Modeling, Identification and Control, Vol. 34, No. 2, 2013, pp.

83–97, ISSN 1890–1328

Relaxed Ziegler-Nichols Closed Loop Tuning of PI

Controllers
Finn Haugen 1 Bernt Lie 1

Telemark University College, Kjolnes ring 56, 3918 Porsgrunn, Norway. E-mail contact:
{finn.haugen,bernt.lie}@hit.no

Abstract

A modification of the PI setting of the Ziegler-Nichols closed loop tuning method is proposed. The modifi-
cation is based on a combination of the Skogestad SIMC tuning formulas for “integrator plus time-delay”
processes with the Ziegler-Nichols tuning formulas assuming that the process is modeled as an “inte-
grator plus time-delay” process. The resulting PI settings provide improved stability margins compared
with those obtained with the original Ziegler-Nichols PI settings. Compared with the well-known Tyreus-
Luyben PI settings, the proposed PI settings give improved disturbance compensation. For processes with
zero or a negligible time-delay, but with some lags in the form of time-constants, tuning based on ultimate
gain and ultimate period may give poor results. Successful PI settings for such processes are proposed.

Keywords: PI controller, tuning, open loop, closed loop, Ziegler-Nichols, Tyreus-Luyben, Skogestad,
relay-tuning, performance, stability, robustness.

1. Introduction where Kcu is the ultimate gain, and Pu is the ultimate

period to be found by the user. A practical, experimen-
The PI (proportional plus integral) controller is prob- tal way to find Kcu and Pu is using relay oscillations,
ably the most frequently used controller function in Åstrøm and Hägglund (1995), cf. Appendix A.
practical applications. The PI controller stems from It is well-known that Ziegler and Nichols closed loop
a PID controller with the D-term (derivative) deac- PI tuning in many cases give relatively fast process
tived to reduce the propagation of amplified random disturbance compensation, but unfortunately poor sta-
measurement noise via the controller, thereby limiting bility margins, seen as poorly damped oscillatory re-
variations in the control signal due to noise. sponses. This is demonstrated in several examples in
Ziegler and Nichols (1942) presented two, now fa- Section 3. Tyreus and Luyben (1992) proposed a now
mous, methods for tuning P, PI, and PID controllers: well-known modification of the Ziegler-Nichols PI set-
The closed loop, or ultimate gain, method, and the tings which typically give improved control system sta-
open loop, or process reaction curve, method. In the bility:
present paper, focus is on closed loop tuning of PI con- Kc = 0.31Kcu (3)
trollers. Ti = 2.2Pu (4)
The PI settings with the Ziegler and Nichols closed
loop method are: In the present paper, another modification of the
Ziegler-Nichols PI settings is proposed to provide ac-
Kc = 0.45Kcu (1) ceptable stability margins and improved disturbance
compensation compared to the Tyreus and Luyben set-
Pu tings. The proposed tuning rules, here denoted the
Ti = (2) Relaxed Ziegler-Nichols (R-ZN) PI settings, are based
1.2

doi:10.4173/mic.2013.2.4
c 2013 Norwegian Society of Automatic Control
 Modeling, Identification and Control

on the open loop tuning rules in the SIMC method In the following, it is assumed that eq. (6) is satisfied
(Simple Internal Model Control) by Skogestad (2004) for the process to be controlled. Thus, an “integrator
applied to an “integrator plus time-delay” process esti- plus time-delay” process is assumed, with the following
mated from the ultimate gain and the ultimate period, transfer function:
Yu (1999). Kip −τ s
∆y(s)
The outline of this paper is as follows: In Section = Hp (s) = e (7)
∆u(s) s
2, the R-ZN PI settings are derived. In Section 3, the
original Ziegler-Nichols (ZN) PI settings, the Relaxed The Skogestad PI settings for this process are:
Ziegler-Nichols PI settings, and the Tyreus-Luyben
(TL) PI settings are applied to two simulation cases 1
Kc = (8)
and to a practical temperature control system of an Kip (Tc + τ )
air heater. In Section 4 an adjustable parameter of the
Ti = cs (Tc + τ ) (9)
R-ZN method is used to tune processes without time-
delay, but with lags. Section 5 contains a discussion, The parameter cs is introduced here. The original PI
and conclusions are given in Section 6. settings in Skogestad (2004) correspond to cs = 4 in
Appendix A reviews the relay experiment of find- eq. (9). For “integrator plus time-delay” processes
ing the ultimate gain and the ultimate period from with an “input” process disturbance, the disturbance
both sinusoidal and triangular oscillations. Appendix compensation appears as unnecessarily slow with cs =
B presents a modification of the Skogestad PI settings 4. To obtain a faster disturbance compensation while
for improved disturbance compensation, used in the retaining acceptable stability margins, a value of cs
derivation of the proposed PI controller setting. Ap- smaller than 4 can be used. It is found that values
pendix C shows abbreviations and nomenclature. around 2 are proper values. Thus, cs = 2 is proposed.
In this paper, the same symbol (letter) will be used The implications of various values of cs are investigated
for variables in time-domain as in the Laplace domain. in Appendix B.
This simplifies the notation. It is assumed that the The user must select a proper value of Tc in eqs.
meaning of the symbol is clear from the context. (8) and (9). Skogestad provides the following rule-of-
MATLAB and SIMULINK (MathWorks, Inc.) are thumb:
used for numerical computations and simulations. Lab- Tc = τ (10)
VIEW (National Instruments, Inc.) is used to imple- With cs = 2 and the rule-of-thumb eq. (10), eqs. (8)
ment the temperature control system for the real air and (9) become
heater. 1
Kc = (11)
2Kip τ
2. Relaxed Ziegler-Nichols PI Ti = 4τ (12)

