Int. J. Automation and Control, Vol. 10, No.

2, 2016 167

Analytical tuning method for cascade control system

via multi-scale control scheme

Jobrun Nandong
Department of Chemical Engineering/CSRI,
Curtin University Sarawak,
98009 Miri, Malaysia
Email: [email protected]

Abstract: Cascade control strategy has been widely used in process industry
owing to its superior disturbance rejection performance over that of the
standard single-loop control. Designing the cascade control system is a
challenging task, especially for an unstable time-delay process. In this paper,
based on the multi-scale control scheme, a new analytical tuning method is
presented. The method is applicable to stable and unstable processes where it
can be implemented through a fixed or variable tuning approach. In the variable
tuning approach, one adjustable tuning parameter is sufficient to obtain desired
cascade control performance with the aid of a set of analytical tuning formulae
providing direct relationship between the controller parameters and model
parameters. Meanwhile, the fixed tuning approach uses a set of empirically
developed rules to determine the controller parameters. Numerical studies
demonstrate the effectiveness of the proposed method compared with several
established cascade control designs reported in the literature.

Keywords: cascade control; robust tuning; PID control; time-delay; unstable

process.

Reference to this paper should be made as follows: Nandong, J. (2016)

‘Analytical tuning method for cascade control system via multi-scale control
scheme’, Int. J. Automation and Control, Vol. 10, No. 2, pp.167–192.

Biographical notes: Jobrun Nandong received his MEng degree in Chemical

Engineering from Imperial College London and a PhD degree in Process
System Engineering from Curtin University. His research interests cover areas
in process modelling and control, generalised multi-scale control theory and
biosystem design. He is the Chair of Intelligent Systems, Design and Control
(ISDCON) Research Group at the Faculty of Engineering and Science, Curtin
University, Sarawak Campus, Malaysia.

1 Introduction

Cascade control strategy has been widely used in industries because of its superior
disturbance rejection performance over that of the standard single-loop feedback control
strategy. Furthermore, the cascade control strategy is relatively simple to implement as
compared to some well-known advanced control techniques reported in the open
literature. The benefits of cascade control have been well-recognised in many process

Copyright © 2016 Inderscience Enterprises Ltd.

168 J. Nandong

control textbooks, e.g., see Smith and Corripio (1985), Marlin (1995) and Skogestad and
Postlethwaite (2007).
In industrial applications, a cascade control system often consists of two or more
nested PID-type controllers, i.e., the system has two or more feedback loops. For the
secondary or inner feedback loop, a simple P or PI controller is often selected. This
secondary controller is cascaded with a more complex PID controller implemented at the
primary or outer feedback loop. The P or PI controller for the secondary loop serves as a
slave controller, which relies on a secondary measurement to compute its control action.
This secondary measurement is selected in such a way that, the occurrence of
disturbances is observable through the measurement. Hence, the slave controller shall be
able to provide an early compensation of the disturbances before they can seriously affect
the main process. Meanwhile, the primary PID controller serves as a master controller,
which provides a remote setpoint to the slave controller. Note that, this master controller
uses another measured variable known as the primary measurement in order to compute
its control action, i.e., the remote setpoint that is sent to the slave controller.
It is interesting to note that, extensive research studies have been reported in the
literature on cascade control, which cover broad areas in engineering applications, such
as, the control of an exothermic catalytic reactor (Caetano et al., 2014), system actuated
by Pneumatic Muscle Actuator (Zhong et al., 2014), hydraulic driven parallel robot
platform (Nedic et al., 2014), industrial boiler combustion process (Chen and Chang,
2013), thermal power plant (Kwon et al., 2013), HVAC system (Legweel et al., 2014)
and distillation composition control (Kano et al., 2003), just to highlight a few works
appeared in recent literature reports.
Although cascade control strategy has long been used in process industry, the
challenge in designing a cascade control system has remained one of the main research
topics in both academia and industry. The effort toward improving cascade control design
techniques remains ongoing in view of the need for well-designed PI/PID cascade control
systems, which are critical to achieving growingly stringent control objectives in
response to increasingly tight product specifications and environmental regulations. It is
worth highlighting that, designing an effective cascade control system can often be a
difficult task especially when the process involved has long dead time and integrating or
unstable mode. In view of this challenge, various cascade control designs have been
proposed by numerous researchers over the last 20 years, where in recent years many
researchers have focused on time-delay unstable or integrating processes. In general, the
existing cascade control designs can be broadly categorised into two major groups:
1 methods that were developed based on the standard or classical cascade control
structure
2 methods that were constructed based on some modified Smith predictor or internal
model control (IMC) structures.
The methods based on the standard cascade control structure are advantageous over those
based on the modified Smith predictor structures because the former are easy to
implement in real processes. In essence, the design methods built for the standard cascade
structure attempt to acquire good tuning values for both secondary and primary PI/PID
controllers as a mean to achieve good overall control performance. On the other hand, the
modified cascade control structures, e.g., the modified Smith predictor cascade control
systems are often overly complex, which in turn might hinder their industrial
 Analytical tuning method for cascade control system 169

implementation. Note that, the modified Smith predictor or IMC cascade control
structures might use non-standard feedback and require several PID-type (or in some
cases are non-PID-type) controllers which are to be designed via some tedious ways.
Based on the standard cascade control structure, one way to design the cascade
controllers involved is via a sequential tuning approach. In this approach, the slave
controller is first designed or tuned, and then followed by tuning of the master controller.
Interestingly, several PID tuning rules that have been developed for SISO processes, for
examples, the Ziegler-Nichols tuning and its variants can be used in the sequential design
of cascade controllers. It has been reported that, the secondary and primary controllers
can in fact be tuned individually (Tan and Liu, 2005). Please note that, a number of
studies have been reported in the open literature which provide procedures for performing
cascade controller tunings. Some of the recent works can be found in Chen et al. (2013),
Azar and Serrano (2014), Hasanien and Muryeen (2013), Jeng (2014), Uma et al. (2009),
Kaya et al. (2007), Garcia et al. (2010), Nandong and Zang (2014b) and the references
therein.
In Jeng (2014), a simultaneous method of tuning cascade controllers based on a
step-response data was proposed. The method can help engineers to improve the tuning
of some existing under-performing controllers without any need for prior information on
the old controller tuning values. Furthermore, an independent design method was
proposed in Azar and Serrano (2014) based on the gain and phase margin (PM)
specifications. Following this method, a few IMC-PID parameters of the secondary and
primary controllers need to be adjusted properly as to meet desired robustness
specifications. Meanwhile, Hasanien and Muryeen (2013) proposed a Taguchi-based
design method for tuning four PI controllers in a cascaded electrical system. This
Taguchi-based method may avoid a cumbersome simultaneous tuning procedure
involving a large number of cascaded controller parameters. It is interesting to point out
that, some researchers have also advocated the incorporation of Smith predictor into
cascade control design, for examples, the works reported in Uma et al. (2009) and Garcia
et al. (2010). Also note that, recently in Nandong and Zang (2014b) a new method was
proposed to designing a cascade control system based on an extension of the basic SISO
multi-scale control (MSC) scheme proposed in Nandong and Zang (2013a, 2013b).
In the present work, we propose a set of empirical tuning rules to help design a
cascade control system for a stable or unstable process with a significant time-delay. The
salient feature of the proposed method lies in its simplicity, which enables a control
engineer to determine good controller parameter values without the need for detailed
process control knowledge, i.e., through a fixed tuning method based on the process
model parameters. It is worth highlighting that, the proposed method ensures both
closed-loop stability and satisfactory performance robustness. Bear in mind that, the
proposed tuning rules are developed with an assumption that a given process is reducible
to either first-order plus dead-time (FOPDT) model for a case of stable process, or
first-order dead-time unstable process (FODUP) form for a case of open-loop unstable
process.
The rest of the paper is structured as follows. Section 2 presents some preliminaries
and followed by Section 3 which provides derivation of several multi-scale control
(MSC-PID) tuning relations for both stable and unstable processes. Section 4
demonstrates the applications of the proposed empirical tuning rules and finally Section 5
highlights some conclusions and future studies.
170 J. Nandong

