You are on page 1of 14

The TQM Magazine

Robust PID controllers by Taguchi's method


John G. Vlachogiannis Ranjit K. Roy
Article information:
To cite this document:
John G. Vlachogiannis Ranjit K. Roy, (2005),"Robust PID controllers by Taguchi's method", The TQM
Magazine, Vol. 17 Iss 5 pp. 456 - 466
Permanent link to this document:
http://dx.doi.org/10.1108/09544780510615951
Downloaded on: 18 March 2016, At: 01:38 (PT)
Downloaded by University College London At 01:38 18 March 2016 (PT)

References: this document contains references to 21 other documents.


To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 947 times since 2006*
Users who downloaded this article also downloaded:
M.J. Ploos van Amstel, David Farmer, (1990),"Controlling the Logistics Pipeline", The International Journal
of Logistics Management, Vol. 1 Iss 1 pp. 19-27 http://dx.doi.org/10.1108/09574099010804518
Robert A. Marshall, Philippe Ruiz, Christophe N. Bredillet, (2008),"Earned value management insights
using inferential statistics", International Journal of Managing Projects in Business, Vol. 1 Iss 2 pp. 288-294
http://dx.doi.org/10.1108/17538370810866386
Jaakko Niskanen, Jussi Rouhento, Heidi Falkenbach, (2011),"European real estate equities: ownership
structure and value of the firm", Journal of European Real Estate Research, Vol. 4 Iss 2 pp. 131-144 http://
dx.doi.org/10.1108/17539261111157307

Access to this document was granted through an Emerald subscription provided by emerald-srm:368933 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for
Authors service information about how to choose which publication to write for and submission guidelines
are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as
providing an extensive range of online products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee
on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive
preservation.
*Related content and download information correct at time of download.
Downloaded by University College London At 01:38 18 March 2016 (PT)
The Emerald Research Register for this journal is available at The current issue and full text archive of this journal is available at
www.emeraldinsight.com/researchregister www.emeraldinsight.com/0954-478X.htm

TQM TQM IMPLEMENTATION


17,5
Robust PID controllers by
Taguchi’s method
456
John G. Vlachogiannis
Department of Informatics and Computer Technology, Technological
Educational Institute of Lamia, Lamia, Greece, and
Ranjit K. Roy
Nutek Inc., Bloomfield Hills, Michigan, USA
Downloaded by University College London At 01:38 18 March 2016 (PT)

Abstract
Purpose – The aim of the paper is the fine-tuning of proportional integral derivative (PID) controllers
under model parameter uncertainties (noise).
Design/methodology/approach – The fine-tuning of PID controllers achieved using the Taguchi
method following the steps given: selection of the control factors of the PID with their levels;
identification of the noise factors that cause undesirable variation on the quality characteristic of
PID; design of the matrix experiment and definition of the data analysis procedure; analysis of the
data; decision regarding optimum settings of the control parameters and predictions of the
performance at optimum levels of control factors; calculation of the expected cost savings under
optimum condition; and confirmation of experimental results.
Findings – An example of the proposed method is presented and demonstrates that given certain
performance criteria, the Taguchi method can indeed provide sub-optimal values for fine PID tuning in
the presence of model parameter uncertainties (noise). The contribution of each factor to the variation of
the mean and the variability of error is also calculated. The expected cost savings for PID under
optimum condition are calculated. The confirmation experiments are conducted on a real PID controller.
Research limitations/implications – As a further research it is proposed the contiguous
fine-tuning of PID controllers under a number of a variant controllable models (noise).
Practical implications – The enhancement of PID controllers by Taguchi method is proposed with
the form of a hardware mechanism. This mechanism will be incorporated in the PID controller and
automatically regulate the PID parameters reducing the noise influence.
Originality/value – Application of Taguchi method in the scientific field of automation control.
Keywords Process control, Taguchi methods, Optimization techniques
Paper type Research paper

