Escolar Documentos
Profissional Documentos
Cultura Documentos
k
j =i + 1
b
ij
X
i
X
j
1
where Y is the predicted response (percent As(V) removal), b
0
is the
offset term and b
i
is the linear effect while b
ii
and b
ij
are the square
and interaction effects, respectively. X
i
and X
j
are the coded values of
independent variables and k is the number of the independent
variables. The software Design Expert (Version 7.0.0, Stat-Ease Inc.,
Minneapolis, USA) and Statistica 5.0 were used for the experimental
design, determination of the coefcients, the data analysis, and the
graph plotting. The validity of the model and the reliability of the
statistical experimental strategies were determined by comparing the
experimental and predicted values. Furthermore, the predicted
arsenic removal efciencies from response function were experimen-
tally veried in an additional study.
3. Results and discussion
3.1. Optimization of parameters for As(V) removal
The removal of As(V) from drinking water was investigated by
precipitation/coprecipitation method using aluminum sulfate as a
coagulant. The effects of process variables such as initial arsenic
concentration, coagulant dose, and pH on the arsenic removal
efciency were investigated using RSM according to BBD. Different
response functions such as linear, interactive, quadratic and cubic
models were used to correlate the experimental data and to obtain the
regression equation. To decide about the adequacy of the models to
represent arsenate removal by aluminum sulfate, three different tests
namely sequential model sum of squares, lack of t tests, and model
summary statistics were carried out in the present study and the
results are presented in Table 3. Cubic model was not recommended
for this system due to insufcient points to estimate the coefcients
for this type of the model. Sequential model sumof squares and model
summary statistics indicated that the quadratic model provided the
best t to experimental data with the lowest standard deviation, the
highest correlation coefcient, adjusted R
2
, predicted R
2
values, and
the lowest p value. Therefore, the quadratic model was chosen for
further analysis.
Experimental data were used for determining of the response
function (Eq. (1)) coefcients for each independent variable. The
response function with the determined coefcients for percent
arsenate removal (Y) is presented by Eq. (2).
Y =172:377 + 64:209 10
3
X
1
+ 1:513X
2
+ 54:251X
3
71
10
6
X
1
X
2
+ 14:340 10
4
X
1
X
3
+ 14:644 10
2
X
2
X
3
5:850 10
5
X
2
1
1:910 10
2
X
2
2
4:113X
2
3
:
2
The predicted values of percent As(V) removal were determined
by the response function with the obtained coefcients. The
experimental results and the predicted values of percentage arsenate
removal using aluminum sulfate are presented in Table 2. The high
value of regression coefcient (R
2
=0.9726) indicates that the model
predictions t satisfactorily with the experimental observations.
The statistical signicance of the response function was checked
by F-test, and the ANOVA results for response surface quadratic model
and model terms are summarized in Table 4. The model F value and
Table 2
The BBD with coded/actual values for the three independent variables and the
comparison of observed and predicted As(V) removal efciency.
Experiment
no.
Actual and coded levels of variables As(V) removal
efciency (%)
Initial As(V)
concentration (X
1
)
Coagulant
dose (X
2
)
pH (X
3
) Observed Predicted
1 1 (1000) 0 (40.25) 1 (4.0) 37.30 40.83
2 1 (10) 1 (80.00) 0 (6.5) 90.00 82.08
3 0 (505) 0 (40.25) 0 (6.5) 95.25 95.49
4 1 (1000) 0 (40.25) 1 (9.0) 84.40 81.40
5 0 (505) 0 (40.25) 0 (6.5) 95.45 95.49
6 0 (505) 0 (40.25) 0 (6.5) 95.64 95.49
7 0 (505) 0 (40.25) 0 (6.5) 94.65 95.49
8 1 (1000) 1 (0.50) 0 (6.5) 14.70 22.62
9 0 (505) 1 (0.50) 1 (9.0) 13.07 8.15
10 1 (10) 0 (40.25) 1 (9.0) 70.00 66.47
11 0 (505) 0 (40.25) 0 (6.5) 96.44 95.49
12 1 (10) 1 (0.50) 0 (6.5) 0.00 8.45
13 0 (505) 1 (0.50) 1 (4.0) 11.68 0.23
14 1 (10) 0 (40.25) 1 (4.0) 30.00 33.00
15 0 (505) 1 (80.00) 1 (9.0) 96.63 100.00
16 0 (505) 1 (80.00) 1 (4.0) 37.03 41.95
17 1 (1000) 1 (80.00) 0 (6.5) 99.10 90.65
Table 3
Adequacy of the model tested for arsenate removal.
