See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/222402204

Optimal tuning of PID controllers for FOPTD, SOPTD and SOPTD with lead
processes

Article in Chemical Engineering and Processing - Process Intensification · February 2008

DOI: 10.1016/j.cep.2006.11.013

CITATIONS READS

92 5,257

3 authors, including:

A. Elkamel
University of Waterloo
654 PUBLICATIONS 13,489 CITATIONS

SEE PROFILE

All content following this page was uploaded by A. Elkamel on 05 October 2018.

The user has requested enhancement of the downloaded file.

 Available online at www.sciencedirect.com

Chemical Engineering and Processing 47 (2008) 251–264

Optimal tuning of PID controllers for FOPTD,

SOPTD and SOPTD with lead processes
C.R. Madhuranthakam, A. Elkamel ∗ , H. Budman
Department of Chemical Engineering, University of Waterloo, 200 University Avenue West, Waterloo, Ontario N2L 3G1, Canada
Received 21 October 2006; received in revised form 27 November 2006; accepted 28 November 2006
Available online 3 December 2006

Abstract
This paper presents the synthesis and analysis of optimal tuning of proportional integral derivative (PID) parameters for different process systems:
first order plus time delay (FOPTD), second order plus time delay (SOPTD) and second order plus time delay with lead (SOPTDLD). This work
involved optimization of the PID control parameters to achieve the minimization of the integral absolute error (IAE). A set of new and generalized
tuning correlations relating the controller parameters to the process parameters was obtained for step changes in set point and load separately.
Simulation results showed that the use of the proposed tuning correlations results in superior closed loop performance compared to other tuning
techniques previously proposed in the literature.
© 2007 Elsevier B.V. All rights reserved.

Keywords: PID control; FOPTD; SOPTD; SOPTD with lead; Multi-objective optimization; IAE; Ziegler–Nichols method; Cohen–Coon method

1. Introduction were proposed for first order plus time delay (FOPTD) system,
as they can explain the behavior of a wide range of processes.
Proportional integral derivative (PID) controllers are still Thus, suitable tuning rules for PID controllers are needed for
widely used today to control many processes because of their processes represented by transfer functions other than FOPDT.
long history of proven operation in addition to the fact that they Also, model based tuning methods were found attractive for
are well understood by many operational, technical and mainte- practitioners because they have only one tuning parameter. How-
nance personnel. Furthermore, an industrial PID controller has ever, many control systems do not provide Smith Predictor or
many extensions that make it a practical tool for operating a Internal Model Control functionality [8]. At the same time, many
chemical process. Many PID controller tuning methods have distributed parameter systems where a state variable is a func-
been proposed in the literature. For example, Ziegler–Nichols tion of more than one independent variable encountered in the
tuning [1], Cohen–Coon tuning [2], Direct Synthesis Method process plants such as packed beds, trickle beds with recycle,
[3], Internal Model Control [4], tuning rules based on the mini- continuously stirred tank reactors (CSTRs) in series, etc., cannot
mization of different error criteria [5], and neural networks based be modeled as FOPTD and instead, are often empirically mod-
methodologies [6,7]. All these methods have their own advan- eled using second order plus time delay (SOPTD) and SOPTD
tages, disadvantages and limitations. Most of the tuning methods with lead (SOPTDLD). Many industrially proven techniques are
now available for fitting these models to plant data [9].
In this paper, a set of tuning correlations is obtained for three
Abbreviations: C–C, Cohen and Coon; CSTR, continuously stirred tank
different processes whose dynamics are modeled with FOPTD,
reactor; FOPTD, first order plus time delay; IAE, integral of absolute error; ISE, SOPTD and SOPTDLD. Most of the chemical processes can
integral of square error; ITAE, integral of time weighted absolute error; ITSE, be effectively modeled by one of these types of models. In this
integral of time weighted square error; PID, proportional integral derivative; work, controller tuning parameters that minimize integral abso-
P–M, proposed method; R/L, Rovira (1981)/Lopez et al. (1967); SOPTD, second lute error (IAE) are investigated. The resulting solutions are then
order plus time delay; SOPTDLD, second order plus time delay with lead; Z–N,
Ziegler–Nichols
summarized in the form of algebraic correlations that can be con-
∗ Corresponding author. Tel.: +1 519 888 4567; fax: +1 519 746 4979. veniently used by practitioners when applying PID control. In
E-mail address: [email protected] (A. Elkamel). the proposed method, the controller parameters obtained from

0255-2701/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2006.11.013
252 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264

fer functions for the processes are described by the following

equations:

FOPTD:
KP exp(−θs)
KP GP = (3)
τ1 s + 1
SOPTD:
Fig. 1. Schematic of the feedback control loop. KP exp(−θs)
KP G P = (4)
(τ1 s + 1)(τ2 s + 1)

Ziegler–Nichols’ method are used as the initial guess for the SOPTDLD:
optimization problem that deals with the minimization of the KP (τ3 s + 1) exp(−θs)
IAE in the output variable. K P GP = (5)
(τ1 s + 1)(τ2 s + 1)