Introduction
Since the late 1950s Dr Genichi Taguchi introduced several new statistical tools and
concepts for quality improvement, relying heavily on the statistical theory of the design
of experiments technique introduced by Sir R.A. Fisher in 1920s in England (Barker,
1986; Box, 1988; Byrne and Taguchi, 1987; Kackar, 1985; Logothetis, 1988; Montgomery,
1997, 1999; Ross, 1996; Roy, 2001; Shoemaker et al., 1991; Taguchi, 1986). Dr Taguchi
introduced his strategy of experimental design for applications in the following areas:
The TQM Magazine
Vol. 17 No. 5, 2005
pp. 456-466 The authors gratefully acknowledge the valuable suggestions given by Professor George
q Emerald Group Publishing Limited
0954-478X
P. Syrcos of the Automation Department of Technological Educational Institute (TEI) of Piraeus,
DOI 10.1108/09544780510615951 Greece.
(1) designing products and processes so that they are robust to environmental Robust PID
conditions; controllers
(2) developing products and processes so that they are robust to component
variation; and
(3) minimising variation around a target value.

In his view, Dr Taguchi considers three stages in a process development (Box et al., 1988; 457
Byrne and Taguchi, 1987; Montgomery, 1997, 1999; Shoemaker et al., 1991; Taguchi, 1986):
(1) system design;
(2) parameter design; and
(3) tolerance design.
Downloaded by University College London At 01:38 18 March 2016 (PT)

In system design, the engineer uses scientific and engineering principles to determine
the basic configuration. Parameter design is an investigation conducted to identify
settings that minimise (or at least reduce) the performance variation. Variation in the
performance characteristic may change with different settings. This variation
increases the product manufacturing cost, the lifetime cost and the cost due to harm
that is caused by the product to the society when the customer “uses” product/process
whose quality characteristics differ from the nominal, namely as social cost (Ferrell
and Reddivari, 1995; Ross, 1996; Roy, 2001). Tolerance design is used to specify the
best tolerances for the system parameters (Box, 1988; Box et al., 1988; Byrne and
Taguchi, 1987; Kackar, 1985; Montgomery, 1997, 1999; Ross, 1996; Shoemaker et al.,
1991).
Considerable advantages could be obtained by implementation of Taguchi’s
approach in the process of automation control. In process control today, proportional
integral derivative (PID) controllers still predominate and are sufficient for most needs
with more than 95 per cent of control loops being of the PID type. A vast amount of
literature is available that addresses the problem of PID controllers tuning (Astrom
and Hagglund, 1995; Luyben, 1990). The most important factor in all of these tuning
methods is the form of the actual transfer function of a plant. If the structure of this
transfer function is significantly different from the one used to derive the tuning
formulas, then there is a substantial amount of system unpredictability. However, this
problem is simplified when there is no structural mismatch between the actual and the
assumed transfer functions. In this case there is no structural uncertainty, but rather
the uncertainty is isolated to the actual value of the parameters used to model the plant.
If the model and the actual plant have the same structure type, then the parameters of
the transfer function become of primary concern. In a study made by Bialkowski
(1993), it was shown that 30 per cent of their control loops functioned poorly due to
incorrect PID controller settings. It is, therefore, evident that a tuning refinement in the
presence of model parameter uncertainties is of great importance.
To date, there is no research about robustness of PID controllers under noise
condition. The quality of PID controllers tuning is the result of a number of parameters.
Some of these are controllable while others are noise factors (Astrom and Hagglund,
1995; Luyben, 1990; Ziegler and Nichols, 1942).
The focus of this paper is the robustness of the PID controllers using the Taguchi’s
method. This application to PID controller’s tuning follows the basic steps listed below:
TQM (1) Selection of the control factors of the PID controllers and their alternative levels.
17,5 (2) Identification of the noise factors that cause undesirable variation on the quality
characteristic.
(3) Design of the matrix experiment (orthogonal array – OA) and definition of the
data analysis procedure.
458 (4) Analysis of the data. Response graphs are plotted to determine preferred levels
of each parameter. Also, an analysis of variance (ANOVA) table can be
generated to determine the statistical significance of the parameters.
(5) Decision regarding optimum settings of the control parameters and predictions
of the performance at optimum levels of control factors.
(6) Calculation of the expected cost savings under optimum condition.
Downloaded by University College London At 01:38 18 March 2016 (PT)

(7) Confirmation of experimental results.

