Vous êtes sur la page 1sur 2

S H O RT C O M M U N I C AT I O N S

Fresenius J Anal Chem (2000) 167 : 201–203 – © Springer-Verlag 2000

J. R. Luna · J. F. Ovalles · A. León · M. Buchheister


A factorial design for optimizing
a flow injection analysis system

Received: 16 June 1999 / Revised: 22 November 1999 /


Accepted: 24 November 1999 Fig. 1 Representation of the flow-injection manifold employed

Abstract The use of a factorial design for the response explo- Experimental
ration of a flow injection (FI) system is described and illus-
trated by FI spectrophotometric determination of paraquat. Reagents. All reagents used were of analytical grade. A stock
Variable response (absorbance) is explored as a function of the solution of paraquat (0.1 µg/mL) was prepared by dissolving
factors flow rate and length of the reaction coil. The present paraquat dichloride (Sigma, St. Louis, MO) in deionized, dis-
study was found to be useful to detect and estimate any inter- tilled water. A working standard solution was prepared by ap-
action among the factors that may affect the optimal conditions propriate dilution of the stock solution. A sample solution of
for the maximal response in the optimization of the FI system, ascorbic acid reagent was prepared by dissolving 2.500 g of
which is not possible with a univariate design. In addition, this ascorbic acid (BDH), 0.100 g of potassium iodate (Merck) and
study showed that factorial experiments enable economy of ex- 1.000 g of EDTA, disodium salt (BDH) in deionized, distilled
perimentation and yield results of high precision due to the use water and the solution was diluted to the final volume of 1 L.
of the whole data for calculating the effects. This solution was stored in a refrigerator (5 °C) during the ex-
perimental work. A 1 mol/L sodium hydroxide (Merck) solu-
tion was prepared from the solid base.
Introduction
Apparatus. A SHIMADZU spectrophotometer model UV-1201
Flow injection analysis (FIA) represents an advancement of with a home-made glass flow cell was employed. A Cole-
wet chemical methods because it offers great advantages such Palmer (model Masterflex L/S) with two peristaltic pumps was
as flexibility, speed, high sampling rate and wide applicability employed to propel the reagents. A Rheodyne (model 7000) in-
in many fields [1–3]. However, the choice of the best set of op- jection valve was used for the sample injection. The Tygon tub-
erational conditions for assembling FIA systems is generally ing of 0.8 mm i.d. was used in all systems. All the connections
complex. The conventional method of optimization involves were made as indicated in Fig. 1. Statgraphics statistical soft-
varying each parameter in turn, while keeping the others con- ware package (version 5.1, Statistical Graphics) was used for
stant until an adequate response is obtained. This type of opti- data treatment.
mization procedure has some limitations, since it is not reason- Operation. The sample solution (a fixed volume, 60 µL) is in-
able to assume that the variables may be independent from each jected into the carrier stream of 1 mol/L sodium hydroxide so-
other. On the other hand, in most cases it may be too demand- lution (Factor FB). After injection, the sample and the ascorbic
ing in terms of experimental work [4–6]. acid reagent (Factor FA) emerge at a confluence point. Subse-
Today, as in chromatography, the optimization of FIA sys- quently, sample and reagent are forwarded by both streams (FA
tems on basis of statistically designed experiments has been and FB) to pass through a certain length of coiled tube (Factor
thoroughly developed and many strategies are available, such FC) where reaction takes place. Paraquat is reduced by ascor-
as the sequential simplex method, factorial design, central com- bic acid reagent to yield a highly colored free radical absorbing
posite design, etc. [7–15]. However, a factorial design has not at 600 nm. The variation of absorbance is thus continuously
been employed in order to find the precise relationship between monitored in the flow cell [16, 17].
the large number of parameters involved and the optimal re-
sponse for the measurement of paraquat by continuos-flow col- Procedures. For the present work the probable experimental re-
orimetry. Therefore, in this work the development of a totally gion is unknown. Thus to start the experiment, a tentative ex-
randomized factorial design has been used to detect and esti- perimental region was selected with the following minimum
mate any interaction among the factors (coil lengths and flow and maximum limits: 1 and 5 mL/min for both flow rates (FA
rates) that may affect the optimal conditions for the maximal and FB) and 30 and 100 cm for the reaction coil length (FC).
response (absorbance) in the flow injection system. Beginning with this tentative region various designs were stud-
ied and discussed to detect and estimate any interaction among
the factors that may affect the maximal response at each step of
optimization.
J. R. Luna (쾷)
Toxicology Laboratory, Faculty of Pharmacy,
Universidad de los Andes, Mérida, 5101, Venezuela Results and discussion
e-mail: lunajr@ciens.ula.ve
Design 1. A 3k fractional factorial design was selected; specifi-
J. F. Ovalles · A. León · M. Buchheister cally, a one-third fraction of a 33 design (or 33–1) which gives
Department of Analysis and Control, Faculty of Pharmacy, nine experimental points [18]. To the fractional factorial design
Universidad de los Andes, Mérida, 5101, Venezuela five points were added in order to cover the whole experimen-
202