2 Preliminaries

2.1 Standard cascade control structure

Figure 1 shows the block diagram of the standard cascade control structure. In Figure 1,
Fr, Gc1, Gc2, Gs and Gp denote the setpoint pre-filter, master (primary) controller, slave
(secondary) controller, secondary process and primary process transfer functions,
respectively; R1, R2, Di, Do, Y2 and Y1 represent the external setpoint, remote setpoint,
input disturbance, output disturbance, secondary variable and primary variable signals,
respectively. For the cascade control scheme to work effectively, it is required that the
secondary loop to be substantially faster than the primary loop. The idea of cascade
control is to quickly remove the effect of disturbances Di and Do by using the slave
controller before the disturbances can seriously affect the primary variable Y1.

Figure 1 Block diagram of the standard cascade control structure (2-level cascade system)

Figure 2 Block diagram of equivalent reduced single-loop structure of 2-level cascade control

Note that, for the purpose of feedback analysis, it is helpful to reduce the cascade control
structure (Figure 1) to an equivalent single-loop structure in Figure 2. In Figure 2, the
closed-loop setpoint transfer function from R2 to Y2
Gc 2 ( s )Gs ( s )
H r 2 ( s) = (1)
[1 + Gc 2 ( s)Gs ( s)]
Meanwhile, the closed-loop disturbance transfer function from Do to Y2
1
H do ( s ) = (2)
[1 + Gc 2 (s)Gs (s)]
and that of from Di to Y2 is written as follows
Gs ( s )
H di ( s ) = (3)
[1 + Gc 2 ( s)Gs ( s)]
 Analytical tuning method for cascade control system 171

The overall closed-loop transfer function from the external setpoint R1 to Y1 is

Fr ( s )Gc1 ( s )Gove ( s )
H r1 ( s ) = (4)
1 + Gc1 ( s )Gove ( s )

whereas the augmented primary plant transfer function is expressed as

Gove ( s ) = H r 2 ( s )G p ( s ) (5)

The overall closed-loop transfer function from Do to Y1 is written as

H do ( s )G p ( s )
H d1 ( s) = (6)
1 + Gc1 ( s )G p ( s )

2.2 Overview of the MSC

The details about the MSC (MSC) scheme can be found in Nandong and Zang (2013a,
2013b). In this preliminary section, only a brief overview of the MSC scheme is
presented. Figure 3 shows the simplest form of MSC scheme, i.e., with a 2-layer
structure. In Figure 3, Fr, P and W indicate setpoint pre-filter, plant and multi-scale
predictor transfer functions; Kj j = 0, 1 denote the sub-controller for the outermost mode
(m0) and inner-layer mode (m1); R, C, U and Y are the setpoint, outermost sub-controller
output, manipulated input and controlled variable signals, respectively.

Figure 3 Block diagram of a 2-layer MSC scheme

R C U Y
Fr + K0 + ‐ K1 P
‐
W

The 2-layer MSC scheme assumes that a given process can be decomposed into a sum of
two basic modes or factors only. The decomposition can be represented by
P ( s ) = m0 ( s ) + m1 ( s ) (7)

Here, m0 and m1 are the outermost and inner-layer modes respectively, which are either
first- or second-order systems with real coefficients. In terms of the speed of responses,
m0 has slower dynamic response than that of m1.
Note that, the fast inner mode is often chosen as the multi-scale predictor, hence
W = m1. The closed-loop inner-layer transfer function from C to U is given by
G1 ( s ) = K1 ( s ) (1 + K1 ( s )W ( s ) ) (8)

An overall MSC controller can then be obtained as follows

K msc ( s ) = K 0 ( s )G1 ( s ) (9)

It is interesting to point out that, selecting a simple P controller form is sufficient for the
inner sub-controller while a PI controller can be chosen for the outermost sub-controller.
172 J. Nandong

For the case of 2-layer MSC scheme, a combination of P and PI sub-controllers will lead
to an overall MSC controller, which can be arranged into a PID controller augmented
with a lag filter

⎛ 1 ⎞⎛ 1 ⎞
K msc ( s ) = K c ⎜1 + + τ D s ⎟⎜ ⎟ (10)
⎝ τ I s ⎠⎝ τ f s +1⎠

where Kc, τI, τD and τf are the controller gain, reset time, derivative time and filter time
constant, respectively.
Nandong and Zang (2014a) developed some mathematical relations for calculating
the PID controller parameters in (10) for a process that can be approximated by the
first-order plus dead time (FOPDT) model, i.e., of the form

K p e −θs
P(s) = (11)
τs + 1

where Kp, τ and θ represent the process gain, time constant and dead time, respectively.
In Nandong and Zang (2014a), the MSC-PID relations are given as follows:

⎛ ( λ − 1) ( λ1 − 1) ⎞ ⎛ 2γτ + θ ⎞ ⎛ ( 2 τ − θ ) ⎞
2
Kc = ⎜ 0 ⎟⎜ ⎟⎜⎜ ⎟⎟ (12)
⎝ γλ1 ⎠ ⎝ 2τ + θ ⎠ ⎝ 4τθK p ⎠

τ I = γτ + 0.5θ (13)

τ D = γθτ ( 2γτ + θ ) (14)

τ f = θ ( 2 λ1 ) (15)

In (12) to (15), λ0, λ1 and γ denote the MSC tuning parameters as defined in Nandong and
Zang (2014a).
Please note that, Nandong (2015) has recently constructed some PID tuning relations
based on the first-order dead time unstable process (FODUP) model given as

K p e −θs
P(s) = (16)
τs − 1

For a process in (16), the controller gain is expressed as follows

⎛ ( λ + 1) ( λ1 − 1) ( 2τ − θ )2 ⎞ ⎛ 2γτ + θ ⎞
K c = ⎜⎜ 0 ⎟⎟ ⎜ ⎟ (17)
⎝ 4 λ1θK p (2 τ − θ ) ⎠ ⎝ γτ ⎠