Conventional methods of PID controllers tuning


The Ziegler and Nichols (1942) tuning method is based on the calculation of the
ultimate period (Pu) and ultimate gain (Ku) of the system. The pair Pu, Ku can be
calculated by letting the process loop gain increase until the system begins sustained
oscillations (Astrom and Hagglund, 1995; Luyben, 1990). This is when the poles are
located on the imaginary axis on the root locus plane or when the open loop transfer
function intersects the negative real axis on the Nyquist curve. However, this method is
difficult to automate and dangerous to implement in real situations because it has to
operate near the region of system’s instability.
An alternative method for tuning is the relay feedback method (Astrom and
Hagglund, 1995). In this method the system is forced to oscillate by introducing a
non-linear feedback of the relay type in order to generate a limit cycle oscillation in the
system. The amplitude of the system oscillations can be controlled since it is
proportional to the relay amplitude. In the PID tuning mode, when the steady state is
reached, the system oscillates at a frequency with period and amplitude close to the
ultimate period (Pu) and ultimate gain (Ku) of the open loop system.
In this study the Taguchi’s approach is applied to refine the near optimum
PID parameters, after a classical relay auto-tuning, which is used to select the optimum
parameters.

Quality characteristic of PID controllers


Clarifying things it is assumed that the model of process and actual plant have the
same transfer function (Astrom and Hagglund, 1995; Luyben, 1990):
K dc
GðsÞ ¼ expð2d · sÞ ð1Þ
T ·s þ 1
where, Kdc is the DC-gain of process model, d is the time delay and T is the time
constant.
The PID controller may be implemented in continuous or discrete time, in a number
of controller structures. The ideal continuous PID controller time is expressed in
Laplace form as follows (Astrom and Hagglund, 1995; Luyben, 1990):
GPID ¼ K P þ sK D þ K I =s ð2Þ
It is comprised of a proportional (KP), an integral (KI/s) and a derivative (sKD) term.
The performance index used is the integral-squared-error (ISE) and is mathematically Robust PID
defined as (Astrom and Hagglund, 1995; Luyben, 1990; Ziegler and Nichols, 1942): controllers
Z 1 Z 1
ISE ¼ eðtÞ2 dt ¼ ð yðtÞ 2 ysp Þ2 dt ð3Þ
0 0

where y(t) and ysp is the output and the desirable output of process model, respectively.
This particular performance criterion is widely used for controllers tuning because
459
its minimisation is related to minimisation of error magnitude and duration (Ziegler
and Nichols, 1942). Thus, in this study, the ISE will be taken as the quality
characteristic to be observed.

Control factors of PID controllers and noise


Downloaded by University College London At 01:38 18 March 2016 (PT)

In the PID controllers, the KP, KI and KD parameters (equation (2)) are the control
factors since they can be changed by the conventional auto-tuner to minimise the
quality characteristic ISE (Astrom and Hagglund, 1995; Ziegler and Nichols, 1942). The
experience reveals that non-linear behaviour of the control factors of a PID can be
determined only if more than two levels are used (Day, 1986; Ziegler and Nichols, 1942).
Therefore, each control factor is analysed in three levels (Day, 1986). Also literature
reveals that between KP, KI and KD there is no significant conjunction (there is no
dependence), which effects on the quality characteristic ISE (Astrom and Hagglund,
1995; Ziegler and Nichols, 1942). So, the interaction effect of these parameters is not
considered for examination in the PID controllers. In this study, after the auto-tuning
method (Ziegler and Nichols, 1942) the initial optimum set of values is determined for
the control factors under ideal conditions, i.e. without considering any noise factor, is
K P ¼ 8; K I ¼ 45 and K D ¼ 0:47: The three levels of three control factors are identified
for study as presented in Table I. These factor levels define the experimental region,
which is also studied and presented in Table I.
Noise factors are those parameters of process model (equation (1)) that cannot be
controlled or are too expensive to control or cannot be identified. In equation (1), the
Kdc (DC-gain), d (delay time) and d/T (dimensionless factor) are generally known to be the
most uncertain factors; even small uncertainties in their values (uncontrollability) lead to
poor control (Astrom and Hagglund, 1995; Luyben, 1990; Ziegler and Nichols, 1942). In this
study, the first noise factor NF1 (DC-gain, Kdc) could be higher (up to 400) than the original
estimate of 360. The second noise factor NF2 (time delay, d ) could be higher (up to 30 s)
than the original estimate of 27 s and the third noise factor NF3 (dimensionless
controllability factor, d/T) could be higher (up to 0.1666) than the original estimate of 0.135.

