Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 131 (2014) 189–194

Contents lists available at ScienceDirect

Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy journal homepage: www.elsevier.com/locate/saa

Synthesis of zinc oxide nanoparticles–chitosan for extraction of methyl orange from water samples: Cuckoo optimization algorithm–artificial neural network Mostafa Khajeh ⇑, Ali Reza Golzary Department of Chemistry, University of Zabol, P.O. Box 98615-538, Zabol, Iran

h i g h l i g h t s

g r a p h i c a l a b s t r a c t

 Organic dye in industrial wastewater

Training, validation and testing mean squared errors for the LM algorithm.

may result in serious environmental problem.  The nanoparticles have large specific area and internal diffusion resistance is absence.  The nanoparticles have a higher efficiency for the removing of analyte.  The ANN–COA was used to optimize the extraction percent of analyte.

a r t i c l e

i n f o

Article history: Received 4 February 2014 Received in revised form 24 March 2014 Accepted 17 April 2014 Available online 26 April 2014 Keywords: Methyl orange Zinc oxide nanoparticles–chitosan Artificial neural network–cuckoo optimization algorithm Water samples

a b s t r a c t In this work, zinc nanoparticles–chitosan based solid phase extraction has been developed for separation and preconcentration of trace amount of methyl orange from water samples. Artificial neural network– cuckoo optimization algorithm has been employed to develop the model for simulation and optimization of this method. The pH, volume of elution solvent, mass of zinc oxide nanoparticles–chitosan, flow rate of sample and elution solvent were the input variables, while recovery of methyl orange was the output. The optimum conditions were obtained by cuckoo optimization algorithm. At the optimum conditions, the limit of detections of 0.7 lg L1was obtained for the methyl orange. The developed procedure was then applied to the separation and preconcentration of methyl orange from water samples. Ó 2014 Elsevier B.V. All rights reserved.

Introduction Organic dyes (ODs) are broadly used in the paper, textile, plastic and leather industry to color commercial products. However, the ODs in industrial wastewater may result in serious environmental ⇑ Corresponding author. Tel.: +98 542 2226562; fax: +98 542 2226765. E-mail address: [email protected] (M. Khajeh). http://dx.doi.org/10.1016/j.saa.2014.04.084 1386-1425/Ó 2014 Elsevier B.V. All rights reserved.

problems due to their poor biodegradability and toxicity that could risk humankind’s health [1–4]. Also, the ODs are very difficult to be removed from wastewater by chemical or biological degradation procedures because of their stability against light, heat and many chemical reagents [1,5–7]. So, it is need to find an efficient way to separate these organic contaminants. Chitosan, a high molecular weight cationic polysaccharide, has attracted considerable interest due to its non-toxicity, antibacterial