Note that, the reset time, derivative time and filter time constant are calculated in the
same ways as in (13), (14) and (15) respectively. In the next section, several empirical
relations for calculating the MSC-PID parameters of cascade control system will be
presented.
 Analytical tuning method for cascade control system 173

3 Cascade control system tuning

3.1 Case A: stable primary process

In this case, both primary and secondary processes are assumed to be open-loop stable
and representable by the FOPDT model. Hence, the primary and secondary processes
should be given as in (18) and (19), respectively:
K p e−θ p s
G p (s) = (18)
τ ps +1

K s e − θs s
Gs ( s ) = (19)
τs s + 1
Here, Kp, τp and θp denote the gain, time constant and dead time of the primary process;
Ks, τs and θs denote the gain, time constant and dead time of the secondary process,
respectively.
Please note that, two sub-cases are considered: case A.1 and case A.2 where a
P-controller and a PI controller are used as the secondary controllers, respectively. For
both sub-cases, a PID controller augmented with a filter given in (10) is used as the
primary (master) controller.

3.1.1 Case A.1: secondary P controller design

In the case where P controller is used as the secondary (slave) controller, it can be shown
that the closed-loop transfer function in (1) for the secondary loop:
K c 2 Gs ( s )
H r 2 ( s) = (20)
1 + K c 2 Gs ( s )

Here, Kc2 denotes the secondary controller gain, i.e., Gc2 = Kc2.
From (20), the characteristic equation (CE) is given as follows.
1 + K c 2 Gs ( s ) = 0 (21)

Let us approximate the dead time of the secondary process by using the first-order
Taylor’s series given as follows
e − θs s = 1 − θs s (22)
After substituting (22) into (20), the CE simplifies to
( τ s − K c 2 K s θs ) s + (1 + K c 2 K s ) = 0 (23)

Now, based on the Routh-Hurwitz stability criterion, i.e., see Ho et al. (1998), all
coefficients of the CE must be positive to ensure closed-loop stability (i.e., corresponding
to the necessary condition), hence:
τ s − K c 2 K s θs > 0 (24)
1 + Kc2 K s > 0 (25)
174 J. Nandong

From (24) and (25), one can conclude that the range of loop gain Kc2Ks ensuring
closed-loop stability should be
−1.0 ≤ K c 2 K s < τ s θs (26)

It is interesting to note that, the upper limit of (26) can be expressed in term of a positive
parameter r as in (27):
τs τ
Kc2 K s = < s ; r >1 (27)
rθs θs
Here, r > 1 ensures a closed-loop stability of the secondary or slave loop.
From (27), one can then specify the value of r to calculate the secondary controller
gain. Hence, the controller gain can be expressed in terms of r and model parameters in
(19), i.e., the gain is expressed as follows.
τs
Kc2 = (28)
rK s θs
Remark 1: Notice that in (28), r is used as a tuning parameter for the secondary
controller. A recommended range of tuning values for r: 1.1 ≤ r ≤ 2.5. When r is too
large, the slave controller will give a poor (sluggish) disturbance (regulatory) response.
On the other hand, if r is too small, the regulatory response will be fast but this may lead
to poor robustness against modelling uncertainties, especially due to a structural
mismatch between the nominal model and actual process.

3.1.2 Case A.1: primary PID controller design

After the secondary controller design is completed, the transfer function in (20) can be
simplified to
K so e −θs s
H r 2 (s) = (29)
τ cs s + 1

where the overall secondary process gain and closed-loop time constant are
Kc2 Ks
K so = (30)
1 + Kc2 Ks

τ s − K c 2 K s θs
τ cs = (31)
1 + Kc2 Ks
Both (30) and (31) can be conveniently written in terms of r and model parameters (19)
leading to the following expressions
1
K so = (32)
r ( θs τ s ) + 1

θs (r − 1)
τ cs = (33)
r ( θs τ s ) + 1
 Analytical tuning method for cascade control system 175

Remark 2: The overall secondary process gain Kso and closed-loop time constant τcs can
be easily calculated after specifying a desired value for r, which represents an adjustable
tuning parameter for the secondary P controller.
Based on Figure 2, the augmented primary process transfer function in (5) can be
expressed in the form of

K p K so e −( θ p + θs ) s
Gove ( s ) = (34)
( τ p s + 1) ( τ cs s + 1)
Further notice that, upon reducing (34) to the FOPDT model by applying the Skogestad
Half-Rule (Skogestad, 2003), the augmented primary process becomes
K ove e−θeff s
Gove ( s ) = (35)
τ eff s + 1

Here, the augmented process gain Kove = KpKso, effective time constant τeff = τp + 0.5 τcs
and effective dead-time θeff = θp + θs + 0.5τcs.
Remark 3: After simplifying (34) to (35), one can then tune the primary PID controller by
using the MSC tuning relations given by (12) to (15). Interestingly, to ease the primary
controller tuning task, the empirical tuning rules given in Table 1 can be used, which
have been obtained through an extensive numerical simulation study.
Table 1 Empirical tuning rules for FOPDT system with λ1 = 20 and γ = 0.8

Range Tuning equation for λ0

0 < θ / τ ≤ 0.4 1.29(θ / τ) + 1.47
0.4 < θ / τ ≤ 0.7 2.62(θ / τ) + 0.72
0.7 < θ / τ ≤ 1.2 5.10(θ / τ) – 1.07
1.0 < θ / τ ≤ 1.2 3.82(θ / τ) + 1.13
1.2 < θ / τ ≤ 1.6 6.05(θ / τ) + 1.01

3.1.3 Case A.2: secondary PI controller design

Let us now consider that the secondary controller is a PI controller given by
( τ Is s + 1)
Gc 2 ( s ) = K c 2o (36)
s
where Kc2o = Kc2 / τIs; Kc2 and τIs denote the controller gain and reset time, respectively.
From the closed-loop transfer function Hr2, the second-order CE can be generalised as
follows.
a2 s 2 + a1s + a0 = 0 (37)
where the coefficients of the CE are given by
a2 = τ s − K c 2o K s τ Is θs (38)

a1 = 1 + K c 2o K p ( τ Is − θ ) (39)
176 J. Nandong

a0 = K c 2o K p (40)

According to the Routh stability necessary criterion, all of the coefficients (38) to (40)
must be positive. Thus, the loop gain upper limit (a2 > 0) is given by
τs τ
K c 2o K s = < s ; r >1 (41)
rθs τ Is θs τ Is
From (41), the controller gain can be expressed in the form of
τs
Kc2 = ; r >1 (42)
rK s θs
Notice that, the controller gain expression is the same as in the case of P-only controller,
i.e., see (28).
Due to the fact that a1 > 0, one will have a lower limit if and only if (τIS – θs) > 0
leading to
−1
K c 2o K s > (43)
τ Is − θs
Otherwise, one will have another upper limit if and only if (τIS – θs) < 0, leading to:
1 1
K c 2o K s = < ; r >1 (44)
r ( θs − τ Is ) ( θs − τ Is )