tuning which may be denoted the modified Skogestad PI set-

tings for “integrator plus time-delay” processes.
2.1. Derivation of the tuning formulas The Skogestad PI settings, also with cs = 2, typically
yield acceptable stability of the control system, while
The following PI controller function is assumed: Ziegler and Nichols PI settings often give poor stability,
Z t with oscillatory responses (as demonstrated in several
Kc
u (t) = uman + Kc e (t) + e (τ ) dτ (5) applications in Section 3). The PI settings, eqs. (11)
Ti 0 and (12), will now be exploited to relax the original ZN
PI settings, eqs. (1)-(2).
Skogestad (2004) has provided PI settings for a num- For an “integrator plus time-delay” process, Kip and
ber of different types of process dynamics, among τ can be estimated from K and P as follows, Yu
cu u
which are “integrator plus time-delay” and “time- (1999), DiRuscio (2010):
constant plus time-delay”. Assuming that Skogestad’s
rule-of-thumb about setting the user-specified closed 2π
Kip = (13)
loop time-constant, Tc , equal to the process time-delay, Kcu Pu
τ , his PI settings for these two process types are actu-
ally identical as long as the relation between the time- Pu
τ= (14)
constant of the “time-constant plus time-delay” pro- 4
cess and the time-delay satisfies As pointed out in Seborg et al. (2004), process param-
eters Kip and τ can be used in any model-based con-
T ≥ 8τ (6) troller tuning method. Here, the (modified) Skogestad

84
 F. Haugen and B. Lie “Relaxed Ziegler-Nichols Closed Loop Tuning of PI Controllers”

PI settings, eqs. (11) and (12), are used. Inserting eqs.example of this: From the user’s perspective, adjusting
(13) and (14) into eqs. (11) and (12) gives Tc , which has a meaningful interpretation, to obtain
the PI settings is a much simpler task than adjusting
Kcu
Kc = = 0.32Kcu (15) Kc and Ti directly.
π

Ti = Pu (16) 2.2. Some derived results

which will be referred to as the (default) Relaxed Estimation of control system response-time
Ziegler-Nichols (R-ZN) PI settings. The control system response-time, Tr , can be estimated
from the ultimate period, Pu , as explained in the fol-
Comparing with ZN and TL lowing. The typical setting of kr = 1 is here assumed.
Compared with the (original) ZN PI settings, eqs. (1)- Then the PI settings are eqs. (6) and (6). Assume that
(2), the gain is smaller and the integral time is some- the setpoint is changed as a step. Then the response
what larger in the R-ZN PI settings, indicating im- in the process output reaches 63% of its final value at
proved stability. time (approximately)
Compared with the TL PI settings, eqs. (3)-(4), the Pu Pu Pu
R-ZN gain is almost the same, while the R-ZN integral Tr ≈ τ + Tc = + = (20)
4 4 2
time is smaller, indicating faster integral action, i.e.
Tr is here the 63% rise-time, or response-time, of the
the control error is brought faster to zero, however,
control system. As an example of eq. (20), see Figure
somewhat reduced stability can be expected.
9 where the response in air heater temperature due to
a setpoint step is plotted. In that example, Pu = 15 s,
Enhanced relaxation giving Tr ≈ Pu /2 = 7.5 s, which is in good accordance
Above, the closed loop time-constant is set equal to with the plotted response in Figure 9.
the (estimated) process time-delay, cf. eq. (10). Par-
ticularly in applications where the process has zero or Retuning the PI controller
negligible time-delay but some lag, the default R-ZN
Equations (18) and (19) can be used to retune a PI
PI settings may result in poor stability (and the ZN
controller safely. Note that the factor (kr + 1) appears
settings may even give instability). Acceptable stabil-
in the denominator of eq. (18) and in the numerator
ity can be obtained with enhanced relaxation of the PI
of eq. (19). For example, assume that it is desired to
settings. To this end, we propose
decrease the present value of Kc by a factor of 2 (to
Tc = kr τ (17) obtain a smoother control signal). This gain reduction
should be acccompanied by an increase of Ti by a factor
where kr ≥ 1 is a relaxation parameter to be set by 2. (This inversely proportional adjustment also follows
the user. The default PI settings, eqs. (6) and (6), are directly from Skogestad’s formulas, eqs. (8) and (9).)
obtained with kr = 1. Enchanced relaxation of the PI
settings is obtained with kr > 1. Using eq. (17) in eqs.
(8) and (9), and setting cs = 2 in (9), give
3. Applications
2 3.1. Overview
Kc = Kc (18)
π (kr + 1) u In the following subsections, PI settings with the (orig-
inal) Ziegler-Nichols closed loop method, the R-ZN
and closed loop method, and the TL method are applied
kr + 1
Ti = Pu (19) to the following three cases:
2
The usefulness of enhanced R-ZN tuning is demon- • A simulated control system for an “integrator with
strated in Section 4. time-delay” process (Section 3.3).
One question may arise: Why not just apply origi-
• A simulated control system for a “time-constant
nal ZN settings and adjust Kc and Ti directly? While
with time-delay” process (Section 3.4).
this is of course an option, we think that it better to
use a meaningful single parameter, kr , to obtain the PI • A practical temperature control system for a lab-
settings. The benefit of reducing the number of con- oratory air heater (Section 3.5). The process
troller parameters to adjust from two to one is actually dynamics is roughly “time-constant with time-
substantial. Skogestad’s tuning method is an excellent delay”.