2. Proposed tuning correlations The error e(t) is defined by Eq. (6) given below:

The ideal PID controller transfer function is defined by Eq. e(t) = R(t) − Y (t) (6)
(1):
  Ciancone and Marlin [13] have recently made an attempt at
1 developing simple tuning rules for FOPTD processes with the
KC GC = KC 1 + + τD s (1)
τI s following goals in mind: (1) minimization of the IAE, (2)
The above transfer function is not physically realizable since assumption of a +25% (correlated) change in the process model
no device can be constructed which truly differentiates an input parameters and (3) compliance with pre-specified limits on the
signal (Luyben and Luyben [10]). Also, an ideal PID controller variation of the manipulated variable. They proposed differ-
causes a derivative kick when used in the feedback control ent charts for dimensionless tuning constants in terms of the
system especially for a step change in the set point. Thus, an fraction dead time defined as θ/(θ + τ 1 ), with the above con-
implemental form of the PID controller that is based on aug- straints. Separate charts were obtained by these authors for both
menting Eq. (1) with a first order filter of the derivative term is set point and load changes. In their method, optimization was
used instead. This can be achieved either by a series or parallel done by continuously varying the controller parameters until the
configuration. Most often, the derivative term multiplied by the objective function was satisfied by the above constraints. Since
filter transfer function is used on the measured variable rather the objective function for minimizing all the constraints was
than on the controller error. This is done to avoid a derivative non-linear, they proposed to optimize each controller parameter
kick occurring following step changes in the set point. For the independently as per the following procedure:
simulations conducted in the present manuscript, a parallel form
of PID controller configuration as shown in Fig. 1 was used. The 1. With the integral time set to infinity and derivative time set
transfer function for the parallel form PID controller is given by to zero, they obtained an optimal proportional gain.
Eq. (2). 2. Using the optimal proportional gain obtained in step 1, they
    optimized the integral time followed by the optimum deriva-
τI s + 1 τD s tive time.
U(s) = KC e(s) − KC Y (s) (2)
τI s ατD s + 1
where U is the controller output, e the error and Y is the controlled In the present work, we attempt to develop simple and useful
variable as shown in Fig. 1. The “derivative filter factor” α value mathematical expressions that can be readily used for tuning
is generally selected in the interval [0.05, 0.2] (Shinskey [11] and an extensive set of processes and as described by the transfer
Luyben [12]) and most often it is preset to 0.1. In the current functions (3)–(5). In order to improve the convergence properties
work, a value of α = 0.1 is used for conducting the comparison of the involved optimization problems in search of the tuning
of the performance of the proposed tuning method for different parameters, the following steps are employed:
process transfer functions. Fig. 1 shows the conventional feed-
back control system used in the present study. In Fig. 1, KC GC,Ser 1. For a unit step change in set point or load, with propor-
is the servo compensator (=KC (1 + (1/τ I s))), KP GP the process, tional control alone, the ultimate gain and ultimate period
KC GC the controller (=KC (1 + (1/τ I s) + (τ D s/(ατ D s + 1)))), R the (the gain for which the system is at the limit of stability and
set point, U the controller output, D the disturbance variable and the corresponding period of oscillation) for a particular pro-
Y is the controlled variable. cess transfer function (FOPTD, SOPTD and SOPTDLD) is
The combined dynamics of the process, final control ele- determined.
ment and the sensor is assumed to be conveniently represented 2. Ziegler–Nichols tuning rules are used to provide an initial
by FOPTD, SOPTD and SOPTDLD. The corresponding trans- guess for the non-linear optimization. The best controller
 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264 253

