Chemical Engineering and Processing 45 (2006) 980–989

Process improvement approach to the saponification reaction

by using statistical experimental design
Nihal Bursali a , Suna Ertunc b , Bulent Akay b,∗
a Yenimahalle Municipality Precidency, 06170 Yenimahalle, Ankara, Turkey
b Ankara University, Faculty of Engineering, Department of Chemical Engineering, 06100 Tandoğan, Ankara, Turkey
Received 22 April 2005; received in revised form 10 January 2006; accepted 28 February 2006
Available online 30 March 2006

Abstract
The purpose of this study was to determine the effective operating parameters and the optimum operating conditions of a batch saponification
process in the frame of the process improvement. Full two-level factorial and face-centered central composite (FCCC) statistical experimental
design methods were used successively. Examined parameters were the main and interaction effects of temperature, agitation rate, initial sodium
hydroxide (NaOH) and ethyl acetate (EtOAc, CH3 COOC2 H5 ) concentrations. Selected process response was the fractional conversion rate of
NaOH (XA ). Temperature and agitation rate were found to have no effect on the response at the 10% selected significance level (α = 0.1). The
dependence of response on the NaOH and EtOAc concentrations was illustrated by a linear second-order polynomial model. Examination of the
residuals served as a diagnostic check of the model and it was found that the model was good enough to fit the experimental data. Optimum
operating conditions at which maximum XA was obtained about 100% were found to be 0.01 mol L−1 NaOH and 0.1 mol L−1 EtOAc by applying
response surface method (RSM). With the use of residual analysis and statistical techniques, more reliable and proper results were obtained at the
process improvement stage of the saponification process.
© 2006 Elsevier B.V. All rights reserved.

Keywords: Process improvement; Statistical experimental design; Response surface methodology; Saponification reaction; Residual analysis

1. Introduction
Experimentation is made to determine the effect of the inde-
pendent variable (factor) on the dependent variable (response)
of a process and a relation between them illustrated by a regres-
sion model by using experimental data. Statistical design of
experiment (DOE) is a well known efficient experimentation
technique and has been applied in a wide range of fields such as
drug and food industry, chemical and biological processes, etc.,
to produce high quality products, to operate more economical
process, to ensure more stable and reliable process [1,2]. The
studies including application of DOE methods have been made
for more than 40 years and the advance of DOE applications has
been assisted by the developments in the field of computer sci-
Abbreviations: ANOVA, analysis of variance; DOE, statistical design of
experiments; CCC, central composite circumscribed design; CCI, central com-
posite inscribed design; EtOAc, ethyl acetate; FCCC, face-centered central
composite design; RSM, response surface method
∗ Corresponding author. Tel.: +90 312 2126720; fax: +90 312 2121546.
E-mail address: bakay@eng.ankara.edu.tr (B. Akay).
ence. Carr et al., applied statistical program planning for process
improvement to reduce the process development time by apply-
ing fractional factorial design [3]. Lind et al. applied response
surface methodology (RSM) and full two-level factorial design
to a chemical process in which antibiotic was produced [4].
Xu et al. used statistically based experimental designs for the
medium optimization of an important medical microorganism
[5]. Andersons, applied design of experiments technique to the
problem of preparing microwave popcorn [6]. More studies are
available in e-journals [7,8].
There are a lot of DOE methods and the selection of them
is made according to objectives and the number of examined
factors [2]. The objectives of the experiment can be classified
as screening, comparing and applying RSM. Screening exper-
iment is applied to determine the most effective factors on the
process response. RSM is generally used to find the optimal con-
dition by using quadratic polynomial model and it is applied in
consequence of a screening experiment.
In this study, the hydrolysis process of an ester called ethyl
acetate (EtOAc, CH3 COOC2 H5 ) in an alkaline condition was
examined in terms of process improvement. As it is well known,

