reach a new steady-state value. Such integrating
processes are typical of level-control loops. A pseudo-integrating process can also exist in, for
example, a temperature loop in which the new equilibrium value is well above the normal
operational region.

In either case, it is no longer possible to determine the gain graphically. However, there is
nonetheless, sufficient information to allow us to determine the controller tuning values. This
is shown in Figure 8.12 in which a tangent is drawn to the point of steepest rise and extended
below the initial equilibrium value to intersect the vertical line representing the step function
change in the PD.

©Mick Crabtree 2007 Page 8.15

Process Control and Instrumentation Loop Tuning

Tangent to point of
steepest rise

Figure 8.12. A tangent is drawn

to the point of steepest rise and
extended below the initial
∆C
equilibrium value to intersect
the vertical line representing the
step function change in the PD.
θ

∆PD

We can now define the following

parameters:

∆PD = step change of the PD.

∆C = vertical distance from the PV equilibrium value to the intersection of

the vertical line representing the step function change in the PD.

θ = apparent deadtime

From this we can calculate:

KP ∆C
=
τ θ ⋅ ∆PD
KP
The value may now be entered directly into the Ziegler Nicholls calculation for the gain
(KC). τ

8.6 Closed loop continuous cycling method (Ziegler-Nichols)

The main disadvantage of the open loop method is that it tries to force the process to look like
a FOPDT model. The closed loop model, on the other hand, makes no assumptions regarding
the form of the process model.

Simply, the closed loop continuous cycling method requires you to find a critical value of K
(called the ultimate gain KU) that will produce continuous oscillations – neither increasing nor
decreasing (Figure 8.13). At this point, we know that the phase shift through the process is
180° and that the total loop gain is 1.

©Mick Crabtree 2007 Page 8.16

 Process Control and Instrumentation Loop Tuning

Continuous constant
amplitude oscillation Figure 8.13. With a gain
(KU) that produces
continuous oscillations –
A neither increasing nor
SP decreasing –. the phase
shift through the process
is 180° and that the total
loop gain is 1.
PU

It now follows that in a system employing proportional action only, if the ultimate gain were
halved then each successive cycle of the oscillation would be half of the previous one and the
result is Quarter Amplitude Damping (QAD) (Figure 8.14).
Quarter
amplitude Figure 8.14.
damping If the ultimate gain is
halved, each successive
A cycle would be half of the
previous one and the
SP 1/4A
result is Quarter
Amplitude Damping.

Tuning method
1. Put controller in P control only – do not use I or D control. This would entail setting
TI to a maximum (or turning integral action off). If integral action is expressed in
Repeats/minute then it needs to be set to zero.
2. Set TD to zero.
3. If necessary set control to ERR = (SP - PV)
4. Set the controller to Automatic
5. Make a small step-change to the SP
6. Increase the value of K until continuous equal-amplitude cycling occurs

Note: A large amplitude is not required, and, in fact, need only be large enough to
distinguish the oscillation from the measurement noises band.

7. Record the value of gain – referred to as the ultimate gain (KU)

8. Measure the period of the cycling (e.g. from one peak to the next). This is called the
ultimate period (PU)
9. Calculate the tuning constants according to Table 8.4.

©Mick Crabtree 2007 Page 8.17

Process Control and Instrumentation Loop Tuning

Table 8.4. Calculation of P, I and D settings according to the closed loop continuous
cycling method (Ziegler-Nichols).

Mode of Proportional Integral time Derivative time

Control (KC) (TI) (TD)
P control
0.5 ∗ K U
- -
PI control PU
0.45 ∗ K U
1.2 -
PID
control 0.6 ∗ K U PU PU
2 8

There are several advantages of this method over the reaction curve method.

¾ Whilst the deadtime (θ) and the response time (τ) can often only be approximated, the
period can be measured very precisely. This means that the quality of the tuning
parameters is much higher.
¾ The loop stays in Auto – thus making it safer for difficult and very fast loops.
¾ The test includes the effects of valve hysterisis and dead-band as well as system non-
linearities that are dependent on direction and rate of change.
¾ The test also includes the dynamics and special features of different controller algorithms.

Despite these advantages there are several reasons why it is difficult to implement the closed
loop test:

¾ In many applications, a sustained oscillation may not be acceptable to supervisory and

operations personnel.
¾ It is sometimes difficult to control the amplitude of the oscillation – with a small change
in set point yielding a larger than expected oscillation amplitude.
¾ Several tests may be required over a relatively long period of time – during which the
production output is out of spec. This again, may not be acceptable to the production
personnel.

8.7 Damped oscillation method

In applications where a continuous oscillation cannot be tolerated, use can be made of the
damped oscillation method.

Tuning method
1. Put controller in P control only – again do not use I or D control
2. If necessary set control to ERR = (SP - PV)
3. Set the controller to Automatic
4. Make a small step-change to the SP
5. Increase the value of K until you obtain a quarter amplitude damped response.
6. Record the value of the gain (KQU) that produced this response as well as the period
(PQU).

©Mick Crabtree 2007 Page 8.18

Process Control and Instrumentation Loop Tuning

Had the gain been increased to a level where continuous equal-amplitude cycling occurs it
would have been noticeably larger than the recorded value (KQU) by a fixed factor –
approximately 1.67.