Experiments
The objective now is the determination of the optimum levels of control factors so
that the PID controllers to be robust to noise factors. The robust design method

Factors PID parameters Range Level 1 Level 2 Level 3


Table I.
A KP 6-10 6 8 10 PID parameters with
B KI 40-50 40 45 50 their ranges and values at
C KD 0.42-0.52 0.42 0.47 0.52 three levels
TQM (Taguchi’s method) uses OAs based on the design of experiments theory to study a
17,5 large number of decision variables with a small number of experiments (Day, 1986;
Montgomery, 1999; Taguchi and Konishi, 1987). The OA, fitting a specific case study
is selected by determining the total degree-of-freedom (DOF) for the factors included
in the study. The selected OA helps in describing the minimum number of
experiments that must be performed to reach a near optimum parameter combination.
460 One DOF is associated with the overall mean regardless of the number of control
factors. To this it is added the DOF associated with each control factor, which is
equal to one less than the number of levels. In this study, the DOF for three control
factors, in each of three levels is: ð3 2 1Þ £ 3 þ 1 ¼ 7: A three-level OA with at least
seven DOF must be selected (Day, 1986; Taguchi and Konishi, 1987). The nearest in
the seven DOF three-level OA is the OA (L9) with nine experimental runs (Day, 1986;
Taguchi and Konishi, 1987).
Downloaded by University College London At 01:38 18 March 2016 (PT)

The selected OA (L9) is shown in Table II. In this array, the columns are mutually
orthogonal. That is, for any pair of columns, all combinations of factor levels occur and
they occur in an equal number of times. The next step is the development of a
procedure, simulating the variation in the quality characteristic ISE, due to the noise
factors. Taguchi (1986) also proposes OA based on simulation to evaluate the mean
and the variance of quality characteristic response resulting from variations of noise
factors. The diversity of noise factors (levels) is studied by crossing the selected OA
(L9) of control factors with an OA of noise factors (N-OA) (Barker, 1986; Box et al., 1988;
Byrne and Taguchi, 1987; Logothetis, 1988; Montgomery, 1997, 1999; Ross, 1996; Roy,
2001; Shoemaker et al., 1991; Taguchi, 1986). Thus, ISE is evaluated for each of the nine
trials against the background of four different combinations of three noise factors (two
levels at each noise factor). The ISE response was computed for each combination of
control and noise matrix experiments using the MATLAB program (version 6.1),
which simulates the above control system equations (Ziegler and Nichols, 1942). For
each combination of control factor levels, the mean and the standard deviation (std) are
shown in Table III.
It must be notified that column C in the control OA9 was left empty since there are
only three factors in this study (Byrne and Taguchi, 1987; Montgomery, 1999; Ross,
1996; Roy, 2001). Table III shows nine experiments that were carried out. In the next
section, results analysis of these experiments will identify the optimum condition,
which is the optimum mean and has minimum variation around the mean due to noise
factors.

Run A B C D

1 1 1 1 1
2 1 2 2 2
3 1 3 3 3
4 2 1 2 3
5 2 2 3 1
6 2 3 1 2
7 3 1 3 2
Table II. 8 3 2 1 3
Orthogonal array OA9 9 3 3 2 1
Downloaded by University College London At 01:38 18 March 2016 (PT)