0255-2701/$ – see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2006.02.010
 N. Bursali et al. / Chemical Engineering and Processing 45 (2006) 980–989
the hydrolysis of a fat or an oil in alkaline condition produces
soap for cleaning purpose and the reaction is called saponifica-
tion. The reaction which occurs in the alkaline condition is also
called saponification. Thus, the hydrolysis of EtOAc to produce
sodium acetate (CH3 COONa) and ethyl alcohol (C2 H5 OH) by
using NaOH is called saponification reaction, although the end
product is not soap. When the related literature on the alkaline
hydrolysis of EtOAc is searched, it is observed that there are
not so many studies on the process improvement stage of this
process. Previously published studies are generally related to
kinetics of this reaction [9].
Although esters are not soluble in alcohol, water and in
similar solvents, the salt obtained at the end of the saponi-
fication reaction is soluble in such kind of solvents. This
property is important in view of the usage area of carboxylic
acid salt obtained at the end of the saponification reaction.
Sodium acetate, obtained as an end product in this study, is
a commercially important carboxylic acid salt, is used in a
large area in industry such as petrol, cosmetic, textile, paint
etc. It is used to remove the insoluble calcium salts in textile,
to intensify the color of paint in paint industry and is used in
food industry being as a tampon and protector. It is also used
as a tampon in haemodialysis and leather industry and used in
the neutralisation of mineral acids. By considering commercial
and industrial importance of this process, optimum operating
condition and operating factors effects were examined for this
process to obtain high purity end product, to operate more
economical and reliable process.
In this work, statistical design of experimental method was
applied on the hydrolysis process of ethyl acetate in an alkaline
condition. This study focused specifically on the improvement of
this hydrolysis process. The optimal condition was determined
and the factor effects were investigated.
This study based on experimentation consists of two stages:
screening experimentation and applying RSM. Screening exper-
imentation aimed to find the most important factors and it was
planned according to full two-level factorial experimental design
method. Fractional conversion rate of NaOH (XA ) was selected
as response. All possible factors thought to have an effect on
the response and their levels were considered at the brainstorm-
ing stage of the study and it was decided to examine the main
and interaction effects of the temperature (◦ C), agitation rate
(rpm), initial NaOH and EtOAc concentrations (mol L−1 ) on
the response. Experimental results were analyzed statistically
by analysis of variance (ANOVA) using Fisher’s F-ratio [1,2].
Optimal values of significant factors were determined as
0.01 mol L−1 NaOH and 0.1 mol L−1 CH3 COOC2 H5 for 100%
of XA by obtaining response surface contours at different levels
of factors by using second-order polynomial model.
2. Material and method
2.1. Reaction kinetic
The stochiometric representation of saponification reaction
between EtOAc and NaOH is given by Eq. (1):
NaOH + CH3 COOC2 H5 → CH3 COONa + C2 H5 OH (1)
981
This irreversible reaction is a second-order overall but first order
with respect to each reactant and rate expression is given by Eq.
(2):
−rNaOH = −rEtOAc = kCNaOH CEtOAc (2)
In this reaction, hydroxyl ions are consumed and acetate ions
are produced. Since hydroxyl ions are more conductive than the
acetate ions, a decrease in the conductivity is observed as the
reaction progresses. The change in conductivity is used to mon-
itor the alkaline hydrolysis of EtOAc. Based on this principle,
a relation between the conductivity of the reaction mixture and
NaOH concentration is obtained as given below [11]:
C − C∞ CNaOH − CNaOH∞
= (3)
C0 − C ∞ CNaOH0 − CNaOH∞
since,
CNaOH∞ → 0 as t→∞ (4)
Eq. (3) is rearranged and given in Eq. (5):
CNaOH C − C∞
= 1 − XA = (5)
CNaOH0 C0 − C ∞
The relation between reaction rate and XA is given by Eq. (6):


CEtOAc0
−rNaOH = kCNaOH CEtOAc = kCNaOH0 (1− XA )
2
XA .
CNaOH0
(6)
2.2. Experimental apparatus
In this study a bench-top batch reactor (Armfield CEM-liquid
phase chemical reactor, Hempshire, England) with temperature,
flow rate and agitation rate control units was used as given
schematically in Fig. 1. Two-liter mixed batch reactor was used
with 1.5 L working volume. Temperature was controlled by cir-
culating hot water through a coil immersed in reactor.
2.3. Analytical methods
The conductivity of the NaOH was measured by using WTW
LF39 (Welheim, Germany) type conductivity meter. The sam-
pling time was determined on the basis of the experimental
design treatments and 10 mL of reaction mixture sample was
taken from the reactor to measure the conductivity.
2.4. Plan of experiments
2.4.1. Screening experiment—full two-level factorial
experimental design
At this stage, factors were studied at their maximum and
minimum levels determined according to our experience about
the process and from the literature research. The number of
experiments for four factors was determined as 24 = 16 accord-
ing to full two-level factorial experimental design method
[2,10]. To simplify the calculations, factors were studied with
982 N. Bursali et al. / Chemical Engineering and Processing 45 (2006) 980–989