In the present work, we assume that the first condition holds, i.e., (τIS – θs) > 0. Thus, the
secondary controller gain can be calculated as in (41). Otherwise, if the upper limit given
in (44) is lower than that given in (41), the controller gain should instead be calculated as
follows.
τ Is
Kc2 = ; r >1 (45)
r ( θs − τ Is )

Bear in mind that, we propose to set the reset time τIS via specifying a parameter η
representing a fraction on the sum of the dead time and time constant:
τ Is = η ( τ s + θs ) ; 0 < η < ηmax (46)

Remark 4: In (46) ηmax > 1 and a value of η in the range of 0.5 < η < 1 is recommended.
When η is too small, the controller is not sufficiently robust against modelling
uncertainties. A too large value of η may lead to a sluggish closed-loop response – due to
insufficient integral action.
Remark 5: For satisfactory tradeoffs between performance and robustness of a PI
controller used in the secondary loop, recommended values of r = 1.6 and η = 0.8 should
be used for 0.1 ≤ θs / τs < 1.0. These settings lead to the values of gain margin (GM) and
PM above 8 dB and 49° respectively. Of course, one may use different values r and η in
order to achieve different performance-robustness requirements.
Recalling the general CE in (37), one can re-write the CE in a more specific form
τ cs2 s 2 + 2ξcs τ c s + 1 = 0 (47)
 Analytical tuning method for cascade control system 177

where the closed-loop time constant τcs and damping factor are ξcs expressed in terms of r
and τIS

τ cs = τ Is θs (r − 1) (48)

rτ Is θs + τ s ( τ Is − θs )
ξcs = (49)
2τ s τ Is θs (r − 1)

Supposed that ξcs > 1, the CE in (47) can be further expressed in terms of two time
constants τcs1 and τcs2 leading to CE of the form

( τ cs1s + 1)( τ cs 2 s + 1) = 0; τ cs1 > τ cs 2 (50)

τ cs
τ cs1 = (51)
( ξcs − ξcs2 − 1 )

τ cs
τ cs 2 = (52)
( ξcs + ξcs2 − 1 )

Based on a PI controller giving the CE in (50), the setpoint closed-loop transfer function
(1) becomes

( τ Is s + 1) e−θs s
H r 2 (s) = (53)
( τ cs1s + 1)( τ cs 2 s + 1)
Remark 6: Notice in (53), the overall gain is always unity, i.e., for a PI controller there is
no offset in the setpoint tracking.

3.1.4 Case A.2: primary PID controller design

By applying the Skogestad half-rule (Skogestad, 2003), the second-order transfer
function (52) can be reduced to the FOPDT model. According to the half-rule,
τIS / τcs1 = Kld. Hence, the approximated FOPDT model becomes
Kld e −θs s
H r 2 (s) = (54)
( τ cs 2 s + 1)
Based on (54), the augmented primary process is given as

K p Kld e −( θ p + θs ) s
Gove ( s ) = (55)
( τ p s + 1) ( τcs 2 s + 1)
Remark 7: It should be noted that (55) can be further simplified to the FOPDT form as
given in (35). Hence, Kove = KpKld, τeff = τp + τ0.5τcs2 and θeff = θp + θs + 0.5τcs2. After the
final model reduction, one can then use the empirical tuning rules given in Table 1 to
obtain the tuning values for the primary controller, i.e., similar to the case A.1 in which P
controller is used as a slave controller.
178 J. Nandong

3.2 Case B: unstable primary process

For this case, consider a process given by the FODUP form as follows
K p e−θ p s
G p (s) = (56)
τ p s −1

The secondary process is assumed to be an open-loop stable process. Also, for the
secondary P or PI controller tuning, the tuning relations developed in the previous (cases
A.1 and A.2) are still applicable to case B. Hence, in this section, we only focus on the
development of tuning relations for the primary PID controller, i.e., the master controller
design. Similar to the Case A, we further sub-divide the case B into two categories: case
B.1 and case B.2 which correspond to using a P and PI controllers respectively, in the
secondary loop.

3.2.1 Case B.1: primary PID controller design

Here, a P-only controller is used as the slave controller, which results in the augmented
primary process as given by

K p K so e −( θs + θ p ) s
Gove ( s ) = (57)
( τ p s − 1) ( τcs s + 1)
where τcs denotes the closed-loop time constant of the secondary process, i.e., as defined
in (31).
Remark 8: The model in (57) is often called the second-order dead time unstable process
(SODUP). With respect to this SODUP form, direct use of the Skogestad
half-rule to reduce (57) to the FODUP form is not advisable – the approximated model
can lead to an unsatisfactory closed-loop performance/robustness.
In the present work, we propose a general form of model reduction to reduce the SODUP
to the FODUP form where the latter can be used for controller tuning purposes.
Analogous to the FOPDT form in (35) for stable process, the desired (FODUP) form for
unstable process is given by
K ove e−θeff s
Gove ( s) = (58)
τ eff s − 1

It is worth highlighting that, we propose some general relations for the effective dead
time and time constant calculations:
θeff = θs + θ p + hd τ cs (59)

τ eff = τ p + ht τ cs (60)

Note that, the parameters hd and ht in (59) and (60) respectively describe the contribution
of the stable mode in (57) to the delay and lag time components in (58). Interestingly, for
a case where hd = ht = 0.5, the rule is similar to the Skogestad half-rule for a stable
process, which is not applicable to an unstable process.
 Analytical tuning method for cascade control system 179

Remark 9: In the case of unstable primary process, both hd and ht have to be determined
appropriately. This is important if one seeks to obtain good controller tuning values based
on the approximated model in (58).
In the following section, we present a simple way to determine appropriate values for hd
and ht such that, the controller designed based on the FODUP form (58) is robust enough
against modelling uncertainties.

3.2.2 Construction of relationship between hd and ht

Note that, the work in De Paor (1985) shows that a well-tuned P-only controller can only
stabilise a FODUP system if and only if the dead-time (θ) to time constant (τ) ratio is less
than unity, i.e.:
θ
<1 (61)
τ
Meanwhile, for the SODUP form, i.e., G = Kexp(–θs) / (τs – 1)(as + 1) it is known that a
P-only controller can stabilise the process if and only if (Huang and Chen, 1997):
θ a
+ <1 (62)
τ τ
It follows from (61) and (62) that, if the SODUP is to be reduced to the FODUP form, the
upper limit of the original SODUP and the approximated FODUP should be given as
follows
θeff θ a
= + <1 (63)
τ eff τ τ