85
 Modeling, Identification and Control

The PI settings will be compared using quantitive Tr indicates the speed of the response of the control
measures of performance and robustness defined in Sec- system due to a setpoint step change. Tr is approxi-
tion 3.2. mately the time-constant of the control system. ωc is
For easy reference, the various PI settings formulas equal to the phase crossover frequency, ω180d , of the
are summarized in Table 1. In the examples, Kcu and loop brought to marginal stability by a reduction of
Pu are found from the method of relay oscillations de- the phase of the loop while the amplitude is retained,
scribed in Appendix A. as by an increase of the loop time-delay:
1 Pu
Tr = = (24)
Table 1: PI settings formulas. ω180d 2π
ZN R-ZN TL
Kc 0.45Kcu 0.32Kcu 0.31Kcu where Pu [s] is the (ultimate) period of the oscillations
Pu at marginal stability.
Ti 1.2 Pu 2.2Pu

Setpoint tracking versus disturbance compensation

3.2. Measures of performance and For systems where the setpoint is constant, which is
robustness the case in many practical process control systems, it
can be claimed that good disturbance compensation is
The measures used in this paper for comparing the more important than good setpoint tracking. In the ex-
various methods of PI controller tuning can be grouped amples presented in the following sections, disturbance
into performance and robustness measures described in compensation is emphasized.
the detail in the following.
3.2.2. Stability robustness (stability margins)
3.2.1. Performance
Gain margin, GM
IAE at setpoint change
For the cases based on simulations GM is calculated
In the tests the setpoint is changed as a step. The from the loop transfer function, HL (s), using the mar-
setpoint tracking is measured with the IAE (Integral gin function in MATLAB. HL (s) is
of Absolute Error) index calculated over a proper time
interval as Z tf HL (s) = Hc (s)Hp (s) (25)
IAEs = |e| dt (21) where Hc (s) is the controller transfer function, and
ti
Hp (s) is the process transfer function.
where e is the control error, ti is the initial time, se- For the practical case (air heater) an adjustable gain,
lected as the time of the step change, and tf is a proper ∆K, is inserted into the loop (between the controller
final time. A reduced IAEs value indicates improved and the process), see Figure 1. Initially, ∆K = 1. The
setpoint tracking.
Disturbance
IAE at process disturbance change Adjustable Adjustable d Process
Setpoint gain time-delay measurement
Process
ySP ymf
In the tests a process disturbance is changed as a step. Cont- u
DK (t-Dt)
w/actuator
roller and sensor
The disturbance compensation is measured with and filter
Z tf
IAEd = |e| dt (22)
ti
Figure 1: An adjustable gain and time-delay are in-
A reduced IAEd value indicates improved disturbance serted into the loop to find the stability
compensation. margins (gain margin and phase margin)
experimentally.
Response time
(ultimate) value ∆Ku that brings the control system to
The response time, Tr [s], is here defined as the inverse
the stability limit so that the responses are sustained
of the bandwidth defined as the amplitude crossover
oscillations, is found experimentally (by trials). The
frequency, ωc [rad/s]:
gain margin is then
1
Tr = (23) GM = ∆Ku (26)
ωc

86
 F. Haugen and B. Lie “Relaxed Ziegler-Nichols Closed Loop Tuning of PI Controllers”

Phase margin, PM ysp: red. y: blue.

2

For the cases based on simulations PM is calculated 1

from the loop transfer function using the margin func- 0
tion in MATLAB. −1
For the practical case (air heater) an adjustable time- −2
0 5 10 15 20 25 30
delay, ∆τ [s], is inserted into the loop (between the
u
controller and the process), see Figure 1. Initially,
1
∆τ = 0. For each of the tuning methods, the value ∆τu
0.5
that brings the control system to the stability limit, i.e.
0
causing sustained oscillations, is found experimentally.
−0.5
The period, Pu [s], of the oscillations is measured. The
−1
corresponding phase margin is 0 5 10 15 20 25 30

d
∆τu 0.5
PM [deg] = 360 (27)
Pu 0

Equation (27) is derived in Haugen (2012) (Ap- −0.5

pendix 1).
−1
0 5 10 15 20 25 30
t [s]
Proper values of GM and PM
Seborg et al. (2004) propose the following ranges for Figure 2: Responses during relay tuning
proper values of the stability margins:

1.7 = 4.6 dB ≤ GM ≤ 4.0 = 12.0 dB (28) It is interesting that the ultimate gain using a P con-
troller also gives Kcu = 1.57. Hence, the Fourier-series
and approximations used to derive eq. (47) give a very pre-
30o ≤ PM ≤ 45o (29) cise result in this case.
Furthermore, from the plots,
Since poor control system stability must be avoided,
the lower limits of GM and PM can be regarded as Pu = 4.0 s (32)
critical, while the upper limits are not.
Various PI settings are calculated from the above
3.3. Application: Simulated “integrator values of Kcu and Pu using the formulas in Table 1.
plus time-delay” process The PI settings are shown in Table 2.

3.3.1. Process description

3.3.3. Performance and stability robustness of the
The process to be controlled is an “integrator plus control system
time-delay” process:
Figure 3 shows responses in the process output variable
ẏ(t) = Kip u(t − τ ) + Kd d(t) (30) (y) and the controller output (u) with a step change
of the temperature setpoint (ysp ) and a step change of
which has transfer function as in eq. (7). The process the disturbance (d) for the three different PI settings
parameter values are: Kip = 1 s−1 , Kd = 1, τ = 1 s. shown in Table 2.
GM, PM and Tr are calculated from the model.
3.3.2. PI controller tuning from relay oscillations IAEs is calculated time-series over the interval t =
[2 s, 40 s]. IAEd is calculated over t = [40 s, 80 s].
Kcu and Pu are found from relay oscillations. Figure 2 Table 2 summarizes the performance and robustness
shows plots of the sustained oscillations during the re- measures.
lay tuning, cf. Appendix A. From the plots,Atri = 1.0. Below are a number of observations made in Table 2
The square wave in the control signal has amplitude (the abbreviations are as in Table 2):
Asq = 1.
Equation (47) in Appendix A gives • Setpoint tracking:
πAsq π·1 IAEs : ZN and TL are the best, and almost equal,
Kcu = = = 1.57 (31) but ZN suffers from large overshoot.
2Atri 2·1

87
 Modeling, Identification and Control

ysp: Red. y: ZN: Blue; relaxed ZN: Black; TL: Magenta unnecessarily large as it is larger than the higher
4
limit in ineq. (29).
3