Noise factors Noise levels


NF1 360 360 400 400
NF2 27 30 27 30
NF3 0.1666 0.135 0.135 0.1666
Trial KP KI C KD ISE (trial £ noise level) Mean Std S/N ratio
1 6 40 0.42 0.2516 0.2521 0.2099 0.2638 0.24435 0.023648 12.209351
2 6 45 0.47 0.2606 0.2649 0.2148 0.2719 0.25305 0.025922 11.901827
3 6 50 0.52 0.2696 0.2792 0.2213 0.2815 0.2629 0.028208 11.566850
4 8 40 0.52 0.1896 0.1929 0.166 0.1926 0.18527 0.012936 14.627813
5 8 45 0.42 0.2698 0.2604 0.2196 0.2836 0.25835 0.027534 11.718989
6 8 50 0.47 0.2716 0.2678 0.2197 0.2838 0.26072 0.028189 11.638437
7 10 40 0.47 0.1966 0.192 0.17 0.1998 0.1896 0.013453 14.426865
8 10 45 0.52 0.197 0.1959 0.1701 0.2002 0.1908 0.01392 14.371130
9 10 50 0.42 0.3032 0.2772 0.239 0.3186 0.2845 0.034814 10.869852
Note: C: neglected, M ¼ 0:236617
controllers

experiments
Design and results of
Robust PID

Table III.
461
TQM Analysis of experimental results
17,5 There are several approaches to this analysis. One common approach, suggested by
Taguchi’s method, involves graphing the effects and visually identifying the factors, which
appear to be significant. This technique uses a statistical measure of performance called
signal-to-noise (S/N) ratio borrowed from electrical control theory (Astrom and Hagglund,
1995; Barker, 1986; Byrne and Taguchi, 1987; Ross, 1996). Taguchi (1986) stresses the
462 importance of studying also the response variation, using the S/N ratio, resulting in
the minimisation of the quality characteristic variation, due to noise factors. The
smallest-is-best type S/N ratio will be used to analyse the results, since the objective here is
the ISE minimisation (Ross, 1996). Therefore, the S/N ratio is used for that type of response
(Astrom and Hagglund, 1995; Barker, 1986; Byrne and Taguchi, 1987; Ross, 1996):
" #
1 X n
S=N ratio ðdbÞ ¼ 210 · log · ISEi ð4Þ
Downloaded by University College London At 01:38 18 March 2016 (PT)

n i¼1
where, ISEi is the response value for a trial condition repeated n times.
S/N ratios are computed in each of the trial combination and the values are
displayed in the last column of Table III. Since the experimental design is orthogonal, it
is possible to separate out the effect of each factor. Therefore the average S/N of each
level (1, 2, 3) of three control factors (KP, KI and KD) are plotted in Figure 1.
Comparing the Figure 1(a)-(c) it is shown that the line in Figure 1(b) has the largest
gradient. Thus, from Taguchi’s (1986) method it is testified that factor KI, shown in
Figure 1(b), is more prominent than other factors (Montgomery, 1997; Ross, 1996; Roy,
2001). Moreover, level 3 appears to be the best choice for factors KP and KD and level 1
for factor KI since they correspond to the largest average S/N ratio (Figure 1). In the
next sections the significance of each control factor in ISE is calculated and the
predicted ISE mean and their confidence interval (CI) are estimated.

Estimation of control factors significance


Another common method used in statistical ANOVA (Montgomery, 1997, 1999;
Ross, 1996), indicates which factors are statistically significant and how much (in
percentage) each one affects the variance of ISE. With a view to studying the factor
significance, ANOVA using the S/N of the results is performed in Table IV.
Taking Table IV into account it is marked that KI (parameter B) significantly
affects the variability of ISE with percentage 50.79 per cent. Also the KP (parameter A)
and KD (parameter C) affect the variability of ISE with percentages 15.69 per cent
and 32.69 per cent, respectively. The influence of the error variance is calculated at
0.8097 per cent.

Estimation of predicted ISE mean


Utilizing the estimation model of Taguchi based on the average of factorial factors and
equivalent “b-discount factors”, a predicted optimum mean of IŜE is estimated by
equation (Logothetis, 1992; Montgomery, 1997; Ross, 1996):
^ ¼ M · bðM Þ þ ðA3 2 M Þ · bðAÞ þ ðB1 2 M Þ · bðBÞ þ ðC 3 2 M Þ · bðCÞ
ISE ð5Þ
where: M is the total average of ISE, b(A), b(B), b(C) are the b-discount factors of
factors A, B, C, respectively, and defined as (Logothetis, 1992):
Robust PID
controllers