tuning parameters are then obtained with the objective of Table 1

minimizing IAE. First moments (time-scale) for different processes
3. Steps 1 and 2 are repeated for various different combinations Process (KP GP ) Time-scale Fraction dead time
for the process parameters, τ 1 , τ 2 , τ 3 , KP and θ. All process FOPTD θ + τ1 θ/(θ + τ 1 )
parameters are varied from 1 to 50. SOPTD θ + τ1 + τ2 θ/(θ + τ 1 + τ 2 )
4. The controller parameters thus obtained are made dimension- SOPTDLD θ + τ1 + τ2 − τ3 θ/(θ + τ 1 + τ 2 − τ 3 )
less by multiplying/dividing by appropriate scale factors.
5. Using regression techniques, simple and accurate correla-
tions are obtained for the controller parameters as functions following two steps:
of process parameters (in terms of τ 1 , τ 2 , τ 3 and θ) for the
corresponding three processes. Several sets of dimensionless 1. The independent variable for the tuning correlations was
groups are tried for the tuning rules and the ones that lead to selected as the fraction dead time, i.e. the ratio of the dead
the largest coefficients of correlations (R2 ) are retained in the time of the process and the sum of all the time constants
proposed tuning rules. including the dead time.
2. The dependent variables for each process for set point or load
The IAE is defined as: change (separately) was obtained by continuously iterating
 ∞ the different possible dimensionless groups as a function of
IAE = |e(t)| dt (7) τ I and τ D with different combinations of the process param-
0 eters (θ, τ 1 , τ 2 , τ 3 ) until a high degree of correlation, i.e. with
where e(t) is error defined according to Eq. (6). Though the R2 ≥ 0.95 between the independent and the dependent vari-
upper bound on time is infinity, in the simulations the integra- ables were obtained. The tuning correlations obtained from
tion was performed over a sufficiently long time as compared the simulations have an R2 value in the range of 0.97–0.99.
to the closed loop settling time, i.e. after the response reaches
a steady state. Minimum IAE was chosen as a suitable objec- To simplify the correlations further, a time scaling was used
tive function in the optimization because it is an indication of in our proposed method. Every process responds with a dif-
product variability. Minimization of the IAE generally produces ferent “speed”, which can be characterized by the time for a
intermediate responses between fast responses obtained from step response to achieve 63% of its final value. This is also
the minimization of the integral square error (ISE) and slow known as the first moment of the impulse response. For exam-
responses obtained from the minimization of the integral time ple, the method of moments was previously used to find the
weighted absolute error (ITAE) [14]. The process models were characteristic time of the process (Marlin [22]). In this work,
designed and simulated using SIMULINK and the dynamic opti- the first moment was used to normalize the dynamic responses
mization was performed using a combination of the SIMULINK with respect to time in the dimensional analysis. Hence, for the
results with MATLAB programs used to calculate the IAE and to processes considered in the study, the first moment or the time-
interface the results with the optimization routines. Separate tun- scale is shown in Table 1. Dividing all time constants by this
ing rules for set point change and load change were developed time-scale will normalize all the models by the same speed of
for the different processes. The optimized controller parame- response. The range of the variables KP , θ, τ 1 , τ 2 and τ 3 , used in
ters thus obtained were correlated with the process parameters the simulations was 0–50 (for KP it was 1–50, for the SOPTLD
according to the following equations: all parameters are varied from 1 to 20). It was found that there
was no clear relation between the fraction dead time and the
KC = f1 (KP , τ1 , τ2 , τ3 , θ) (8) scaled integral and derivative terms. Hence, the integral and
derivative parameters are scaled with respect to the process time
τI = f2 (KP , τ1 , τ2 , τ3 , θ) (9) constants by trial and error to get a simple algebraic correlation
τD = f3 (KP , τ1 , τ2 , τ3 , θ) (10) with a high R2 . The performances of different processes with the
proposed tuning methods were compared with the conventional
In Eqs. (8)–(10), the unit of θ, τ 1 , τ 2 , τ 3 is time and the unit tuning methods.
of KP depends on the plant input and output. In principle, the
number of possible independent dimensionless numbers that can 3. Simulation results and analysis
be formed for the purpose of representing the tuning rules is very
high. Analysis of control performance using the dimensionless 3.1. FOPTD
groups obtained by using the Pi-theorem for the FOPTD and
SOPTD (for load change in terms of charts) models can be found Simulations were performed by varying the dead time and
elsewhere (Sayeed and Mahdi [15] and Archibald and Tae-Won the time constant in the range from 1 to 50. The ratio of the dead
[16]). time to the time constant of the process (θ/τ 1 ) corresponding
In the present work, to avoid the multiple charts and formulae to the above settings varied from 0.1 to 2. The tuning parame-
for the controller parameters of different processes considered, ters thus obtained for a step change in set point and load were
inspection analysis was performed to get the dimensionless con- separately correlated with the fraction dead time. Fig. 2 shows
troller parameters. This inspection procedure is made of the the comparison of the results obtained from the simulations and
254 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264

Fig. 2. Dimensionless control parameters obtained from model vs. simulation results for first order plus time delay process (described by Eq. (3)): (a, c and e) for
set point change and (b, d and f) for load change.

the tuning model. Fig. 2(a, c and e) correspond, respectively, objective function, i.e. minimization of IAE. His tuning rules
to the proportional gain, integral time and derivative time rela- can be used with θ/τ 1 varying from 0.1 to 1. For processes with
tionship for a unit step change in the set point. Fig. 2(b, d and a ratio θ/τ 1 greater than one, the performance was not robust
f) are the correlations of the tuning parameters corresponding and gives an over stable performance. This is further discussed
to a unit step change in the load. The tuning models are shown in the following case studies section. Also, Lopez et al. [18]
in Table 2. Rovira [17] developed tuning rules with the same developed separate tuning rules for FOPTD processes for unit

Table 2
Proposed tuning relations for FOPTD process
Tuning parameter Set point change Load change