2
The low stability margins with ZN are apparent
in the oscillatory responses with the ZN settings,
1
see Figure 3.
0
0 10 20 30 40 50 60 70 80

u: ZN: Blue; relaxed ZN: Black; TL: Magenta

2
Comments and conclusions

1 The Ziegler-Nichols PI settings give poor control loop

stability margins. The TL and the R-ZN settings give
0
acceptable stability margins. With emphasis on dis-
−1 turbance compensation rather than setpoint tracking,
0 10 20 30 40 50 60 70 80
the R-ZN settings are better than the TL settings.
d
0.5

0 3.4. Application: Simulated

−0.5
“time-constant plus time-delay”
process
−1
0 10 20 30 40 50 60 70 80
t [s] 3.4.1. Process description
The process to be controlled is a “time-constant plus
Figure 3: Responses with various PI settings. time-delay” process (assuming the time-delay is at the
input-side):

Table 2: Controller settings and performance and ro- T ẏ(t) = −y(t) + Ku(t − τ ) + Kd d(t) (33)
bustness measures for simulated control sys-
tem for ”integrator plus time-delay” process The process parameter values are: K = 8, Kd = 8, τ =
with different PI settings. 1 s.
ZN R-ZN TL The time-constant being 8 times the time-delay
Kc 0.71 0.50 0.49 makes the Skogestad PI settings for a “time-constant
Ti [s] 3.3 4.0 8.8 plus time-delay” process become identical with the
IAEs 7.9 8.1 8.0 settings for an “integrator plus time-delay” process.
IAEd 2.8 4.5 9.0 Therefore, the condition for using Skogestad tuning
GM 1.9 2.7 3.1 for “integrator plus time-delay” processes, ineq. (6),
GM [dB] 5.4 8.8 9.7 is satisfied.
PM [deg] 24.9 34.1 48.6
Tr [s] 1.3 s 1.8 2.0 3.4.2. PI controller tuning from relay oscillations
The ultimate gain and the ultimate period are found
Tr : ZN is the best, while R-ZN and TL do not from relay oscillations. Figure 4 shows plots of the
differ much. sustained oscillations during the relay tuning. The re-
sponse in y are approximately triangular, so eq. (47)
• Disturbance compensation: is used to calculate Kcu . From Figure 4,Atri = 0.94,
Asq = 1. Equation (47) gives
IAEd : ZN is clearly best. R-ZN is in turn clearly
better than TL as the R-ZN has a value which is πAsq π · 0.94
50% of the value of TL. Kcu = = = 1.48 (34)
2Atri 2·1
• Stability robustness (margins): The ultimate gain using a P controller gives Kcu = 1.65
GM: ZN is poor, and actually below the lower limit which differs somewhat from 1.48. Still, Kcu = 1.48 is
in ineq. (28). R-ZN and TL do not differ much used to stick to relay tuning, and using 1.48 rather
and have acceptable values. than 1.65 is safe (conservative) regarding control loop
stability.
PM: Again ZN is poor, and below the lower limit Furthermore, from Figure 4,
in ineq. (29). R-ZN gives a somewhat small, but
acceptable, value. TL gives large value, possibly Pu = 3.78 s (35)

88
 F. Haugen and B. Lie “Relaxed Ziegler-Nichols Closed Loop Tuning of PI Controllers”

ysp: red. y: blue. ysp: Red. y: ZN: Blue; relaxed ZN: Black; TL: Magenta
1
3
0.5
2
0
1
−0.5

−1 0
0 5 10 15 20 25 30 0 10 20 30 40 50 60 70 80

u u: ZN: Blue; relaxed ZN: Black; TL: Magenta

1 2

0.5
1
0
0
−0.5

−1 −1
0 5 10 15 20 25 30 0 10 20 30 40 50 60 70 80

d d
0.5 0.5

0 0

−0.5 −0.5

−1 −1
0 5 10 15 20 25 30 0 10 20 30 40 50 60 70 80
t [s] t [s]

Figure 4: Responses during relay tuning Figure 5: Responses with various PI settings.

Various PI settings are calculated from the above Below are a number of observations made in Table 2
values of Kcu and Pu using the formulas in Table 1. (the abbreviations are as in Table 2):
The PI settings are shown in Table 2. • Setpoint tracking:
IAEs : TL is best.
3.4.3. Performance and stability robustness of the Tr : ZN is best, while R-ZN and TL do not differ
control system much.
Figure 5 shows responses in the process output variable • Disturbance compensation:
(y) and the controller output (u) with a step change
IAEd : ZN is clearly best. R-ZN is in turn clearly
of the temperature setpoint (ysp ) and a step change of
better than TL. R-ZN has a value which is 45 %
the disturbance (d) for the three different PI settings
of the value of TL.
shown in Table 3.
GM, PM and Tr are calculated from the model. IAEs • Stability robustness (margins):
is calculated over the interval t = [2 s, 40 s]. IAEd is GM: Strictly, all settings give acceptable values,
calculated over t = [40 s, 80 s]. Table 3 summarizes but ZN is on the lower limit.
the performance and robustness measures.
PM: ZN is poor, and below the lower limit in ineq.
(29). R-ZN gives a somewhat small, but accept-
Table 3: Controller settings and performance and ro- able, value. TL gives a large value, possibly unnec-
bustness measures for simulated control sys- essarily large as it is larger than the higher limit
tem for ”time-constant plus time-delay” pro- in ineq. (29).
cess with different PI settings.
ZN R-ZN TL Comments and conclusions
Kc 0.75 0.53 0.52 The Ziegler-Nichols PI settings give poor control loop
Ti [s] 3.2 3.8 8.3 stability as the PM is too small. The rest of the com-
IAEs 6.1 5.8 4.3 ments are identical with those for the “integrator plus
IAEd 2.1 3.6 7.9 time-delay” case in Section 3.3: The TL and the R-ZN
GM 1.7 2.6 3.0 settings give acceptable stability margins. With em-
GM [dB] 4.8 8.1 9.7 phasis on disturbance compensation rather than set-
PM [deg] 22.3 32.2 60.1 point tracking, the R-ZN settings are better than the
Tr [s] 1.2 1.7 1.9 TL settings.