463
Downloaded by University College London At 01:38 18 March 2016 (PT)

Figure 1.
Effects of each control
factor on S/N ratio

Factor Sum square DOF F-ratio Effect (per cent)

A 2.674782 2 340.2757 15.69933


B 8.636447 2 1098.696 50.7938
C 5.562273 2 707.6114 32.69709 Table IV.
Error 0.113979 29 0.809779 ANOVA for S/N ratio
Total 16.98748 35 100 of ISE

bðPÞ ¼ 1 2 1=F P ð6Þ


where, FP is the F-ratio of factor P ¼ {A3 ; B1, C3} is the best experimental condition for
controllable factors. The b(M) is the overall b-factor defined as (Logothetis, 1992):
bðM Þ ¼ 1 2 V e =SS ð7Þ
where: SS is the total sum square of trials and Ve is the variance due to error.
TQM From Tables III and IV it is shown that total SS ¼ 16:98748; M ¼ 0:236617;
V e ¼ 0:113979: As a final result, the predicted mean estimation of IŜE is 0.173774.
17,5
Estimation of predicted confidence interval
The CI for the predicted mean of IŜE in a confirmation run can be calculated using the
following equation (Logothetis, 1988; Montgomery, 1997; Ross, 1996; Roy, 2001):
464 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Fð1; a; DOFe Þ · V e
CI ¼ ð8Þ
neff
where: F(1, a, DOFe) is the F-ratio required for a ¼ risk; confidence ¼ 1 2 a; DOFe is
the degree-of-freedom of pooled error, Ve is the pooled error variance and neff is the
effective sample size (Logothetis, 1988; Montgomery, 1997; Ross, 1996; Roy, 2001):
Downloaded by University College London At 01:38 18 March 2016 (PT)

N
neff ¼ P ð9Þ
1 þ ðDOFp · bðPÞÞ
p

where: N is the total number of trials, DOFp the degree-of-freedom of factor


P ¼ {A3 ; B1 ; C 3 } and b(P) is the b-discount factor (Logothetis, 1992) of control factor P.
For interval confidence of 99 per cent of predicted ISE, the next parameters are
equal with: F(1, 1 per cent, 29Þ ¼ 7:6 (tabulated in Ross (1996)), the error variance
V e ¼ 0:113979 and the effective size of samples neff ¼ 5:643176: So the CI is computed
CI ¼ 0:027671: The 99 per cent CI of the predicted optimum is:
^ 2 CIÞ # ISE
ðISE ^ # ðISE
^ þ CIÞ ) 0:1641 # ISE
^ ¼ 0:173774 # 0:20144 ð10Þ

Expected cost savings at optimum condition


The variability of ISE at optimum condition is estimated using the Qualitek-4 (2002)
program. In this study, the factors A-C significantly affect the variability around the
mean of IŜE. Contributions of optimum condition are shown in Table V.
The significant levels of control factors A3, B1 C3 have a total contribution of
2.7221 db. As the current grand average of performance was 12.5923 db, the expected
result at the optimum condition was estimated to be 15.3144 db. Therefore, the
expected improvement was 17.7747 per cent.
In order to calculate the expected cost savings, the Taguchi’s loss function has been
used (Ross, 1996; Roy, 2001). Savings in dollars ($) can be calculated when S/N ratio is
known. On account of the lack of the current performance status (in terms of S/N), the
cost savings in relation to the average performance loss (average S/N of all trials) can
be calculated using the next formula (Qualitek-4, 2002; Roy, 2001):

Factor Level description Level Contribution (db)

KP 10 3 0.6303
KI 45 1 1.1623
Table V.
KD 0.52 3 0.9295
Main contribution of
control factors at Notes: Total contribution from all factors ¼ 2:7221; Current grand average of
optimum condition performance ¼ 12:5923; Expected results at optimum condition ¼ 15:3144
L ¼ {1 2 10½ðS=NÞ1 2ðS=N Þ2 =10 } · 100 per cent of L1 ¼ 46:569 cents=$1 loss ð11Þ Robust PID
controllers
where L1 is the per cent loss before the experiment (S/N)1 is the average S/N
performance of all trials and (S/N)2 is the optimum performance.

