chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606

Contents lists available at ScienceDirect

Chemical Engineering Research and Design

journal homepage: www.elsevier.com/locate/cherd

Maximum sensitivity based analytical tuning rules

for PID controllers for unstable dead time processes

K. Ghousiya Begum a , A. Seshagiri Rao b,∗ , T.K. Radhakrishnan a

a Department of Chemical Engineering, National Institute of Technology, Tiruchirappalli 620 015, Tamilnadu, India
b Department of Chemical Engineering, National Institute of Technology, Warangal 506 004, Telangana State, India

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, maximum sensitivity and internal model control (IMC) based proportional-
Received 6 August 2015 integral-derivative (PID) controllers are designed for unstable first-order plus-dead-time
Received in revised form 20 (UFOPDT) processes. The designed controller parameters are functions of the UFOPDT model
February 2016 parameters and the IMC closed loop tuning parameter. The tuning parameter plays a vital
Accepted 3 March 2016 role and determines the closed loop performance and robustness of the designed con-
Available online 9 March 2016 troller. Systematic guidelines are provided for selection of this tuning parameter based
on maximum sensitivity. Analytical tuning rules are developed for the controller param-
Keywords: eters for different time delay to time constant ratios with desired level of robustness.
IMC control These controller settings allow the operator to deal with the closed-loop control system
H2 minimization performance-robustness trade-off by specifying the robustness level (maximum sensitiv-
PID controller ity). Simulation studies have been carried out on various UFOPDT processes to explain the
Unstable process advantages of the proposed analysis.
Maximum sensitivity © 2016 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.

1. Introduction unstable systems in engineering processes and provided an

excellent overview of controller design techniques for unsta-
Open loop unstable processes arise frequently in the chemical ble processes. They developed tuning rules based on equating
(distillation, polymerization reactors, heat exchangers, etc.) coefficient method for first order unstable processes with
and biological (fermentation) processes and are fundamen- time delay. Arrieta et al. (2011) proposed PID tuning based on
tally difficult to control than that of the stable processes and servo or regulatory operations. Nasution et al. (2011) proposed
the difficulty increases when there exist a time delay. Time PID controller design based on IMC method and H2 mini-
delays occur frequently in process control problems, because mization and Maclaurin series approximation and obtained
of the distance velocity lags, recycle loops, and composition improved performances over many previous methods. They
analysis loops. The performance specifications that are usu- have developed five different desired closed loop transfer
ally achieved for stable systems are difficult to achieve for functions and designed the controllers correspondingly based
unstable systems. The closed loop response for such pro- on these five desired closed loop transfer functions and rec-
cesses exhibits large overshoots and settling times. As has ommended one. Shamsuzzoha and Lee (2008) showed that
been widely reported, PID controllers are with no doubt, the PID controllers in series with lead/lag compensators provide
controllers most extensively used in the process industry. improved closed loop performances when compared to that of
The PID controller design methods for unstable pro- PID controller alone for UFOPDT processes. Further, complex
cesses have been addressed by many researchers (Rao and control structures have also been developed such as modified
Chidambaram, 2006, 2012). Sree et al. (2004) and Sree and Smith predictor (Rao et al., 2007; Uma et al., 2010; Normey-
Chidambaram (2006) described occurrence and existence of Rico and Camacho, 2008), modified IMC (Tan et al., 2003) and

∗
Corresponding author. Tel.: +91 870 2462633.
E-mail address: [email protected] (A.S. Rao).
http://dx.doi.org/10.1016/j.cherd.2016.03.003
0263-8762/© 2016 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
594 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606

two-degrees-of-freedom control structures (Liu et al., 2005; controller with a lead-lag filter for UFOPDT processes. Their
Tan, 2010) with more number of controllers and also with more method is briefly discussed here. The UFOPDT process transfer
complexity in the design of controllers for UFOPDT processes function is considered as
to improve the closed loop performance. In most of these con-
trollers, there exists performance-robustness trade-off and kP e−s
Gp (s) = (1)
the tuning parameter should be selected in such a way that the P s − 1
resulting controllers give both nominal as well as robust per-
According to IMC principle, the IMC controller QC is equiv-
formance as a function of peak value of the sensitivity function
alent to
i.e. maximum sensitivity (Ms). Moreover, PID controller can-
not provide stabilized responses when the time delay to time
QC = Q̃C F (2)
constant ratio is greater than 1.2 for unstable systems. It can
be observed that complex control schemes such as modified
where F is a filter which is used for altering the robustness
Smith predictor and two degrees of freedom structures where
of the controller. The filter structure should be selected such
there are more than two controllers involved are not desir-
that the IMC controller QC is proper and realizable and also
able for practical purposes and these modified schemes also
the control structure is internally stable. In addition to these
are not applicable when the time delay to time constant ratio
requirements, it should be selected such that the resulting
(To ) exceeds 1.2. Hence, keeping the simplicity into account,
controller provides improved closed loop performances. Q̃C
properly designed PID controller is better than these mod-
is designed for a specific type of step input disturbance (v)
ified schemes. However, the designed PID controller should
to obtain H2 optimal performance (Morari and Zafiriou, 1989)
provide good nominal and robust closed loop responses and
and is based on the invertible portion of the process model.
smooth manipulated variable responses. Alfaro et al. (2010)
The process model and the input are divided as
have developed analytical equations for PI controller for stable
systems based on Ms. Arrieta and Vilanova (2012) developed
PID tuning rules for stable systems for servo/regulatory prob- Gm = Gm− Gm+ and v = v− v+ (3)
lems based on Ms. They formulated an optimization problem
with constraints to design the controller. However, such rules where the subscript “−” refers to minimum phase part and “+”
do not exist for unstable systems. refers to non-minimum phase part. The Blaschke product of
Based on this motivation, in the present work, an attempt RHP poles of Gm and v are defined as
is made to develop analytical tuning rules for PID controllers
as a function of Ms for unstable processes using IMC–H2 
k
−s + p 
k̃
−s + p
i i
minimization theory. This approach explicitly considers the bm = and bv = (4)
s + p̄i s + p̄i
control system performance-robustness trade-off aiming to i=1 i=1

