Salehi2016 PDF
Salehi2016 PDF
Salehi2016 PDF
DOI 10.1007/s13369-016-2109-3
Abstract The easily separable and regenerable magnetic to indicate the prediction potential of CCD and ANN. The
activated carbon was synthesized for adsorption of toxic MAPE (%) of 0.59 and 0.38 was found for CCD and ANN
cationic dye, crystal violet, from aqueous solution. The syn- respectively. The obtained results indicated higher capabil-
thesized magnetic activated carbon was characterized by ity and accuracy of ANN in prediction. The experimental
SEM–EDX. The magnetic property of sorbent was evalu- data were found to be properly fitted to the Langmuir and
ated by VSM method. The obtained saturation magnetization Freundlich models which indicates that the sorption takes
of 41.56 emu g−1 showed facile separation of sorbent after place on a heterogeneous material and the sorption capacity
adsorption process. The effect of five parameters of pH, tem- of 12.59 mg g−1 was achieved.
perature, time, initial dye concentration and sorbent amount
on adsorption (%) were investigated. The percentage of Keywords Adsorption · Crystal violet · Magnetic activated
adsorption was mathematically described as a function of carbon · Central composite design · Artificial neural network
experimental parameters and was estimated by central com-
posite design (CCD). The maximum adsorption percent of
99.5 ± 0.2 was obtained experimentally which was close to 1 Introduction
the percent of CCD prediction of 99.90 %. The same design
was used for a three-layer artificial neural network model In recent years releasing toxic dyes by different industries
(ANN). The predicted data of CCD versus ANN showed the (textile, printing, cosmetic, food, and etc.) into water sources
linear agreement with regression value (R 2 ) of 0.9994 which such as rivers, lakes, and seas has been a considerable con-
confirmed the ideality of CCD and ANN. The results of two cern due to water pollution and aquatic living organism death.
models were compared in terms of coefficient of determi- Water pollution is one of the most threatening factors of
nation (R 2 ) and mean absolute percentage error (MAPE) human health. Aggregation of dyes in water causes cancer,
genetic mutation, digestive disorders, and poisoning. Crystal
violet (Scheme 1) is one of the most common cationic dyes
which is utilized in all dye industries. Crystal violet is very
Electronic supplementary material The online version of this
toxic to aquatic organisms. It is harmful to the embryo and
article (doi:10.1007/s13369-016-2109-3) contains supplementary
material, which is available to authorized users. can cause genetic mutation. Accumulation of crystal vio-
let in body can lead to gastrointestinal tract irritation with
B Abolfazl Semnani nausea, vomiting, hypermobility, abdominal pain, and respi-
a.semnani1341@gmail.com
ratory problems [1,2]. Therefore, due to the dangerous health
1 Department of Chemistry, Faculty of Science, Shahrekord effect of crystal violet, removal of such toxic dye becomes
University, P.O. Box 115, Shahrekord, Iran necessary. Various techniques such as oxidative degradation
2 Department of Mechanical Engineering, Najafabad Branch, [3], electrocoagulation [4], coagulation [5], membrane filtra-
Islamic Azad University, Isfahan, Iran tion [6] and ozonation [7] have been reported for toxic dyes
3 Department of Chemistry, Payame Noor University, removal from aqueous solutions. However, adsorption grabs
P.O.Box 19395-4697, Tehran, Iran scientists’ attention among all removal methods due to the
123
Arab J Sci Eng
2.1 Regents
advantages of simplicity, low-cost adsorbents, easy industri-
alization, facile accessibility. Zeolite, active carbon, alumina, All the chemicals [i.e., crystal violet (C25 N3 H30 Cl),
are widely used for removal of different dyes. Application of hydrochloric acid, sodium hydroxide, deionized water, iron
magnetic sorbents has made a revolution in adsorption tech- (II) sulfate heptahydrate, and activated carbon] used in this
niques in recent years due to the advantages of easier and study were analytical grade with the highest purity available
faster separation of sorbent from the sample solution by an and they were purchased from Merck.