Fig. 1. Experimental system.

their codified values. The relation between codified and natural Maximum, minimum and average levels of the factors are
values of factors is given below: codified as (+1), (−1) and (0), respectively as given in
Table 1. The center point is equal to the average value of a
Ui,natural − Ui,average factor.
Xi = (7)
U Experimental design matrix constructed according to stan-
Ui,max + Ui,min dard order rule was given in Table 2 with the codified values
Ui,average = (8) of the factors. Experiments were executed randomly to provide
2
protection against the extraneous factors, which could effect the
Ui,max − Ui,min measured response. Randomization of experiments was made
Ui = (9)
2 by using Gnumeric program of Linux operation system. In all

Table 1
Natural and codified values of factors
Natural Codified

Factors Max Center Min Max Center Min

Agitation rate (rpm) 250 150 50 +1 0 −1

Temperature (◦ C) 40 30 20 +1 0 −1
Sodium hydroxide concentration (mol L−1 ) 0.1 0.055 0.01 +1 0 −1
Ethyl acetate concentration (mol L−1 ) 0.1 0.055 0.01 +1 0 −1
 N. Bursali et al. / Chemical Engineering and Processing 45 (2006) 980–989 983

Table 2 Table 3
Experimental design matrix for full two-level factorial design with three center FCCC experimental design matrix
points
N X3 X4 XA
N Random Standard X1 X2 X3 X4 XA
order order 1 −1 −1 0.11
2 −1 −1 0.6
1 0 0 0 0 0.847 3 −1 −1 0.36
2 1 10 1 −1 −1 1 0.946 4 −1 −1 0.65
3 2 6 1 −1 1 −1 0.1 5 +1 −1 0.1
4 3 9 −1 −1 −1 1 0.97 6 +1 −1 0.1
5 4 1 −1 −1 −1 −1 0.11 7 +1 −1 0.0
6 5 13 −1 −1 1 1 0.79 8 +1 −1 0.1
7 6 5 −1 −1 1 −1 0.1 9 −1 +1 0.97
8 7 7 −1 1 1 −1 0.0 10 −1 +1 0.95
9 8 11 −1 1 −1 1 0.92 11 −1 +1 0.92
10 0 0 0 0 0.847 12 −1 +1 0.99
11 9 8 1 1 1 −1 0.1 13 +1 +1 0.79
12 10 4 1 1 −1 −1 0.65 14 +1 +1 0.86
13 11 15 −1 1 1 1 0.9 15 +1 +1 0.9
14 12 14 1 −1 1 1 0.86 16 +1 +1 0.98
15 13 12 1 1 −1 1 0.99 17 0 0 0.85
16 14 16 1 1 1 1 0.98 18 0 0 0.85
17 15 2 1 −1 −1 −1 0.6 19 0 0 0.67
18 16 3 −1 1 −1 −1 0.36 20 −1(−α* ) 0 0.93
19 0 0 0 0 0.67 21 +1(+α* ) 0 0.55
22 0 −1(−α* ) 0.18
23 0 +1(+α* ) 0.95
experimental runs, the reaction time was considered as 5 min. To 24 0 0 0.77
estimate experimental error and to detect the quadratic effects
of factors three center point runs at the beginning, middle and
end of the 16 experiments were placed in the design matrix. In In the above equations SSeffect is called the sum of square of a
this table the last column represents the experimental data. factor effect and is calculated by using Yates’ Algorithm. SSerror
represents the sum of squares of error and it is calculated from
2.5. RSM-FCCC experimental design the replicated center point experiments as given below:

One of the RSM designs called face-centered central com- SSerror = (yC,i − ȳC )2 (13)
posite (FCCC) design was used to identify the second-order center point
polynomial model [2]. Design matrix given in Table 2 for four
factors for screening experiment was used to construct the FCCC Also center point data are used to investigate the existence of
design matrix for only two factors by augmenting adding star the curvature over the region of exploration [1,2]. For a single-
points (±α* ). “α* ” value is given ±1 for FCCC design for two degree-of-freedom, sum of squares for curvature is given by Eq.
factors [2]. FCCC Design matrix was given in Table 3. (14):