89
 Modeling, Identification and Control

3.5. Application: Practical temperature ymf

39
control system
38

3.5.1. Process description 37

[Deg C]
36
Figure 6 shows an air heater laboratory station. The
35

34
120 140 160 180 200 220 240 260 280 300 320

[V]
2

120 140 160 180 200 220 240 260 280 300 320
t [s]

Figure 7: Open loop step response in filtered tempera-

ture, ymf , due to a step in the heater control
signal, u.

Figure 6: Temperature control system for an air heater 3.5.2. PI controller tuning from relay oscillations
(laboratory rig)
Kcu and Pu are found from relay oscillations. Figure
8 shows plots of the sustained oscillations during the
temperature of the air outlet is controlled by adjusting relay tuning. The oscillations in temperature (process
the control signal (voltage) to the heater. The temper- measurement) looks more sinusoidal than triangular.
ature is measured with a Pt100 element. A measure- Therefore, Kcu is calculated using eq. (45).
ment filter with time-constant 0.5 s is used to attenuate From Figure 8, Asin = 0.75 o C and Asq = 2.5 V.
measurement noise. The National Instruments USB- Equation (45) gives
6008 is used as analog I/O device. The control system
is implemented in LabVIEW (National Instruments) 4Asq 4 · 2.5 V V
running on a PC. The fan rotational speed, and the air Kcu = = o
= 4.24 o (36)
πAsin π · 0.75 C C
flow, can be adjusted manually with a potentiometer.
Changes of the air flow comprises a process disturbance From Figure 8,
giving an impact on the temperature. The measured Pu = 15.0 s (37)
voltage drop across the potensiometer is represented
by the variable F in percent. Thus, F represents the Various PI settings are calculated from the above
air flow disturbance. 1 values of Kcu and Pu using the formulas in Table 1.
The PI settings are shown in Table 4. Both standard R-
The nominal operating point of the system is tem-
ZN and enhanced R-ZN tuning are applied, with kr = 1
perature at 35 o C and air flow F = 50 %.
and kr = 2, respectively.
Figure 7 shows the open loop, or process, step re-
sponse in the filtered temperature, ymf , due to a step
in the heater control signal, u. The response indicates 3.5.3. Performance and stability robustness of the
that the process dynamics is roughly “time-constant control system
with time-delay”, with time-constant ≈ 37 s and time- Figures 9, 10, 11, and 12 show responses in the air
delay ≈ 3 s which is about 8% of the time-constant. temperature (ymf ) and the controller output (u) due
to a step change of the temperature setpoint (ysp ) and a
1 Additional information about the air heater is available at step change of the disturbance (d) for the four different
Haugen (2013). PI settings shown in Table 4.

90
 F. Haugen and B. Lie “Relaxed Ziegler-Nichols Closed Loop Tuning of PI Controllers”

ysp: Red. ymf: Blue. ysp: Red. ymf: Blue.

37 37
36.5

[Deg C]
36 36
35.5
[Deg C]

35 35
34.5
80 100 120 140 160 180 200 220 240 260 280 300
34
u
5
33 4
100 110 120 130 140 150 160 170 180 190 200
3

[V]
2
u
1
5 0
80 100 120 140 160 180 200 220 240 260 280 300
4
d (air flow)
3 100
[V]

[%]
1 50

0
0
100 110 120 130 140 150 160 170 180 190 200 80 100 120 140 160 180 200 220 240 260 280 300
t [s] t [s]

Figure 8: Responses during relay tuning Figure 9: Responses with Ziegler-Nichols PI settings

Performance and stability robustness measures are Tr : ZN is clearly best. It gives fast control. R-ZN
calculated from the time-series as explained in Sec- with kr = 1 is also relatively fast.
tion 3.2. IAEs is calculated over the interval
t = [100 s, 180 s]. IAEd is calculated over t = • Disturbance compensation:
[200 s, 280 s]. Table 4 summarizes the performance IAEd : ZN and R-ZN with kr = 1 are much better
and robustness measures. than both R-ZN with kr = 2 and TL. R-ZN with
kr = 1 give only 36% of that of TL. Relaxed ZN
Table 4: Controller settings and performance and ro- with kr = 2 is also clearly better than TL.
bustness measures for practical temperature
• Stability robustness (margins):
control system for different PI settings.
R-ZN R-ZN GM: ZN is poor, and actually below the lower limit
ZN TL in ineq. (28). R-ZN with kr = 1 is small, but just
kr = 1 kr = 2
Kc 1.91 1.35 0.90 1.32 within the limits.
Ti [s] 12.5 15.0 22.5 33.0 PM: Again ZN is poor, and below the lower limit
IAEs 16.2 12.3 10.6 10.3 in ineq. (29). R-ZN with kr = 1 is small, but
IAEd 4.3 4.9 7.5 11.8 just within the limits. TL has a large value, pos-
GM 1.5 1.8 2.8 2.6 sibly unnecessarily large since it is larger than the
GM [dB] 3.5 5.1 8.9 8.3 higher limit in ineq. (29). R-ZN with kr = 2 has
∆τu [s] 1.6 2.6 7.7 5.7 a very large value.
Pu [s] 24.0 31.0 50.0 39.0 The low stability margins with ZN are apparent in
PM [deg] 24.0 30.2 55.4 52.6 the oscillatory responses with the ZN settings. R-
Tr [s] 3.8 4.9 8.0 6.2 ZN with kr = 1 seems to give acceptable stability
as seen from time-series. R-ZN with kr = 2 and
Below are a number of observations made in Table 4 TL both give smooth, but slow, responses.
(the abbreviations are as in Table 4):
• Setpoint tracking: Comments and conclusions
IAEs : TL and R-ZN with kr = 1 and with kr = 2 The Ziegler-Nichols PI settings give poor control loop
do not differ much and are clearly better than ZN stability. The TL and the R-ZN settings both with
which is due to the large overshoot and oscillatory kr = 1 and kr = 2 give acceptable stability margins,
response with ZN. though R-ZN with kr = 1 gives small margins. R-ZN

91
 Modeling, Identification and Control

ysp: Red. ymf: Blue. ysp: Red. ymf: Blue.