Source Sum of squares df Mean square F pNF Remark
Sequential model sum of squares
Mean 66261.33 1 66261.33
Linear 13033.18 3 4344.39 5.79 0.0097
2FI 867.54 3 289.18 0.33 0.8073
Quadratic 8271.32 3 2757.11 30.95 0.0002 Suggested
Cubic 621.89 3 207.30 490.72 b0.0001 Not recommended
Residual 1.69 4 0.42
Total 89056.95 17 5238.64
Lack of t tests
Linear 9760.75 9 1084.53 2567.35 b0.0001
2FI 8893.20 6 1482.20 3508.75 b0.0001
Quadratic 621.89 3 207.30 490.72 b0.0001 Suggested
Cubic 0.00 0 Not recommended
Pure error 1.69 4 0.42
Source Model summary statistics
Std. dev. R
2
Adjusted R
2
Predicted R
2
Press Remark
Linear 27.40 0.5717 0.4729 0.3347 15165.48
2FI 29.82 0.6098 0.3757 0.0438 23793.09
Quadratic 9.44 0.9726 0.9375 0.5634 9952.80 Suggested
Cubic 0.65 0.9999 0.9997 Not recommended
44 M. Bilici Baskan, A. Pala / Desalination 254 (2010) 4248
very low probability value (0.0001) indicated that the model was
statistically signicant and model equation can adequately be used to
describe the arsenate removal under a wide range of operating
conditions. The p values are used to estimate whether F is large
enough to indicate statistical signicance and used to check the
signicance of each coefcient. p values lower than 0.05 indicate that
the model and model terms are statistically signicant [48]. In this
case, the model equation and all regression coefcients are signicant
except for one linear coefcient (X
1
) and two cross-product coef-
cients (X
1
X
2
and X
1
X
3
). Although they are nonsignicant, they are
still considered in Eq. (2) because it is a hierarchical model. Fur-
thermore the study of contour plots and the regression equation
highlight certain interactions between the selected parameters as
well as their individual effects on arsenate removal. The analysis
showed that the form of the model chosen to explain the relationship
between the factors and the response is correct and can be used to
navigate the design space.
The high value of R
2
(0.9726) indicated a high dependence and
correlation between the observed and the predicted values of response.
The value of adjusted R
2
(0.9375) suggested that the total variation of
about 94% for arsenate removal was attributed to the independent
variables and only about 6%of the total variationcannot be explained by
the model. The value of the adequate precision is a measure of the signal
(response) to noise (deviation) ratio. A ratio greater than 4 is desirable.
In the present study, the ratio was found to be 14.899, which indicates
the adequate signal. Therefore the quadratic model can be used for the
arsenate removal by coagulation process.
Variations of percent arsenate removals with the initial arsenate
concentration at different pH and constant aluminum sulfate
concentration of 40 mg L
1
are depicted in Fig. 1. pH value was
prominently effective for arsenate removal efciency. As it can be
seen in Fig. 1, percent arsenate removal increased with increasing pH
and the optimum pH range for arsenate removal by coagulation was
68. In the high acidic pH, the arsenate removal efciency declined
sharply. Similarly, high alkaline pH above 8 caused lower efciencies.
The obtained optimum pH range fromthis study is in good agreement
with the values given in the literature [28,38]. The effect of the pH
range on the arsenate removal efciency is related to the solubility of
the amorphous hydroxide solid of aluminum. Arsenate removal ef-
ciency increased in the pH range of 6 to 8 because of the fact that the
amorphous hydroxide solid is stable in this pH range.
According to the results of the BBD method for aluminum sulfate,
arsenate removal was highly pH-dependent and the optimum pH
range was 68. Fig. 2 shows the variations of percent arsenate re-
movals with initial arsenate concentrations at different Al
2
(SO
4
)
3
concentrations and constant pH of 7. Aluminum sulfate concentration
is important in determining the level of arsenic removal as shown in
gure. When aluminum sulfate is added to arsenic contaminated
water, aluminum sulfate dissociates and forms aluminum hydroxide,
which coprecipitates with arsenic. The possible chemical equations of
aluminum sulfate coagulation are as follows [49,50]:
Al
2
SO
4
3
:18H
2
O2Al
3
3SO
2
4
18H
2
O Aluminum sulfate dissolution
2Al
3
6H
2
O2AlOH
3
6H
Aluminum precipitation
H
2
AsO
4
AlOH
3
AlAs complex Coprecipitation
During coagulation, arsenic is removed through three main mech-
anisms including precipitation, coprecipitation, and adsorption. The
formation of the insoluble compounds Al(AsO
4
) is seen in rst step. In
coprecipitation, the soluble arsenic species are incorporated into a
growing metal hydroxide phase via inclusion, occlusion, or adsorp-
tion. Finally, the soluble arsenic is binded electrostatically to the
external surfaces of the insoluble metal hydroxide. All three of these
mechanisms can independently contribute towards arsenic removal.
Table 4
ANOVA results of the second-order polynomial equation for percentage arsenate
removal using aluminum sulfate.
Source Sum of squares df Mean square F value p value
Model 22172.04 9 2463.56 27.65 0.0001
X
1
258.78 1 258.78 2.90 0.1321
X
2
10033.07 1 10033.07 112.63 b0.0001
X
3
2741.33 1 2741.33 30.77 0.0009
X
1
X
2
7.84 1 7.84 0.088 0.7753
X
1
X
3
12.60 1 12.60 0.14 0.7180
X
2
X
3
847.10 1 847.10 9.51 0.0177
X
1
2
867.86 1 867.86 9.74 0.0168
X
2
2
3834.89 1 3834.89 43.05 0.0003
X
3
2
2781.93 1 2781.93 31.23 0.0008