2.5.1. Analysis of variance (ANOVA) nF nC (ȳF − ȳC )2

SScurvature = . (14)
The method used to compare the magnitude of estimated nF + n C
effects of factors with the magnitude of experimental error is
called analysis of variance (ANOVA). If the magnitude of a 2.5.2. Graphical residuals analysis
factor effect is large when compared with experimental error, it Graphical residual analysis is an important and efficient
is decided that the changes in the selected response cannot occur model validation test method and was used to check the ade-
by chance and those changes in the response can be considered quacy of the constructed models in this study [2]. The objective
to be the effects of the factors. The factors causing a variation in behind this method was to check the following assumptions that
the response are called significant. In this study, Fisher’s F-test made at the beginning stage of the experimental designs.
was used in the analysis of variance and given in Eq. (10):
MSSeffect 1. Errors are normally distributed with mean zero and constant
F= (10) variance.
MSSerror
2. Errors are random and distributed independently.
SSeffect
MSSeffect = (11)
d.f.1
Design-Expert 6.0.10 trial software (Stat-Ease Inc., Min-
SSerror neapolis, USA) was used in this study to examine the residual
MSSerror = (12)
d.f.2 graphics [12].
984 N. Bursali et al. / Chemical Engineering and Processing 45 (2006) 980–989

Table 4
Application of Yates’ Algorithm to determine factor effects and sum of squares
Treatment combination Y (1) (2) (3) (4) (4) ÷ 24-1 Estimate of effect (4)2 (4)2 ÷ 24 SSeffect

(1) 0.11 0.71 1.72 2.02 9.57 1.17 – –

X1 0.6 1.01 0.3 7.35 1.076 0.134 1.16 0.0725
X2 0.36 0.2 3.82 0.88 0.43 0.054 0.185 0.0115
X1 X2 0.65 0.1 3.53 0.196 0.004 5 × 10−4 1.6 × 10−5 1 × 10−6
X3 0.1 1.91 0.78 0.2 −1.71 −0.21 2.92 0.1825
X1 X3 0.1 1.91 0.1 0.23 −0.576 −0.072 0.33 0.02
X2 X3 0.0 1.65 0.046 −0.1 −0.17 −0.02 0.029 1.8 × 10−3
X 1 X2 X3 0.1 1.88 0.15 0.104 0.216 0.027 0.047 2.53 × 10−3
X4 0.97 0.49 0.3 −1.42 5.33 0.67 28.41 1.77
X 1 X4 0.946 0.29 −0.1 −0.29 −0.684 −0.086 0.47 0.03
X 2 X4 0.92 0 0 −0.68 0.03 3.75 × 10−3 9 × 10−4 5.62 × 10−5
X 1 X2 X4 0.99 0.1 0.23 0.104 0.204 0.026 0.0416 2.6 × 10−3
X 3 X4 0.79 0.024 −0.2 −0.4 1.13 0.14 1.28 0.08
X 1 X3 X4 0.86 0.07 0.1 0.23 0.784 0.098 0.615 0.039
X2 X3 X4 0.9 0.07 0.094 0.3 0.63 0.079 0.4 0.025
X1 X2 X3 X4 0.98 0.08 0.01 −0.084 −0.384 −0.048 0.147 9.18 × 10−3

3. Results and analysis region were found to be significant at the α = 0.1 level. At this
stage, it was decided to construct and identify a second-order
3.1. Screening experiment polynomial model.