The ultimate gain (KU) that would have given rise to equal amplitude cycling may thus be
estimated as:

KU (estimated) = 1.67* KQU

In a similar manner had a continuous equal-amplitude cycling occurred the recorded ultimate
period (PU) would be shorter than that actually measured by a factor of approximately 1.1.

Consequently, we can estimate the ultimate period as:

PU (estimated) = 0.9 * PQU

These estimated values, may now be applied and directly to the formula given in Table 8.4.

8.8 Alternative tuning constants

In both the continuous and damped oscillation methods the settings give rise to a closed loop
reaction that approximates to Quarter Amplitude Damping. However, as we have seen, in
many processes this is not desirable and little or no overshoot can be tolerated – at the
expense of a much slower response.

The formulae, according to Pessen, still make use of the closed loop continuous cycling or
damped oscillation methods according to Ziegler-Nichols but, as shown in Table 8.5, have
been adapted to cater for different response requirements.

Pessen 1, for example is used where no overshoot (even for start-up) is permitted. Typically
this might be tank level or a pressure vessel. And Pessen 2 is used where a slight amount of
overshoot is permitted.

Table 8.3. Comparison of tuning methods.

Closed loop Pessen 1 Pessen 2

Ziegler-Nichols
KC
0.6 ∗ KU 0.2 ∗ KU 0.33 ∗ KU

TI PU PU PU
2 3 2
TD PU PU PU
8 2 3

©Mick Crabtree 2007 Page 8.19

Process Control and Instrumentation Loop Tuning

8.9 Flow loops

Flow loops are a special case. Since they are inherently stable, it is not possible to carry out
the closed loop cycling method because a continuous oscillation cannot be sustained. On the
other hand, because the reaction of most flow loops is so rapid, use of the open loop reaction
method is barely useful because there is virtually no dead time.

Fortunately, flow loops are pretty consistent in their dynamics and, assuming a process gain
of around about 1, it is possible to apply the following default values:

KC = 0.8
TI = 0.1 minutes
Note: The above assumptions can only be made on the basis of having previously ascertained
that the final control element is not subject to excessive deadband, hysterisis or stick-slip and
that the transmitter is correctly spanned and that the process variable noise is minimal.

8.10 Level control systems

In listing the top 20 mistakes made in the field of process instrumentation, Greg McMillan
and Stan Weiner, two of the world’s great authorities in the field, put ‘tight control of level in
a surge tank’ at number 5.

As they point out, the purpose of the surge tank volume is to absorb changes in flow and for
the level to roll with the punches. Keeping the level within a few per cent of set point passes
them on and defeats the purpose of the tank. Indeed, most oscillations in a process can be
traced back to an over zealous person tuning a level controller to an excessively tight set of
low and high level alarm limits. This is a pity. Tight control is not easy to accomplish and
requires that the control loop should be engineered rather than just tuned.

However, level is rarely a factor in quality control and the storage capacity of the vessel is a
mostly used to absorb upsets. This is called averaging level control and is a fairly easily
implemented.

Averaging level control

The rules given below have been suggested by David St. Clair.

©Mick Crabtree 2007 Page 8.20

Process Control and Instrumentation Loop Tuning

First decide how much the level may vary. In a surge tank this may be 60 to 80%. Next,
determine the following parameters:

UL% = Allowed upper limit (%)

LL% = Allowed lower limit (%)
SP% = Setpoint (%)
V = Volume between 0 and 100% (m3)
Q = Maximum flow rate (m3/min) when valve is fully open

Then calculate:

KC1 = 100% /(UL% - SP%)

KC2 = 100% /(SP% - LL%)
KC = Larger of KC1 and KC2
TI = 4V/ (Q.KC)

These rules are adequate for the vast majority of level loops. The formula for the integral
time, TI assumes linear components throughout and no lags except the integrating process
itself. The point is to use gain to keep the level within limits and enough integral action to
satisfy performance needs.

Let's look at an example, as illustrated in Figure 8.15.

Figure 8.15. Example of averaging level control in, for example, a

surge tank.

8m
100 m3

4m

UL% = Allowed upper limit (%) = 70

LL% = Allowed lower limit (%) = 40
SP% = Setpoint (%) = 60
V = Volume = 100 m3
Q = Maximum flow rate = 5 m3/min

We can now calculate:

KC1 = 100% /(UL% - SP%) = 100/ (70 – 60) = 10

KC2 = 100% /(SP% - LL%) = 100/ (60 – 40) = 5
KC = Larger of KC1 and KC2 = 10
TI = 4V/ (Q.KC) = 4.100/5. 10 = 8 minutes

©Mick Crabtree 2007 Page 8.21

Process Control and Instrumentation Loop Tuning

But here’s another thought. If control to a set point is not required and a proportional offset
can be tolerated, then it is possible to dispense with integral action altogether and use only
proportional control.

With the manual reset set to 50%, the controller gain should thus be set to:

100
KC =
2 * ∆Lmax

where ∆Lmax is the maximum allowable deviation from setpoint as a percentage of full-scale.

Thus, for a maximum deviation of 5% the equation becomes:

100
KC = = 10
2 *5
With this arrangement, if a controller output of 50% is required, the level will be at setpoint

©Mick Crabtree 2007 Page 8.22