0.4967
 θ −1.2299 0.5249
 θ −1.2787
KC
KP θ+τ KP θ + τ1
 θ 1 −1.317   θ  
τI 0.6739θ θ −1.9167 + 2.1356
 θ + τ1 θ + τ1
 θ 2  θ    θ 2  θ 
τD τ1 1.138 + 0.1992 τ1 1.1321 + 0.1788
θ + τ1 θ + τ1 θ + τ1 θ + τ1
 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264 255

cesses were considered. The processes are represented by

GP1 (s), GP2 (s) and GP3 (s) and are described by Eqs. (11)–(13),
respectively.
1
KP1 GP1 (s) = exp(−s) (11)
5s + 1
4
KP2 GP2 (s) = exp(−10s) (12)
10s + 1
1
KP3 GP3 (s) = exp(−14s) (13)
7s + 1
In the process GP1 (s), the ratio of the dead time to time constant
is 0.2, for GP2 (s) the ratio is unity, while the last process GP3 (s)
has a ratio equal to 2.
Fig. 4 shows the closed loop response due to a set point change
and a load change for GP1 (s), GP2 (s) and GP3 (s) processes using
Fig. 3. The ratio between integral time and derivative time as a function of different tuning methods. It clearly shows that the Z–N and C–C
fraction dead time for FOPTD process. tuning parameters would give more oscillations before reaching
the steady state value thus resulting in a higher IAE. For pro-
cesses GP1 (s) and GP2 (s), the response obtained from Rovira
step change in the load variable. In their work, a similar restric-
[17] was similar to that of the response obtained by using the
tion was imposed on the range of θ/τ 1 ratio. Recently, Sayeed
proposed tuning rules. For load changes, similar results to the
and Mahdi [15] developed tuning rules for the FOPTD process
ones obtained with the method proposed by Lopez et al. [18]
with a derivative filter (used on the controller error) in the PID
were calculated, though the IAE’s obtained with the proposed
controller. But, the controller parameters obtained from their
method were less than those with Lopez method. However, in
rules give unstable response when the configuration given by
the case of large dead time ratio (i.e. with the process GP3 (s)),
Eq. (2) was used for the PID controller. In the present work,
the proposed method of tuning parameters gave a better response
tuning models were developed for θ/τ 1 varying from 0.1 to 2 for
than the other methods. The Z–N and C–C methods resulted in
the parallel form of the PID controller. The general conclusions
highly oscillatory responses and the other two methods did not
for the models obtained are similar to those of Ziegler–Nichols,
perform well too. The controller parameters obtained from dif-
Rovira [17] and Lopez et al. [18]. Thus, the controller gain is
ferent methods and the corresponding performances are shown
inversely proportional to the process gain. Also, the optimum
in Table 3.
controller gain decreases with an increase in the θ/τ 1 ratio. The
In Table 3, Tr is the rise time, Ts the settling time (based on
controller gain needed for a load change is slightly greater than
time required to reach 5% of the steady state value) and Os is the
the gain needed for a set point change which is evident from
overshoot for different servo responses. The overshoot was very
the coefficients in the corresponding models shown in Table 2.
low compared to the other methods when the proposed tuning
Furthermore, with an increase in the dead time the integral time
relations are used for designing the PID parameters. Also, the
decreases and the derivative time increases.
settling time obtained by using the proposed tuning relations was
The ratio of integral time to derivative time is very important
reduced by two to three times as compared to those obtained with
in assessing the relative importance of the derivative time with
the other tuning methods.
respect to integral time. Many PID controllers using a series
´
configuration, work with a τ I /τ D greater than 4 (Åstrom and 3.3. SOPTD
Hägglund [19]). Fig. 3 shows the τ I /τ D versus fraction dead
time for both set point and load changes. There is a signifi- Many process systems may be described by second order
cant variation in the τ I /τ D ratio at lower fraction dead time. processes with time delay. For example, two blending tanks in
The figure clearly shows that for 0.1 < θ/(θ + τ 1 ) < 0.8 the τ I /τ D series/parallel, two CSTRs in series with first order dynamics for
ratio obtained from the proposed tuning rules stays at a constant each CSTR, etc., can be properly represented by second order
value of 5 ± 2 for load changes. On the other hand, for set point plus time delay dynamics. Simulations were performed for crit-
change and for processes without a significant dead time the ically damped and over damped second order process plus time
ratio increases with a decrease in dead time. A similar trend for delay dynamics described by Eq. (4). Since, many combinations
set point change was found by Rovira [17] with τ I /τ D = 32 for for the dead time and the two process time constants were possi-
θ/τ 1 = 0.1. ble, the ratios τ 1 /θ and τ 1 /τ 2 were varied from 1 to 50. In all the
simulations, τ 1 was always greater than or equal to τ 2 and the
3.2. FOPTD case studies dead time was never greater than the sum of τ 1 and τ 2 . Though it
is always possible to approximate a SOPTD model by a FOPTD
To compare the performance of the proposed tuning method model and tuning methods based on the approximated FOPTD
with other conventional methods, three different FOPTD pro- model can be then used, Weigand and Kegerreis [20] proved that
256 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264

Fig. 4. Closed loop response for first order plus time delay process: (a, c and e) set point change and (b, d and f) load change for processes GP1 (s), GP2 (s) and GP3 (s),
respectively.

tuning methods based on full SOPTD models lead to superior results cannot be compared with the proposed method because
results. Archibald and Tae-Won [16] performed simulations for the controller in our study has parallel configuration and also the
finding the optimum tuning parameters for SOPTD processes dynamics of the disturbance variable D(t), is unit step change
with first order disturbance dynamics. The controller defined by rather than a first order exponential disturbance. Since, the opti-
Eq. (1) was used in their study. As discussed in Section 1, they mization was performed with the initial settings obtained from
have produced many charts based on different normalized pro- Z–N method, the results obtained from the proposed method
cess and control parameters with first order disturbance. Their were initially compared with Z–N method only. Also, there is
 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264 257