Note that, θeff and τeff are as expressed in (59) and (60) respectively.
Following the assumed relation in (63), one can then readily obtain the following
relation
τ ⎛ θ + hd a ⎞
ht = ⎜ − 1⎟ (64)
a⎝ θ+a ⎠
Remark 10: Based on an extensive numerical study, we suggest to set hd = 2/3, i.e.,
henceforth, referred to as the Two-Third Rule of SODUP approximation. From (64), ht
can be calculated after specifying the value of hd = 2/3.
After approximating the SODUP (57) to FODUP (58) via the proposed Two-Third Rule,
one can then use the MSC tuning relations given in (13) to (15) and (17) to tune the
primary controller.
Remark 11: In order to simplify the primary PID controller tuning task via the MSC
tuning relations, a set of empirical tuning rules based on the FODUP system given in
(Nandong, 2015) can be adopted. These empirical tuning rules are given in Table 2.
180 J. Nandong

Table 2 Empirical tuning rules for FODUP system

Range Parameters
λ0 = 0.1, γ = 1, aI = 1, ad = 0.05
λ1 = –4.751(θ / τ) + 3.169
0 < θ / τ ≤ 0.25
τrl = τ / λ1
τrg = (τ / λ1) (1 + 5 θ / τ)
λ0 = 0.05, aI = 1, af = 0.03
λ1 = –0.933(θ / τ) + 2.236
0.25 < θ / τ ≤ 0.4 γ = [(τ + θ) / τ]2
τrl= τ / λ1
τrg = (τ + θ) (1 + 1.8 θ / τ)
λ0 = 0.05, aI = 1.2, af = 0.03
λ1 = –1.2(θ / τ) + 2.344
0.4 < θ / τ ≤ 0.5 γ = [(τ + θ) / τ]2.4
τrl = τ / λ1
τrg = (τ + θ) (1 + 2.4 θ / τ)
λ0 = 0.05, aI = 1.2, af = 0.03
λ1 = –1.082(θ / τ) + 2.287
0.5 < θ / τ ≤ 0.6 γ = [(τ + θ) / τ]2.7
τrl = τ / λ1
τrg = (τ + θ) (1 + 2.7 θ / τ)

Bear in mind that, in order to effectively use the PID tuning rules given in Table 2, one
needs to use the modified relations for the reset time and filter time constant which are
suggested to be as follows.
τ I = aI ( γτ eff + 0.5θeff ) (65)

⎛ θeff ⎞
τ f = af ⎜ ⎟ (66)
⎝ 2 λ1 ⎠
Also note that, to avoid an excessive overshoot following a setpoint change, it is
recommended to use a setpoint pre-filter of a lead-lag form
τ rl s + 1
Fr ( s ) = (67)
τ rg s + 1

where τrl and τrg denote the lead and lag time constants, respectively; both can be
calculated using the empirical equations given in Table 2.
Further note that, the summary of all MSC-PID tuning relations for cases A and B are
presented in Tables 3 and 4, respectively.
 Analytical tuning method for cascade control system 181

Table 3 The cascade controller tuning relations for case A

K p exp ( −θ p s ) K s exp ( −θs s )

G p ( s) = Gs ( s ) =
τ ps + 1 τss + 1

Case A.1 Case A.2

Secondary controller: Gc2 = Kc2 Secondary controller:
τs ⎛ 1 ⎞
Kc = ; r = 1.5 Gc 2 ( s ) = K c 2 ⎜1 + ⎟
rK pθs ⎝ τ Is s ⎠

τs
Kc = ; r = 1.6
rK pθs

τ Is = η ( τ s + θs ) ; η = 0.8

⎛ 1 ⎞⎛ 1 ⎞
Primary controller: Gc1 ( s ) = K c ⎜1 + + τ D s ⎟⎜ ⎟
⎝ τ I s ⎠⎝ τ f s + 1⎠

⎛ ( λ − 1) ( λ1 − 1) ⎞ ⎛ 2γτ eff + θeff ⎞ ⎛ (2τ eff − θeff ) 2 ⎞

Kc = ⎜ 0 ⎟⎜ ⎟⎜ ⎟
⎝ γλ1 ⎠ ⎝ 2 τ eff + θeff ⎠⎝ 4 τ eff θeff K ove ⎠

τ I = γτ eff + 0.5θeff τ D = γθeff τ eff ( 2γτ eff + θeff ) τ f = θeff ( 2 λ1 )

$ ( τI ε ) s + 1
Setpoint pre-filter (optional): Fr = ; ε = [3 4]
τI s + 1
#
MSC tuning parameters: [ λ0 λ1 γ ]

Case A.1 Case A.2

⎛
K ove = K p ⎜
1 ⎞ K p τ Is ( ξcs − ξcs2 − 1 )
⎟ K ove =
⎝ r ( θ s τ s ) + 1 ⎠ τ Isθs (r − 1)

⎛ θ (r − 1) ⎞ ⎛ τ Isθs (r − 1) ⎞
τ eff = τ p + 0.5 ⎜ s τ eff = τ p + 0.5 ⎜
⎟
⎝ r ( θs τ s ) + 1 ⎠ ⎜ ⎟⎟
⎝ ξ cs + ξ cs − 1 ⎠
2

⎛ θ (r − 1) ⎞ ⎛ τ Isθs (r − 1) ⎞
θeff = θ p + θs + 0.5 ⎜ s ⎟ θeff = θ p + θs + 0.5 ⎜ ⎟⎟
⎝ r ( θs τ s ) + 1 ⎠ ⎜
⎝ ξ cs − ξ cs − 1 ⎠
2

rτ Isθs + τ s ( τ Is − θs )
ξcs =
2τ s τ Isθs (r − 1)

Notes: $The filter should be used in case of an excessive overshoot occurs, otherwise, it is
not needed.
#
The MSC tuning parameters are calculated using the empirical equations in
Table 1.
182 J. Nandong

Table 4 The cascade controller tuning relations for case B

K p exp ( −θ p s ) K s exp ( −θs s )

G p ( s) = Gs ( s ) =
τ ps −1 τss + 1

⎛ 1 ⎞⎛ 1 ⎞
Primary controller: Gc1 ( s ) = K c ⎜ 1 + + τ D s ⎟⎜ ⎟
⎝ τI s ⎠⎝ τ f s + 1⎠

⎞ ⎛ ( 2 τ eff − θeff ) ⎞
2
⎛ ( λ + 1) ( λ1 − 1) ⎞ ⎛ 2γτ eff + θeff θ
Kc = ⎜ 0 ⎟⎜ ⎟ ⎜⎜ 4 τ θ K ⎟ ; τ eff > eff
⎟
⎝ γλ1 ⎠ ⎝ 2 τ eff − θeff ⎠⎝ eff eff ove ⎠ 2

τ I = aI ( γτ eff + 0.5θeff ) τ D = γθeff τ eff ( 2γτ eff + θeff ) ⎛ θeff ⎞

τ f = af ⎜ ⎟
⎝ 2 λ1 ⎠

$$ τ rl s + 1
Setpoint pre-filter (optional): Fr = ; τ rl < τ rg
τ rg s + 1
#
MSC tuning parameters: [ λ0 λ1 γ ] $$
Modification factors: [ aI af ]
‡ ‡
Case B.1 Case B.2
⎛
K ove = K p ⎜
1 ⎞ K p τ Is ( ξcs − ξcs2 − 1 )
⎟ K ove =
⎝ r ( θ s τ s ) + 1 ⎠ τ Isθs (r − 1)