obtain a smooth response to both disturbance and set point

where pi and p̄i are the ith RHP pole and its conjugate respec-
step changes and at the same time to guarantee a minimum
tively. Based on this, the H2 optimal controller is derived
robustness level. The distinctive feature of the resulting tuning
by using the following formula given by Morari and Zafiriou
procedure is that the designer may select one of four different
(1989).
robustness levels in the range 2 ≤ Ms ≤ 10.5 for UFOPDT pro-
cess models with time delay to time constant ratio (To ) in the −1 −1
Q̃C = bm (Gm− bv v− ) {(bm Gm+ ) bv v− }|∗ (5)
range 0.10 ≤ To ≤ 0.9. For clear illustration, the paper is orga-
nized as follows. Controller design is addressed in Section 2
followed by the proposed Ms based tuning rules in Section 3. where {. . .} |* is defined as the operator that operates by omit-
Simulation studies are explained in Section 4 followed by dis- ting all terms involving the poles of (Gm+ )−1 after taking the
cussion in Section 5 and finally conclusions are presented in partial fraction expansion. This idea is applied successfully
Section 6. by Nasution et al. (2011) and derived IMC based PID con-
troller. The same derivation for obtaining the IMC controller
Q̃C for UFOPDT processes is given here for clear understand-
2. Controller design
ing. Considering perfect model case i.e. Gp = Gm , first, split the
process model and input into minimum and non-minimum
The closed-loop control structure of IMC is shown in Fig. 1, phase parts as
where Gp (s) is the transfer function of the unstable process,
Gm (s) is the corresponding transfer function model and QC is −kP
Gm− = and Gm+ = e−s (6)
the transfer function of the IMC controller. Here, the controller P (−s + (1/P ))
design is addressed for UFOPDT processes. Recently, Vanavil
et al. (2014) have developed analytical tuning formulas for PID −kP
v− = and v+ = 1 (7)
P (−s + (1/P ))s

d Then, the Blaschke product is obtained as

u + y
r + Qc + Gp (−s + (1/P )) −s + (1/P )
-
bm = and bv = (8)
(s + (1/P )) s + (1/P )
+
-
Gm Substituting all expressions in Eq. (5), one will get,

(P s − 1) /P
Q̃C = {(e − 1)P s + 1} (9)
Fig. 1 – IMC control scheme. kP
 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606 595

To get the final form of the IMC controller, here, the filter is Substituting all the terms, one can obtain
selected as
{(e/P − 1)P s + 1}(˛s + 1)(P s − 1)
(˛s + 1)
GC =  3
 (17)
F= (10) kP (s + 1) − {(e/P − 1)P s + 1}(˛s + 1)e−s
3
(s + 1)
Vanavil et al. (2014) simplified the above form into a PID
Therefore, the IMC controller is obtained as with lead-lag controller. In this work, it is simplified as a PID
controller. To simplify this expression to a PID controller form,
(P s − 1) /P (˛s + 1) Maclaurin series or Laurent series can be used. To do that, let
QC = {(e − 1)P s + 1} 3
(11)
kP (s + 1) us define J(s) = sGc (s). Expand J(s) using Maclaurin series expan-
sion to obtain the controller Gc as
According to IMC principle, the closed loop transfer func-
 

tion between output (y) and set point (r) for perfect model 1 J 2
Gc (s) = J(0) + J (0)s + s + ··· (18)
conditions (i.e. Gp = Gm ) from Fig. 1 is s 2!

y
H= = QC Gm (12) By considering this as a PID controller in the form
r
 1

Substituting QC from Eq. (11), one can obtain Gc (s) = kc 1 + s + d s (19)
i

{(e/P − 1)P s + 1}(˛s + 1)e−s the PID controller parameters are obtained as
H= 3
(13)
(s + 1)
kc = J (0), i = J (0)/J(0) and d = J (0)/2J (0) (20)
This is the desired closed loop trajectory for the filter cho-
sen as per Eq. (10). If the filter structure changes, then the where
expression for H also changes as per Eq. (12) and hence the
final controller. Here,  is the closed loop tuning parameter.
The value of ˛ is obtained from the conditions of internal J(0) = 1/pm (0)D(0)
stability for IMC structure. The conditions to be followed for J (0) = −[pm (0)D(0) + pm (0)D (0)]/[pm (0)D(0)]
2

internal stability are

pm (0)D(0) + 2pm (0)D (0) + pm (0)D (0) 2J (0)
J (0) = J (0) +
pm (0)D(0) + pm (0)D (0) J(0)
Condition 1: QC must be stable and should cancel the right
D(0) = 3 − pA (0); D (0) = [62 − pA (0)]/2; D (0) = [63 − p
A (0)]/3
half plane poles of Gm .
Condition 2: QC Gm should be stable. pA (0) = ˛ −  + P x
Condition 3: (1 − Gm QC ) at the RHP poles of the process should pA (0) =  2 − 2˛ + 2P x(˛ − )
be zero.
pA (0) = − 3 + 3˛ 2 + 3P x 2 − 6˛P x

The first two conditions are satisfied from the above design pm (0) = −kP ;
procedure and third condition can be applied as pm (0) = −kP (P − ˛ − P x);
2
pm (0) = −kP (2P x(P − ˛) + 2˛P + 2(P − ˛ − P x) )
(1 − QC Gm )|s=1/P = 0 (14)
x = (e(/tp) − 1)
(21)
Substituting QC from Eq. (11) into Eq. (14), the value of ˛ is
obtained as
In which ˛ is given in Eq. (15). These are the final expres-
2
sions for the controller parameters obtained based on IMC
˛ = {(/P ) + 3(/P ) + 3} (15) method and H2 minimization. If the process model is known,
then one has to select proper value of  and then obtain the
Now, Fig. 1 is converted in to a unity feedback control sys- controller parameters. However, selection of  is very impor-
tem as shown in Fig. 2 and the corresponding unity feedback tant for unstable processes and there should be systematic
controller GC is obtained as guidelines for selection of .