190

M. Khajeh, A.R. Golzary / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 131 (2014) 189–194

activity, good mechanical property and biocompatibility ability. So, it has been used in environmental, biomaterial, medicinal, and other industrial fields. In recent years, hybrid materials based on chitosan have been developed [8]. Also, nanoparticles are interesting materials as sorbents for extraction of analytes. Semiconductor nanoparticles have much attention because of their novel electrical, mechanical and optical properties. Among various semiconductor nanaoparticles, zinc oxide nanoparticles are the most frequently studied due to of their applied aspects. Due to the large specific area, nanoparticle sorbent has a higher efficiency for the extraction of target compounds [9–12]. In this study hybrid of chitosan and zinc oxide nanoparticles was used as sorbent for extraction of methyl orange from water samples. Recently an alternative modeling system, artificial neural networks (ANNs), has been applied for representing non-linear functional relationships between variables. The capability of an ANN to learn and generalize the behavior of any complex and non-linear process makes it a powerful modeling tool [13,14]. In this study, an evolutionary optimization algorithm was applied for the first time in chemical modeling, which was inspired by lifestyle of a bird family called cuckoo. Specific egg laying and breeding of cuckoos is the basis of this novel optimization algorithm. The aims of this work are: (i) to use of zinc oxide nanoparticles– chitosan as sorbent for separation and preconcentration of methyl orange from water samples, (ii) to obtain a predictive model based on ANN method for prediction of the recovery of methyl orange, (iii) to optimize the recovery of methyl orange using combination of ANN and cuckoo optimization algorithm (COA) methods, and (iv) to develop a simple and cheap method for extraction of analyte. Finally, UV–Vis spectrophotometry was employed for analysis of methyl orange. Materials and methods Reagents and samples The methyl orange and HPLC grade of solvent including acetonitrile, chloroform, methanol and ethanol were obtained from Merck (Darmstadt, Germany). Reagent grade zinc oxide was obtained from Merck. A stock solution of methyl orange was prepared by dissolving the proper amount of it in double distilled water. Dilute solutions were prepared by an appropriate dilution of the stock solution in double distilled water. Apparatus The measurements were performed using a UV–Vis (UV-2100 RAY Leigh, Beijing, China) by monitoring the absorbance changes at a wavelength of maximum absorbance (470 nm). All experiments were performed in triplicate and the means of values were used for optimization. The pH was determined with a model 630 Metrohm pH meter with combined glass–calomel electrode. Preparation of zinc oxide nanoparticles – chitosan One gram of zinc oxide was dissolved in 100 mL of 1% acetic acid where it changed to zinc cation. Also, 1 g of chitosan was added to this solution and then the mixture was sonicated for 30 min. After that, 1 mol L1 NaOH drop by drop was added until the solution attained pH 10. This mixture was heated in water bath at 60 °C for about 3 h. Finally, it was filtered and washed with deionized water several times and dried in an oven at 50 °C for 1 h [15]. The amount of zinc in the chitosan–zinc oxide nanoparticles composite was 0.1 g g1. The representative scanning electron

Fig. 1. SEM image of chitosan–zinc nanoparticle.

microscopy (SEM) image of chitosan–zinc oxide nanoparticle is shown in Fig. 1. Procedure A plastic syringe was used as a cartridge and filled with different amount of dried synthesized chitosan–zinc oxide nanoparticles (according to experimental design). The cartridge was treated with 5 mL nitric acid and methanol. Then, the cartridge was washed with deionized water. A portion of aqueous sample solution containing 1 mg L1 of methyl orange was prepared. After that, the pH were adjusted at 3–10 by the dropwise 0.1 mol L1 NaOH and HCl, and then, passed through the cartridge with flow rate in the range 2–8 mL min1. Subsequently, analyte retained on the sorbent, were eluted with 1–3 mL of eluent (acetonitrile). The eluent was analyzed for the determination of analyte concentration. The efficiency of chitosan for the extraction of methyl orange without modification was 25% and it is lower than chitosan–zinc oxide nanoparticles. Thereby, in this study, chitosan–zinc oxide nanoparticles were used as adsorbent. Definition of the ANN model In this research, Neural Network Toolbox V7.12 of MATLAB mathematical software was used to predict the recovery of methyl orange in water samples. The universal method theory suggested that a neural network (NN) with a single hidden layer with a suitably large number of neurons could map any input to any output to an arbitrary degree of accuracy [16,17]. Thereby, the model used in this work was equipped with a single hidden layer. The input–output relationship of back-propagation neural network (BPNN) can be described as a composite mapping [13] that is represented by follow equation:

O ¼ g ½LW  ðIW  P þ b1 Þ þ b2 

ð1Þ

where O is the output, P is the input vector composed of the normalized network inputs. IW and b1 are the connection weights and biases of hidden layer. LW and b2 are the connection weights and biases of output layer. Symbol and g denote transfer function of hidden and output layers. The most widely used transfer functions are the tan-sigmoid (tansig) and linear transfer function (purelin), which is presented in follows eq.

PurelinðsumÞ ¼ sum tansigðsumÞ ¼

1  expðsumÞ 1 þ expðsumÞ

ð2Þ ð3Þ

M. Khajeh, A.R. Golzary / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 131 (2014) 189–194

The tansig was used as transfer function between input and hidden layer, while purelin was used as transfer function between hidden and output layer. Training of ANN by means of BP algorithm is an iterative optimization process where the mean-squared-error (MSE, below equation), the error between the predicted data and experimental data, is minimized by adjusting the weights and biases appropriately.