τ eff = τ p + ht τ cs τ eff = τ p + ht τ cs 2

2 2
θeff = θ p + θs + τ cs θeff = θ p + θs + τ cs 2
3 3
⎛ θ (r − 1) ⎞ ⎛ τ Isθs (r − 1) ⎞
τ cs = ⎜ s ⎟ τ cs 2 = ⎜ ⎟⎟
⎝ r ( θs τ s ) + 1 ⎠ ⎜
⎝ ξ cs + ξcs − 1 ⎠
2

Two-third rule: Two-third rule:

τ p ⎛ θ p + θs + ( 2τ cs 3) ⎞ τ p ⎛ θ p + θs + ( 2 τ cs 2 3) ⎞
ht = ⎜ − 1⎟ ht = ⎜ − 1⎟
τ cs ⎝ θ p + θs + τ cs ⎠ τ cs 2 ⎝ θ p + θs + τ cs 2 ⎠
Notes: $$Table 2 give the empirical equations to calculate τrl and τrg, MSC tuning
parameters and modification factors.
‡
The secondary controller is tuned in the same way as in case A.

4 Illustrative examples

To demonstrate the effectiveness of the proposed cascade controller tuning method, we

present four numerical examples which include two stable and two unstable primary
processes. For all examples considered, the empirical tuning rules given in Tables 1 or 2
are used to simplify the overall cascade controller tuning task. Furthermore, the
simulation study is performed using MATLAB/Simulink.
 Analytical tuning method for cascade control system 183

4.1 Example 1 – case A

Consider an example in Jeng (2014) where the primary process is
10(−5s + 1)e −5 s
G p (s) = (68)
(30 s + 1)3 (10s + 1) 2

and the secondary process is

3e −3s
Gs ( s ) = (69)
13.3s + 1
For comparing the performance robustness of different cascade control systems, we
consider perturbed primary and secondary process models given as in (70) and (71)
respectively:
15(−10s + 1) 2 e−10 s
G p~ ( s) = (70)
(30 s + 1) 2 (10s + 1)
2

3.5e−4 s
Gs~ ( s ) = (71)
11s + 1
After approximating the nominal primary process model (68) via the Skogestad half-rule,
the simplified FOPDT model is obtained for the primary process
10e−75 s
G p ( s) = (72)
45s + 1
It should be noted that, the approximated FOPDT model (72) is used in the cascade
controller tuning. In this example, case A.2 tuning procedure is chosen, i.e., the slave
controller takes the form of PI controller.
The secondary (PI) controller is obtained by using the recommended default settings:
r = 1.6 and η = 0.8, which leads to (73)
⎛ 1 ⎞
Gc 2 ( s ) = 0.821⎜1 + ⎟ (73)
⎝ 9.78s ⎠
The controller (73) gives Kld = 1.27 and τcs2 = 3 leading to and for the augmented primary
process. The primary controller is obtained using the rules given in Table 1, which results
in the controller
⎛ 1 ⎞⎛ 1 ⎞
Gc1 ( s ) = 0.0726 ⎜1 + + 18.1s ⎟⎜ ⎟ (74)
⎝ 81.9s ⎠⎝ 1.35s + 1 ⎠
The primary controller in (74) gives the overall GM, PM, Ms (maximum peak of
sensitivity function) of 9.6 dB, 66o and 1.34, respectively.
Note that, the controller settings for another method can be referred directly to Jeng
(2014). Here, we compare the performance of the proposed cascade control scheme with
that of Jeng (2014) on the basis of sequential step changes of 1 unit in the setpoint (R1),
–0.5 units in the input disturbance (Di¬) and –1 unit in the output disturbance (Do).
Figure 4 shows the closed-loop responses under the nominal condition while Figure 5
184 J. Nandong

shows the responses under the perturbed condition given in (70) and (71). The integral
absolute values (IAEs) are also presented in Figures 4 and 5, which clearly demonstrate
that the proposed cascade tuning method produces superior performance to that of Jeng
(2014).

Figure 4 Closed-loop responses under the nominal condition (Example 1) (see online version
for colours)

Figure 5 Closed-loop responses under the perturbed condition (Example 1) (see online version
for colours)

4.2 Example 2 – case A

In the previous example 1, we have compared the proposed method with another method
based on the standard cascade control structure, and obviously the proposed method has
produced improved performance robustness. In this example 2, we compare the
effectiveness of the proposed method with that of a more complex 2-degree-of-freedom
(2-DoF) cascade control scheme proposed in Alfaro et al. (2009). In this example, we
adopt one of the examples reported in Alfaro et al. (2009) where the primary and
secondary processes are given as in (75) and (76):
 Analytical tuning method for cascade control system 185

e−1.5 s
G p (s) = (75)
5s + 1

e−0.3s
Gs ( s ) = (76)
s +1

To evaluate the control system performance robustness, we assume the following

perturbed models in (77) and (78):

1.1(−0.5s + 1)e−1.3s
G p~ ( s ) = (77)
(6 s + 1)( s + 1)

1.1e −0.6 s
Gs~ ( s ) = (78)
(1.5s + 1)(0.2s + 1)

By applying the proposed method (Case A.2), the secondary and primary controllers are
obtained as follows.

⎛ 1 ⎞
Gc 2 ( s ) = 2.085 ⎜1 + ⎟ (79)
⎝ 1.04s ⎠

⎛ 1 ⎞⎛ 1 ⎞
Gc1 ( s ) = 1.715 ⎜1 + + 0.766 s ⎟⎜ ⎟ (80)
⎝ 5.014s ⎠⎝ 0.047 s + 1 ⎠

For this example, we found that the overshot following the setpoint change is fairly high.
Hence, a setpoint pre-filter is used given as

1.67 s + 1
Fr = (81)
5.01s + 1

We compare the performance of the proposed cascade control system with that of the
2-DoF cascade control system of Alfaro et al. (2009). Note that, the latter control scheme
has four controllers. The performance of the different control schemes are compared on
the basis of sequential step changes in R1 Di and Do with magnitudes of 1 unit, –5 units
and –5 units, respectively. Figures 6 and 7 show the closed-loop responses of the
different control systems under the nominal and perturbed conditions, respectively. As in
the previous example 1, the proposed control scheme substantially outperforms the more
complex cascade control scheme of Alfaro et al. (2009). This means that a more complex
cascade control scheme will not necessarily produce a better performance than that of the
standard cascade control scheme. A more important factor contributing to a good cascade
control performance lies in an effective controller tunings where the proposed method
basically helps to meet this objective.
186 J. Nandong

Figure 6 Closed-loop responses under the nominal condition (Example 2) (see online version
for colours)

Figure 7 Closed-loop responses under the perturbed condition (Example 2) (see online version
for colours)