QC
GC = (16) 2.1. Guidelines for selection of 
1 − QC Gm

It is well-known that there is always a trade-off in selecting

d the desired closed-loop tuning parameter (). For stable pro-
cesses, fast speed of response and good disturbance rejection
+ are favoured by choosing a small value of  and stability and
r u + y
Gc Gp robustness are favoured by a large value of . However, this
+
-
statement is not always true for unstable processes. Hence,
the choice of  is very important when dealing with unstable
processes. To have clear understanding for selection of , in
the present work, a systematic analysis is carried out using
Fig. 2 – Simple unity feedback control. Ms as the performance index.
596 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606

Ms is the inverse of smallest distance from the Nyquist 14

curve to the critical point (−1,0) in Nyquist plot. Ms value sets a
lower bound for the Gain margin (GM) and Phase margin (PM). 12
The lower bound for GM and PM in terms of Ms are related as
To=0.1
(Skogestad and Postlethwaite, 2005) 10
To=0.2
GM ≥ Ms/(Ms − 1); PM ≥ 2 sin−1 (1/2Ms) To=0.3

c
8
As Ms value increases, the lower bound for GM and PM
decreases. In fact, the recommended minimum values of GM 6
and PM are 1.7 and 35 degrees for a typical process. However,
for unstable processes, the lower bound for phase margin is 4
lesser than 35. This is the limitation for unstable processes.
If the process is delay dominant (ratio of time delay to time 2
2 2.5 3 3.5 4 4.5
constant) is more (>1.2), it is difficult to achieve robust closed
Maximum Sensivity
loop responses because of the limitations of the controller for
controlling a delay dominant unstable process. Hence, one can (a)
select  value based on Ms which provides useful information
3.5
about the lower limits of GM and PM (Anusha and Seshagiri
Rao, 2012). More details about Ms can be found in Skogestad
and Postlethwaite (2005). In the present work, Ms values are To=0.4
3 To=0.5
plotted against the tuning parameter , controller parameters
To=0.6
and from this understanding, one can select the controller c
parameters based on the required level of robustness.
2.5

3. Tuning rules based on Ms

2
In the present work, Ms values are plotted against the con-
troller parameters and from this plot one can select the Ms
value and obtain the controller parameters based on the 3 3.5 4 4.5 5 5.5 6 6.5
required level of robustness. The designer may resolve the Maximum Sensivity
performance-robustness trade off by selecting the appropri-
ate robustness allowed for the control system according to the (b)
expected variation of the controlled process parameters. This 2.1

will give the designer a closed loop control system behaviour 2

with the highest speed that can be obtained for the speci-
1.9 To=0.7
fied minimum robustness. Three models were considered for
To=0.8
carrying out this analysis such as (1) Model 1: kP = 1,  = 0.4, 1.8 To=0.9
 P = 1; (2) Model 2: kP = 4,  = 2,  P = 4; (3) Model 3: kP = 1,  = 0.5,
1.7
c

 P = 5; which are extensively published models in the literature

(Shamsuzzoha and Lee, 2008; Tan et al., 2003; Liu et al., 2005; 1.6
Park et al., 1998).
1.5

3.1. Tuning equations 1.4

1.3
The controller parameters (kc ,  i ,  d ) equations obtained in Eq. 5 6 7 8 9 10
(20) are functions of To (= / p ) and / p . Based on the simpli- Maximum Sensivity
fied expressions obtained and using the process model gain
kP and time constant  P and the normalized dead time To , the (c)
controller normalized parameters as considered as
Fig. 3 – Normalized gain variation with Ms. (a) normalized
dead times 0.1–0.3; (b) normalized dead times 0.4–0.6; (c)
kc = kc ∗ kP (22)
normalized dead times 0.7–0.9.
i
i = (23)
P shows the graph between Ms and normalized proportional
d gain kc . As the normalized dead time To increases the effect of
d = (24) Ms in terms of the tuning parameter , the controller gain kc
P
decreases. Accordingly, as the robustness degree is increased
These normalized controller parameters values are plot- the normalized controller gain kc decreases. This can be visu-
ted against the Ms values for three ranges of normalized dead alized from Fig. 3a–c drawn for the normalized dead times in
times 0.1–0.3, 0.4–0.6, 0.7–0.9. Initially a graph between Ms the range of 0.1–0.9.
and the tuning parameter  is drawn. The maximum and Fig. 4 shows the graph between Ms and normalized integral
minimum values of Ms are selected in order to analyze the per- time i . However, the behaviour of the integral time is entirely
formance and robustness trade off. For that particular range different when compared to normalized controller gain kc . As
of Ms, the controller parameter’s behaviour is analyzed. Fig. 3 the normalized dead time To increases, the effect of Ms in
 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606 597

4.5 0.13

To=0.1 0.12
4
To=0.2 0.11
3.5 To=0.3
0.1
3
0.09

d
2.5 0.08
i

2 0.07

0.06
1.5
0.05
1
0.04
0.5
0.03
2 2.5 3 3.5 4 4.5
0
2 2.5 3 3.5 4 4.5 Maximum Sensivity
Maximum Sensivity (a)
(a)
0.28
9
To=0.4 0.26
8 To=0.4
To=0.5
0.24 To=0.5
7 To=0.6
To=0.6
0.22
6
d 0.2
5
i

0.18
4
0.16
3

2 3 3.5 4 4.5 5 5.5 6 6.5

1 Maximum Sensivity
3 3.5 4 4.5 5 5.5 6 6.5 (b)
Maximum Sensivity
(b) 0.44

22 0.42
To=0.7 To=0.7
20 0.4
To=0.8 To=0.8
18 To=0.9 0.38 To=0.9
d

16 0.36
14
0.34
i

12
0.32
10
0.3
8 5 6 7 8 9 10

6 Maximum Sensivity
(c)
4
5 6 7 8 9 10
Maximum Sensivity Fig. 5 – Normalized derivative time variation with Ms. (a)
normalized dead times 0.1–0.3; (b) normalized dead times
(c)
0.4–0.6; (c) normalized dead times 0.7–0.9.
Fig. 4 – Normalized integral time variation with Ms. (a)
normalized dead times 0.1–0.3; (b) normalized dead times
0.4–0.6; (c) normalized dead times 0.7–0.9.
Using the information given in Figs. 3–5, a set of equations
for the controller’s normalized parameters have been devel-
oped as a function of the robustness index Ms for different
terms of the tuning parameter , the integral time i decreases.
ranges of the normalized model dead times. Those specific
Similarly, as the robustness degree is increased, the integral
tuning expressions are obtained as
time i increases. This can be observed from Fig. 4a–c which
are drawn for the normalized dead times in the range 0.1–0.9.
Fig. 5a–5c illustrates the graph between Ms and normalized kc = a1 exp(b1 ∗ Ms) + c1 exp(d1 ∗ Ms) (25)
derivative time d . As the normalized dead time To increases,
the effect of Ms in terms of the tuning parameter , on the i = a2 exp(b2 ∗ Ms) + c2 exp(d21 ∗ Ms) (26)
derivative time d slightly increases. Further, as the robust-
ness degree is increased, the derivative time d decreases with
respect to the normalized dead times in the range 0.1–0.9. d = a3 exp(b3 ∗ Ms) + c3 exp(d3 ∗ Ms) (27)
598 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606