37 37
36.5 36.5
[Deg C]

[Deg C]
36 36
35.5 35.5
35 35
34.5 34.5
80 100 120 140 160 180 200 220 240 260 280 300 80 100 120 140 160 180 200 220 240 260 280 300

u u
5 5
4 4
3 3
[V]

[V]
2 2
1 1
0 0
80 100 120 140 160 180 200 220 240 260 280 300 80 100 120 140 160 180 200 220 240 260 280 300

d (air flow) d (air flow)

100 100
[%]

[%]
50 50

0 0
80 100 120 140 160 180 200 220 240 260 280 300 80 100 120 140 160 180 200 220 240 260 280 300
t [s] t [s]

Figure 10: Responses with R-ZN PI settings with kr = Figure 11: Responses with R-ZN PI settings with kr =
1. 2.

with kr = 1 give clearly the best disturbance compen- cases to obtain proper stability. Note that for processes
sation, and since the stability margins are within the with a noteable time-delay the R-ZN PI settings with
acceptable limits, it gives the prefered PI settings in the default value kr = 1, i.e. eqs. (6)-(6), should be
this application. used.
If it is important with smooth responses, both TL Assume that the process is an integrator without
and R-ZN with kr = 2 can be used. Among these two, any time-delay but with two lags in the form of time-
we prefer the latter because it gives best disturbance constant terms where one of the time-constants is one
compensation, and because the R-ZN settings are ad- tenth of the other. Specifically, the following process
justable, while the TL settings are fixed. transfer function model is assumed:
1  
y(s) = Kip u(s) + Kd d(s) (38)
4. Relaxed tuning for processes s(T1 s + 1)(T2 s + 1)

with no time-delay but with lags where u is control variable and d is disturbance. Time-
constant T1 may represent a process lag due to e.g.
Closed loop PI tuning with the standard Ziegler- dynamics of a heating element or a valve or a pump or
Nichols method, the TL method, or even the R-ZN represent inhomogeneous conditions in a tank, while
tuning method with the default setting kr = 1 may T2 may represent the time-constant of a measurement
not work well if the process has no, or negligible time- filter. The integrator, 1/s, may represent e.g. energy
delay, however, some lag is assumed. The resulting or material balance. The following parameter values
stability may be very poor. Such cases may occur in are assumed: Kip = 1 s−1 Kd = 1, T1 = 1 s, and
e.g. temperature control, Haugen et al. (2013) and bio- T2 = 0.1 s. In less extreme cases the difference between
gas flow control of bioreactors, Haugen and Lie (2013). the two time-constants are less, and there may also be
However, enhanced R-ZN tuning with a proper kr > 1 a non-zero time-delay.
seems to work well. An explanation of the resulting The relay method is used, giving Kcu = 10.24 and
poor stability is that, due to the lack of a time-delay, Pu = 2.02 s. The three PI tuning methods mentioned
the phase characteristic is relatively flat around the in the beginning of the present section are tested. Fig-
critical frequencies, making the phase margin small. ure 13 shows their responses. With TL tuning and
Now, an extreme case is assumed, and enhanced R- R-ZN tuning with kr = 1 the control system is stable,
ZN PI tuning is used. The value of kr that is found but the stability is poor. With Ziegler-Nichols tuning,
useful in this case may be used in other less extreme the system is unstable!

92
 F. Haugen and B. Lie “Relaxed Ziegler-Nichols Closed Loop Tuning of PI Controllers”

ysp: Red. ymf: Blue. ysp: Red. y: ZN: Blue

37 2
36.5 1.5
[Deg C]

36
1
35.5
35 0.5

34.5 0
80 100 120 140 160 180 200 220 240 260 280 300 0 10 20 30 40 50 60 70 80

u u
5 2
4
1
3
[V]

2
0
1
0 −1
80 100 120 140 160 180 200 220 240 260 280 300 0 10 20 30 40 50 60 70 80

d (air flow) d
0.5
100

0
[%]

50
−0.5

0 −1
80 100 120 140 160 180 200 220 240 260 280 300 0 10 20 30 40 50 60 70 80
t [s] t [s]

Figure 12: Responses with TL PI settings Figure 14: Responses with PI controller tuned with the
R-ZN method with kr = 4.
ysp: Red. y: ZN: Blue; relaxed ZN: Black; TL: Magenta
3

2.5

2
Table 5: Controller settings and performance and ro-
bustness measures for simulated PI control
1.5
system for an ”integrator with two lags” pro-
1 cess with R-ZN tuning with kr = 4.
0.5
Kc 1.3
Ti [s] 5.2
0
GM 6.7
−0.5 GM [dB] 16.5
−1
PM [deg] 29.4
0 5 10 15
t [s]

Figure 13: Responses with various PI settings.