external magnet. In this research the toxic cationic dye of
crystal violet was adsorbed by magnetic activated carbon. 2.2 Apparatus
In order to save time and money, statistical and math-
ematical models have been used to estimate and simulate The pH measurements were done with 86502-pH/ORP (AZ
the conditions in different processes especially in chemistry. Instrument, Taiwan) using a combined glass calomel elec-
Response surface methodology (RSM) and artificial neural trode. Ultrasonic bath (Sonorex digitec, DT255H, Germany)
network (ANN) are two powerful models which are widely and oven (Universal model UF 55, Germany) both were
applied to predict the processes in chemistry. Artificial neural utilized for magnetic activated carbon preparation. IKA
network is a representative of human brain system in which overhead stirrer (model RW 28 Digital Package, Germany)
the interconnected neurons and their exchange messages with glass stirrer was applied to stir the magnetic sorbent
between each other are considered. ANN is successfully sample solution during adsorption process. Ultrospec 3100
applied for both linear and nonlinear systems [8,9]. Among pro UV-Visible spectrophotometer (Amersham Biosciences,
all statistical models, ANN is one of the most prevalent USA) was carried out for data analyzing. A vibration-sample
models for nonlinear analysis. Determination of complicated magnetometer (Meghnatis Daghigh Kavir Co.) was used to
nonlinear relationships between input variables and predic- characterize the magnetic properties of synthesized magnetic
tion of output variables can be gained by ANN model. ANN activated carbon.
has been reported for adsorption of crystal violet [10]. RSM
deals with a group of mathematical and statistical techniques 2.3 Preparation of Magnetic Activated Carbon
in order to fit predicted model to the experimental data
obtained in relation to experimental design and obtain the Magnetic activated carbon was prepared according to the
optimum parameters. RSM is a powerful method for predic- proposed procedure [20]. Briefly, 2.78 g of iron (II) sulfate
tion of large number of processes which is applied in two heptahydrate was dissolved in deionized water. Then 0.5 g of
common designs of Box–Behnken design and central com- fine activated carbon powder was added to the solution. Ten
posite design [11]. Many studies have been reported in which milliliters solution of NaOH 2.5 mol L−1 was added drop-
RSM and ANN both were applied to predict and optimize the wise to the mixture. The mixture was stirred at 100 ◦ C for
process [12–16]. The results in these studies showed higher 1 h. The obtained magnetic sorbent was washed with deion-
potential and ability of ANN in prediction and optimization ized water for several times. Finally, the obtained magnetic
comparing to RSM. For instance, Khayet et al. [17] applied activated carbon was dried at 50 ◦ C in the oven.
RSM and ANN modeling and optimization of desalination
process by reverse osmosis. They showed that RSM was not 2.4 Adsorption Procedure
efficient and useful the proposed method of reverse osmo-
sis over a wide range of salt concentration in feed solution. Ten milliliters of crystal violet solution was adjusted at the
However, ANN was completely and powerfully successful in proper values of pH, temperature, time, initial dye concen-
123
Arab J Sci Eng
the predicted conditions of CCD experimental design runs. Y = b0 + bi xi + bii xi2 + bi j xi x j (4)
After the certified time, the magnetic activated carbon was i=1 i=1 i j
where m is the adsorbent mass (g) and v is the volume of In order to minimize the mean-squared error, training net-
adsorbate solution (L). work is applied on any ANN modeling. To this aim, feed
Finally the percentage of adsorption percentage can be forward back propagation (FFBP) was used as training algo-
stated according to the following equation (Eq. 3): rithm in this study. In this process, the final calculated outputs
are compared with the experimentally obtained outputs, and
Dye removal percentage (%) = (Ci − Ce ) /Ci × 100 (3) the errors are estimated. These errors are then propagated
backwards and used for adjusting the weights of neurons. It
2.5 Experimental Design Models should be mentioned that the weights of the trained network
are stored, and can be applied later for the outputs predic-
2.5.1 Response Surface Methodology tion of a different set of inputs. Thirty-two experimental data,
obtained from CCD, were applied as input data in ANN mod-
There are various kinds of statistical methods to analyze eling for five input variables of pH, temperature, time, initial
chemical processes. Some of these methods are used to dye concentration and amount of sorbent. Twenty-six and
predict the conditions in a process. Response surface method- six data points were derived to training and test sets respec-
ology (RSM) is one of the most common factorial designs tively. Data points in ANN were normalized in the range of
used for statistical analysis. RSM is based on the fit of empir- [−0.9, 0.9] by using tan-sigmoid transfer function as follows:
ical models to the experimental data obtained from experi-
mental design [11,22]. RSM is composed of two designs X norm = 1.8 (X − X min ) / (X max − X min ) − 0.9 (6)
of two designs of central composite design (CCD) and
Box–Behnken design (BBD). Experimental design of cen- where X is variable, X min , X max are minimum and maximum
tral composite design (CCD) as a partial factorial design was values respectively.