for the normalized dead time in the range of (i) 0.1–0.3, (ii)
Table 1 – Constants values for calculating the
normalized proportional gain kc . 0.4–0.6 and (iii) 0.7–0.9 respectively for the above three cases.
This broad classification allows a qualitative specification of
To a1 b1 c1 d1
the control system robustness.
0.1 7.64 0.123 −766.4 −3.094
0.2 4.611 0.07708 −56.02 −1.762 3.2. Simplified tuning rules
0.3 2.475 0.1184 −9297 −4.467
0.4 2.469 0.04955 −66.01 −1.897
0.5 2.029 0.043282 −34.38 −1.591
In Section 3.1, the lower and upper values for Ms are con-
0.6 1.756 0.03586 −18.84 −1.322 sidered as 2 and 10.5. Note that Ms = 2 and Ms = 10.5 are two
0.7 1.763 0.01296 −4.176 −0.7112 different significant cases where the robustness is of primary
0.8 1.559 0.0109 −4.705 −0.7219 concern (Ms = 2) and where the aggressiveness is more impor-
0.9 1.354 0.01304 −6.461 −0.7983 tant (Ms = 10.5). The controller parameters are plotted against

Table 2 – Constants values for calculating the 14

normalized integral time  i .

Ms =2.625
12 Ms =3.25
To a2 b2 c2 d2 Ms =3.875
Ms =4.5

0.1 4.024e+010 −15.42 0.8914 −0.2878

10
0.2 3028 −4.433 1.328 −0.1651
0.3 2.122e+013 −14.68 3.66 −0.2664
0.4 3.274e+005 −5.087 3.559 −0.1305 8

c
0.5 3.074e+004 −3.493 5.278 −0.1296 k,
0.6 9049 −2.572 7.589 −0.1296 6
0.7 889.9 −1.443 7.475 −0.05575
0.8 8765 −1.718 12.16 −0.06257
4
0.9 2.712e+004 −1.664 22.25 −0.08384

2
Table 3 – Constants values for calculating the derivative 0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3
To
time  d .
(a)
To a3 b3 c3 d3

0.1 0.02527 0.109 −4.873e+007 −13.11 Ms=3.875

3.4
0.2 0.06492 0.0532 −38.96 −3.96 Ms=4.75
0.3 0.08932 0.0737 −2.459e+008 −11.03 3.2 Ms=5.625
0.4 0.1446 0.02943 −234.4 −3.79 Ms=6.5
3
0.5 0.1871 0.02589 −41.49 −2.722
0.6 0.2327 0.02119 −13.34 −2.04 2.8
c

0.7 0.2979 0.00851 −1.228 −1.037

2.6
0.8 0.3459 0.00726 −3.066 −1.122
0.9 0.3885 0.008236 −7.318 −1.19 2.4

2.2
where ai , bi and ci (i = 1,2,3) are the constants and the values
2
of these constants are given in Tables 1–3 for Eqs. (25)–(27) as
a function of the corresponding To . 1.8
0.4 0.45 0.5 0.55 0.6
To
3.1.1. Robustness specific simple tuning rules (b)
The above tuning equations allows the design of the con-
2.1
trol system with any desired robustness level in the following Ms=6.375
range to analyze the robustness-performance trade-off. 2 Ms=7.75
Ms=9.125
Ms = [2,4.5] with normalized dead times from 0.1 to 0.3, 1.9 Ms=10.5
Ms = [3,6.5] with normalized dead times from 0.4 to 0.6,
Ms = [5,10.5] with normalized dead times from 0.7 to 0.9 1.8
c

1.7
However, for unstable processes, it is difficult to achieve
the robustness levels within small values of Ms. To develop 1.6
the tuning rules for different robustness levels, three sets of Ms
values are considered indicating different levels of robustness. 1.5
Usually, for unstable time delay processes, Ms = [4.5, 6.5, 10.5]
1.4
is recognized as the minimum acceptable robustness level or 0.7 0.75 0.8 0.85
the minimum degree of robustness. In the present work, three To
ranges of Ms values are considered (c)

Ms = [3.875, 5.625, 9.125] for low level of robustness, Fig. 6 – Normalized gain variation with To . (a) To between
Ms = [3.25, 4.75, 7.75] for medium level of robustness, 0.1 and 0.3; (b) To between 0.4 and 0.6; (c) To between 0.7
Ms = [2.625, 3.875, 6.375] for high level of robustness and 0.9.
 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606 599

0.13
Ms=2.625
1.8 Ms=3.25 0.12
1.6 Ms=3.875
0.11
Ms=4.5
1.4
0.1
1.2
0.09
i

1
0.08

d
0.8
0.07
0.6 Ms=2.625
0.06 Ms=3.25
Ms=3.875
0.4
Ms=4.5
0.05
0.2
0.1 0.15 0.2 0.25 0.3
0.04
To
(a) 0.03
0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3
5.5
To
Ms=3.875
5 Ms=4.75 (a)
Ms=5.625 0.3
4.5
Ms=6.5 Ms=3.875
4 Ms=4.75
Ms=5.625
3.5
i

Ms=6.5
0.25
3
d

2.5

2
0.2
1.5
0.4 0.45 0.5 0.55 0.6
To
(b) 0.15
14 0.4 0.45 0.5 0.55 0.6
Ms=6.375 To
Ms=7.75 (b)
12 Ms=9.125 0.44
Ms=10.5

10 0.42
i

0.4
8
0.38
d

6 0.36

0.34 Ms=6.375
4 Ms=7.75
0.7 0.75 0.8 0.85
0.32 Ms=9.125
To Ms=10.5
(c) 0.3
0.7 0.75 0.8 0.85
Fig. 7 – Normalized integral time variation with To . (a) To To
between 0.1 and 0.3; (b) To between 0.4 and 0.6; (c) To (c)
between 0.7 and 0.9.
Fig. 8 – Normalized derivative time variation with To . (a) To
between 0.1 and 0.3; (b) To between 0.4 and 0.6; (c) To
Ms for different values of To and then the optimal coefficients
between 0.7 and 0.9.
are developed for different values of To based on regression
analysis. In this section, the controller parameters are plot- are considered and shown in Figs. 6–8. This information is
ted against To for different values of Ms. Here, four different used to obtain the new PID tuning rules as a function of nor-
robustness levels such as malized dead time based on regression techniques and are
given below.
Ms = {2.625, 3.25, 3.875, 4.5} for normalized dead times from 
0.1 to 0.3, kc = a1 + b1 Toc1 (28)
Ms = {3.875, 4.75, 5.625, 6.5} for normalized dead times from 
0.4 to 0.6, i = a2 + b2 Toc2 (29)
Ms = {6.375, 7.75, 9.125, 10.5} for normalized dead times from 
0.7 to 0.9, d = a3 + b3 Toc3 (30)
600 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606