Fig. 3 Plot of the relationship between three levels of factor


FB on the dependent variable mean absorbance following
Fischer’s LSD procedure

Fig. 2 Experimental regions in initial region = complete cube The corresponding analysis of variance (ANOVA) is pre-
and reduced region = shadow cube. The vertices are denoted by sented in Table 1. The probability values observed in all indi-
dots (쎲). The whole cube with vertexes in (1, 1, 30), (1, 5, 30), cate that eight effects are statistically significant (p < 0.05).
(5, 1, 30), (5, 5, 30), (1, 1, 100), (1, 5, 100), (5, 1, 100), (5, 5, However, the significant level due to lack-of-fit is < 0.05, a
100) and the shadow cube in (1, 1, 30), (1, 3, 30), (3, 1, 30), (3, non-significant value which indicates that the second-order
3, 30), (1, 1, 100), (1, 3, 100), (3, 1, 100), (3, 3, 100) model is not an adequate approximation to the data. This sug-
gests that a third design must be established.
tal region more uniformly. The obtained response variable was Design 3. A 22 factorial design with four additional central
arranged in decreasing order. The actual regions of highest ab- points was selected. This design will allow the approach to the
sorbance were as follows: Region 1: FA = low levels, FB = low corresponding response surface by a polynomial of first-order
levels and FC = high levels; Region 2: FA = low levels, FB = and to measure the lack-of-fit. The used center point was FA =
low levels and FC = low levels, and Region 3: FA = low levels, 0.45, FB = 0.6 and FC = 100.
FB = medium levels and FC = high levels. According to this In spite of the fact that the polynomial first-order model
classification, the highest response variable was found in re- considers the above three factors, this model was of little inter-
gion 1. Moreover, the corresponding analysis of variance est due to the domain in the variability of the factor FB, as
(ANOVA) of the data allowed rejection of about 75% of the shown in Fig. 3, which is based on Fisher’s least significant dif-
initial region as can be seen from the Figure 2. ference (LSD). Nevertheless, this result allowed separation of
Fisher’s LSD analysis in two groups of different data: one
Design 2. This new design was constructed using the Box and formed by the four experiments where the FB takes the level of
Behnken design [7, 13, 16], which allows estimation of the pa- 0.5 and the other one formed by the eight experiments where
rameters in a second-order model. In this case, the point of the the FB takes values of 0.6 and 0.7. The first group allowed con-
region 1 (FA = 1, FB = 1 and FC = 100) was used as the center sideration of a polynomial of first-order without the domain of
of the new design, with increments of ∆FA = 0.5, ∆FB = 0.5 the variability of the FB, since this was fixed at 0.5, where in-
and ∆FC = 20. deed the maximum absorbance is shown. The corresponding
polynomial function of first-order is given by: Absorbance =
(267.5) - (70.0FA) + (1.0FC). The estimated model showed a
Table 1 Anova table for the Box-Behnken designa downward trend for levels of FA and an upward trend for lev-
Effect SS d.f. MS F-ratio P-value els of FC when levels of FB were maintained constant. Under
the above conditions the variability of the absorbance was ex-
FA 7564.5 1 7564.5 22693.50 0.0000 plored by decreasing the factor FA from 0.50 to 0.20. This ap-
FB 220.5 1 220.5 661.50 0.0015 proach offered an optimal region (maximum absorbance) for
values varying from 0.45 to 0.30.
FC 4050.0 1 4050.0 12150.00 0.0001
FA*FB 9025.0 1 9025.0 27075.00 0.0000
FA*FC 16.0 1 16.0 48.00 0.0202 Conclusion
FB*FC 625.0 1 625.0 1875.00 0.0005
FA2 433.3 1 433.3 1300.00 0.0008 The present work shows the application of factorial designs on
FB2 3860.1 1 3860.1 11580.31 0.0001 a FIA system. The factorial optimization of the analytical sys-
FC2 0.1 1 0.1 0.31 0.6349 tem allows one to find the optimum conditions (two different
Lack of fit 1823.0 3 607.7 1823.00 0.0005 flow rates and one incubation coil length) for the maximal re-
Pure error 0.7 2 0.3 sponse (spectrophotometric absorbance for the measurement of
paraquat). This approach offers a double advantage over the
R-squared = 93.36% univariate design for system optimization. First, it allows one
R-squared (ajusted for d.f.) = 81.42% to study the simultaneous influence of three factor levels on the
Durbin-Watson statistic = 1.8094 response variable, and second, it enables economy of experi-
mentation to find effective knowledge of the system’s behav-
a The data adjusted to this model fit the function: Abs = ior. Since the use of factorial design was found to be well
416.667 – 184.833FA – 66.8333FB – 0.408 333FC + suited to the optimization of the FIA system, the possibility of
190.0FAFB + 0.2FAFC + 1.25FBFC – 43.3333FA2 – using it in other FIA systems is hypothesized.
129.333FB2 + 0.000 416 667FC2