4.3 Example 3 – case B

In this example, the processes reported in Uma et al. (2009) where the primary process is
unstable while the secondary process is stable are used as the case study to demonstrate
the effectiveness of the proposed scheme. The primary and secondary processes are given
as follows
e−4 s
G p (s) = (82)
20 s − 1
2e−2 s
Gs ( s ) = (83)
20 s + 1
 Analytical tuning method for cascade control system 187

Also consider the perturbed models represented by

(−0.5s + 1)e−3s
G p~ ( s ) = (84)
(25s − 1)(4 s + 1)

2e− s
Gs~ ( s ) = (85)
(18s + 1)(2 s + 1)

For this example, the case B.1 tuning method is used which leads to the controllers and
setpoint pre-filter given in (86):

⎡ ⎛ 1 ⎞⎤
⎢ 3.05 ⎜1 + + 3.0 s ⎟ ⎥
⎡ c1 ⎤
G ( s ) ⎝ 45.44 s ⎠⎥
⎢G ( s ) ⎥ = ⎢
⎢ c2
⎥ ⎢ 3.33 ⎥ (86)
⎢ ⎥
⎣⎢ Fr ( s ) ⎦⎥ ⎢ 10s + 1 ⎥
⎣⎢ 41.65s + 1 ⎦⎥

The performance of the proposed cascade control system (86) is compared with that of
the modified Smith predictor and generalised cascade multi-scale control (CMSC)
schemes proposed by Uma et al. (2009) and Nandong and Zang (2014b), respectively.
Note that, the modified Smith predictor-based cascade control structure is the most
complex (non-standard form which is difficult to implement industrially) and involve
complicated controller design procedures. Meanwhile, the generalised CMSC scheme is
only slightly complex relative to the standard cascade control structure but the existing
design procedure is quite general and hence might be quite difficult to apply industrially.
Hence, the proposed method aims to resolve such complication through the use of
empirical tuning rules which requires minimum process control knowledge for its
implementation.
Sequential step changes in R1, Di and Do are applied to the controlled process with
respective magnitudes of 1 unit, 2 units and 2 units. Figures 8 and 9 show the closed-loop
responses for the three different control schemes under the nominal and perturbed
conditions, respectively. The generalised CMSC provides the best nominal performance,
followed by the proposed scheme and the worst performance is given by that the complex
modified Smith predictor scheme of Uma et al. (2009). Interestingly, under the perturbed
condition the proposed scheme demonstrates the best performance robustness. The
modified Smith predictor scheme provides a very poor performance even though it has a
more complex structure. Again, this example shows that a complex cascade control
structure could not ensure excellent performance/robustness compared with the simple
standard cascade control scheme commonly used in industries. It is very important for the
cascaded controllers to be tuned correctly in order for the control system to give good
performance robustness.
188 J. Nandong

Figure 8 Closed-loop responses under the nominal condition (Example 3) (see online version
for colours)

Figure 9 Closed-loop responses under the perturbed condition (Example 3) (see online version
for colours)

4.4 Example 4 – case B

For this example, the nominal primary and secondary processes are given as
e−3s
G p (s) = (87)
10s − 1
e− s
Gs ( s ) = (88)
5s + 1
Meanwhile, the perturbed models are as follows
 Analytical tuning method for cascade control system 189

1.1e−3.2 s
G p~ ( s ) =
(9.5s − 1)(0.1s + 1)

0.9e−1.1s
Gs~ ( s ) =
(5s + 1)(0.1s + 1)

Both case B.1 and case B.2 tuning procedures are applied. Based on the case B.1, the
following PID/P cascade control system is obtained:

⎡ ⎛ 1 ⎞⎛ 1 ⎞⎤
2.91⎜1 + + 1.95s ⎟ ⎜ ⎟⎥
⎡Gc1 ( s ) ⎤ ⎢ ⎝ 30.43 s ⎠ ⎝ 0.035 s + 1 ⎠⎥
⎢ G ⎥= ⎢
⎢ c2 ⎥ ⎢ 3.33 ⎥ (89)
⎢⎣ Fr ⎥⎦ ⎢ 5.34 s + 1 ⎥
⎢ ⎥
⎣⎢ 28.66 s + 1 ⎦⎥
Meanwhile, based on the case B.2, a PID/PI scheme is:

⎡ ⎛ 1 ⎞⎛ 1 ⎞⎤
⎢ 2.13 ⎜⎝1 + 31.13s + 2.02 s ⎟⎜ ⎟
⎠⎝ 0.037 s + 1 ⎠ ⎥⎥
⎡Gc1 ( s ) ⎤ ⎢
⎢ G ⎥=⎢ ⎛
3.13 ⎜1 +
1 ⎞ ⎥
(90)
⎢ c2 ⎥ ⎢ ⎟ ⎥
⎝ 4.8s ⎠
⎢⎣ Fr ⎥⎦ ⎢ ⎥
⎢ 5.34s + 1 ⎥
⎢⎣ 29.41s + 1 ⎥⎦

For comparison purpose, we also use the well-cited cascade control design method
proposed by Lee et al. (2002). The following PID/PID control system is produced
through the method:

⎡ ⎛ 1 ⎞⎛ 1 ⎞⎤
⎢ 2.22 ⎜⎝1 + 40.75s + 2.31s ⎟⎠ ⎜⎝ 2.38s + 1 ⎟⎠ ⎥
⎡Gc1 ( s ) ⎤ ⎢ ⎥
⎢ G ⎥=⎢ ⎛
4.41 ⎜1 +
1
+ 0.31s ⎟
⎞ ⎥
(91)
⎢ c2 ⎥ ⎢ ⎝ 2.73 s ⎠
⎥
⎢⎣ Fr ⎥⎦ ⎢ ⎥
⎢ 1 ⎥
⎢⎣ 38.34 s + 1 ⎥⎦

The control performances of the three different schemes are compared on the basis of
sequential step changes of 1 unit in R1, –2 units in Di and –2 units in Do. The
closed-loop responses under the nominal condition are shown in Figure 10. Notice that,
the proposed PID/PI scheme provide improved performance over the proposed PID/P
scheme. Interestingly, both of the proposed schemes outperform the cascade PID/PID
scheme designed via the Lee et al. (2002). Figure 11 displays the responses under the
perturbed condition. In this situation, the proposed PID/P scheme shows a more robust
performance than the proposed PID/PI scheme. The cascade control system designed via
the Lee et al. method produced very oscillatory response – not desirable in practice.
190 J. Nandong

Figure 10 Closed-loop responses under the nominal condition (Example 4) (see online version
for colours)

Figure 11 Closed-loop responses under the perturbed condition (Example 4) (see online version
for colours)

5 Conclusions

A new systematic method for tuning cascade PID/PI or PID/P control system has been
presented. It has been shown that to achieve improved performance robustness for
unstable time-delay processes, it is better to use a simple P controller in the secondary
loop, rather than a more complex PI or PID controller. But for improved nominal
performance, one might consider using the PI controller for the secondary loop. One
conclusion that can be drawn from the numerical studies is that, a good tuning method is
essential for an effective cascade control system. Some of the modified cascade control
structures tend to be too complicated and might not be readily realisable in industrial
environment. Complicated modified cascade structures may result in only small
improvement over the standard cascade control. The proposed method has been shown to
be able to provide an effective way to tuning standard cascade control systems which
 Analytical tuning method for cascade control system 191

produce improved performance. The proposed method can even produce better
performance than that of some complex modified cascade control structures. In future
works, we will further develop simple tuning methods based on the standard cascade
control structure for more complex high order processes, which have both stable and
unstable modes.