Table 4 – Constants values for calculating the From the given UFOPDT model, note down the
normalized proportional gain kc . dead time to time constant ratio (T0)
Ms a1 b1 c1 To range

2.625 −1.361 1.629 −0.8557 0.1–0.3 Select desired level of robustness by choosing Ms
3.25 −1.927 2.139 −0.7932 0.1–0.3
3.875 −1.571 1.985 −0.8443 0.1–0.3
4.5 −0.4606 1.435 −0.9814 0.1–0.3
Use tables 1-3 & equations 25-27 OR tables 4-6 &
3.875 −0.4157 1.454 −0.9159 0.4–0.6
equations 28-30 for design of the PID controller
4.75 −0.07606 1.27 −1.006 0.4–0.6
5.625 0.03487 1.226 −1.056 0.4–0.6
6.5 −2.619e−005 1.286 −1.064 0.4–0.6
6.375 −0.1915 1.468 −0.9516 0.7–0.9 Plot a graph between λ vs. Ms to understand the
7.75 0.1079 1.224 −1.12 0.7–0.9 maximum and minimum robustness levels
9.125 0.4028 0.9682 −1.365 0.7–0.9
10.5 0.6546 0.7522 −1.667 0.7–0.9
No
Satisfied with desired level of robustness selected
Table 5 – Constants values for calculating the
normalized integral time  i . Yes
Analyse the closed loop responses
Ms a2 b2 c2 To range

2.625 0.3049 25.8 2.355 0.1–0.3

Fig. 9 – Steps involved in the present design procedure.
3.25 0.2229 17.11 2.13 0.1–0.3
3.875 0.121 9.834 1.759 0.1–0.3
Control acon, u Closed loop output, y
3
4.5 0.002252 5.801 1.38 0.1–0.3
3.875 1.35 27.28 3.855 0.4–0.6 2
4.75 0.9958 16.54 3.155 0.4–0.6
5.625 0.8769 14.56 3.124 0.4–0.6 1
6.5 0.8054 13.5 3.202 0.4–0.6
6.375 2.747 20.1 5.752 0.7–0.9 0
0 2 4 6 8 10
7.75 1.846 16.15 4.701 0.7–0.9 Time
9.125 1.215 14.05 4.079 0.7–0.9 10

10.5 0.5584 12.52 3.491 0.7–0.9

0

Table 6 – Constants values for calculating the derivative

time  d . -10
0 2 4 6 8 10
Ms a3 b3 c3 To range Time

2.625 −0.02617 0.327 0.7381 0.1–0.3 Fig. 10 – Closed loop and control action responses for
3.25 −0.02253 0.3428 0.7676 0.1–0.3 Example 1 for To = 0.3, dashed line – Ms = 2.625, solid line –
3.875 −0.009337 0.3764 0.895 0.1–0.3
Ms = 3.25 and dotted line – Ms = 4.5.
4.5 0.001879 0.4245 1.032 0.1–0.3
3.875 −0.06847 0.4709 0.7797 0.4–0.6
4.75 −0.1411 0.5567 0.6588 0.4–0.6
4. Simulation studies
5.625 −0.01715 0.4601 0.9782 0.4–0.6
6.5 −0.02703 0.4715 0.9243 0.4–0.6 The proposed analysis is explained in a more detailed manner
6.375 −0.1075 0.558 0.7936 0.7–0.9 by considering different examples in this section.
7.75 −0.0002367 0.4618 1.046 0.7–0.9
9.125 0.06296 0.4064 1.264 0.7–0.9
4.1. Example 1
10.5 0.1012 0.3752 1.44 0.7–0.9

Consider the following UFOPDT model for different values for

where constants ai , bi and ci (i = 1, 2, 3) of these new equations the time delay.
are shown in Tables 4–6.
The method of obtaining the other controller parameters G(s) = kP e−s /(P s − 1); ␪ = {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9};
using Eqs. (28)–(30) are also shown in Figs. 7 and 8 which
depicts the behaviour of controller parameters for different kP = P = 1
values of normalized dead times as function of a particular Ms
value. The new tuning Eqs. (28)–(30) are more easy to use than The PID controller parameters are derived based on Eqs.
the original ones (25)–(27). Fig. 6 implies that as the normalized (28)–(30) for different robustness levels and are shown in
dead time To increases, the controller gain kc decreases and as Table 7. Simulation studies have been carried out by consider-
Ms increases, kc increases. From Fig. 7, it can be understood ing unit step change in the set-point followed by a disturbance
that as the normalized dead time To increases, the integral of magnitude 0.5 at t = 5 s for different To values. The cor-
time i increases and as Ms increases, i decreases. Similarly, responding closed loop responses are shown in Fig. 10 for
from Fig. 8 it can be understood that as the normalized dead three different values of Ms for To = 0.3. It can be observed
time To increases, the derivative time d increases and as Ms from Fig. 10 that for high value of Ms (= 4.5), the closed
increases, d decreases. response becomes faster but provides some oscillations for
For clear illustration, the present methodology is given in both set point tracking and disturbance rejection where as low
Fig. 9 as a flow sheet. value of Ms (= 2.625) provides smooth set point tracking and
 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606 601

Control acon, u Closed loop output, y

Table 7 – Controller parameters for different values of Ms 4
& To for Example 1.
To Ms kc i d 2