Discussion and confirmation of results


This paper described a method of optimisation of PID controller’s performance following 465
the Taguchi robust design approach. In this study, the values of control factors ðK P ¼ 10;
K I ¼ 40; K D ¼ 0:52Þ in the fine-tuning of the PID controllers under noise factors were
achieved. The ISE mean was estimated at 0.174 with a std of 0.011 and S/N of 15.177. The
above optimum results are different from the initial given by Ziegler and Nichols (1942)
ðK P ¼ 8; K I ¼ 45; K D ¼ 0:47 with ISE mean of 0.226, with a std of 0.02 and S/N of 12.88).
Conclusively, the study produced a decrease of 0.052 of the ISE mean (about 23 per cent
Downloaded by University College London At 01:38 18 March 2016 (PT)

more precision in control procedure) and a slight reduction in std. In the above optimum
levels of PID control factors, the mean performance of IŜE mean was estimated to be
0.173774 with a CI between 0.14610 and 0.20144. The confirmation tests at the optimum
condition resulted in an average IŜE of 0.18873, which was within the predicted CI.
Furthermore, the contribution of each factor to the variation of the mean and the variability
of ISE was also calculated. KI was elected as the most significant factor, affecting the
variability of ISE with a percentage of 50.79 per cent. Also, as the current S/N average of all
trials was 12.5923 db, the expected result at the optimum condition was estimated to be
15.3144 db. Therefore, the expected improvement in reduction of ISE was 17.7747 per cent.
The expected cost savings for PID controllers under optimum condition were calculated at
46.569 cents for every dollar of loss before the Taguchi’s experimental design.
Finally, the confirmation experiments were conducted on a real PID controller in the
Informatics and Energy Informatics (IEI) Lab of Technological Educational Institute
(TEI) of Lamia. More specifically, the optimum settings of PID parameters ðK P ¼ 10;
K I ¼ 40; K D ¼ 0:52Þ recommended by this investigation for various actual plants,
under various size of noise were tested. The average ISE of PID was found to be 0.1887,
which was within the predicted CI (equation (10)).

Conclusions and further research


This paper implemented the Taguchi’s method (robust design method) in the scientific
field of automation control. Specifically, the control factors (decision variables)
affecting the performance of PID controllers under noise condition are regulated.
Regulating these values much more quality and fine-tuning of PID controllers are
achieved. As a further research it is proposed the enhancement of PID controllers by
Taguchi’s method with the form of a hardware mechanism. The novel mechanism will
be incorporated in the PID controller as shown in the diagram of Figure 2.

Figure 2.
Hardware enhancement of
PID controllers by
Taguchi’s mechanism
TQM Then, this hardware Taguchi’s mechanism will automatically regulate the PID
parameters reducing the noise influence.
17,5
References
Astrom, K.J. and Hagglund, T. (1995), PID Controllers, 2nd ed., ISA, Research Triangle Park, NC.
Barker, T.B. (1986), “Quality engineering by design: Taguchi’s philosophy”, Quality Progress,
466 Vol. 19 No. 12, pp. 33-42.
Bialkowski, W.L. (1993), “Dreams versus reality: a view from both sides of the gap, keynote
address, control systems’92”, Whistler, British Columbia, 1992, Vol. 94 No. 11.
Box, G.E.P. (1988), “Signal-to-noise ratios performance criteria and transformations”,
Technometrics, Vol. 30 No. 1, pp. 1-40.
Box, G.E.P., Bisgaard, S. and Fung, C.A. (1988), “An explanation and critique of Taguchi’s
Downloaded by University College London At 01:38 18 March 2016 (PT)

contributions to quality engineering”, Quality and Reliability Engineering International,