The data given in Table 2 was used to estimate the main and
interaction effects and sum of squares of the factors by using
Yates’ Algorithm as illustrated in Table 4.
To compare the magnitude of the curvature in the studied
region, sum of squares of curvature was calculated from Eq.
(14). The average values of 3 runs at the center point and 16 fac-
torial points of the design were calculated being 0.788 and 0.586,
respectively. Sum of square of curvature was calculated being
0.103 by using Eq. (14). To compare the magnitude of estimated
factor effects given in Table 4 and curvature with the experimen-
tal error, variance analysis was made and given in Table 5.
As can be seen from Table 5, main effects of initial concentra-
tions of NaOH and EtOAc and also the curvature over the studied
3.2. RSM and FCCC design
A second-order polynomial model between significant fac-
tors and response was selected to illustrate the dependence of
response on the significant factors as given in Eq. (15):
Y = b0 + b33 X32 + b44 X42 + b3 X3 + b4 X4 + b34 X3 X4 (15)
To estimate the model parameters, FCCC design matrix was
constructed as given in Table 3. Since temperature and agitation
rate were found to be insignificant factors at the end of the
screening phase, 16 screening experiments were considered
as a full two-level factorial experimental design for 2 factors

Table 5
Variance analysis of factor effects
Sources of SSeffect d.f. MSSeffect F Critical F-value
variation

X1 0.0725 1 0.0725 6.9 F0.1(1,2) = 8.526

X2 0.0115 1 0.0115 1.1
X3 0.1825 1 0.1825 17.38
X4 1.77 1 1.77 168.6
X1 X2 1.10−6 1 1.10−6 9.52 × 10−5
X1 X3 0.02 1 0.02 1.9
X1 X4 0.03 1 0.03 2.85
X2 X3 1.8.10−3 1 1.8 × 10−3 0.17
X2 X4 5.62 × 10−5 1 5.62 × 10−5 5.35 × 10−3
X3 X4 0.08 1 0.08 7.6
X1 X2 X3 2.53 × 10−3 1 2.53 × 10−3 0.24
X1 X2 X4 2.6 × 10−3 1 2.6 × 10−3 0.25
X1 X3 X4 0.039 1 0.039 3.71
X2 X3 X4 0.025 1 0.025 2.38
X1 X2 X3 X4 9.18 × 10−3 1 9.18 × 10−3 0.87
Curvature 0.103 1 0.103 9.8
Error 0.021 2 0.0105
Total 2.37 18
 N. Bursali et al. / Chemical Engineering and Processing 45 (2006) 980–989 985

Table 6
Variance analysis of Eq. (16)
Source of SSeffect d.f. MSSeffect F-value Critical F-value
variation

Model 2.55 5 0.51 36.42 F0.1(5,18) = 2.19

X3 0.24 1 0.24 17.14
X4 2.07 1 2.07 147.86 F0.1(1,18) = 3.007
X32 1.4 × 10−4 1 1.4 × 10−4 0.01
X42 0.1 1 0.1 7.14 F0.1(3,15) = 2.49
X3 X4 0.079 1 0.079 5.64
Residuals 0.25 18 0.014
Lack-of-fit 0.020 3 0.0067 0.45
Pure error 0.23 15 0.015
Total 2.8 23

replicated 4 times and were used in the design of FCCC. By

using least square parameter estimation method, factor effects
were identified as given by Eq. (16):
Y = 0.77 − 0.12X3 + 0.34X4 − 6.6 × 10−3 X32
− 0.18X42 + 0.07X3 X4 (16)
The variance analysis of Eq. (16) was made in Table 6. The
quadratic effect of NaOH and lack-of-fit term were insignificant
at the α = 0.1 significance level. The quadratic effect was
ignored and included in lack-of-fit term. The second-order
model was rearranged as given by Eq. (17):
Y = 0.77 − 0.12X3 + 0.34X4 − 0.18X42 + 0.07X3 X4 (17)
Variance analysis of Eq. (17) was given in Table 7. The
lack-of-fit term was not significant at the α = 0.1 significance
level after moving X32 term and Eq. (17) was found to be
adequate to represent our data with calculated 0.9 R2 value.
The validation of Eq. (17) was checked by using graphical
residual analysis. The residuals were calculated from the fol- Fig. 2. Normal graph of residuals.
lowing equation:
residuals = Yi,obser − Yi,pred (18)
In Fig. 2 normal graph of residuals was given. As can be seen all
residuals lie on a straight line which shows that residuals were
distributed normally. But only one residual deviated from the
straight line.
In Fig. 3, the predicted responses were plotted versus residu-
als to check the assumption that the random errors are distributed

Table 7
Variance analysis of Eq. (17)
Source of SSeffect d.f. MSSeffect F-value Critical F-value
variation