0.1 2.625 10.3239 0.4188 0.0336

0.1 3.25 11.3618 0.3498 0.0360 0
0 10 20 30 40 50
0.1 3.875 12.3004 0.2922 0.0386 Time
0.1 4.5 13.2878 0.2441 0.0413 5
0.2 2.625 5.0960 0.8877 0.0735
0
0.2 3.25 5.7411 0.7782 0.0771
0.2 3.875 6.1553 0.7005 0.0798 -5
0.2 4.5 6.5026 0.6317 0.0825
0.3 2.625 3.3020 1.8192 0.1083 -10
0 10 20 30 40 50
0.3 3.25 3.6320 1.5398 0.1135 Time
0.3 3.875 3.9156 1.3036 0.1188
0.3 4.5 4.2166 1.1037 0.1244 Fig. 12 – Closed loop and control action responses for
0.4 3.875 2.9493 2.1473 0.1620 Example 1 for To = 0.8, dashed line – Ms = 6.375, solid line –
0.4 4.75 3.1161 1.9148 0.1633
Ms = 7.75 and dotted line – Ms = 9.125.
0.4 5.625 3.2611 1.7082 0.1706
0.4 6.5 3.4069 1.5238 0.1751
0.5 3.875 2.3273 3.2349 0.2058
point tracking and disturbance rejection performances but are
0.5 4.75 2.4742 2.8537 0.2115 slow. Hence, it can be noted that Ms = 4.75 provides a trade-off
0.5 5.625 2.5838 2.5462 0.2164 between robustness and performance. Similarly, Fig. 12 shows
0.5 6.5 2.6871 2.2731 0.2214 the closed loop responses for To = 0.8 and it can be observed
0.6 3.875 1.9055 5.1568 0.2477 from Fig. 12 that high value of Ms (= 9.125) increases, the closed
0.6 4.75 2.0468 4.2980 0.2565
response becomes faster but provides some oscillations for
0.6 5.625 2.1374 3.8276 0.2620
both set point tracking and disturbance rejection where as
0.6 6.5 2.2134 3.4366 0.2670
0.7 6.375 1.8700 5.3291 0.3129 low value of Ms (= 6.375) provides smooth set point tracking
0.7 7.75 1.9324 4.8649 0.3178 and disturbance rejection performances but are slow. Hence,
0.7 9.125 1.9780 4.4962 0.3219 it can be noted that Ms = 7.75 provides a trade-off between
0.7 10.5 2.0176 4.1630 0.3257 robustness and performance. It can be understood from these
0.8 6.375 1.6240 8.3137 0.3599 Figs. 10–12 that as To increases, the peak value increases for
0.8 7.75 1.6789 7.5020 0.3654
set point as well as for disturbance rejection responses which
0.8 9.125 1.7156 6.8716 0.3695
indicates that the model dead time to time constant ratio has
0.8 10.5 1.7456 6.3040 0.3733
0.9 6.375 1.4315 13.7083 0.4057 an adverse effect over the closed-loop system performance
0.9 7.75 1.4847 11.6863 0.4134 and as this ratio increases the closed performance degrades.
0.9 9.125 1.5207 10.3602 0.4187 To analyze the robustness further, an analysis is carried
0.9 10.5 1.5512 9.2267 0.4236 out here based on Ms. Fig. 13 shows the variation of Ms with
respect to  with normalized dead times To = 0.30, 0.6, 0.8.
disturbance rejection performances but are slow. Hence, it can There exist a large value for Ms corresponding to  = 0.2, 0.48,
be noted that Ms = 3.25 provides a trade-off between robust- 0.6 after which the Ms decreases.
ness and performance and a compromise can be achieved It can be observed from Fig. 13 that one should not select
between robustness and performance for this value of Ms.  without proper analysis. Hence  should be selected in the
Fig. 11 shows the closed loop responses for To = 0.6 and it range of 0.25–0.7 for normalized dead time To = 0.30, 0.6–1.2 for
can be observed from Fig. 11 that for a high value of Ms (= normalized dead time To = 0.60 and in the range of 0.89–1.25
6.5), the closed response becomes faster but provides some for normalized dead time To = 0.8. Within this range of , the
oscillations for both set point tracking and disturbance rejec- maximum value for Ms will be 4.5, 6.5 and 10.5. If  is selected
tion where as low value of Ms (= 3.865) provides smooth set
600
Closed loop output, y

3
500
Maximum Sensivity (Ms)

2
400
1

0 300
0 5 10 15 20 25
Time
5
Control acon, u

200

0 100

-5 0
0 5 10 15 20 25 0 0.5 1 1.5 2 2.5 3
Time λ

Fig. 11 – Closed loop and control action responses for Fig. 13 – Variation of Ms for different values of  for
Example 1 for To = 0.6, dashed line – Ms = 3.875, solid line – Example 1 with normalized dead times To = 0.30 (dashed
Ms = 4.75 and dotted line – Ms = 6.5. line), 0.6 (solid line), 0.8 (dash dotted line).
602 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606

4.5
10

Maximum Sensivity (Ms)

Maximum Sensivity (Ms)

4
9

3.5 8

3 7

2.5 6

5
2 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25
0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65
λ
λ
Fig. 16 – Maximum and minimum values of Ms for selected
Fig. 14 – Maximum and minimum values of Ms for selected
range of  for Example 1 with normalized dead times
range of  for Example-1 with normalized dead times
To = 0.8.
To = 0.3.

Control acon, u Closed loop output, y

3
6.5
2
6
Maximum Sensivity (Ms)

1
5.5
0
5 0 5 10 15
Time
5
4.5

4
0

3.5
-5
3 0 5 10 15
0.6 0.7 0.8 0.9 1 1.1 1.2
λ Time

Fig. 17 – Closed loop and control action responses for

Fig. 15 – Maximum and minimum values of Ms for selected
Example 2 for To = 0.1, dashed line – Ms = 2.625, solid line –
range of  for Example 1 with normalized dead times
Ms = 3.25, dotted line – Ms = 3.875 and dash-dotted line –
To = 0.6.
Ms = 4.5.

outside this range the closed loop performance is not good or The controller parameters are calculated from Eqs. (28)–(30)
is not stable. Note that the minimum value of Ms achievable and are given in Table 8. Simulation studies have been carried
in this range of  is 2, 3, 5 as shown in Figs. 14–16. Hence the out by considering unit step change in the set-point followed
Ms ranges were selected as 2–4.5 for To = 0.3, 3–6.5 for To = 0.6, by a disturbance of magnitude 0.5 at t = 7.5 s for different To
5–10.5 for To = 0.8. values. The corresponding closed loop responses are shown
in Fig. 17 for four levels of robustness (high, medium, low,
4.2. Example 2 minimum) for different values of Ms for To = 0.1. It can be
observed from Fig. 17 that for high value of Ms (= 4.5), the
Here, UFOPDT models with different process gains, time closed response becomes faster but provides more oscillations
constants and time delay are considered whose parameters for both set point tracking and disturbance rejection where as
are given below. low value of Ms (= 2.625) provides smooth set point tracking,
disturbance rejection and also control action performances
but are slow. Hence, Ms = 4.5 is less robust and Ms = 2.625 is
(i) kp = 4,  p = 4,  = 0.4 (To = 0.1) more robust. It can be noted that Ms = 3.25 provides a trade-
(ii) kp = 1.8,  p = 1.049,  = 0.298 (To = 0.284) off between robustness and performance. Fig. 18 shows the
(iii) kp = 2,  p = 1.247, ␪ = 0.691 (To = 0.554) closed loop responses for To = 0.284 and it can be observed from

Table 8 – Controller parameters for different values of Ms & To for Example 2.