PN MSE ¼

i¼1 ðY t

 Y N Þ2

N

ð4Þ

where Yt is the target output, YN is the predicted output and N is the number of points. The algorithm changes the weights in each layer to decrease the MSE. This process is repeated several times to the error between the predicted and the experimental data satisfies certain error criterion. There are many variations of BP algorithm for training NNs. During training step the weights and biases are iterative updated by Levenberg–Marquardt (LM) technique until the convergence to the certain value is achieved. To prevent numerical overflows and preserve the interpretation of the weights, the experimental data required to be normalized within a uniform range of [0–1] according to following relationship:

xnorm ¼

ðx  xmin Þ ðxmax  xmin Þ

ð5Þ

where x is variable, xmax is maximum value and xmin is minimum value. Cuckoo optimization algorithm (COA) Like other evolutionary algorithms, COA starts with an initial population of cuckoos. They have some eggs to lay in host birds’ nests. Some of the eggs, which are more similar to the host bird’s eggs have the chance to grow up and become a mature cuckoo. But, others are detected by host birds and will kill. The eggs that grow reveal the suitability of the nests in that area. The more eggs survive in an area, the more profit is gained in that area. So the position where eggs survive will be the term that COA is going to optimize. Cuckoo search for the most appropriate area to lay eggs for maximum survival rate of eggs. After eggs remained grow and turn into a cuckoo mature, they make some societies. Each society has its habit region to live in; the best habitat for all societies will be the destination of cuckoos in other societies. Then, they immigrate to the best habitat. They will inhabit everywhere near the best habitat. Considering the number of eggs each cuckoo has and also the cuckoos distance to the goal point (best habitat), some egg laying radii is dedicated to it. Therefore, cuckoo begins to lay eggs in some random nests inside her egg laying radius. This process continues until the best position with maximum profit value is obtained and most of the cuckoo population is gathered around the same position [18,19]. In Nvar dimensional optimization problem, a habitat is an array of 1  Nvar, representing current living position of cuckoo. This array is defined as follows:

Habitat ¼ ½X 1 ; X 2 ; . . . ; X Nvar 

ð6Þ

Each of the variable values (X1, X2,. . ., XNvar) is floating number. The profit of a habitat is obtained by evaluation of profit function fp at a habitat of (X1, X2,. . ., XNvar). So

Profit ¼ fp ðhabitatÞ ¼ fp ðX 1 ; X 2 ; . . . ; X Nvar Þ

ð7Þ

As it seen COA is algorithms that maximize a profit function. In order to use COA in cost minimization problems, easily maximize the following profit function:

Profit ¼ CostðhabitatÞ ¼ fc ðX 1 ; X 2 ; . . . ; X Nvar Þ

191

ð8Þ

To start optimization algorithm, a candidate habitat matrix of size Npop  Nvar is generated. After that, some randomly produced number of eggs is supposed for each of these initial cuckoo habitats. In nature, each cuckoo lays from 5 to 20 eggs, which these values are used as the lower and upper limits of egg dedication to each cuckoo at various iterations. Another habit of real cuckoos is that they lay eggs in a maximum distance from their habitat, so called ‘‘Egg Laying Radius (ELR)’’, which is proportional to the total number of egg, number of current cuckoo’s eggs and also variable limits of varhi and varlow. So ELR is defined as follow:

ELR ¼ a 

Number of cuckoos eggs  ðv ar hi  v ar low Þ Total number of eggs

ð9Þ

where a is an integer, supposed to handle the maximum value of ELR [18]. Each cuckoo’s begins laying eggs randomly in some other host birds’ nests in her ELR. Results and discussion Optimization of procedure Predictive modeling with ANN Based on preliminary experiments, five variables including pH, mass of adsorbent, volume of elution solvent (V), flow rate of sample and elution solvent were selected in this work. Table 1 is shown the main factors and recovery of methyl orange. The performance of an ANN model depended on the data set for its training. In this work 41 various experiments were carried out to develop the ANN model. 71% was used for the training set and 29% was used for the validation and test set. The split of data into three subsets including training, validation and test was performed to estimate the performance of the neural network for prediction of unseen data that were not used for training set. In this way the generalization ability of the ANN can be evaluated. The number of neurons in hidden layer was established by training various architectures and selecting the optimal one based on minimization of performance function (MSE, mean squared errors) and improved the generalization ability of the topology. The optimal topology of the ANN model developed in this work involves a feed forward neural network (FFNN) with 5 inputs, one hidden layer with 11 neurons and one output layer including single neuron (Fig. 2). This FFNN architecture can be noted as ANN (5:11:1), referring to the number of inputs and the number of neurons in the hidden layer and output layer, respectively. The most widely used criteria including MSE for training, validation; testing and all data sets are 0.0001, 0.0021, 0.0014 and 0.0006, respectively. Fig. 3 shows the assessment of the MSE during training process using Levenberg– Marquardt (LM) algorithm for methyl orange. The training phase has been considered satisfactory terminated. A regression analysis between ANN predicting data and the experimental data were carried out, as shown in Fig. 4. The goodness of fit carried out between the experimental data and ANN output. In addition, the analysis of variance (ANOVA) has been used to validate statistically the constructed ANN model. The mathematical relationships employed to calculate the statistical estimators for ANOVA test were reported previously [20,21]. The results of ANOVA are summarized in Table 2. According to this table, the Fvalue is quite high and the P-value is smaller than 0.0001. Thereby, the model is significant. COA The COA technique was applied to optimize the input space of ANN model with objective of maximization of recovery of methyl orange. The values of COA-specific parameters used in the

192

M. Khajeh, A.R. Golzary / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 131 (2014) 189–194

Table 1 Experimental data was used for ANN modeling. No. pH Mass of sorbent (g) Training 1 6.5 2 6.5 3 3 4 10 5 6.5 6 6.5 7 6.5 8 3 9 6.5 10 10 11 6.5 12 3 13 10 14 6.5 15 6.5 16 6.5 17 6.5 18 6.5 19 6.5 20 10 21 6.5 22 3 23 3 24 10 25 10 26 6.5 27 6.5 28 10 29 6.5

a

Hidden layer

Sample V Eluent flow R% R% flow rate (mL) rate (experimental) (predicted) (mL min1) (mL min1) a

0.3 0.2 0.3 0.3 0.3 0.3 0.3 0.4 0.3 0.3 0.3 0.3 0.4 0.2 0.3 0.2 0.4 0.3 0.3 0.2 0.2 0.2 0.3 0.3 0.3 0.2 0.4 0.3 0.3

5 5 5 5 8 8 8 5 2 2 2 5 5 5 5 8 5 2 8 5 5 5 8 5 5 5 5 8 5

3 1 3 2 3 2 1 2 2 2 1 1 2 2 3 2 1 2 2 2 3 2 2 3 1 2 2 2 2

1 2 2 1 2 1 2 2 3 2 2 2 2 1 3 2 2 1 3 2 2 2 2 2 2 3 1 2 2

63.8 23.8 47.6 83.5 74.3 66.7 57.1 92 38.1 75.7 66.7 71.4 65.4 23.8 63.8 14.3 62.9 57.1 47.6 47.6 66.6 76.2 93.8 41.3 52.4 33.3 63.8 23.8 55.6

Validation 1 6.5 0.3 2 3 0.3 3 6.5 0.4 4 3 0.3 5 10 0.3 6 6.5 0.4

5 5 5 2 5 5

1 2 2 2 2 3

1 1 3 2 3 2

57.1 71.4 72.5 71.4 40.4 75.1

61.7 77.6 72.6 75.7 40.4 75

Testing 1 6.5 2 3 3 6.5 4 6.5 5 6.5 6 6.5

2 5 8 2 2 5

2 2 2 2 3 1

2 3 2 2 2 3

14.3 67.7 47.2 54.4 52.6 59.8

21.6 67.7 47.2 54.4 52.6 59.8

0.2 0.3 0.4 0.4 0.3 0.3

1 2 Input layer

63.6 23.8 47.8 83 72.8 67.7 57.4 92.7 37.4 75.5 66.3 71.4 65.3 24.4 62.9 16.4 63 57.1 47.4 46.6 65.7 77.3 93.7 41 54.5 33.5 63.8 22.2 55.7

3 4

1

5

2

6

3

7

4

8

5

9

Output layer

1

10 11

bias bias Fig. 2. Optimal ANN structure.