applied to predict the present study conditions. MINITAB
software version 16.0 was used for the experimental design
to optimize the dominant parameters of pH (in the range of 3 Results and Discussion
4.5–10.5 with 0.5 intervals), temperature (25–65 ◦ C with ten
intervals), time (5–45 min with ten intervals), initial dye con- 3.1 Characterization of Magnetic Activated Carbon
centration (5–45 mg L−1 with ten intervals) and the amount
of sorbent (15–75 mg with 15 intervals) on dye removal per- The SEM image of magnetic activated carbon is shown in
centage. The coded setting data for parameters were −2, −1, Fig. 1, which indicates the nanostructure of the sorbent. The
0, +1, +2. Three replications at the center of the design chemical composition of magnetic activated carbon, taken
were applied to determine the precision of the method. A by EDX, is shown in Fig. 2. The vibrating sample magne-
full quadratic equation was used to depict the interaction of tometer analysis (VSM) was applied to prove the magnetic
the factors in the process as the following (Eq. 4): properties and the required force to move magnetic activated
123
Arab J Sci Eng
Fig. 1 The SEM image of magnetic activated carbon. a 500 nm scale, b 200 nm
123
Arab J Sci Eng
Table 1 Experimental matrix for five-level–five factors CCD for adsorption of crystal violet
Exp. no. Factors Adsorption (%)
pH Temperature (◦ C) Time (min) Initial dye con. (mg L−1 ) MAC amount (mg) Obtained Predicted
CCD ANN
variables in the model are all <0.05, which are indicative predicted and experimental responses. The adjusted R 2 was
explanation for ideal prediction of CCD for the process with 98.74 %, which proved the competence of the developed
a linear relationship between the variables. The p values for model to predict the proposed process.
all variables in the model which are >0.05, reveal the non-
linear relationship between the interest variables and for all p
values >0.05, the related variables would be removed from 3.3 Effect of Variables on Dye Removal
the model equation for the process. In each run, the pre-
dicted results were obtained by substituting the coefficients The effect of five parameters of pH, temperature, time, initial
and numerical values of the variables in Eq. 6. The linear dye concentration, and amount of magnetic activated carbon
regression coefficients (R 2 ) of 0.9953 for the predicted data on dye removal and their interactions were investigated. The
versus experimental data proved good agreements between obtained results from CCD modeling are shown in Fig. 4.
123
Arab J Sci Eng
Fig. 4 The surface plots: showing the effect of five interest parameters on the percent of crystal violet adsorption at optimum conditions of 9.5,
25 ◦ C, 36 min, 19.8 mg L−1 , and 52.8 mg, for pH, temperature, time, initial dyes concentration, and adsorbent mass
3.3.1 The Effect of pH tic interaction and attraction of cationic dye (crystal violet)
on the negative surface of the adsorbent. At low pH, there
pH is a representative of acidity or basicity of the solution will be a tough competition between proton and the cationic
and is a significant factor in the dye adsorption process due dye for adsorbing on the magnetic activated carbon.
to the nature of cationic or anionic base of dyes. The effect of
pH was studied in the range of 4.5–10.5, using hydrochloride
3.3.2 The Effect of Temperature
acid and sodium hydroxide. The results in Fig. 4e–h shows
that as the pH solution increased to 9.5, the adsorption (%)
Temperature is one of the most indicative factors in adsorp-
increased and remained constant up to 9.5. Therefore, the
tion. Generally, considering the type of dye (cationic or
optimum pH of 9.5 was chosen for the process. It seems
anionic dyes), adsorption can represent exothermic or
that at basic conditions (high pH), the surface of adsorbent
endothermic [24]. While the increase in temperature leads
becomes negatively charged which intensifies the electrosta-
to adsorption percent enhancement, the adsorption process is
123
Arab J Sci Eng
3.4 ANN Modeling CCD and ANN are as two powerful prediction models
which widely applied in many chemical processes with linear
The input variables of trained ANN are pH, temperature and nonlinear multivariate regression problems [17]. ANN
(Temp.), time, initial dye concentration (Conc.), and amount and CCD were successfully utilized for adsorption of crys-
of magnetic activated carbon (MAC) and the output is tal violet by magnetic activated carbon. In order to assess
adsorption efficiency of crystal violet. The number of lay- the suitability of ANN and CCD in the proposed process,
ers, neurons of each layer, and their interconnects are shown the MAPE (%) values and the determination coefficients
123
Arab J Sci Eng
Table 2 Comparative results of the experiment and predicted values by ANN and CCD for test data set
No. Factors Adsorption (%) APE (%)
pH Time (min) Temp. (◦ C) Initial dye con. (mg L−1 ) MAC amount (mg) Exp. ANN CCD ANN CCD
123
Arab J Sci Eng
Fig. 7 The adsorption isotherms; Langmuir isotherm plots (a, b) and Freundlich isotherm plot (c) obtained at optimum conditions
Kf 1/n N R2 Sm Sb
concentrations in the range of 5–500 mg L−1 at optimum Freundlich isotherm was also applied for this study consid-
conditions. Langmuir isotherm consists of the monolayer ering Eq. 10 as below:
adsorption on a homogenous adsorbent and has a rational
basis. Each vacant site can only be occupied by one ion [27].