Table 3
Tuning parameters and performance characteristics for FOPTD
Process Method Set point change Load change

KC τI τD Tr Ts Os IAE KC τI τD IAE

KP1 GP1 (s) Z–N 6.39 2.31 0.37 1.7 17.6 1.83 4.4 6.39 2.31 0.37 0.61
C–C 6.92 2.27 0.35 1.7 35.7 1.97 7.7 6.92 2.27 0.35 0.88
R/L 4.4 7.0 0.4 2.2 4.7 1.1 1.8 6.32 1.71 0.39 0.66
P–M 4.5 7.0 0.37 2.1 4.7 1.12 1.8 5.26 1.93 0.7 0.42
KP2 GP2 (s) Z–N 0.43 19.37 3.1 16.4 78.4 1.43 23.49 0.43 19.37 3.1 49.83
C–C 0.4 18.09 3.08 16.8 66.3 1.38 21.02 0.4 18.09 3.08 46.16
R/L 0.27 16.39 3.48 20.6 20.6 1.05 15.96 0.36 11.39 4.82 39.73
P–M 0.28 16.7 4.33 20.0 36.8 1.04 16.28 0.34 11.4 3.56 37.59
KP3 GP3 (s) Z–N 1.15 24.0 3.84 22.2 163.4 1.38 37.7 1.15 24.0 3.84 27.3
C–C 0.88 16.78 3.88 24.5 66.8 1.19 23.7 0.88 16.78 3.88 19.3
R/L 0.59 14.58 4.59 54.2 54.2 – 24.7 0.76 13.40 7.42 20.87
P–M 0.8 15.6 4.28 25.6 53.7 1.11 22.3 0.82 11.91 4.19 16.05

Table 4
Proposed tuning relations for SOPTD process
Tuning parameter Set point change Load change

0.5723
 θ
−1.0409 0.6202
 θ
−0.9931
KC
KP θ + τ1 + τ2 KP θ + τ1 + τ2
 −1.6501   2   
τI τ1 θ θ θ
0.2476(θ + τ1 + τ2 ) (θ + τ1 ) 13.81 − 14.906 + 4.566
θ θ + τ1 + τ2 θ + τ1 + τ2 θ + τ1 + τ2
τD τ1
 θ
−1.4636  θ
−1.4849
0.0943 0.0921
(τ2 + θ)θ θ + τ1 + τ2 θ + τ1 + τ2

very scarce literature on optimization of controller tuning param- result is similar to the result obtained for FOPTD process. A
eters obtained for SOPTD and SOPTD with lead processes with ´
very similar trend was also observed by Åstrom and Hägguland
minimization of IAE as the objective function. Furthermore, [19] for lower fraction dead time.
the optimal controller tuning parameters vary with the objective
function used in optimization. Fig. 5 shows the validation of
the tuning models with the simulation data for set point and load 3.4. Case studies for SOPTD
changes. The corresponding tuning models are shown in Table 4.
The conclusions from these models were generally similar to The performance of three different processes with SOPTD
those obtained for FOPTD models except for the derivative time. dynamics were analyzed by using the controller parameters
Once again the controller gain is inversely proportional to the obtained from the proposed method and was compared with
process gain. With the increase in the dead time, the integral time parameters obtained from Z–N method. The process transfer
decreases. Contrary to the FOPTD process, with an increase in functions for the three processes (with small dead time, dead
the dead time the derivative time also increases. Fig. 6 shows the time equal to the dominant process time constant and dominant
integral time to derivative time ratio as a function of fraction dead dead time) are described by the following equations:
time for SOPTD processes. It clearly shows that a constant τ I /τ D
ratio was obtained for load change while for set point change 1
KP4 GP4 (s) = exp(−2s) (14)
at very small fraction dead time, the τ I /τ D ratio increases. This (15s + 1)(2s + 1)

Table 5
Tuning parameters and performance characteristics for SOPTD
Process Method Set point change Load change

KC τI τD Tr Ts Os IAE KC τI τD IAE

KP4 GP4 (s) Z–N 7.44 8.25 1.32 5.5 53.0 1.77 13.32 7.44 8.25 1.32 1.73
P–M 5.97 26.9 1.62 6.6 21.8 1.22 6.67 5.74 6.56 1.71 1.33
KP5 GP5 (s) Z–N 2.04 21 3.36 16.6 87.6 1.5 24.44 2.04 21 3.36 13.04
P–M 1.57 22.83 4.38 19.3 44.5 1.1 17.15 1.77 14.58 3.88 9.39
KP6 GP6 (s) Z–N 0.88 6.75 4.32 10.5 167.4 1.65 29.65 0.88 6.75 4.32 27.21
P–M 0.88 13.5 2.16 11.1 55.0 1.4 15.14 0.68 7.82 2.82 12.68
258 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264

Fig. 5. Dimensionless control parameters obtained from model vs. simulation results for SOPTD (described by Eq. (4)): (a, c and e) for set point change and (b, d
and f) for load change.