Mean of three replicates.

optimization simulation were maximum number cuckoo = 20; minimum number cuckoo = 5; maximum number of eggs = 4; minimum number of eggs = 2 and ELR = 0.5. Optimum condition has been selected after the evaluation of COA for 5 iterations, to achieve a good recovery for target compound. The optimized process conditions are as follows: the pH of solution is 4.4, the amount of adsorbent is 0.39 g, sample flow rate is 2 mL min1, elution volume is 1.2 mL and the elution flow rate is 1.3 mL min1. The model prediction of the percent of extraction under the optimized conditions was 101.8 using COA. The experimental response was 98.9%. Relative importance of each input variable The relative importance of each input variable was calculated by the connection weight partitioning methodology [14], which is presented in follows equation:

   PmjIW ij j jLW j j jIW kj j  k¼1   V¼ Pm Ph jIW ij j P LW m j j¼1 j¼1 Ph

j¼1

k¼1

jIW kj j

ð10Þ

Fig. 3. Training, validation and testing mean squared errors for the LM algorithm.

where V is the relative effect of the input variable I, m and h were the neuron numbers in the input and hidden layer, respectively. This method of calculating the relative importance of each input variable on the output will be useful to identify key variables. According to the Eq. (10), the relative importance of each input variables shown in Fig. 5. It suggests that the pH had significant effect on the extraction recovery.

Evaluation of method performance In this study, under the optimum condition, the linear range was obtained between 10 and 1000 lg L1 with a correlation coefficient of 0.999. The limit of detection (LOD) for the determination of methyl orange was studied under the optimum condition. The LOD is evaluated using 3(Sd)blank/m was 0.7 lg L1 where Sd is the standard deviation of the blank signals and m is the slope of the calibration curve. The relative standard deviation (RSD%) of the

193

M. Khajeh, A.R. Golzary / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 131 (2014) 189–194

Validation

Training 100

R² = 0.962

R² = 0.9982 Predicted

Predicted

80 60 40 20 0 0

20

40

60

80

100

Actual

Actual Testing

All R² = 0.995

Predicted

Predicted

R² = 0.9907

Actual

Actual

Fig. 4. Predicted values by ANN versus experimental values for methyl orange (according to Table 1).

Table 2 Analysis of variance (ANOVA). Degree of freedom

Sum of squares

Mean of squares

Fvalue

P-value

Model Residual Total

5 35 40

14664.37 152.03 14816.4

2932.87 4.34

675.2

98.5%) in the sample volume range of 25–250 mL. After that, the percent of extraction of analyte was decreased. In this study, the preconcentration factor was 208 for 250 mL sample volume due to the elution volume of 1.2 mL. The maximum sorption capacity of the adsorbent particles towards methyl orange was determined after saturation of the adsorbent with methyl orange at pH 4.4. The maximum adsorption (corresponding to a 500 mg L1 methyl orange initial concentration), which represents saturation of active points (which are available for methyl orange) on the adsorbent, was 481 mg g1. This method was compared with other procedure (Table 3) and according to these results; the maximum sorption capacity of this method is better than those obtained from most other methods.

Real sample analysis To evaluate the accuracy and applicability of the proposed procedure for real water samples, the extraction and determination of the methyl orange in various water samples were carried out. To assess the effects of matrix, the water samples were spiked with methyl orange at a concentration of 100 and 200 lg L1. Table 4 pH

Mass of sorbent

Sample flow rate

Volume

Eluent flow rate

Fig. 5. The relative effects of input variables on output factor.

ten replicate determination was

Synthesis of zinc oxide nanoparticles-chitosan for extraction of methyl orange from water samples: cuckoo optimization algorithm-artificial neural network.

In this work, zinc nanoparticles-chitosan based solid phase extraction has been developed for separation and preconcentration of trace amount of methy...
864KB Sizes 0 Downloads 3 Views