References
Alfaro, V.M., Vilanova, R. and Arrieta, O. (2009) ‘Robust tuning of two-degree-of-freedom
(2-DoF) PI/PID based cascade control systems’, J. Process Control, December, Vol. 19,
No. 10, pp.1658–1670.
Azar, A.T. and Serrano, F.E. (2014) ‘Robust IMC-PID tuning for cascade control systems with gain
and phase margin specifications’, Neural Comput. and Applic., October, Vol. 25, No. 5,
pp.983–995.
Caetano, R., Lemos, M.A., Lemos, F. and Freire, F. (2014) ‘Modeling and control of an exothermic
reaction’, Chem. Eng. J., February, Vol. 238, No. 15, pp.93–99.
Chen, J. and Chang, Y.H. (2013) ‘Performance design of image-oxygen based cascade control
loops for boiler combustion processes’, Ind. Eng. Che. Res., December, Vol. 52, No. 6,
pp.2368–2378.
Chen, X., Li, J., Yang, J. and Li, S. (2013) ‘A disturbance observer enhanced composite cascade
control with experimental studies’, Int. J. Control, Automation, and Systems, June, Vol. 11,
No. 3, pp.555–562.
De Paor, A.M. (1985) ‘A modified Smith predictor and controller for unstable process with time
delay’, Int. J. Control, Vol. 41, No. 4, pp.1025–1036.
Garcia, P., Santos, T., Normey-Rico, J.E. and Albertos, P. (2010) ‘Smith predictor-based control
schemes for dead-time unstable cascade processes’, Ind. Eng. Chem. Res., October, Vol. 49,
No. 22, pp.11471–11481.
Hasanien, H.M. and Muryeen, S.M. (2013) ‘A Taguchi approach for optimum design of
proportional-integral-derivative controllers in cascaded control scheme’, IEEE Transactions
on Power Systems, May, Vol. 28, No. 2, pp.1636–1644.
Ho, M.T., Datta, A. and Bhattacharyya, S.P. (1998) ‘An elementary derivation of the
Routh-Hurwitz criterion’, IEEE Transactions Automatic Control, March, Vol. 43, No. 3,
pp.405–409.
Huang, H.P. and Chen, C.C. (1997) ‘Control-system synthesis for open-loop unstable process with
time delay’, IEE Proc. Control Theory Appl., July, Vol. 144, No. 4, pp.334–346.
Jeng, J.C. (2014) ‘Simultaneous closed-loop tuning of cascade controllers based directly on
set-point step-response data’, J. Process Control, May, Vol. 24, No. 5, pp.652–662.
Kano, M., Showchaiya, N., Hasebe, S. and Hashimoto, I. (2003) ‘Inferential control of distillation
compositions: selection of model and control configuration’, Control Eng. Pract., August,
Vol. 11, No. 8, pp.927–933.
Kaya, I., Tan, N. and Atherton, D.P. (2007) ‘Improved cascade control structure for enhanced
performance’, J. Process Control, January, Vol. 17, No. 1, pp.3–16.
Kwon, O., Jung, W. and Heo, H. (2013) ‘Steam temperature controller with LS-SVR-based
predictor and PID gain scheduler in thermal power plant’, J. Mech. Sci. Technol., February,
Vol. 27, No. 2, pp.557–565.
Lee, Y., Oh, S. and Park, S. (2002) ‘Enhanced control with a general cascade control structure’,
Ind. Eng. Chem. Res., May, Vol. 41, No. 11, pp.2679–2688.
Legweel, K.M., Lazic, D.V., Ristanovic, M.R. and Sajic, J.V. (2014) ‘The performance of
PIP-cascade controller in HVAC system’, Therm. Sci., Vol. 18, No. Suppl. 1, pp.213–220.
Marlin, T.E. (1995) Process Control, McGraw-Hill, New York.
192 J. Nandong

Nandong, J. (2015) ‘Guaranteed stable PID controller tuning rules for first-order dead-time
unstable processes’, IEEE 10th Conference on Industrial Electronics and Applications
(ICIEA), 15–17 June 2015, pp.1443–1448.
Nandong, J. and Zang, Z. (2013a) ‘Novel multiscale control scheme for nonminimum-phase
processes’, Ind. Eng. Chem. Res., May, Vol. 52, No. 24, pp.8248–8259.
Nandong, J. and Zang, Z. (2013b) ‘High-performance multi-scale control scheme for stable,
integrating and unstable time-delay processes’, J. Process Control, November, Vol. 23,
No. 10, pp.1333–1343.
Nandong, J. and Zang, Z. (2014a) ‘Multi-loop design of multi-scale controllers for multivariable
processes’, J. Process Control, May, Vol. 24, No. 5, pp.600–612.
Nandong, J. and Zang, Z. (2014b) ‘Generalized multi-scale control scheme for cascade processes
with time-delays’, J. Process Control, July, Vol. 24, No. 7, pp.1057–1067.
Nedic, N., Prsic, D., Dubonjic, L., Stojanovic, V. and Djordjevic, V. (2014) ‘Optimal cascade
hydraulic control for a parallel robot platform by PSO’, Int. J. Adv. Manuf. Technol., May,
Vol. 72, Nos. 5–8, pp.1085–1098.
Skogestad, S. (2003) ‘Simple analytic rules for model reduction and PID controller tuning’, J.
Process Control, June, Vol. 23, No. 4, pp.291–309.
Skogestad, S. and Postlethwaite, I. (2007) Multivariable Feedback Control: Analysis and Design,
Wiley, New York.
Smith, C.A. and Corripio, A.B. (1985) Principles and Practice of Automatic Process Control,
Wiley, New York.
Tan, W. and Liu, J. (2005) ‘Robust analysis and PID tuning of cascade control systems’, Chem.
Eng. Comm., January, Vol. 192, No. 9, pp.1204–1220.
Uma, S., Chidambaram, M. and Rao, A.S. (2009) ‘Enhanced control of unstable cascade processes
with time delays using modified Smith predictor’, Ind. Eng. Che. Res., February, Vol. 48,
No. 6, pp.3098–3111.
Zhong, J., Fan, J., Zhu, Y., Zhao, J. and Zhai, W. (2014) ‘One nonlinear PID control to improve the
control performance of manipulator actuated by a pneumatic muscle actuator’, Advances in
Mechanical Engineering, May, Vol. 6, pp.1–19.