1
KP5 GP5 (s) = exp(−7s) (15)
(7s + 1)2
2
KP6 GP6 (s) = exp(−5s) (16)
(5s + 1)(3s + 1)
The closed loop response due to a unit step change in set point
and load, respectively, are shown in Fig. 7. The controller param-
eters used and the corresponding performance IAE calculations
are shown in Table 5. Fig. 7(e and f) shows the relative advantage
of the proposed methodology over the Z–N method. The large
oscillations resulting with the Z–N tuning will not only result in
a high IAE but also in highly oscillatory control actions that will
lead to wear of the control element. On the other hand, using the
tuning parameters obtained from the proposed method would
give a better performance with reduced oscillations. Table 5
shows that lower IAE values were obtained by using the pro-
posed method of tuning. Also, the overshoot and the settling Fig. 6. The ratio between integral time and derivative time as a function of
fraction dead time for SOPTD process.
time were smaller as compared to the Z–N tuning method.
 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264 259

Fig. 7. Closed loop response for SOPTD: (a, c and e) set point change for processes 1, 2 and 3, respectively, and (b, d and f) load change for processes 1, 2 and 3,
respectively.

3.5. SOPTDLD part of the product stream is a conventional technique practiced

by many industries. A scarce amount of literature is available on
Another frequently encountered process in chemical engi- tuning PID controllers for these types of processes. Tuning the
neering is the second order plus time delay with lead. The most PID controller for these types of processes is a challenging task
common examples of processes that can be represented by these as the lead time constant (τ 3 in Eq. (5)) has very strong influence
types of models are processes with recycle streams (Seborg et on the closed loop response. The closed loop response will vary
al. [9] and Bequette [21]). To enrich product quality, recycling a based on the relative values of τ 1 and τ 3 . In the simulations,
260 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264

Fig. 8. Dimensionless control parameters obtained from model vs. simulation results for second order plus time delay with lead process (described by Eq. (5)): (a, c
and e) for set point change and (b, d and f) for load change.

different combinations of the process parameters are considered more than that of set point change. The integral time and deriva-
with the range being 1–20. Furthermore, the simulations were tive time are modeled in a similar fashion and were found to
performed such that τ 1 > τ 3 . Fig. 8 shows the validity of the increase with a decrease in the fraction dead time. The models
proposed tuning models with respect to simulation results. For obtained for SOPTDLD are shown in Table 6. The τ I /τ D ratio
SOPTDLD processes, the controller gain is inversely propor- as a function of fraction dead time is shown in Fig. 9. For a set
tional to the process gain. The controller gain for load change is point change and at very low values of fraction dead time, the

Table 6
Proposed tuning relations for SOPTDLD process
Tuning parameter Set point change Load change

0.5254
 θ
−1.2206 0.4938
 θ
−1.3594
KC
KP θ + τ1 + τ2 − τ3 KP θ + τ1 + τ2 − τ3
 θ
−1.4569 τ2 (τ1 + τ2 − τ3 )
 θ
−2.4858
τI 0.5657θ 12.585
θ + τ1 + τ2 − τ3 θ θ + τ1 + τ2 − τ3
τD θ
 θ
−2.9057  θ
−2.9185
6.7572 6.465
τ2 (τ1 + τ2 − τ3 ) θ + τ1 + τ2 − τ3 θ + τ1 + τ2 − τ3
 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264 261

performance characteristics obtained for the two methods are

shown in Table 7. From Table 7, we can conclude that by using
the proposed method of tuning, the overshoot and the settling
time are significantly reduced in addition to the minimization of
IAE compared to the conventional Z–N method.

4. Robustness, performance and control effort

There are many mathematical definitions for performance

and control effort in the literature. The definitions for robustness,
performance and control effort used to compare the proposed
method with different existing methods are similar to those used
by Foley and co-workers [8]. The robustness of a control system
is usually evaluated by randomly varying the process parame-
ters while keeping the control parameters constant. This is done
Fig. 9. The ratio between integral time and derivative time as a function of to study the controller performance in situations where there
fraction dead time for SOPTDLD process. are uncertainties or errors in the process model. To study the
robust stability for the different processes discussed in the ear-
τ I /τ D ratio goes to infinity. Nevertheless, for fraction dead time lier sections, the process parameters, i.e. KP , θ, τ 1 , τ 2 and τ 3 are
greater than 0.1 the τ I /τ D ratio is constant (around 5) for both set simultaneously increased until the system becomes condition-
point and load changes, which is favorable for many industrial ally stable (the closed response will have sustained oscillations).
controllers. Then robustness was defined as the largest deviation in the model
parameters for which the response becomes conditionally stable.
3.6. Case studies for SOPTDLD This is mathematically represented by the following equation:

KP max − KP
To compare the performance of the processes with the Robustness = × 100% or
proposed model for control parameters, three processes with KP
different process parameter configurations are analyzed. The τ1 max − τ1
× × 100% or
transfer functions for the three processes GP7 (s), GP8 (s) and τ1
GP9 (s) are described by Eqs. (17)–(19), respectively. τ2 max − τ2
× × 100% or
6(s + 1) τ2
KP GP7 (s) = exp(−5s) (17)
(7s + 1)(3s + 1) τ3 max − τ3
× × 100% or
4(5s + 1) τ3
KP GP8 (s) = exp(−10s) (18) θmax − θ
(15s + 1)(7s + 1) × × 100% (20)
θ
15(7s + 1)
KP GP9 (s) = exp(−10s) (19)
(20s + 1)(5s + 1) The subscript ‘max’ refers to the value that corresponds to the
conditionally stable response.
The closed response due to step change in the set point and
The performance is defined as follows:
load are shown in Fig. 10 for the three processes, GP7 (s), GP8 (s)
and GP9 (s). Fig. 10 clearly shows that the response obtained
σe,M
2
by the proposed method is superior to the one obtained using Performance = × 100% (21)
Z–N settings. The controller parameters and the corresponding σe2

Table 7
Tuning parameters and performance characteristics for SOPTDLD
Process Method Set point change Load change

KC τI τD Tr Ts Os IAE KC τI τD IAE

KP7 GP7 (s) Z–N 0.39 12.6 2 9.8 42.5 1.5 13 0.39 12.6 2 32.5
P–M 0.27 14.31 2.93 12.5 12.5 1.04 9.4 0.32 8.32 2.5 28.1
KP8 GP8 (s) Z–N 0.63 22.25 3.56 17.4 73.1 1.52 23.8 0.63 22.25 3.56 35.7
P–M 0.47 27.5 4.78 20.2 40.1 1.1 17.1 0.5477 15.04 4.1 30.53
KP9 GP9 (s) Z–N 0.17 19.81 3.17 16.2 78.3 1.52 23.1 0.17 19.81 3.17 125.8
P–M 0.12 25.8 4.53 19.8 47.1 1.0 17.2 0.14 12.41 3.79 107.5
262 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264

Fig. 10. Closed loop response for SOPTD with lead process: (a, c and e) set point change for processes 1, 2 and 3, respectively, and (b, d and f) load change for
processes 1, 2 and 3, respectively.

where σe2 is the variance of the error and σe,M2 is the minimum σU
2 (which also represents the valve adjustments) to the mini-

control error variance achieved by using parallel PID structure mum variance PID controller, σU,M
2 . The definition for control
with α = 0.05. The performance of PID controller, either series or effort is then given by the following equation:
parallel configuration will be bounded by 0% (unstable control
σU
2
system) and 100% (best control system). Similarly, the control Control effort = × 100% (22)
effort is defined as the ratio of the variance of the control error, σU,M
2
 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264 263

Table 8 0.1 to 2 while other conventional methods can be used only up to

Comparison of robustness, performance and control effort for different processes a ratio of 1. Tuning rules for critically damped and over damped
corresponding to a unit step change in set point
SOPTD processes were developed. Different case studies for
Process Method Set point change different FOPTD, SOPTD and SOPTD with lead process were
Robustness Performance Control effort analyzed. From the comparison of the results obtained using the
proposed tuning method and various other methods for tuning
KP1 GP1 (s) Z–N 22 82.9 143.5
C–C 11 63.26 196.92
PID controllers, it can be concluded that the proposed tuning
P–M 94 99.43 106.3 method gives a better response in all cases. Finally, the robust
stability, performance and control effort were evaluated for the
KP2 GP2 (s) Z–N 35 99.21 103.03
C–C 43 99.6 102.1 case studies with unit step changes in the set point. It was shown
P–M 88 99.47 101.67 that the controller tuned with the proposed method is highly
KP3 GP3 (s) Z–N 31 97.29 111.58
robust and ensures best achievable controller performance, com-
C–C 57 99.75 104.79 pared to other conventional tuning methods. Also, the proposed
P–M 65 99.86 104.79 sets of PID tuning rules are simple, accurate and efficient for
KP4 GP4 (s) Z–N 27 83.97 145.49 practitioners.
P–M 83 97.18 119.11
KP6 GP5 (s) Z–N 36 97.19 112.4 Acknowledgements
P–M 75 99.52 109.37
KP6 GP6 (s) Z–N 33 86.53 152.53 The financial support of the Natural Sciences and Engi-
P–M 38 98.91 105.45 neering Research Council (NSERC) of Canada is gratefully
KP7 GP7 (s) Z–N 41 99.6 101 acknowledged.
P–M 97 100 101
KP8 GP8 (s) Z–N 40 99.2 103.5 Appendix A. Nomenclature
P–M 87 99.99 102
KP9 GP9 (s) Z–N 37.5 99.7 99.96
P–M 94 100.9 100.6 D load (or disturbance) variable
e control error, R–Y
GC feedback controller transfer function
GC,Ser servo compensator
Table 8 shows the comparison of the robustness, performance GP process transfer function
and control effort calculated for different processes using dif- KC proportion gain of PID controller
ferent control parameters obtained from various tuning methods KP gain of the process
for unit step changes in the set point. The results show that the Os overshoot
control system designed by using the proposed method is highly r set point
robust compared to those obtained by using other methods. In s Laplace operator
almost all case studies (except for KP6 GP6 (s)), the robustness tss time to reach steady state
measure obtained with the proposed method is over 85%. The Tr rise time
robustness achieved with other methods is very low and this Ts settling time (to reach 5% of the steady state value)
would entail frequent tuning of the PID parameters for model U controller output
uncertainties and changes in levels of operating conditions. The Y controlled variable (or process variable)
control effort using the proposed method was always lower than
those obtained for other methods, but relatively the difference Greek symbols
was not high except for some cases. Small control effort would α derivative filter factor
cause smooth increments in the control valve. θ dead time
σe2 control error variance
5. Conclusions σe,M
2 minimum error variance achievable using parallel PID
control with α = 0.05
PID tuning relations (with parallel form) for FOPTD, SOPTD σU
2 variance of controller output (also manipulative vari-
and SOPTD with lead processes are developed for unit step able)
change in set point and load (separately). The derivative fil- σU,M
2 variance of controller output using parallel PID control
ter factor α was preset to 0.1 in all simulations. The tuning with α = 0.05
relations were derived based on the minimization of IAE. Gener- τD derivative time of the PID controller
alized tuning rules for PID controller, as functions of the process τI integral time of the PID controller
parameters were developed. The proposed tuning rules were τ1 process time constant
compared with conventional tuning methods. For FOPTD pro- τ2 process time constant
cess, the tuning rules were developed for θ/τ 1 ratio varying from τ3 process time constant
264 C.R. Madhuranthakam et al. / Chemical Engineering and Processing 47 (2008) 251–264