(1/n)
The Langmuir equation is as follows [28]: Q e = K f Ce (12)
Ce /Q e = 1/ (Q m b) + Ce /Q m (10)
This can be linearized as following:
where Q m is the saturation adsorption capacity (mg g−1 ) and
b is the constant related to the free energy of adsorption were
obtained from the slope and the intercept of the equation. Log Q e = log K f + 1/n log Ce (13)
The plots of Q i versus Ce , and Ce /Q e versus Ce are shown
in Fig. 7. The favorability of crystal violet adsorption onto
the magnetic activated carbon can be calculated by dimen- where the K f is Freundlich constant related to the adsorbent
sionless constant called separation factor, R L (Eq. 9): capacity, and n is the constant indicative to the intensity of
the adsorption process. The values of the constants, n and
R L = 1/ (1 + bC0 ) (11) K f , were calculated from the slope and the intercepts of the
equation and shown in Table 3. The Langmuir and Freundlich
In this equation C0 (mg g−1 ) is the highest initial ions con- isotherm plots are shown in Fig. 7. The Freundlich inten-
centration in adsorption isotherm studies. The value of R L sity constant of crystal violet was as n > 1 which show the
implies the unfavorable (R L > 1), linear (R L = 1), favor- significant adsorption of crystal violet in the proposed proce-
able (0 < R L < 1), and irreversible (R L = 0) types of dure even at high ion concentrations. The diagrams showed
the adsorption isotherm [2,9,29]. As the results indicate in models were ideally fitted to the data of adsorption of crys-
Table 1, the values of R L for two ions are 0 < R L < 1 which tal violet. Mean correlation coefficient of 0.981 was achieved
are favorable for adsorption isotherm. for Langmuir isotherm plots. Moreover, Freundlich isotherm
The Freundlich isotherm is an empirical equation which could satisfactorily depict crystal violet adsorption onto mag-
is widely employed for numerous adsorption processes. The netic activated carbon.
123
Arab J Sci Eng
123
Arab J Sci Eng
uous reactor with subcritical acetone using Purolite PD206 as 26. Bingöl, D.; Hercan, M.; Elevli, S.; Kılıç, E.: Comparison of the
catalyst. Fuel Process. Technol. 124(0), 206–211 (2014). doi:10. results of response surface methodology and artificial neural net-
1016/j.fuproc.2014.03.007 work for the biosorption of lead using black cumin. Bioresour.
23. Shakeri, S.; Ghassemi, A.; Hassani, M.; Hajian, A.: Investigation Technol. 112(0), 111–115 (2012). doi:10.1016/j.biortech.2012.02.
of material removal rate and surface roughness in wire electrical 084
discharge machining process for cementation alloy steel using arti- 27. Sohrabnezhad, S.; Pourahmad, A.: Comparison absorption of new
ficial neural network. Int. J. Adv. Manuf. Technol. (2015). doi:10. methylene blue dye in zeolite and nanocrystal zeolite. Desalina-
1007/s00170-015-7349-y tion 256(1), 84–89 (2010)
24. Salleh, M.A.M.; Mahmoud, D.K.; Karim, W.A.W.A.; Idris, 28. Faghihian, H.; Moayed, M.; Firooz, A.; Iravani, M.: Synthesis of a
A.: Cationic and anionic dye adsorption by agricultural solid novel magnetic zeolite nanocomposite for removal of Cs+ and Sr2+
wastes: a comprehensive review. Desalination 280(1), 1–13 from aqueous solution: kinetic, equilibrium, and thermodynamic
(2011) studies. J. Colloid Interface Sci. 393(0), 445–451 (2013). doi:10.
25. Elemen, S.; Akçakoca Kumbasar, E.P.; Yapar, S.: Modeling the 1016/j.jcis.2012.11.010
adsorption of textile dye on organoclay using an artificial neural 29. Mohan, D.; Chander, S.: Single, binary, and multicomponent
network. Dyes Pigment. 95(1), 102–111 (2012). doi:10.1016/j. sorption of iron and manganese on lignite. J. Colloid Interface
dyepig.2012.03.001 Sci. 299(1), 76–87 (2006). doi:10.1016/j.jcis.2006.02.010
123