Model 2.54 4 0.635 48.84 F0.1(4,19) = 2.266

X3 0.24 1 0.24 18.46
X4 2.07 1 2.07 159.2 F0.1(1,19) = 2.99
X42 0.15 1 0.15 11.54
X3 X 4 0.079 1 0.079 6.08 F0.1(4,15) = 2.361
Residuals 0.25 19 0.013
Lack-of-fit 0.020 4 0.005 0.34
Pure error 0.23 15 0.015
Total 2.8 23
Fig. 3. The predicted responses vs. residuals.
986 N. Bursali et al. / Chemical Engineering and Processing 45 (2006) 980–989

Table 8
Variance analysis of Eq. (19)
Source of SSeffect d.f. MSSeffect F-value Critical F-value
variation

Model 2.4 4 0.6 98.35 F0.1(4,14) = 2.395

X3 0.33 1 0.33 54.1
X4 1.68 1 1.68 275.40 F0.1(1,18) = 3.007
X42 0.12 1 0.12 19.67
X3 X 4 0.14 1 0.14 22.95
Residuals 0.11 18 0.0061
Lack-of-fit 0.023 4 0.0057 0.81
Pure error 0.098 14 0.007
Total 2.5 22

The variance analysis of Eq. (19) was given in Table 8. From

Fig. 4. Graph of residuals vs. sodium hydroxide concentration levels.
with mean zero and constant variance. All residuals were dis-
tributed between −1.6 and +1.6 without having any systematic
structure but only one residual was different from the others.
In Figs. 4 and 5, the graphs of residuals versus NaOH
and EtOAc concentrations levels were given respectively. From
those figures, an outlier data was determined at minimum levels
of both factors. Except that outlier, random errors distributed
independently from the factor levels.
From the residual graphs, the outlier data was determined
being the first experiment in Table 3 and was ignored. At this
stage the quadratic model parameters were recalculated and
given by Eq. (19):
Y = 0.77(±0.034) − 0.14(±0.02)X3 + 0.32(±0.02)X4
+ 0.16(±0.039)X42 + 0.096(±0.021)X3 X4
Table 8 it was clear that the lack-of-fit term was not important
after moving outlier data. The standard error associated with
each coefficient was calculated by using Design Expert 6.0.10
trial software.
The validation of Eq. (19) was reanalyzed by using residuals.
The normal graph of residuals, residuals versus predicted values
and factor levels were given in Figs. 6–9, respectively. Since
the residuals were distributed normally with constant variance,
mean zero and independently, it was concluded that Eq. (19) was
good to fit the experimental data. In other words, the underlying
assumptions about the errors were satisfied. The calculated R2
value for Eq. (19) was 0.95.
At the end of the study, response surface contours were
obtained by using Eq. (19) at different levels of initial NaOH
and EtOAc concentrations and given in Fig. 10. The maxi-
mum predicted XA was obtained about 100% at approximately
0.01 mol L−1 NaOH and 0.1 mol L−1 CH3 COOC2 H5 concen-
trations. The numbers on the corner and edge sides of Fig. 10
shows the number of experimental runs that made at these points
as can be seen from Table 3. At the bottom side on the left corner
(19)

Fig. 5. The graph of residuals vs. ethyl acetate concentration levels. Fig. 6. Normal graph of residuals of Eq. (19).
 N. Bursali et al. / Chemical Engineering and Processing 45 (2006) 980–989 987

Table 9
Results of confirmatory experiments
Number of NaOH concentration EtOAc concentration Experimental Predicted from
experiment (mol L−1 ) (mol L−1 ) Eq. (19)

1 0.03 1 0.93 0.95

2 0.07 1 0.91 0.91
3 0.01 1 0.92 0.92
4 0.01 1 0.94 0.92

Fig. 7. Graph of residuals vs. predicted values from Eq. (19). Fig. 9. Graph of residuals vs. ethyl acetate concentration levels.

the number of experiment was 3 since one of the experimental

run at minimum levels of NaOH and EtOAc concentration was
found to be an outlier and removed from the experimental data.
Some experiences are realized to verify the optimum pre-
dicted values by the model. In order to be sure about the fitness
of predicted and observed responses, four confirmatory experi-
ments were made and given in Table 9.

Fig. 10. Response surface contours.