Subscripts [11] F.G. Shinskey, Process Control Systems, third ed., McGraw-Hill, New
max max value for which the response becomes condition- York, 1988.
ally stable [12] W.L. Luyben, Process Modeling, Simulation and Control for Chemical
Engineers, McGraw-Hill, New York, 1973.
1–9 indices to represent different processes [13] R. Ciancone, T. Marlin, Tune controllers to meet plant objectives, Control
5 (1992) 50–57.
References [14] A.B. Corripio, Tuning of Industrial Control Systems, Instrument Society
of America, 1990.
[1] J.G. Ziegler, N.B. Nichols, Optimum settings for automatic controllers, [15] T. Sayeed, T. Mahdi, Optimal tuning of PID controllers for first order plus
Trans. ASME 64 (1942) 759–768. time delay models using dimensional analysis, in: The Fourth Interna-
[2] G.H. Cohen, G.A. Coon, Theoretical consideration of related control, tional Conference on Control and Automation (ICCA’03), June 2003, pp.
Trans. ASME 75 (1953) 827–834. 942–946.
[3] D. Chen, D.E. Seborg, PI/PID controller design based on direct synthesis [16] G.H. Archibald, K. Tae-Won, Effect of disturbance dynamics on optimum
and disturbance rejection, Ind. Eng. Chem. Res. 41 (19) (2002) 4807–4822. control of second order plus dead time processes, Instrum. Chem. Petrol.
[4] D.E. Rivera, M. Morari, S. Skogestad, Internal model control. 4. PID Ind. 20 (April) (1988) 69–79.
controller design, Ind. Eng. Chem. Process Des. Dev. 25 (1986) 252–265. [17] A.A. Rovira, Ph.D. Dissertation, Louisiana State University, Baton Rouge,
[5] A. Andrášik, A. Mészáros, S.F. de Azevedo, On-line tuning of a neural 1981.
PID controller based on plant hybrid modelling, Comput. Chem. Eng. 28 [18] A.M. Lopez, P.W. Murrill, C.L. Smith, Controller tuning relationships
(2004) 1499–1509. based on integral performance criteria, Instrum. Technol. 14 (November)
[6] C.H. Lee, C.C. Teng, Calculation of PID controller parameters by using a (1967) 57.
´
[19] K.J. Åstrom, T. Hägglund, Revisiting the Ziegler–Nichols step response
fuzzy neural network, ISA Trans. 42 (3) (2003) 391–400.
[7] Y. Yeong-Koo, T.I. Kwon, A neural PID controller for the pH neutralization method for PID control, J. Process Control 14 (2004) 635–650.
process, Ind. Eng. Chem. Res. 38 (3) (1999) 978–987. [20] W.A. Weigand, J.E. Kegerreis, Comparison of controller-setting techniques
[8] W. Micheal, M.W. Foley, R.H. Julien, B.R. Copeland, A comparison of as applied to second-order dead time processes, Ind. Eng. Chem. Process
PID controller tuning methods, Can. J. Chem. Eng. 83 (August) (2005) Des. Dev. 11 (1) (1972) 86–90.
712–722. [21] B.W. Bequette, Process Control—Modeling, Design and Simulation, Pren-
[9] D.E. Seborg, T.F. Edgar, D.A. Mellichamp, Process Dynamics and Control, tice Hall, 2003.
second ed., Wiley, New York, 2004. [22] T. Marlin, Process Control, Mc Graw-Hill, NY, 1995.
[10] M.L. Luyben, W.L. Luyben, Essentials of Process Control, McGraw-Hill,
New York, 1997.

View publication stats