To = 0.1 To = 0.284 To = 0.554

Ms kc i d Ms kc i d Ms kc i d

2.625 2.58 1.67 0.13 2.625 1.90 1.71 0.10 3.875 1.04 5.17 0.28
3.25 2.84 1.40 0.14 3.25 2.15 1.46 0.11 4.75 1.11 4.44 0.29
3.875 3.07 1.17 0.15 3.875 2.31 1.25 0.11 5.625 1.16 3.96 0.30
4.5 3.32 0.97 0.16 4.5 2.48 1.07 0.12 6.5 1.20 3.54 0.30
 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606 603

Closed loop output, y

Control acon, u Closed loop output, y

3 3

2 2

1 1

0 0
0 5 10 15 0 5 10 15 20 25
Time Time
2
5
Control acon, u

0 -2

-4
-5 0 5 10 15 20 25
0 5 10 15 Time
Time
Fig. 19 – Closed loop and control action responses for
Fig. 18 – Closed loop and control action responses for Example 2 for To = 0.554, dashed line – Ms = 3.875, solid line
Example 2 for To = 0.284, dashed line – Ms = 2.625, solid line – Ms = 4.75, dotted line – Ms = 5.625 and dash-dotted line –
– Ms = 3.25, dotted line – Ms = 3.875 and dash-dotted line – Ms = 6.5.
Ms = 4.5.

Fig. 18 that for high value of Ms (= 4.5), the closed response the model dead time to time constant ratio has an adverse
becomes faster but provides some oscillations for both set effect over the closed-loop system performance and as this
point tracking and disturbance rejection where as low value ratio increases the closed performance degrades. To analyze
of Ms (= 2.625) provides smooth set point tracking and dis- closed loop system robustness, simulation studies are carried
turbance rejection performances but are slow. Here, it can out for uncertainty of +10% in dead time and the correspond-
be noted that Ms = 3.875 provides a trade-off between robust- ing closed loop responses are shown in Fig. 20. It can be
ness and performance. Similarly, Fig. 19 shows the closed loop observed that less value of Ms provides more robust closed
responses for To = 0.554 and it can be observed from Fig. 19 loop responses than those of high Ms values.
that a high value of Ms (= 6.5) provides faster closed response To analyze the robustness further, an analysis is carried
becomes with more oscillations for both set point tracking out here based on Ms. Fig. 21 shows the variation of Ms with
and disturbance rejection, whereas low value of Ms (= 3.875) respect to ␭ with normalized dead times To = 0.1, 0.284, 0.554.
provides smooth set point tracking and disturbance rejec- There exist a large value for Ms corresponding to  = 0.17,
tion performances but are slow. Hence, it can be noted that 0.15, 0.48 after which the Ms decreases. The range of Ms for
Ms = 4.75 provides a trade-off between robustness and perfor- selected ranges of  are calculated from this analysis (figures
mance. It can be understood from these Figs. 17–19 that as not given). Note that the maximum and minimum values of
To increases, the peak value increases for set point as well Ms for selected range of  with normalized dead times To = 0.1
as for disturbance rejection responses which indicates that and 0.224 are the same as already discussed in Example 1.

3.5
Closed loop output, y

2.5

1.5

0.5

0
0 5 10 15 20 25
Time
2

1
Control action, u

-1

-2

-3

-4
0 5 10 15 20 25
Time

Fig. 20 – Closed loop and control action responses for Example 2 for To = 0.554 for uncertainty of +10% in dead time, dashed
line – Ms = 3.875, solid line – Ms = 4.75, dotted line – Ms = 5.625 and dash-dotted line – Ms = 6.5.
604 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606

150 200

180
Maximum Sensivity (Ms)

160

Maximum Sensitivity (Ms)

100
140

120

100
50
80

60

40
0
0 0.5 1 1.5
λ 20 X: 21.79
Y: 2.624

Fig. 21 – Maximum sensitivity (Ms) for different values of  0

0 5 10 15 20 25 30
for Example 2 with normalized dead times To = 0.1 (dotted λ
line), 0.284 (dashed line), 0.554 (solid line).
Fig. 23 – Maximum sensitivity (Ms) for different values of .

4.3. Case study: control of exit concentration in a and exit concentratio is the controlled variable. Linearization
chemical reactor of the manipulated variable around this operating condi-
tion C = 1.316 gives the unstable transfer function model as
A isothermal continuous stirred tank reactor is considered 3.433/(103.1 s). For this particular case, the time delay is con-
here which exhibits multiple steady state solutions. The sidered as 20 s. Hence, the unstable transfer function model is
mathematical model of the reactor is given as (Sree and Chi- obtained as,
dambaram, 2006)
3.433e−20s
dC Q k1 C Gp =
= (Cf − C) − 103.1s − 1
dt V 2
(k2 C + 1)
For this UFOPDT model, the proposed methodology is
where Cf is the inlet concentration, Q is the inlet flow rate, V applied. Ms is considered as 2.62 and corresponding to this Ms,
is the volume of the reactor, C is the exit concentration, k1 & the controller is designed using Eqs. (28)–(30) and the PID con-
k2 are the kinetic parameters. The operating parameters are troller settings are obtained as kc = 1.5342,  i = 87.35,  d = 7.35.
given as Q = 0.0333 L/s, V = 1 L, k1 = 10 L/s, and k2 = 10 L/mol. This With these controller settings the proposed method is sim-
process exhibits three steady states. By considering a nom- ulated for a unit step change in the set point at time t = 0
inal value of Cf = 3.288 mol/L, the steady states are obtained and a disturbance of magnitude 0.5 at t = 350 s respectively.
as C = 1.7673, 1.316 and 0.01424 mol/L. Out of the three steady The corresponding closed loop response and control action
states, there is one unstable steady state at C = 1.316 mol/L. responses are shown in Fig. 22 for perfect model conditions.
Feed concetration is considered as the manipulated variable To analyze the effect of uncertainties, perturbation of +20%

2.5
Closed loop output, y

1.5

0.5

0
0 100 200 300 400 500 600 700
Time
2
Control action, u

-1

-2
0 100 200 300 400 500 600 700
Time

Fig. 22 – Closed loop and control action responses, solid line – perfect model, dashed line – uncertainty of +20% in dead time.
 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606 605