However, the value of 29.4o is here regarded as
acceptable since it is for an assumed extreme case.
By trial-and-error it is found that R-ZN tuning with With kr = 5 PM = 34.3o which is within the range
kr = 4 works well. Hence, with kr = 4 in (18) and (19) given by ineq. (29), but simulations indicate that
the PI settings become the control system becomes unnecessarily sluggish
with kr = 5 applied for less extreme cases.
Kc = 0.13Kcu (39)

and
• How can one know that a process has one or more
Ti = 2.5Pu (40)
lags and no or negligible time-delay, so that the
Figure 14 shows simulated responses. Table 5 shows enhanced relaxed tuning should be applied? Phys-
PI settings and stability margins. ical insight may be useful: If the sensor or actuator
is located close to the main process (which can be
Comments and conclusion: e.g. a reactor vessel), the time-delay may be neg-
ligible compared to time-constant lags. A process
• GM is large, but is accepted here. step response test is also an option, but then an
• PM is small and just outside the acceptable range open loop controller tuning method, as the Skoges-
where 30.0o is the critical limit, cf. ineq. (29). tad method (2003, 2004), may be applied directly.

93
 Modeling, Identification and Control

5. Discussion The proposed PI settings have an adjustable param-

eter which can be used to obtain enhanced relaxation
The proposed new set of PI settings are based on tun- which is useful for processes with zero or negligible
ing rules derived by Ziegler and Nichols (1942), tuning time-delay but some lags (time-constants).
rules derived by Skogestad (2004), and the modeling
of the process as an “integrator plus time-delay” ac-
cording to Yu (1999). The validity and applicability Acknowledgments
of the proposed PI settings rely on assumptions made
by these authors. The sensitivity of the present results Telemark University College has provided practical
with respect to such assumptions has not been inves- and economical support for the work with this paper.
tigated here. However, two simulation tests and one Thanks to colleague David Di Ruscio for useful discus-
practical test indicate that the proposed tuning works sions.
as assumed.
In the simulations it is assumed that the process
disturbance is an input disturbance as it acts on the
A. Finding the ultimate gain and
process at the same place, dynamically, as the control period from relay oscillations
signal does. In most practical processes the main dis-
turbances are actually input disturbances. We have not In the original Ziegler-Nichols closed loop method the
investigated the consequences for our results of moving user must find, typically by trial-and-error, the ulti-
the disturbance to the process output. mate controller gain value, Kcu , of a P controller which
It is found that for processes with no, or a negligible makes the responses in the control system become sus-
time-delay, but with some lags in the form of time- tained oscillations. The user must also read off the
constants, R-ZN tuning with kr = 1 may give poor ultimate period, Pu , of the oscillations. Kcu and Pu
stability (This applies to ordinary ZN and TL tuning, are then used to calculate the PI settings with the fol-
too). However, proper stability may be obtained with lowing formulas:
enhanced relaxation of the tuning, and kr = 4 seems to
be a proper value at least for processes without time- Kc = 0.45Kcu (41)
delay but with two lags with one being one tenth of the
Pu
other. The conditions that make the selection kr = 4 Ti = (42)
1.2
unsuccessful have not been investigated, but for pro-
cesses where the time-constants are closer, the PI set- Åstrøm and Hägglund (1995) introduced a relay,
tings with kr = 4 will certainly be safe (conservative). or on-off, controller to replace the P controller in
the tuning phase, thereby avoiding the possibly time-
consuming trial-and-error procedure as the oscillations
6. Conclusions come automatically. During the relay tuning the con-
trol signal is a square wave.
The main result of this paper is a proposed new set of Kcu can be estimated from the relay oscillations as
PI settings which uses the same information as in the follows. Assume that the amplitude of the square wave
Ziegler-Nichols closed loop method, namely knowledge is
about the ultimate gain, Kcu , and the ultimate period, uon − uoff
Asq = (43)
Pu : The proposed settings are: 2
where uon and uoff are the values of the controller out-
Kc = 0.32Kcu put when the relay is in the on- and off-state, respec-
tively. The square wave is approximated by its funda-
Ti = Pu mental sinusoidal component of its Fourier series. The
These settings are modifications, or relaxations, of the fundamental sinusoid is known to have amplitude
original Ziegler-Nichols PI settings, and they give im-
4Asq
proved control system stability. In this paper, the pro- Asq,F = (44)
posed setting have been successfully applied to two sim- π
ulated control systems and to a practical temperature
Sinusoidal oscillations
control system of an air heater.
Comparing with the TL PI settings, which also are With relay-based oscillations, for many practical pro-
based on knowledge of the ultimate gain and the ul- cesses the filtered process measurement is approxi-
timate period, the proposed PI settings give clearly mately sinusoidal. Assume that the measurement has
better disturbance compensation. amplitude Asin . The control error, which is the input

94
 F. Haugen and B. Lie “Relaxed Ziegler-Nichols Closed Loop Tuning of PI Controllers”

to the relay, then also has amplitude Asin . The equiv- ysp: Red. y: c = 1.5: Blue; c = 2: Black; c = 4: Magenta
4
alent gain of the relay function, which is used as the
3
ultimate gain in eq. (41), is
2
Asq,F 4Asq Asq 1
Kcu = = = 1.27 (45)
Asin πAsin Asin 0
0 10 20 30 40 50 60 70 80

The ultimate period, Pu , needed in eq. (42) is the u: c = 1.5: Blue; c = 2: Black; c = 4: Magenta
1.5
period of the oscillations.
1

0.5
Triangular oscillations
0
If the process dynamics is pure “integrator plus time- −0.5
0 10 20 30 40 50 60 70 80
delay” the relay-based oscillations in the process mea-
d
surement are not sinusoidal, but triangular. Let Atri 0.5
be the amplitude of these triangular oscillations. The
0
fundamental sinusoidal component of the triangular os-
cillation is known to have amplitude −0.5

8Atri −1
0 10 20 30 40 50 60 70 80
Atri,F = (46)
π2 t [s]

The equivalent gain of the relay function, which is used Figure 15: Simulations of control systems for an “in-
as the ultimate gain in eq. (41), is tegrator with time-delay” process with Sko-
Asq,F πAsq Asq gestad controller tuning with cs = 1.5, cs =
Kcu = = = 1.57 (47) 2, and cs = 4.
A tri,F 2A tri Atri

The ultimate period, Pu , in eq. (42) is the period of

the oscillations. • cs = 4 which is Skogestad’s original value.
If the process dynamics is “time-constant plus time-
delay” with the time-constant being much larger than Table 6 shows a number of characteristics of the sim-
the time-delay, and without other process dynamics ulated control system for the three values of cs . GM,
(lags), the relay-based oscillations appear more trian- PM and Tr are calculated from the model, cf. Sec-
gular than sinusoidal. In these cases, eq. (47) can be tion 3.2. IAEs is calculated time-series as explained in
used. Section 3.2 over the interval t = [2 s, 40 s]. IAEd is
calculated over t = [40 s, 80 s].