Vol. 4 No. 2, pp. 123-31.
Byrne, D.M. and Taguchi, G. (1987), “The Taguchi approach to parameter design”, Quality
Progress, Vol. 20 No. 12, pp. 19-26.
Day, A. (1986), Orthogonal Fractional Factorial Designs, Wiley Eastern, New Delhi.
Ferrell, W. and Reddivari, V. (1995), “Higher quality products with better tuned controllers”,
Computers and Industrial Engineering, Vol. 29 Nos 1/4, pp. 321-5.
Kackar, R.N. (1985), “Off-line quality control, parameter design and the Taguchi method”,
Journal of Quality Technology, Vol. 17 No. 4, pp. 176-209.
Logothetis, N. (1988), “The role of data transformation in the Taguchi analysis”, Quality and
Reliability Engineering International, Vol. 4 No. 1, pp. 49-61.
Logothetis, N. (1992), Total Quality Management from Deming to Taguchi and SPC,
Prentice-Hall, Hemel Hempstead.
Luyben, W.L. (1990), Process Modelling, Simulation and Control for Chemical Engineers,
McGraw-Hill, International Edition, New York, NY.
Montgomery, D.C. (1997), Introduction to Statistical Quality Control, 3rd ed., Wiley, New York,
NY.
Montgomery, D.C. (1999), “Experimental design for product and process design and
development”, Journal of the Royal Statistical Society D, Vol. 48 No. 2, pp. 131-9.
Qualitek-4 (2002), Software for Automatic Design and Analysis of Taguchi Experiments, Nutek,
Inc., Bloomfiled Hills, MI, available at: www.rkroy.com/wp-q4w.html
Ross, P.J. (1996), Taguchi Techniques for Quality Engineering, McGraw-Hill, New York, NY.
Roy, R.K. (2001), Design of Experiments Using the Taguchi Approach: 16 Steps to Product and
Process Improvement, Wiley, New York, NY.
Shoemaker, A.C., Tsui, K.L. and Wu, C.F.J. (1991), “Economical experimentation methods for
robust design”, Technometrics, Vol. 33 No. 4, pp. 415-27.
Taguchi, G. (1986), Introduction to Quality Engineering, Asian Productivity Organisation
(Distributed by American Supplier Institute Inc.), Dearborn, MI.
Taguchi, G. and Konishi, S. (1987), Orthogonal Arrays and Linear Graphs, American Supplier
Institute Inc., Dearborn, MI.
Ziegler, J.G. and Nichols, N.B. (1942), “Optimum settings for automatic controllers”, Transaction
ASME, Vol. 64, pp. 759-65.
This article has been cited by:

1. John Manuel Fernandes, Theo van NiekerkImplementation of an industrial controller for a ball-on-wheel
system 129-134. [CrossRef]
2. Matloub Hussain, Paul R. Drake, Dong Myung Lee. 2012. Quantifying the impact of a supply
chain's design parameters on the bullwhip effect using simulation and Taguchi design of experiments.
International Journal of Physical Distribution & Logistics Management 42:10, 947-968. [Abstract] [Full
Text] [PDF]
3. W.J. Zhang, J.W. Li, B. ZettlClassification of design theories and methodologies for effective industrial
applications 1255-1260. [CrossRef]
4. Abdul-Wahid A. Saif, Muneeb A. AkramA Robust-Gain-Scheduled methodology for process parameter
design and control with application to a carbonated beverage filling process 1670-1674. [CrossRef]
5. B. Younes, A. Fotheringham, R. Mather. 2011. Factorial Optimisation of the Effects of Melt Spinning
Downloaded by University College London At 01:38 18 March 2016 (PT)

Conditions on Biodegradable As-spun Aliphatic-Aromatic Co-Polyester Fibres. International Polymer


Processing 26, 150-163. [CrossRef]
6. Xiaoming Chen, Jili Tao, Yong Zhu, Xiaoying LiuOptimized fuzzy control for LED constant current
driver design 1124-1127. [CrossRef]
7. Božidar Bratina, Nenad Muškinja, Boris Tovornik. 2010. Design of an auto-associative neural network
by using design of experiments approach. Neural Computing and Applications 19, 207-218. [CrossRef]
8. Zhijun Sun, Huafeng Li, Weiqing HuangControl of multiple ultrasonic motors with robust parameter
design 1827-1830. [CrossRef]
9. Jiju Antony, Daniel Perry, Chengbo Wang, Maneesh Kumar. 2006. An application of Taguchi method of
experimental design for new product design and development process. Assembly Automation 26:1, 18-24.
[Abstract] [Full Text] [PDF]