4. Conclusions and discussion

The objective of this study was to improve the saponifica-

tion process with respect to both operational efforts, consumed
raw materials and obtained yield. As a result of the screen-
ing experiment, temperature and agitation rate were found to
Fig. 8. Graph of residuals vs. sodium hydroxide concentration levels. be insignificant operating parameters whose selected maximum
988 N. Bursali et al. / Chemical Engineering and Processing 45 (2006) 980–989
and minimum ranges were 20–40 ◦ C and 50–250 rpm, respec-
tively in this study. To reach the maximum XA , initial NaOH
and EtOAc concentration values were determined as 0.01 and
0.1 mol L−1 , respectively by using RSM. These experimental
results were really more reliable and applicable for us to operate
the underlying process under these conditions.
In this study, two types of experimental design methods
were used for screening and response surface identification.
Both methods were applied successively, which was one of the
advantage of the application of statistical experimental design
technique. Both methods were found to be efficient and easy to
determine important operating parameters, optimum conditions
and to interpret the results of experiments and to improve the
process in a short time. At the screening experimentation stage,
effects of the factors were determined by using Yates’ Algo-
rithm [1,2,10]. Temperature and agitation rate were found to
be the insignificant factors and initial NaOH and EtOAc con-
centrations and their quadratic effects were found to be the
significant factors. At the second phase, process optimization
was proposed and RSM was applied by using face-centered cen-
tral composite (FCCC) design [1,2]. A second-order polynomial
model was identified by using least-square parameter estimation
method. The validation of quadratic model was confirmed by
using graphical residual analysis such as normal graph of resid-
uals, graphs of residuals versus predicted values and factors [2].
It was concluded that the second-order model was adequate to
fit the data.
When the objective of an experiment is to make a screen
between the factors, which are less than 5, Full two-level fac-
torial and fractional factorial experimental designs are recom-
mended. There are three different response surface methods.
These are central composite circumscribed (CCC) design, cen-
tral composite inscribed (CCI) design and face-centered central
composite design (FCCC) [2]. In this study, fractional facto-
rial design was not selected since some of the factor effects
could have been confused. FCCC design was chosen, since the
selected minimum levels of factors was not suitable to apply the
other central composite designs. The level selection at the brain-
storming stage is really very important, especially in the study
of process improvement.
Graphical residual analysis was applied in this study to check
the adequacy of the equations and the underlying assumptions,
to determine the outlier data, to have more reliable and accu-
rate results. There are different numerical methods to test the
equation adequacy such as R2 statistic, which is the measure of
the variance in the response explained by the model. But these
numerical methods does not merely guarantee the agreement of
the model with the data as can be seen from our results. In addi-
tion to numerical methods, graphical residual analysis must be
done. The calculated R2 value was 0.9 before graphical resid-
ual analysis. With the result of graphical analysis, calculated R2
value was improved. In this study calculated R2 value was 0.95
for Eq. (19). This meant that our final equation could explain
the 95% of variation in the response. Our graphical and numer-
ical results were in agreement with each other. On the basis of
the confirmatory runs and Table 9, it is noted that Eq. (19) fits
the experimental data and explains the examined saponification
process. The maximum predicted conversion rate was calculated
0.98 from Eq. (19) at determined optimum condition.
On the basis of Eq. (19) and Fig. 10 it was concluded that
absolute value of linear effect of EtOAc was more significant
than its quadratic effect. The negative quadratic effect of EtOAc
was pronounced above 0.055 mol L−1 by indicating the decrease
in conversion rate with the increasing EtOAc concentration.
The positive linear effect of EtOAc was pronounced below the
0.055 mol L−1 and causing an increase in conversion rate. The
mutual effect of NaOH and EtOAc was found to be the least
significant effect of all the other factor effects. However, its F-
value was found to be 22.95 by indicating a significant effect of
mutual effect. The mutual effect can be seen in an elliptical shape
in proportional with its significance in Fig. 10. The negative first
order effect of NaOH was clear especially at higher concentra-
tion of it. Increasing NaOH concentration from 0.01 mol L−1 to
the 0.1 mol L−1 decreased the conversion rate.
In the hydrolysis of esters, the alkali plays a role as a catalyst.
So, it is expected to find that the reaction occurs at very low OH−
concentration. In Fig. 10, low NaOH concentration was utilized
to obtain 0.98 predicted NaOH conversion. This optimal value is
valid between 0.01 and 0.1 mol L−1 NaOH concentration levels.
In this study to provide a general confidence and harmony
in the whole of the statistical analysis, significance level was
not changed and selected 0.1 and was determined at the brain
storming stage of this study by considering used experimental
methods, apparatus and examined reaction.
Acknowledgement
The authors gratefully acknowledge Ankara University Sci-
entific Research Projects for providing financial support (Grant
No. 2004 07 45 024).
Appendix A. Nomenclature
bi regression coefficient
C specific conductivity at time t (␮S)
CEtOAc ethyl acetate concentration (mol L−1 )
CNaOH sodium hydroxide concentration at time t (mol L−1 )
CNaOH,0 sodium hydroxide concentration at time t = 0
(mol L−1 )
CNaOH,∞ sodium hydroxide concentration at time t = ∞
(mol L−1 )
C0 specific conductivity at time t = 0 (␮S)
C∞ specific conductivity at time t = ∞ (␮S)
d.f.1 degrees of freedom for effect
d.f.2 degrees of freedom for error
F Fisher’s F-value
k second-order rate constant (L mol−1 s−1 )
MSSerror mean sum of square of error
MSSeffect mean sum of square of effect
n replication number
nC number of the center point experiment
nF number of factorial design point
N number of experiment
 N. Bursali et al. / Chemical Engineering and Processing 45 (2006) 980–989 989