B. Impact of the proposed

Table 6: Results with c = 1.5, c = 2, and c = 4 in the
parameter cs in the modified (modified) Skogestad PI tuning formulas for
Skogestad PI settings an “integrator with time-delay” process.
cs = 1.5 cs = 2 cs = 4
Simulations are used to investigate the implications of Kc 0.5 0.5 0.5
using various values of parameter cs in eq. (9). Ti [s] 3 4 8
The process to be controlled is an “integrator with IAEs 9.6 8.1 7.8
time-delay” process given by eqs. (30) with Kip = 1 IAEd 4.5 4.5 8.0
s−1 , Kd = 1 and τ = 1 s. The PI controller is tuned GM 2.6 2.7 3.0
with the (modified) Skogestad tuning formulas, eqs. PM [deg] 26.9 34.1 46.9
(8) and (9). Tr [s] 1.7 1.8 1.9
Figure 15 shows simulations for the following values
of cs :

• cs = 1.5 which is the the value corresponding to Comments and conclusions

the IMC settings for an “integrator with time-
delay” process by Chien and Fruehauf (1990). In Table 6, PM = 26.9 for cs = 1.5 which is regarded
as a poor value since it is lower than the lower limit in
• cs = 2 which is the value used in the present paper. ineq. (29).

95
 Modeling, Identification and Control

With cs = 2 and cs = 4 the stability margins are Pu [s]: Period of sustained oscillations.
acceptable.
IAEd with cs = 2 is 56% of IAEd with cs = 4, in- T [s]: Process time-constant.
dicating a considerable improved disturbance compen-
sation with cs = 2. This is also clearly seen in the Tc [s]: Closed loop time-constant.
simulations.
Ti [s]: Controller integral time.
We prefer cs = 2 over cs = 4 in the Skogestad PI
settings formulas for “integrator plus time-delay” pro- Tr [s]: Response-time, or 63% rise time of step re-
cesses since the disturbance compensation is improved. sponse.

τ [s]: Process time-delay.

C. Abbreviations and nomenclature
u: Control signal (controller output).
C.1. Abbreviations
uman : Manual control signal (control bias).
GM: Gain margin.

IAE: Integral of absolute error. y: Process output measurement.

PI: Proportional plus integral (control). ysp : Setpoint

PM: Phase margin.

R-ZN: Relaxed Ziegler-Nichols.

ZN: Ziegler-Nichols (original method).

References
TL: Tyreus-Luyben. Åstrøm, K. J. and Hägglund, T. PID Controllers: The-
ory, Design and Tuning. ISA, 1995.
SIMC: Simple Internal Model Control.
Chien, I. L. and Fruehauf, P. S. Consider IMC Tun-
C.2. Nomenclature ing to Improve Controller Performance. Chem. Eng.
Progress, 1990. Oct:33–41.
Asin : Amplitude of sinusoidal wave in control error or
in process (output) measurement. DiRuscio, D. On Tuning PI Controllers for In-
tegrating Plus Time Delay Systems. Modeling,
Asq : Amplitude of square wave in control signal. Identification and Control, 2010. 31(4):145–164.
doi:10.4173/mic.2010.4.3.
Atri : Amplitude of triangular wave in control error or
in process (output) measurement. Haugen, F. The good gain method for simple ex-
Au : Amplitude of the on-off control signal. perimental tuning of pi controllers. Modeling,
Identification and Control, 2012. 33(4):141–152.
cs : Parameter introduced in the integral time settings doi:10.4173/mic.2012.4.3.
in the Skogestad method.
Haugen, F. Air heater. http://home.hit.no/~finnh/
d is process disturbance. air_heater, 2013.
∆: Deviation from operating point. Haugen, F., Bakke, R., and Lie, B. Temperature con-
e: Control error. e = ysp − y. trol of a pilot anaerobic digestion reactor. Submitted
to Modeling, Identification and Control, 2013.
kr : The relaxation parameter in the Relaxed Ziegler-
Nichols method. Haugen, F. and Lie, B. On-off and pid control of
methane gas production of a pilot anaerobic diges-
K is process gain. tion reactor. Submitted to Modeling, Identification
and Control, 2013.
Kc [s]: Controller proportional gain.

Kd is disturbance gain. Seborg, D. E., Edgar, T. F., and Mellichamp, D. A.

Process Dynamics and Control. John Wiley and
Kip [s]: Process integrator gain. Sons, 2004.

96
 F. Haugen and B. Lie “Relaxed Ziegler-Nichols Closed Loop Tuning of PI Controllers”

Skogestad, S. Simple analytic rules for model re- Yu, C. C. Autotuning of PID Controllers. Springer
duction and pid controller tuning. Modeling, Verlag, 1999.
Identification and Control, 2004. 25(2):85–120.
doi:10.4173/mic.2004.2.2. Ziegler, J. and Nichols, N. Optimum settings for au-
tomatic controllers. Trans. ASME, 1942. 64(3):759–
Tyreus, B. D. and Luyben, W. L. Tuning PI Con- 768. doi:10.1115/1.2899060.
trollers for Integrator/Dead Time Processes. Ind.
Eng. Chem, 1992. 31(31).

97