1.42

1.4

Exit Concetration, C
1.38

1.36

1.34

1.32

1.3
0 100 200 300 400 500 600 700
Time
3.4
Feed Concentration, Cf

3.35

3.3

3.25

3.2

0 100 200 300 400 500 600 700

Time

Fig. 24 – Closed loop and control action responses obtained by simulation on the nonlinear model, solid line – perfect
model, dashed line – uncertainty of +20% in dead time.

in dead time is considered and the corresponding closed loop 6. Conclusions

responses are also shown in Fig. 22. It can be observed that the
proposed method works well and able to provide good closed In the present work, analytical tuning rules have been devel-
loop responses. The normalized dead time is 0.194. For under- oped for PID controllers for UFOPDT processes. The proposed
standing the maximum and minimum values of Ms values, tuning rules allow the designer to design closed loop control
Ms values are plotted against the tuning parameter, . Fig. 23 systems with a specified low, medium, or high robustness level
shows the variation of Ms with . It can be observed that low measured using the maximum sensitivity. The performance-
value of  gives higher values of Ms and vice versa. Further, robustness trade-off is addressed. Once the UFOPDT model
the designed controller is applied by simulation directly on is known, based on the dead time to tome constant ratio,
the original nonlinear system by giving a step change in the the controller parameters can be selected from the developed
set point from 1.316 to 1.34 at time t = 0 and a step disturb- tuning relations for a desired level of robustness. The applica-
ance of magnitude 0.034 at t = 350 s respectively. Fig. 24 shows bility of the present method is discussed with two examples
the closed loop responses for perfect model parameters. Again for different values of time delay to time constant ratios and
the closed loop responses are good for both set point tracking also with a case study of CSTR.
and disturbance rejection. To analyze the performance of the
designed controller for uncertainties, perturbations of +20%
References
in dead time is considered and the corresponding closed loop
responses are also shown in Fig. 24. It can be observed that the
Alfaro, V.M., Vilanova, R., Arrieta, O., 2010. Maximum sensitivity
proposed method performs well even for uncertainties on the based robust tuning for two-degree-of-freedom
nonlinear system simulation. proportional-integral controllers. Ind. Eng. Chem. Res. 49 (11),
5415–5423.
Anusha, A.V.N.L., Seshagiri Rao, A., 2012. Design and analysis of
5. Discussion IMC based PID controller for unstable systems for enhanced
closed loop performance. In: IFAC Conference on Advances in
The proposed methodology and tuning rules are useful for PID control, Brescia, Italy.
Arrieta, O., Vilanova, R., 2012. Simple servo/regulation
design of PID controllers for UFOPDT processes. The tuning
proportional-integral-derivative (PID) tuning rules arbitrary
rules obtained in Eqs. (25)–(27) are useful for the cases where To Ms-based robustness achievement. Ind. Eng. Chem. Res. 51 (6),
varies between 0.1 and 0.9 and to design the PID controller for 2666–2674.
desired Ms values. However, the tuning rules obtained in Eqs. Arrieta, O., Vilanova, R., Visioli, A., 2011.
(28)–(30) are useful for the cases where the PID controller need Proportional-integral-derivative tuning for servo/regulation
to be designed for different rages of To and to achieve any of the control operation for unstable and integrating processes. Ind.
Eng. Chem. Res. 50 (6), 3327–3334.
four Ms. In this paper, the tuning rules are provided for specific
Liu, T., Zhang, W., Gu, D., 2005. Analytical design of two degree of
ranges of To and Ms. Moreover, the proposed methodology can
freedom control scheme for open-loop unstable processes
be extended to other ranges of To and Ms. Also, the method- with time delay. J. Process Control 15 (5), 559–572.
ology can be extended to second order unstable systems after Morari, M., Zafiriou, E., 1989. Robust Process Control. Prentice
validation with multiple plants with uncertainty. Hall, Englewood Cliffs, NJ.
606 chemical engineering research and design 1 0 9 ( 2 0 1 6 ) 593–606

Nasution, A.A., Cheng, J., Huang, H.-P., 2011. Optimal H2 IMC-PID Shamsuzzoha, M., Lee, M., 2008. Analytical design of PID
controller with set point weighting for time-delayed unstable controller for integrating and first order unstable processes
process. Ind. Eng. Chem. Res. 50, 4567–4578. with time delay. Chem. Eng. Sci. 63 (10), 2717–2731.
Normey-Rico, J.E., Camacho, E.F., 2008. Simple robust dead-time Skogestad, S., Postlethwaite, I., 2005. Multivariable Feedback
compensator for first-order plus dead-time unstable Control: Analysis and Design. John Wiley & Sons, New York.
processes. Ind. Eng. Chem. Res. 47, 4784–4790. Sree, R.P., Srinivas, M.N., Chidambaram, M., 2004. A simple
Sree, R.P., Chidambaram, M., 2006. Control of Unstable Systems. method of tuning PID controllers for stable and unstable
Narosa Publishers, New Delhi. FOPTD systems. Comput. Chem. Eng. 28, 2201–2218.
Park, J.H., Sung, S.W., Lee, I.B., 1998. An enhanced PID control Tan, W., 2010. Analysis and design of a double two-degree of
strategy for unstable process. Automatica 34 (6), 751–756. freedom control scheme. ISA Trans. 49 (3), 311–317.
Rao, A.S., Chidambaram, M., 2006. Control of unstable processes Tan, W., Marquez, H.J., Chen, T., 2003. IMC design for unstable
with two RHP poles, a zero and time delay. Asia Pacific J. processes with time delays. J. Process Control 13 (3),
Chem. Eng. 1, 63–69. 203–213.
Rao, A.S., Chidambaram, M., 2012. PI/PID controllers for Uma, S., Chidambaram, M., Rao, A.S., 2010. Set-point weighted
integrating and unstable systems. In: Vilanova, R., Visioli, A. modified Smith predictor with PID filter controllers for
(Eds.), PID Control in the Third Millennium. Springer, pp. non-minimum phase (NMP) integrating processes. Chem. Eng.
75–111. Res. Des. 88 (5), 592–601.
Rao, A.S., Rao, V.S.R., Chidambaram, M., 2007. Simple analytical Vanavil, B., Anusha, A.V.N.L., Perumalsamy, M., Rao, A.S., 2014.
design of modified Smith predictor with improved Enhanced IMC-PID controller design with lead-lag filter for
performance for unstable first order time delay (FOPTD) unstable and integrating processes with time delay. Chem.
processes. Ind. Eng. Chem. Res. 46, 4561–4571. Eng. Commun. 201, 1468–1496.