−rEtOAc ethyl acetate consumption rate (mol L−1 s−1 ) References

−rNaOH sodium hydroxide consumption rate (mol L−1 s−1 )
SScurvature sum of square of curvature [1] D.C. Montgomery, Design and Analysis of Experiments, 3rd ed., John
Wiley & Sons, New York, 1991, pp. 270–569.
SSerror sum of square of error [2] NIST/SEMATECH e-Handbook of Statistical Methods, 2005.
SSeffect sum of square of effect http://www.itl.nist.gov/div898/handbook/.
Ui,average average value of ith factor [3] J.M. Carr, E.A. McCracken, Statistical program planning for process
Ui,max maximum value of ith factor development, Chem. Eng. Prog. 56 (11) (1960) 56–61.
Ui,min minimum value of ith factor [4] E.E. Lind, J. Goldin, J.B. Hickman, Fitting yield and cost response
surfaces, Chem. Eng. Prog. 56 (11) (1960) 62–68.
Ui,natural natural value of ith factor [5] C.-P. Xu, S.-W. Kim, H.-J. Hwang, J.-W. Yun, Application of statistically
XA fractional conversion rate of sodium hydroxide based experimental designs for the optimization of exo-polysaccharide
Xi codified value of ith factor production by Cordyceps milltaris NG3, Biotechnol. Appl. Biochem. 36
X1 codified value of agitation rate (2002) 127–131.
X2 codified value of temperature [6] M.J. Anderson, H.P. Anderson, Applying DOE to microwave popcorn,
Process Ind. Quality (1993) 30–32.
X3 codified value of initial NaOH concentration [7] S.E. Kruger, I.C. Silva, J.M.A. Rebello, Factorial design of experiments
X4 codified value of initial EtOAc concentration applied to reliability assessment in discontinuity mapping by ultrasound,
ȳC average response at the center point NDT Net 3 (11) (1998).
yC,i ith response at the center point [8] B.V. Mehta, H. Ghulman, R. Gerth, Extrusion die design: a new method-
ȳF average response of the factorial design points ology of using design of experiments as a precursor to neural networks,
JOM-e 51 (9) (1999).
Y process response [9] E.M. Terry, J. Stieglitz, J. Am. Chem. Soc. 49 (1927) 2216–2222.
Yi,obser. ith observation in experimental data [10] D. Cuthbert, Applications of Statistics to Industrial Experimentation,
Yi,pred. ith predicted value of response from fitted model John Wiley & Sons, New York, 1976, pp.17–20, 72–126.
Ui incremental value of ith factor [11] http://www.uni-regensburg.de/Facultaeten/nat Fak IV/Organische-
Chemie/Didaktik/Keusch/cassy est-e.htm.
[12] http://www.statease.com, Minneapolis, USA, 2005.
Greek letters
α significance level
α* star point of FCCC design